may 26 , 2021

1

Response Letter

Manuscript ID: BMJ-2021-064733

Title: Use of Antipsychotic Medications and Cholinesterase Inhibitors and the Risk of Falls and Fractures: self-

controlled case series

Committee Members: Dr. John Fletcher (Chair), Dr. Gary Collins (Statistician), Dr. Tiago Villanueva, Dr. Di

Wang, Dr. Wim Weber, Dr. Nazrul Islam, Dr. Elizabeth Loder, Dr. Helen Macdonald, Dr. David Ludwig, Dr.

Joseph Ross, Dr. Jessica Kimpton, Dr. Timothy Feeney

May 26th, 2021

Dear Editors and Reviewers,

Thank you very much for your consideration of our paper. We appreciate your thorough and constructive

reviews. We have considered all comments carefully and incorporated the suggestions to revise our manuscript.

Our point-by-point responses to the reviewers’ comments are provided below. To facilitate your review, we have

provided a summary of tables for the additional sensitivity analyses at the beginning of this letter. Following the

reviewers’ suggestion, we have replaced the terms “dementia” and “behavioural and psychological symptoms of

dementia (BPSD)” with the more up-to-date terminology “major neurocognitive disorder (NCD)” and

“neuropsychiatric symptoms of NCD”, respectively.

We believe that we have improved our manuscript sufficiently to be accepted for publication in the British

Journal of Medicine.

Kind Regards,

Edward Chia-Cheng Lai, PhD

Associate Professor, School of Pharmacy, Institute of Clinical Pharmacy and Pharmaceutical Sciences,

Medical College, National Cheng Kung University, No. 1, University Road, East District, Tainan City, 701, Taiwan,

886-235-3535 ext 4554, [email protected]

2

Summary of Tables for Additional Sensitivity Analyses

Review

Committee Rev. 1 Rev. 2 Rev. 3 Rev. 4

Supp. Table

3-1

Stratified by sex Comment

#4 - - - #8

Supp. Table

3-2

Stratified by age groups (65-74 years, 75-

84 years and 85+ years) - - - - -

Supp. Table

3-3

Stratified by patients’ anticholinergic

burden (Anticholinergic Drug Scale 2+ or

0-1) - #6 - - -

Supp. Table

3-4

Stratified by cumulative dose of

antipsychotic medications - #5-2 - - -

Supp. Table

3-5

Restricted outcomes to outpatient settings

only - #3 - - -

Supp. Table

3-6

Re-selected study populations without

exclusion of schizophrenia, bipolar

disorder, and depression

Comment

#2 - #9 - #2

Supp. Table

3-7

Excluded patients who died during the

study period/within 3 months after the

event - - - - -

Supp. Table

3-8

Restricted to specific codes for the

outcome definition Comment

#3 - - #2 -

Supp. Table

3-9

Counted all episodes of falls and fractures

- - - - #6

Supp. Table

3-10

Redefined the length of the pre-exposure

period from 14 days to 7, 21 and 28 days Comment

#5 - - #1-2 -

Supp. Table

3-11

Focusing on individual antipsychotic

medications - #5-1 - - -

Supp. Table

3-12

Further defined the periods within 1-14

days after initiation of exposure and within

1-14 days after discontinuation of

treatment

- - - - #4

3

Response to the review committee

Comment 1:

The numbers in the flowchart does not add up. For example, 11915 were excluded in the final step, but reasons

were given for 2288 and 4613 patients. Please revise and correct the numbers. Please also add the reasons for

exclusion.

Thank you for pointing this out. We have revised the flowchart and added the reasons for each exclusion

criterion. Please refer to the revised Figure 1.

Revision:

(METHODS, Study Population)

We excluded patients with a record of antipsychotic medications or ChEIs, or an episode of fall or fracture from

2003 to 2005 (washout period) to ensure that the whole study population consisted of incident drug users without

a history of falls or fractures. We also excluded patients with underlying schizophrenia and bipolar disorder to

ensure that antipsychotic medications were used to treat the neuropsychiatric symptoms of major NCD.

(Figure 1: Selection flowchart of the study population)

Comment 2:

In the methods section, it says "We also excluded patients with a record of schizophrenia, bipolar disorder, and

depression because these mental disorders might have confounded the result estimation." However, confounding

may not be justified for exclusion because confounders could be adjusted for in the analysis. Please present the

analysis after adjusting for these, and/or present the results stratified by these conditions.

Thank you for the suggestion. We have added the suggested analyses and revised the manuscript accordingly.

Briefly, we found the results to be consistent with the main analysis, although some of the subgroups had

relatively small numbers and produced a wide range of confidence intervals. Please review the results below. The

full version of the sensitivity analyses is available in Supplementary Table 3-6.

Revision:

4


We also excluded patients with underlying schizophrenia and bipolar disorder to ensure that antipsychotic

medications were used to treat the neuropsychiatric symptoms of major NCD.

(METHODS, Sensitivity Analyses, 2nd paragraph)

To examine the effect of excluding schizophrenia and bipolar disorder, we conducted sensitivity analyses without

exclusion of these mental conditions. We then further stratified this population by schizophrenia, bipolar disorder,

and depression to examine the individual effects from these conditions (Supplementary Table 3-6).

(RESULTS, Sensitivity Analyses)

Only 513 patients were excluded from the main analysis due to schizophrenia or bipolar disorder, and when we

re-selected the study population without excluding schizophrenia and bipolar disorder, the adjusted IRR of the

pre-exposure period was 6.07 (5.60, 6.57). The adjusted IRRs of the pre-exposure period were 2.94 (0.70, 12.37)

for those with schizophrenia, 3.24 (1.76, 5.99) for those with bipolar disorder, and 4.13 (3.49, 4.90) for those

with depression. Of note is the large confidence interval of the schizophrenia group, which is due to its limited

sample size (n=98).

(Supplementary Table 3-6)

Events, n

Follow-up time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)

Re-selected study patients without

excluding schizophrenia and bipolar

disorder (n=15,791)

Non-exposure 10,541 126,870.74 8.31 (8.15, 8.47) Reference

Pre-exposure 670 1,299.57 51.56 (47.76, 55.57) 6.07 (5.60, 6.57)

Cholinesterase inhibitor alone 1,829 19,437.05 9.41 (8.99, 9.85) 1.17 (1.10, 1.24)

Antipsychotic medication alone 1,435 13,941.15 10.29 (9.77, 10.84) 1.32 (1.24, 1.42)

Combination 1,316 12,567.72 10.47 (9.92, 11.05) 1.33 (1.24, 1.43)

With schizophrenia (n=98)

Non-exposure 60 711.20 8.44 (6.50, 10.78) Reference

Pre-exposure 2 8.70 22.99 (3.85, 75.95) 2.94 (0.70,12.37)

Cholinesterase inhibitor alone 3 59.73 5.02 (1.28, 13.67) 0.66 (0.16, 2.72)

Antipsychotic medication alone 21 210.63 9.97 (6.34, 14.98) 1.55 (0.80, 2.99)

Combination 12 101.84 11.78 (6.38, 20.03) 1.79 (0.75, 4.28)

Without schizophrenia (n=15,693)


Pre-exposure 668 1,290.97 51.74 (47.93, 55.78) 6.08 (5.61, 6.59)



Combination 1,304 12,465.88 10.46 (9.90, 11.04) 1.33 (1.24, 1.43)

With bipolar disorder (n=426)

Non-exposure 277 3,257.41 8.50 (7.55, 9.55) Reference

Pre-exposure 11 37.04 29.70 (15.62, 51.62) 3.24 (1.76, 5.99)



Combination 34 435.18 7.81 (5.50, 10.79) 0.84 (0.55, 1.30)

Without bipolar disorder (n=15,365)


Pre-exposure 659 1,262.54 52.20 (48.32, 56.30) 6.15 (5.67, 6.67)



Combination 1,282 12,132.54 10.57 (10.00, 11.16) 1.35 (1.26, 1.45)

With depression (n=4,743)


Pre-exposure 144 399.33 36.06 (30.52, 42.32) 4.13 (3.49, 4.90)

Cholinesterase inhibitor alone 435 5,049.01 8.62 (7.83, 9.45) 1.03 (0.91, 1.16)

5

*Note: this is a truncated version for inclusion in the letter only. The manuscript displays a full table.

Comment 3:

Please add the elaboration of the ICD codes in the Supplementary Appendix. One Editor highlighted that some of

the ICD codes are for non-specific conditions. Please justify using all the codes. Also, add the results after

excluding the non-specific codes for the outcome.

Thank you for the recommendation. We have added the ICD codes in Supplementary Table 1. We selected all

possible diagnosis codes in order to ensure we captured all events and covariates from the database. We have

revised the manuscript and cited some references listed below to justify the codes we used in the study, and they

were based on previous studies.

In addition, to examine the robustness of our study results, we have conducted a sensitivity analysis focusing on

specific diagnosis codes for falls (i.e., falls from a different level and falls from the same level) and hip fractures

(given that hip fractures are mainly caused by falls). We found the results remained consistent with the main

analysis. Please review the results below and the full version in Supplementary Table 3-8.

Revision:

(METHODS, Data Sources)

Many major disease diagnoses in the NHID have been validated by previous studies, including ischemic stroke

[1], epilepsy [2], hypertension [3], diabetes [3], hyperlipidaemia [3], coronary artery disease [3], atrial fibrillation

[3], heart failure [4], Parkinson’s disease [5], major NCD [5], schizophrenia [6], bipolar disorder [6], and

depression [6]. The diagnosis codes for osteoarthritis [7], osteoporosis [7], cataract [7], falls [8], and fractures [7]

have not been validated, but they were selected based on previous studies and expert opinions from a psychiatrist

and a geriatrician.


Moreover, we carried out a sensitivity analysis and selected specific diagnostic codes for falls and fractures (i.e.,

falls from a different level [ICD-9: E880-E884; ICD-10: W00.1, W00.2, W05-W17], falls from the same level

[ICD-9: E885-E886; ICD-10: W00.0, W01-W04, W18], and hip fractures [ICD-9: 820; ICD-10: S72]) to

examine the validity of the outcomes. The reason for selecting hip fractures was that more than 95% of hip

fractures are related to falls [9] (Supplementary Table 3-8). .

(DISCUSSION, Strengths and Weaknesses)

Third, we selected all diagnosis codes related to falls and fractures to ensure we had captured all possible

outcome events from the database. However, some of the codes, such as E888 in ICD-9 and W19 in ICD-10 (i.e.,

unspecified falls), might not have been specific enough to reflect the relationship between exposures and

outcomes in this study. To evaluate the potential impact from these non-specific outcomes, we conducted a

Antipsychotic medication alone 465 5,029.55 9.25 (8.43, 10.11) 1.12 (1.00, 1.27)

Combination 378 4,287.04 8.82 (7.96, 9.74) 1.06 (0.93, 1.20)

Without depression (n=11,048)


Pre-exposure 526 900.25 58.43 (53.59, 63.58) 6.95 (6.34, 7.61)



Combination 938 8,280.68 11.33 (10.62, 12.07) 1.48 (1.37, 1.61)

6

sensitivity analysis selecting only specific codes for falls and fractures. The results were consistent with the main

analysis.

Newly cited references:

1. Hsieh C-Y, Chen C-H, Li C-Y, Lai M-L. Validating the diagnosis of acute ischemic stroke in a National Health Insurance

claims database. Journal of the Formosan Medical Association. 2015;114(3):254-9.

2. Chen CC, Chen LS, Yen MF, Chen HH, Liou HH. Geographic variation in the age‐and gender‐specific prevalence and

incidence of epilepsy: Analysis of Taiwanese National Health Insurance–based data. Epilepsia. 2012;53(2):283-90.

3. Sung S-F, Hsieh C-Y, Lin H-J, Chen Y-W, Yang Y-HK, Li C-Y. Validation of algorithms to identify stroke risk factors in

patients with acute ischemic stroke, transient ischemic attack, or intracerebral hemorrhage in an administrative claims

database. International journal of cardiology. 2016;215:277-82.

4. Lin YS, Chen TH, Chi CC, Lin MS, Tung TH, Liu CH, et al. Different implications of heart failure, ischemic stroke, and

mortality between nonvalvular atrial fibrillation and atrial flutter—a view from a national cohort study. Journal of the

American Heart Association. 2017;6(7):e006406.

5. Liu C-C, Sun Y, Lee P-C, Li C-Y, Hu SC. Risk of dementia after Parkinson’s disease in Taiwan: a population-based

retrospective cohort study using National Health Insurance claims. BMJ open. 2019;9(3):e025274.

6. Wu C-S, Kuo C-J, Su C-H, Wang SH, Dai H-J. Using text mining to extract depressive symptoms and to validate the

diagnosis of major depressive disorder from electronic health records. Journal of affective disorders. 2020;260:617-23.

7. Wei MY, Luster JE, Chan C-L, Min L. Comprehensive review of ICD-9 code accuracies to measure multimorbidity in

administrative data. BMC health services research. 2020;20:1-11.

8. Chien WC, Lai CH, Chung CH, Lin CH. A retrospective population-based data analyses of unintentional fall mortality and

hospitalisation in Taiwan during 2005-2007. Int J Inj Contr Saf Promot. 2013;20(1):50

9. Parkkari J, Kannus P, Palvanen M, Natri A, Vainio J, Aho H, et al. Majority of hip fractures occur as a result of a fall and

impact on the greater trochanter of the femur: a prospective controlled hip fracture study with 206 consecutive patients.

Calcified tissue international. 1999;65(3):183-7


Comment 4:

Please add the results stratified by sex as suggested by the reviewers.

Thank you for the suggestion. We have added the suggested analysis accordingly, and the results were consistent

with the main analysis. Please review the results below and the full version in Supplementary Table 3-1.

Events, n

Follow-up time,

person-years,

sum

Incidence rate (95%CI),

100 person-years

Adjusted IRR

(95%CI)

Redefine the outcome by specific

codes, all events (n=5,458)


Pre-exposure 379 453.89 83.50 (75.40, 92.23) 10.16 (9.10,11.35)



Combination 537 4,321.32 12.43 (11.41, 13.51) 1.67 (1.49, 1.87)

falls (n=621)


Pre-exposure 73 51.14 142.70 (112.70, 178.50) 10.77 (8.19,14.16)



Combination 92 404.43 22.75 (18.44, 27.77) 1.49 (1.09, 2.04)

fractures (n=5,149)


Pre-exposure 357 428.79 83.26 (74.95, 92.24) 10.41 (9.29,11.66)



Combination 496 4,115.68 12.05 (11.03, 13.15) 1.70 (1.51, 1.91)


**Patients could have both a fall and a fracture.

7

Revision:

(METHODS, Sensitivity Analyses, 1st paragraph)

We stratified the patients by sex (male or female) and age group (65 to 74 years, 75 to 84 years, and 85 years or

above) to minimize the effects of sex and age (Supplementary Table 3-1 and 3-2).

(DISCUSSION, Meaning of the study, 2nd paragraph)

From our sensitivity analyses, we identified some groups who might have a higher risk of outcomes at the

baseline. For example, patients with advanced age or male sex had higher risks of falls and fractures.



Comment 5:

"We defined a 14-day pre-exposure period before study drug initiation over concerns about protopathic bias."

This 14-day time period was not justified in the manuscript. Please add the results using at least two other time-

points (21- and 28-day) to check the robustness of the findings.

Thank you for the suggestion. There is no clear evidence from the literature on the exact duration that is adequate

for assessing baseline risk in the pre-exposure period. We assumed 14 days because many elderly patients with

chronic conditions were followed up every two weeks, based on our experience.

To test the assumption, we have added the suggested sensitivity analyses with different lengths of pre-exposure

period (7, 21, and 28 days). Please review the results below and the full version in Supplementary Table 3-10.

Revision:


We further redefined the pre-exposure period to include various lengths of 7, 21 and 28 days to test the adequacy

of a 14-day pre-exposure period in the main analysis (Supplementary Table 3-10).


Events, n

Follow-up

time,

person-years,

sum

IR (95%CI),

/100 person-years Adjusted IRR (95%CI)

Stratification by sex

Male (n=5,088)


Pre-exposure 285 421.32 67.64 (60.13, 75.85) 7.63 (6.73, 8.65)



Combination 443 3,591.78 12.33 (11.22, 13.52) 1.48 (1.31, 1.67)

Female (n=10,190)


Pre-exposure 372 833.58 44.63 (40.26, 49.34) 5.35 (4.80, 5.95)



Combination 827 8,445.56 9.79 (9.14, 10.48) 1.29 (1.18, 1.41)

8

When we redefined multiple lengths of pre-exposure periods as 7, 21 and 28 days, the adjusted IRRs of these pre-

exposure periods were 9.49 (8.64, 10.43), 4.91 (4.56, 5.30) and 4.43 (4.14, 4.75), respectively.

(DISCUSSION, Meaning of the Study, 2nd paragraph)

Furthermore, we have tested various lengths of pre-exposure period from 7 days to 28 days. The IRR was the

highest when the length of the pre-exposure period was defined as 7 days (IRR: 9.49), and it decreased as the

duration of the pre-exposure period lengthened - 14 days (IRR: 6.17), 21 days (IRR: 4.91), and 28 days (IRR: 4.43).

Based on these results, it can be concluded that a 7-day pre-exposure duration probably represents a period of rapid

deterioration. On the other hand, a duration of more than 21 days possibly captures a relatively stable status.

Therefore, our decision to use 14 days appears to be adequate.



Comment 6:

Could you also please add the following information in a separate table (Table 2) as a complementary to Figure 3

(for each row, i.e., for each adjusted IRR estimates in Figure 3a, 3b, and 3c, PLUS the estimates for the non-

exposure period):

Number of patients followed up;

Total, median, and IQR of the follow-up time (with the unite of measurement);

Crude incidence rate and 95% CI (with unit);

Crude incidence rate ratio and 95% CI.

Events, n

Follow-up time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)

Redefine the pre-exposure period to

7 days prior to exposure (n=15,278)


Pre-exposure 474 580.21 81.69 (74.58, 89.30) 9.49 (8.64,10.43)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.32 (1.23, 1.42)

**Main analysis



Pre-exposure 657 1,254.90 52.35 (48.46, 56.47) 6.17 (5.69, 6.69)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.35 (1.26, 1.45)




Pre-exposure 792 1,924.81 41.15 (38.35, 44.09) 4.91 (4.56, 5.30)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.37 (1.27, 1.47)



Pre-exposure 949 2,591.75 36.62 (34.34, 39.00) 4.43 (4.14, 4.75)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.39 (1.29, 1.49)

9

Thank you for the recommendation. We have revised the manuscript accordingly. We have replaced Figure 3 with

a table (Table 2) so as to provide more detailed information. Please review the revision. Thank you very much.

10

Response to reviewers

Reviewer 1: Professor Thomas R. E. Barnes

Overall comment:

This is a succinctly written report of a large case series that has investigated a clinically relevant prescribing issue

in the elderly. The finding of an increased risk of falls/fractures that warranted hospitalisation during periods of

exposure to antipsychotic medication (increase of approximately 40%), cholinesterase inhibitors (approximately

20%), and co-prescription of both (approximately 50%), compared with periods of no such exposure, is

plausible. Less so is the 5-fold increase in risk in the ‘pre-exposure’ period, although it leads the authors to

conclude that the increased risk of falls and fractures is largely the result of assumed ‘underlying’ BPSD, rather

than treatment with these medications and, indeed, they surmise that medications may actually lower the risk. It

would be helpful if the authors could provide reassurance that the analysis and interpretation of the data

adequately addresses putative sources of bias, as follows:

Thank you very much for the supportive comment. We have revised the manuscript accordingly on the basis of

your comments and suggestions. Please review our point-by-point responses below. In addition, we have added

analyses comparing the risk of outcome between the exposed period and the pre-exposure period (Table 2),

where the results support our hypothesis that the targeted medications were associated with a lower risk during

the exposure period. To facilitate your review, we have also summarized the additional sensitivity analyses at the

beginning of this letter.

Comment 1:

Could the high risk of severe falls/fractures in the pre-exposure period be partly a consequence of the eligibility

criteria tending to identify people who had come to the attention of medical services and received a prescription

for antipsychotic medication and/or a cholinesterase inhibitor because of serious falls/fractures?

We appreciate the important question. Yes, it is possible that due to the eligibility criteria (i.e., patients with

neuropsychiatric symptoms of dementia who are exposed to cholinesterase inhibitors and antipsychotic

medications), a high risk of severe falls/fractures was observed in the pre-exposure period.

We consider the risk of falls and fractures may be more attributable to patients’ underlying disease, rather than

caused by medications, because the pre-exposure periods showed much higher risks than the exposure periods.

Nonetheless, we have highlighted the importance of this potential confounding by indication in the revised

manuscript.

Revision:

(DISCUSSION, Principal Findings, 1st paragraph)

From this population-based SCCS, we found that exposures to ChEIs and antipsychotic medications were both

associated with higher risk of falls and fractures compared with the non-exposure period. However, the results

should be interpreted carefully. The 14-day pre-exposure period revealed exceptionally high incidence rates of

falls and fractures, indicating that patients may have already been at a high risk of outcome events before

receiving the medications. The observed higher risks during the exposure periods, as compared with the non-

11

exposure period, may be due to the neuropsychiatric symptoms in addition to the medication use. This implied

that the patients’ condition might not have been fully restored to a steady status, despite receiving treatment.

(DISCUSSION, Principal Findings, 2nd paragraph)

However, the result that the exposure periods were associated with elevated risks of falls and fractures should be

interpreted carefully. Our results showed that patients may have already been at high risk before receiving the

treatment, implying that the interconnection among patients’ underlying conditions, drug effects on relieving

neuropsychiatric symptoms, and side effects of ChEIs and antipsychotic medications has increased the difficulty

of delineating the exact contributing factors and quantifying the magnitude of the risk of falls and fractures that

each factor poses. Similar findings have been reported in previous studies.…. These findings suggested that the

observed higher risks of outcomes during the exposure periods might not be attributable to the medications alone.

Comment 2:

The exposure, non-exposure and pre-exposure periods are of different durations. If this issue is adequately

addressed by the adjustment of the IRRs by one-year bands using regression models, this should be explicit.

Although different lengths of time were used to define the three periods, they remain comparable because the

SCCS was based on the comparison of incidence “rates”; that is, the numbers of cases divided by the observation

duration.

Considering that there might be age-related factors in SCCS, we also broke down the study periods not only by

exposure status but also by one-year intervals, as presented below (for example, the time periods were classified

as 12 different periods). The concept is to compare exposure and non-exposure within the same year, and an

overall estimate incorporating the estimates from each yearly band is generated with Poisson regression model.

This approach helps to control age-related variables in the SCCS, following a suggestion from a previous study

[10]. We have revised the manuscript and made this explicit.

Revision:

(METHODS, Statistical Analysis and Covariates, 2nd paragraph)

Therefore, we adopted a one-year boundary to split the study duration in order to take into account possible age-

related effects when estimating the effects. That is, we partitioned the observation period not only by the

exposure status, but also by one-year intervals (e.g., the first, second and third years etc.) This time-varying

method to adjust for age effects has been described in detail elsewhere [10].

Newly cited reference:

10. Petersen I, Douglas I, Whitaker H. Self-controlled case series methods: an alternative to standard epidemiological study

designs. BMJ. 2016;354:i4515.

Comment 3:

12

The difference in the risk of severe falls/fractures between the pre-exposure period, non-exposure period and

periods on various drug regimens seems to be largely attributed to an interaction between medication effects and

assumed BPSD. But were these periods in different settings for some patients and, if so, could the change in

setting be relevant? For example, could the relatively high risk of falls found during the pre-exposure period be

because patients were more likely to be at home at that time, with trip hazards such as rugs, stairs, walking with

bare feet/in socks, etc., and the risk lower during periods in hospital or other healthcare settings which might

have support rails, walking frames, staff support and other interventions designed to minimise the risk of falls?

Thank you for the insightful comment. Yes, it is possible that different care settings might have different baseline

risks that could influence our analysis. We have incorporated this valuable insight into the discussion section of

the revised manuscript.

In addition, we have also added an analysis by restricting outcomes to only those recorded at outpatient settings

with an effort to better reflect the risks occurred in the community where interventions to prevent falls and

fractures might be insufficient. Please review the results below and the full version in Supplementary Table 3-5.

Revision:


Furthermore, patients in different care settings may have different baseline risks of falls and fractures. To

understand the impact of care settings, we conducted an additional analysis by restricting outcomes to only those

recorded at outpatient settings with an effort to better reflect the risks experienced in the community, where

interventions to prevent falls and fractures might be insufficient (Supplementary Table 3-5).


Fifth, patients living in different care settings might have different baseline risks to begin with, and this should be

considered in the SCCS. For example, the higher risk of outcomes during the pre-exposure period might be

because the patients were living at home, with more trip hazards from rugs, stairs, and walking. Similarly, the

lower risks during the medication treatment periods could partially be due to the support from healthcare

facilities. To assess the possibility of the care setting effect, we conducted an additional analysis by including

only outcome events recorded at outpatient settings. The results showed that IRRs were smaller after limiting

outcomes to only those occurred in the community, but the risk for the pre-exposure period remained higher than

the non-exposure period.



Events, n

Follow-up time,

person-years,

sum

IR (95%CI),


Restriction to only outcomes at

outpatient settings (n=7,914)


Pre-exposure 141 646.13 21.82 (18.44, 25.66) 2.54 (2.14, 3.01)



Combination 495 6,131.50 8.07 (7.39, 8.81) 1.02 (0.91, 1.13)

13

Comment 4:

Patients experiencing minor falls or related symptoms when starting medication may have tended to discontinue

medication (and it seems that if medication was stopped within 14 days of starting, this does not qualify as an

exposure period) and so the data on the risk of severe falls in the ‘continuous’ exposure periods may tend to refer

to those patients who tolerated the medication well initially.

Thank you for the question. We wish to clarify that once a medication was initiated, the duration up to the day of

discontinuation was included in the computation of the exposure period. That means that even if the medication

was stopped within 14 days of starting, it was still considered as an exposure period.

However, as the comment points out, some patients might discontinue the medications due to minor falls or

related symptoms. As such, those who continue treatment/exposure might represent a group of patients who

tolerate the medication well, and this might affect the evaluation of outcomes. We have highlighted this in the

limitation section of the revised manuscript. In addition, we have also conducted a post-hoc analysis to assess the

discontinuation rate of medications after the event occurrence. Please see below for the results.

Revision:


Fourth, some patients might have discontinued the medications due to minor falls or related symptoms. This

meant that those who continued the treatment might represented a group of patients who tolerated the

medications well, and thus could have affected the evaluation of the outcomes. Therefore, we performed a post-

hoc analysis to understand the extent of medication discontinuation after a fall or a fracture. We found that only

7.9% of patients discontinued their medications after events; thus, medication discontinuation and its subsequent

impact on outcome evaluation might be limited.

Comment 5-1:

The comparison of risk of falls between FGAs and SGAs refers to a redundant distinction (FGAs and SGAs are

heterogeneous and overlapping in terms of pharmacology and side effect profiles), particularly as only one FGA,

haloperidol, was prescribed. Rather, the paper should refer to the individual antipsychotic medications

prescribed.

Thank you for this comment. We have removed the subgroup analyses for FGAs and SGAs. Alternatively, we

have added the subgroup analyses for individual antipsychotic medications. We selected haloperidol, risperidone,

olanzapine, and quetiapine for the subgroup analyses because they were most commonly used for elderly patients

in Taiwan. Please review the results below and the full version in Supplementary Table 3-11.

(METHODS, Sensitivity Analyses, 3rd paragraph)

Each antipsychotic medication has different affinities to alpha-adrenergic receptors, histamine receptors, and

dopamine receptors, which can lead to varying degrees of effects that possibly provoke falls and fractures, such

as heart rate reduction, vasodilation, orthostatic hypotension, blurred vision, sedation, and EPS [11]. Therefore,

we conducted subgroup analyses to evaluate individual antipsychotic medications separately. We selected

haloperidol, risperidone, olanzapine, and quetiapine for the subgroup analyses because they are the most

commonly used antipsychotic medications for elderly patients in Taiwan (Supplementary Table 3-11).

14


In the analyses for individual antipsychotic medications, we found that the adjusted IRRs were 3.75 (3.15, 4.47),

1.15 (0.99, 1.35), 1.35 (1.00, 1.82), and 1.16 (1.07, 1.26) under exposure to haloperidol, risperidone, olanzapine,

and quetiapine, respectively.


One of the likely explanations for the higher risk of haloperidol could be due to its greater extrapyramidal side

effects compared with other antipsychotic medications. Confounding by indication could be another explanation

because patients with positive symptoms (e.g., agitation) may be more likely to receive haloperidol.


11. UpToDate [Internet]. Second-generation antipsychotic medications: Pharmacology, administration, and side effects; c2020.

Available from: https://www.uptodate.com/contents/second-generation-antipsychotic-medications-pharmacology-

administration-and-side-effects [cited 2020 Jul 27].



Comment 5-2:

It is a pity that, in this elderly group, the investigators did not look at any dosage effect, even if the dosages were

generally within the recommended range.

Thank you very much for the comments. We have added analyses by stratifying cumulative dose of antipsychotic

medications of study patients to evaluate dosage effects. Please review the results below and the full version in

Supplementary Table 3-4.

Events, n

Follow-up

time,

person-years,

sum

IR (95%CI),


Analyses for individual antipsychotic

medication

Haloperidol (n=4,745)


Pre-exposure 367 368.10 99.70 (89.89, 110.30) 12.11 (10.84,13.54)



Combination 93 340.37 27.32 (22.18, 33.32) 4.26 (3.37, 5.37)

Risperidone (n=5,119)


Pre-exposure 98 393.72 24.89 (20.32, 30.20) 2.74 (2.23, 3.35)



Combination 210 1,907.57 11.01 (9.59, 12.58) 1.32 (1.13, 1.54)

Olanzapine (n=1,314)


Pre-exposure 30 97.09 30.90 (21.23, 43.55) 3.72 (2.57, 5.37)



Combination 62 469.79 13.20 (10.21, 16.81) 1.77 (1.31, 2.39)

Quetiapine (n=12,851)


Pre-exposure 481 1,013.41 47.46 (43.36, 51.85) 5.40 (4.91, 5.93)



Combination 964 9,884.87 9.75 (9.15, 10.38) 1.16 (1.07, 1.26)

15

Revision:


To examine the dosage effect, we stratified the study population based on their cumulative doses of antipsychotic

medications (above or below the median) using the ratio of prescribed daily dose (PDD) to defined daily dose

(DDD) (Supplementary Table 3-4).

(METHODS, Statistical Analysis and Covariates, 1st paragraph)

PDD is calculated as the sum of actual doses of each medication, while DDD is a standard unit widely applied

for the crosswalk of drugs with different strengths, developed by the World Health Organization [12]. It is an

assumed average maintenance dose per day for each medication.

(RESULTS Sensitivity Analysis)

The adjusted IRRs of the pre-exposure period were 5.73 (4.54, 7.24) and 5.45 (4.54, 6.53) among patients who

had higher and lower cumulative doses of antipsychotic medications, respectively.


Lastly, our study did not evaluate the dose-response relationship between the drugs and the risk of falls and

fractures. However, we did compare the dosages of antipsychotic medications from our study population with

those reported in guidelines and previous studies, and the dosages were within the suggested ranges [13-15].

Moreover, the stratification analysis by PDD/DDD ratio indicated the dosages of antipsychotic medications did

not have a differential impact on the risks of falls and fractures.


12. Leucht S, Samara M, Heres S, Davis JM. Dose equivalents for antipsychotic drugs: the DDD method. Schizophrenia bulletin.

2016;42(suppl_1):S90-S4.

13. Masopust J, Protopopová D, Vališ M, Pavelek Z, Klímová B. Treatment of behavioral and psychological symptoms of

dementias with psychopharmaceuticals: a review. Neuropsychiatric disease and treatment. 2018;14:1211.

14. Haw C, Yorston G, Stubbs J. Guidelines on antipsychotics for dementia: are we losing our minds? Psychiatric Bulletin.

2009;33(2):57-60.

15. Lin Y-T, Hwang T-J, Shan J-C, Chiang H-L, Sheu Y-H, Hwu H-G. Dosage and duration of antipsychotic treatment in

demented outpatients with agitation or psychosis. Journal of the Formosan Medical Association. 2015;114(2):147-53.


Events, n

Follow-up

time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)

Stratification by cumulative dose of

antipsychotic medications by PDD/DDD ratio

Higher than the median value (n=3,232)


Pre-exposure 78 279.17 27.94 (22.24, 34.68) 5.73 (4.54, 7.24)



Combination 975 4,295.59 22.70 (21.31, 24.16) 5.38 (4.79, 6.05)

Equal to or lower than the median value

(n=3,247)


Pre-exposure 129 285.56 45.17 (37.87, 53.49) 5.45 (4.54, 6.53)



Combination 295 1,797.32 16.41 (14.62, 18.37) 2.12 (1.82, 2.47)

16


Comment 6:

It seems that the investigators had information on all the medication prescribed for the people in the study, so the

cumulative anticholinergic burden of the full medication regimen for each of the individual patients would be

calculable from the data collected. Given that anticholinergic effects in the elderly include an increased

likelihood of falls as well as an adverse effect on cognitive function, an analysis of whether there was an

association between anticholinergic burden and falls/fractures would be worthwhile.

Thank you very much for the suggestion. To address the issue, we have added sensitivity analyses stratifying the

study population by patients’ anticholinergic drug scale (ADS), which is a widely used measure for

anticholinergic burden. The results remained consistent to the main analysis. Please review the results below and

the full version in Supplementary Table 3-3.

Revision:


We also evaluated patients’ characteristics when the outcome event occurred, including their age on the event

date, settings of the event being diagnosed, mortality after the event, type of exposure, anticholinergic burden

measured by the anticholinergic drug scale (ADS), and the ratio of prescribed daily dose to defined daily dose

(PDD/DDD ratio) for ChEIs and antipsychotic medications (Table 1). ADS is a widely used measure developed

through expert opinions, and it is a four-point scale ranging from 0 to 3 for no-known activity to high

anticholinergic activity [16].


Anticholinergic burden (i.e., the cumulative effect of taking multiple medications with anticholinergic activities)

is a critical issue that may also be associated with falls and fractures in the elderly [17]. High anticholinergic

burden has been reported to adversely affect cognitive and physical functions, and may also increase the risk of

falls, hospitalization, and death [18-21]. To account for the potential impact, we included the anticholinergic drug

scale (ADS) to classify patients by their anticholinergic burden for the sensitivity analyses (Supplementary Table

3-3).


The adjusted IRRs of the pre-exposure period were 5.30 (4.74, 5.94) and 7.41 (6.60, 8.33) among patients with

an ADS of 0 to 1 and 2 or above points, respectively.


Consistent with previous studies that anticholinergic effect in the elderly increases the likelihood of falls and

cognitive deterioration [22-23], we observed much higher risks during the pre-exposure period in patients with

higher anticholinergic burden.


16. Carnahan RM, Lund BC, Perry PJ, Pollock BG, Culp KR. The anticholinergic drug scale as a measure of drug‐related

anticholinergic burden: associations with serum anticholinergic activity. The Journal of Clinical Pharmacology.

2006;46(12):1481-6.

17. Salahudeen MS, Duffull SB, Nishtala PS. Anticholinergic burden quantified by anticholinergic risk scales and adverse

outcomes in older people: a systematic review. BMC geriatrics. 2015;15(1):1-14.

17

18. Tune LE. Anticholinergic effects of medication in elderly patients. The Journal of clinical psychiatry. 2001;62(suppl 21):11-4.

19. Lowry E, Woodman RJ, Soiza RL, Mangoni AA. Associations between the anticholinergic risk scale score and physical

function: potential implications for adverse outcomes in older hospitalized patients. Journal of the American Medical

Directors Association. 2011;12(8):565-72.

20. Uusvaara J, Pitkala KH, Kautiainen H, Tilvis RS, Strandberg TE. Association of anticholinergic drugs with hospitalization

and mortality among older cardiovascular patients. Drugs & aging. 2011;28(2):131-8.

21. Cao YJ, Mager D, Simonsick E, Hilmer S, Ling S, Windham B, et al. Physical and cognitive performance and burden of

anticholinergics, sedatives, and ACE inhibitors in older women. Clinical Pharmacology & Therapeutics. 2008;83(3):422-9.

22. Green AR, Reifler LM, Bayliss EA, Weffald LA, Boyd CM. Drugs Contributing to Anticholinergic Burden and Risk of Fall or

Fall-Related Injury among Older Adults with Mild Cognitive Impairment, Dementia and Multiple Chronic Conditions: A

Retrospective Cohort Study. Drugs Aging. 2019;36(3):289-97.

23. Wilson NM, Hilmer SN, March LM, Cameron ID, Lord SR, Seibel MJ, et al. Associations between drug burden index and

falls in older people in residential aged care. J Am Geriatr Soc. 2011;59(5):875-80.



Comment 7:

Lastly, for several reasons, perhaps ‘antipsychotic medication’ could be substituted for ‘antipsychotics’

throughout the paper. Further, the subheading ‘Principle findings’ at the beginning of the Discussion section,

should be ‘Principal findings’.

Thank you very much, we have revised these terms accordingly. We appreciate your constructive suggestions,

which have helped substantially to improve our manuscript.

Events, n

Follow-up

time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)

Stratification by anticholinergic burden indicator

Anticholinergic drug scale 2+ (n= 6,579)


Pre-exposure 326 546.27 59.68 (53.46, 66.43) 7.41 (6.60, 8.33)



Combination 672 5,664.66 11.86 (10.99, 12.79) 1.67 (1.51, 1.85)

Anticholinergic drug scale 0-1 (n=8,699)


Pre-exposure 331 708.62 46.71 (41.88, 51.95) 5.30 (4.74, 5.94)



Combination 598 6,372.69 9.38 (8.65, 10.16) 1.11 (1.01, 1.23)

18

Reviewer 2: Dr. Rónán O'Caoimh

Overall comment:

This study, utilising a self-controlled case series design, examines whether confounding by indication may

explain the apparent association between psychotropic medications (anti-psychotics and Acetylcholinesterase

inhibitors - ChEIs) and falls/fractures in PwD. The authors suggest that BSPD and other ‘unstable mental and

cognitive conditions’ rather than these medications could account for increased risk of falls/fractures in people

with dementia. This is based on a higher adjusted IRR for these outcomes before compared with after exposure to

these medications. In addition, the results suggest that falls overall (injurious and non-injurious) were not

significantly associated with the prescription of ChEIs or antipsychotics (all types) but that falls resulting in

fractures and hospitalisations were associated these medications, albeit not as strongly as the risk during the pre-

exposure period.

The paper is generally well-written and a self-controlled case series is an appropriate design to investigate

this question in a clearly defined and apparently robust dataset. I suspect the findings are indeed as a result of

reverse causality and also an element of confounding by indication (more severe dementia and/or the effects of

higher doses of antipsychotics – neither of which were available in the study). The likelihood that this is the case

is a more plausible explanation than the theory that these medications somehow mitigate falls in PwD and should

be more explicitly stated in the discussion section. A few comments and some major and minor points to note and

address:

Thank you for the critical comment. We have revised the manuscript based on the comments; specifically, we

have explicitly expressed the possible explanation our findings in the discussion section (please refer to the

comments #19 to #27). Our point-by-point responses to further comments are provided below. We have also

summarized all additional sensitivity analyses at the beginning of this letter.

Comment 1: Abstract

As with the paper, it is generally well-written and clear. I suggest providing more information on the patients

who were included in this study within the abstract. Apart from being aged 65 and older, readers would be

unclear whether they have mild cognitive impairment (MCI) or dementia, and if they have dementia, how this

was diagnosed/confirmed – given that these are registry data, where diagnosis can be variable, more clarity

would provide readers with context.

Thank you for the suggestion. All users of cholinesterase inhibitors were reviewed by neurologists or

psychiatrists based on Mini-Mental State Examination (MMSE) score and the reimbursement guidelines of

Taiwan’s National Health Insurance program. We have confirmed the diagnosis of dementia by using the records

of drug prescriptions. We have added more information about this in the revised abstract to avoid confusion.

Revision:

(ABSTRACT, Participants)

We included older adults (aged 65 or older) who were newly prescribed with antipsychotic medications and

ChEIs and suffered an incident fall or fracture from outpatient, inpatient, emergency room, and contracted

pharmacy records. The diagnosis of major NCD was determined from the prescriptions of ChEIs and was

19

reviewed by experts based on the Mini-Mental State Examination (MMSE) score and the reimbursement

guidelines of Taiwan’s National Health Insurance program.

Comment 2: Abstract

In the conclusion, I suggest avoiding statements such as “The use of ChEIs and antipsychotics may in fact

mitigate the risk”. This is not clearly the case and while future studies could examine such a hypothesis, this

study only found a relatively lower risk after exposure and not for all falls but for more serious events i.e.

fractures and injurious falls leading to hospitalisation, which could be due to other causes (see my comment

below in the discussion). I agree that this should be mentioned and debated in the discussion section but it is not

the main conclusion to be drawn from the study.

Thank you for the suggestion. We have revised the conclusion of the abstract and have removed “The use of

ChEIs and antipsychotics may in fact mitigate the risk” from the abstract as you suggested.

Revision:

(ABSTRACT, Conclusion)

The incidence of falls and fractures was especially high in the pre-exposure period, suggesting that some factors

other than the medications, such as underlying diseases, should be taken into consideration when evaluating the

association between the risk of falls and fractures, and the use of ChEIs and antipsychotic medications. The

exposure periods were also associated with higher risk of falls and fractures compared with the non-exposure

period, although the magnitude was much lower than during the pre-exposure period. Prevention strategies and

close monitoring of risk of falls are still necessary until there is evidence that patients have restored to a steady

status.

Comment 3: Introduction

I would avoid the term ‘anti-dementia’ medications and describe ChEIs as cognitive enhancers or use a similar

term. They treat symptoms but are not as per se ‘anti-dementia’ drugs which implies a preventative or curative

function.

Thank you very much for the suggestion. We have replaced “anti-dementia medications” with “cognitive

enhancers” throughout our paper accordingly.


In the second paragraph (and in the abstract), it is a little unclear what the authors mean by concerns over

‘protopathic bias’. This should define bias resulting when treatment itself (usually initiated in response to the

very first symptoms/signs of the disease) subsequently causes the outcome of interest i.e. reverse causation. Here,

there is concern that the development of BPSD led to the prescription of medications that increased the risk of

subsequent falls, rather than concern that the first signs of falling led to the prescription of medication. The text

should be more explicit.

Thank you very much for the important comment. We have revised the manuscript accordingly.

Revision:

20

(ABSTRACT, Main outcome measures)

Moreover, we defined a 14-day pre-exposure period before study drug initiation over concerns about

confounding by indication.

(INTRODUCTION, 2nd Paragraph)

Although several studies and guidelines have suggested that the use of ChEIs and antipsychotic medications may

be associated with the risk of falls and fractures, some previous studies have drawn divergent

conclusions….Confounding by indication may partly explain the conflicting results from these studies. For

example, patients with neuropsychiatric symptoms of major NCD may manifest with depression, irritability,

agitation, and hallucinations that could lead to prescription of antipsychotic medications, and both the symptoms

and the treatments could increase the risk of subsequent falls and fractures. This confounding effect is especially

likely when the events are observed within a short period of time right before patients initiate a treatment in

response to the neuropsychiatric symptoms of major NCD. However, only few studies have addressed this issue

while evaluating the association between ChEIs, antipsychotic medications, and the risk of falls and fractures

[24-25].


24. Pratt N, Roughead EE, Ramsay E, Salter A, Ryan P. Risk of hospitalization for hip fracture and pneumonia associated with

antipsychotic prescribing in the elderly. Drug safety. 2011;34(7):567-75.

25. Brännström J, Lövheim H, Gustafson Y, Nordström P. Antipsychotic Drugs and Hip Fracture: Associations Before and After

the Initiation of Treatment. Journal of the American Medical Directors Association. 2020;21(11):1636-42. e6.


There are a few minor grammatical and syntax errors throughout the introduction and the rest of the paper. I have

just highlighted a few here but the paper should be reviewed for these e.g. P4, line 40 “With the advent of aging

society,”, line 43-44 “family caregivers may request antipsychotics to control the BPSD in patients.” and on P5

line 27 “The study period for this study was from” and line 51”no matter measured or unmeasured”.

Thank you for the correction. We have enlisted professional English editing services to ensure there is no typo or

grammatical and syntax error.


In the 3rd paragraph, specify why antipsychotics are not recommended by the FDA and hence what is meant by

“unintended outcomes” in the same paragraph, Line 46-47.

Thank you for the question. We have revised the manuscript to make it more explicit.

Revision:

(INTRODUCTION, 1st paragraph)

Due to potential adverse effects of antipsychotic medications, including orthostatic hypotension, sedation, blurred

vision and extrapyramidal symptoms (EPS), the U.S. Food and Drug Administration (FDA) has suggested that a

complete assessment of the risk of falls be undertaken before initiating treatment [26].

(INTRODUCTION, 3rd paragraph)

Even though the use of antipsychotic medications may be associated with falls and fractures, as reported by the

FDA and the Beers Criteria, clinicians and family caregivers may still prescribe antipsychotic medications to

21

control neuropsychiatric symptoms in patients with major NCD. Therefore, understanding the risk profiles of

patients receiving ChEIs and antipsychotic medications has become increasingly important to prevent the

occurrence of falls and fractures.


26. UpToDate [Internet]. Second-generation antipsychotic medications: Pharmacology, administration, and side effects; c2020.

Available from: https://www.uptodate.com/contents/second-generation-antipsychotic-medications-pharmacology-

administration-and-side-effects [cited 2020 Jul 27].

Comment 7: Methods

While it is implied that it is reliable/accurate, how precise is the National Health Insurance Database in providing

exact dates of outcomes (falls/fractures) and prescriptions?

Thank you for the critical questions. We have the exact dates of the prescriptions for both inpatient and outpatient

claims. However, for those who had falls or fractures during hospitalization, the outcome date may be a few days

or weeks later than the actual date of event occurrence because the diagnoses may only have been registered at

release from inpatient care. We may thus have underestimated the risk during exposure period for those who had

falls or fractures during hospitalization and then discontinued the medication (the event was misclassified to non-

exposure period). We have addressed this issue in the revised manuscript.

Revision:


Seventh, we did not have the exact outcome dates for those who had falls or fractures during hospitalization

because the diagnoses were registered at the discharge date, and thus the outcome dates we analysed may have

been later than the dates of actual event occurrences. Therefore, we may have underestimated the risk during the

exposure period for those who had falls or fractures during hospitalization and then discontinued the medications.

Comment 8: Methods

In Fig 1 or in the text, I cannot find details on the proportions excluded because of psychotic disorders or

depression. Please included.

Thank you for the comment. We have revised the flowchart and added the reasons for each exclusion. Please

review the revised Figure 1.

Revision:

(METHODS, Study population)






22

Comment 9: Methods

While I appreciate the reasons to exclude those with pre-existing indications for an antipsychotic, I am unclear as

to why the authors excluded patients with depression. How was depression defined in this case – as a previous

episode of psychotic depression requiring an antipsychotic or any diagnosis of depression, even mild? There is a

marked difference between these. I am unclear how a diagnosis of mild depression in a patient with dementia

could confound the results in this study design? Depression is a very common symptom associated with dementia

and the relationship is complex. Falls and depression also have a complex relationship. However, a self-

controlled case series because such time invariant factors are cancelled out should be a suitable approach to

analyse such data. This is a limitation and it would be worthwhile examining if the inclusion of those with

depression (disregarding those with previous psychotic depression for example), would influence the results of

the study. It is an important sub-analysis.

We appreciate the important question. We defined depression by the records of diagnosis (ICD-9: 296.2, 296.3,

300.4, 311; ICD-10: F32, F33, F341) during the study periods. Yes, we do agree with you about the complex

relationship between depression, dementia, and falls. Therefore, we have re-run the analysis including individuals

with depression. Moreover, we have conducted sensitivity analyses after stratifying patients by depression to

examine the impact of depression on our results. These results were consistent with the main analysis. Please

review the results below and the full version in Supplementary Table 3-6.

Revision:



medications were used to treat the neuropsychiatric symptoms of major NCD. The study selection flowchart is

presented in Figure 1.


23



Comment 10: Methods

Table 1 provides a great deal of information on the participants, which is welcome. However, much more detail is

required about the diagnosis of dementia itself and how that is made in the National Health Insurance Database.

As said above, how it is diagnosed/what classification and the proportion subtypes of dementia etc. should be

provided. I see in the limitations section that stage/grade of dementia was not available but surely there is some

data on the diagnosis and how it is defined in the dataset.

Thank you for the question. As mentioned in Comment #1, we identified the diagnosis of dementia based on

prescriptions of cholinesterase inhibitors that could only be prescribed after review by licensed neurologists or

psychiatrists. We have also provided information regarding the subtypes of dementia using diagnosis codes that

have been adopted by previous studies in Taiwan [26-27]. Please review Supplementary Table 1 for the ICD

codes.

Revision:

Events, n

Follow-up

time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)

Patients comorbid with depression

(n=4,743)

Non-exposure 3,321 38,500.25 8.63 (8.34, 8.92) Reference Pre-exposure 144 399.33 36.06 (30.52, 42.32) 4.13 (3.49, 4.90) Cholinesterase inhibitor alone 435 5,049.01 8.62 (7.83, 9.45) 1.03 (0.91, 1.16) Antipsychotic medication alone 465 5,029.55 9.25 (8.43, 10.11) 1.12 (1.00, 1.27) Combination 378 4,287.04 8.82 (7.96, 9.74) 1.06 (0.93, 1.20)

Patients without depression (n=11,048) Non-exposure 7,220 88,370.49 8.17 (7.98, 8.36) Reference Pre-exposure 526 900.25 58.43 (53.59, 63.58) 6.95 (6.34, 7.61) Cholinesterase inhibitor alone 1,394 14,388.04 9.69 (9.19, 10.21) 1.22 (1.14, 1.31) Antipsychotic medication alone 970 8,911.60 10.88 (10.22, 11.59) 1.43 (1.32, 1.56) Combination 938 8,280.68 11.33 (10.62, 12.07) 1.48 (1.37, 1.61)

24


Patient baseline characteristics were assessed based on the covariates extracted from within one year before 1st

Jan 2006, including age, sex, subtype of major NCD, comorbidities, and co-medications (Table 1). Subtypes of

major NCD included Alzheimer's disease, Parkinson’s dementia, vascular dementia, mixed type (i.e., those who

had been diagnosed with more than one subtype of major NCD), others or unspecified [26-27]. The subtypes of

major NCD and comorbidities were defined using ICD codes listed in Supplementary Table 1.

(RESULTS, Patient Characteristics at Baseline)

We identified 15,278 patients eligible for the SCCS (Figure 1), with a mean age at baseline of 74.5 (SD: 5.5)

years, of whom 66.7% were female. Although all diagnoses of major NCD had been reviewed by licensed

neurologists or psychiatrists, most of the study populations were recorded as the unspecified major NCD

(69.2%).


27. Lin, Pei-Chao, et al. "Prevalence of pain-related diagnoses in patients with dementia: a nationwide study." Journal of pain

research 11 (2018): 1589.

28. Wang, Hao-Kuang, et al. "Increased risk of hip fractures in patients with dementia: a nationwide population-based study."

BMC neurology 14.1 (2014): 1-8.

Comment 11: Methods

P5 line 46 typo “We applied theself-controlled ”.

Thank you. We have corrected the typo.

Comment 12: Methods

The authors should provide the ICD codes used in an appendix/supplemental file.

Thank you for the recommendation. We have provided the ICD codes in Supplementary Table 1.

Comment 13: Methods

Define FGA or SGA where they are first used – P6 line 20.

Thank you for the recommendation. Following a reviewer’s suggestion to remove the classifications of “FGA”

and “SGA” (Prof Thomas R. E. Barnes, Reviewer 1, Comment 5-1) due to their heterogeneous and overlapping

pharmacologic features, we have revised the analyses accordingly and no longer include groupings by “FGA”

and “SGA” in the analysis. Instead, we have conducted the analyses by the individual medications.

Comment 14: Methods

The authors mention a 14-day period in the abstract – the “pre- exposure period before study drug initiation over

concerns about protopathic bias.”. I presume this relates to the pre-exposure category in the ‘Outcome Events and

Exposure periods’ subsection – this should be specified in the text as it is not clearly mentioned again after

appearing in the abstract and should be described in the methods.

Thank you for the recommendation. We have added the definition and the reason for including the pre-exposure

period in our manuscript. And as you suggested in Comment 4, we have replaced “protopathic bias” with

“confounding by indication” (The neuropsychiatric symptoms could lead to prescription of medications and also

increased risk of subsequent falls and fractures.) Please review the revised statement:

25

Revision:

(METHODS, Outcome Events and Exposure periods)

We classified the study time into five discrete periods: 1) 14-day pre-exposure period prior to the exposure to

drugs, 2) exposure to ChEIs alone, 3) exposure to antipsychotic medications alone, 4) exposure to a combination

of ChEIs and antipsychotic medications, and 5) non-exposure when neither ChEIs nor antipsychotic medications

were given. The 14-day pre-exposure period was designed to evaluate the increased incidence of falls and

fractures due to neuropsychiatric symptoms of major NCD before the initiation of medications, and to benchmark

the risk magnitude during the exposure periods.

Comment 15: Methods

Describe why the PDD/DDD ratio was considered important to evaluate. Define DDD and the ratio. This will not

be familiar to all readers of the BMJ.

Thank you for the recommendation. The PDD/DDD was used for the crosswalk of dosages of drugs with

different strengths, using a standardized unit developed by the World Health Organization. We have added the

definition of the PDD/DDD ratio and also have specified the reason in the revised manuscript.

Revision:


We also evaluated patients’ characteristics when the outcome event occurred, including their age on the event

date, settings of the event being diagnosed, mortality after the event, type of exposure, anticholinergic burden

measured by the anticholinergic drug scale (ADS), and the ratio of prescribed daily dose to defined daily dose

(PDD/DDD ratio) for ChEIs and antipsychotic medications (Table 1).…. PDD is calculated as the sum of actual

doses of each medication, while DDD is a standard unit widely applied for the crosswalk of drugs with different

strengths, developed by the World Health Organization [29]. It is an assumed average maintenance dose per day

for each medication.


29. Leucht S, Samara M, Heres S, Davis JM. Dose equivalents for antipsychotic drugs: the DDD method. Schizophrenia bulletin.

2016;42(suppl_1):S90-S4.

Comment 16-1: Results

Avoid the term ‘slightly’ – I presume from the CIs that these were statistically significant differences – is that

correct?

Thank you for the recommendation. We have removed ‘slightly’ from the revised manuscript.

Comment 16-2: Results

From Fig. 3, “all” falls (injurious and non-injurious) were not significantly associated with all antipsychotics but

falls that resulted in fractures and hospitalisations were (i.e. injurious). If so, this should be made clear and then

discussed.

26

Thank you very much for the recommendation. In the previous version, we found the IRR of falls for

antipsychotics was 1.33 (95% CIs: 0.84, 2.10). Unfortunately, we believe the likely reason for the non-

significance is the limited case number (n=302).

We have revised the selection criteria for study patients (i.e., without excluding those with depression) as

suggested in Comment 9. The case numbers have thus increased (n=766) and the updated IRR of falls under

antipsychotics is 1.36 (95% CIs: 1.02, 1.82), which reaches statistical significance.

Comment 17: Results

Clarify what is meant by ‘No interaction with ChEIs was found for antipsychotics, FGAs, or SGAs’. The method

and rationale for analysing this is poorly described in the statistics section in the methods.

Thank you for the clarification question. At first we hypothesized that the combined use of ChEIs and

antipsychotic medications may have synergistic effects that greatly increase the risk of falls and fractures. We

therefore examined the statistical interaction in the regression model, and the results showed that there was no

such synergistic effect. We agree that interaction is a further topic requiring deeper rationale before analysing. To

avoid confusion, we have removed the interaction part from the revised manuscript.

Comment 18: Results

There are some important findings in the tables and figures, Fig 3 in particular, that should be mentioned in more

detail in the results sections, which is overly concise/brief.

Thank you for the suggestion. We have added more detailed description specifically for Figure 3 in the results

section. Moreover, we have replaced Figure 3 with a table (Table 2) so as to provide more detailed information,

as shown below.

Revision:

(RESULTS, Evaluation of the Risk of Falls and Fractures)

Compared with the non-exposure period, the risk of falls and fractures was higher under exposure to ChEIs alone

(adjusted IRR; 95% CI: 1.17; 1.10 to 1.24; E-value 1.62), antipsychotic medications alone (1.33; 1.24 to 1.43; E-

value 1.99), and combinations (1.35; 1.26 to 1.45; E-value 2.04). The risk was even higher during the pre-

exposure period (6.17; 5.69 to 6.69; E-value 11.82) compared with the non-exposure period. In the analysis of

falls, the adjusted IRRs were 0.91 (0.71, 1.18), 1.36 (1.02, 1.93), 1.36 (1.02, 1.82) and 10.39 (8.08, 13.37) for the

exposure to ChEIs alone, antipsychotic medications alone, combination, and the pre-exposure period,

respectively. In the analysis of fractures, the adjusted IRRs were 1.18 (1.11, 1.26), 1.34 (1.24, 1.43), 1.35 (1.25,

1.45) and 6.11 (5.62, 6.63) for the exposure to ChEIs alone, antipsychotic medications alone, combinations, and

pre-exposure period, respectively. We included 7,364 cases with hospitalized falls or fractures, and the results

were also consistent with the main analysis (Table 2).

(Table 2)

Events, n

Follow-up time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI),

All events (n=15,278)


Pre-exposure 657 1,254.90 52.35 (48.46, 56.47) 6.17 (5.69, 6.69)

27



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.35 (1.26, 1.45)

Falls (n=766)


Pre-exposure 84 62.84 134.70 (107.30, 164.70) 10.39 (8.08,13.37)



Combination 114 518.55 21.98 (18.22, 26.31) 1.55 (1.17, 2.05)

Fractures (n=14,874)


Pre-exposure 627 1,222.01 51.31 (47.41, 55.45) 6.11 (5.62, 6.63)



Combination 1,215 11,759.94 10.33 (9.76, 10.93) 1.35 (1.25, 1.45)

Hospitalized events (n=7,364)


Pre-exposure 516 608.77 84.76 (77.68, 92.32) 10.20 (9.28,11.21)



Combination 775 5,905.85 13.12 (12.22, 14.07) 1.73 (1.57, 1.90)


Comment 19: Discussion

In the discussion, the authors suggest again that BPSD may have led to the falls rather than the medications to

treat them (or cognitive enhancers). I would avoid the term ‘very likely’ as causation has not been proven i.e. it is

not clear that the BPSD caused the falls (nor cognition in the case of ChEIs), only that these medications have

less impact/play a smaller role than has been suggested in other studies. Comparison with these studies here

would provide context and contrast.

Thank you very much for the comment. We have revised the manuscript and removed “very likely” from the

sentence. We agree that the observed risks could possibly result from BPSD, medications, or both. We have

revised the manuscript and highlighted these points. Moreover, we have also compared our study with previous

studies as suggested, and added a new paragraph with more details in the discussion section. Please review our

revisions.

Revision:

(DISCUSSION, Principal findings, 1st paragraph)






exposure period, may be due to the neuropsychiatric symptoms in addition to the medication use. This implied

that the patients’ condition might not have been fully restored to a steady status, despite receiving treatment.

(DISCUSSION, Principal findings, 2nd paragraph)

Previous studies have reported that both ChEIs and antipsychotic medications were associated with falls and

fractures, with the magnitude of risk increasing up to 18% [29] to 63% [30] among patients receiving ChEIs, and

up to 21% [31] to 54% [32] among patients receiving antipsychotic medications. Consistent with these previous

28

studies, our findings suggest that compared with the non-exposure period, exposures to ChEIs and antipsychotic

medications led to a 17% and 33% increase, respectively, in the risk of falls and fractures. However, the result

that the exposure periods were associated with elevated risks of falls and fractures should be interpreted carefully.

Our results showed that patients may have already been at high risk before receiving the treatment, implying that

the interconnection among patients’ underlying conditions, drug effects on relieving neuropsychiatric symptoms,

and side effects of ChEIs and antipsychotic medications has increased the difficulty of delineating the exact

contributing factors and quantifying the magnitude of the risk of falls and fractures that each factor poses. Similar

findings have been reported in previous studies. For example, Brännström et al. [34] reported that the highest risk

of hip fractures occurred before initiation of antipsychotic medications (OR: 9.09; 95% CI: 7.00 to 11.81 within -

16 to -30 days and OR: 5.84; 95% CI: 4.42 to 7.71 within -1 to -15 days), rather than after receiving treatment

(OR: 4.31; 95% CI: 3.05 to 6.10 within 1 to 15 days). Pratt et al. [35] also found that the risk of hospitalization

for hip fractures was the highest one week before antipsychotic medications initiation (IRR 10.99; 95% CI: 7.94

to 15.21), and the risks reduced within one week after antipsychotic medications initiation (IRR 1.04; 95% CI:

0.40 to 2.70). These findings suggested that the observed higher risks of outcomes during the exposure periods

might not be attributable to the medications alone.


30. Gill SS, Anderson GM, Fischer HD, Bell CM, Li P, Normand S-LT, et al. Syncope and Its Consequences in Patients With

Dementia Receiving Cholinesterase Inhibitors: A Population-Based Cohort Study. Archives of Internal Medicine.

2009;169(9):867-73.

31. Epstein NU, Guo R, Farlow MR, Singh JP, Fisher M. Medication for Alzheimer's disease and associated fall hazard: a

retrospective cohort study from the Alzheimer's Disease Neuroimaging Initiative. Drugs & aging. 2014;31(2):125-9.

32. Johnell K, Jonasdottir Bergman G, Fastbom J, Danielsson B, Borg N, Salmi P. Psychotropic drugs and the risk of fall injuries,

hospitalisations and mortality among older adults. International journal of geriatric psychiatry. 2017;32(4):414-20.

33. Koponen M, Taipale H, Lavikainen P, Tanskanen A, Tiihonen J, Tolppanen AM, et al. Antipsychotic Use and the Risk of Hip

Fracture Among Community-Dwelling Persons With Alzheimer's Disease. J Clin Psychiatry. 2017;78(3):e257-e63.


the Initiation of Treatment. Journal of the American Medical Directors Association. 2020;21(11):1636-42.e6.

35. Pratt N, Roughead EE, Ramsay E, Salter A, Ryan P. Risk of hospitalization for hip fracture and pneumonia associated with



Similarly, the authors should avoid unfounded inferences that these medications ‘may in fact’ reduce risk. This is

a hypothesis that could be derived for future testing but it is clearly not a fact. If the side effects of antipsychotic

medications resulted in immobility and drowsiness, as well as cerebrovascular events or progression in dementia,

this could explain a reduction in falls – this would not be a desirable approach.

Thank you for the comment. We have removed “in fact” from the sentence.

Yes, we agree that some side effects of antipsychotic medications may also reduce the risk of falls and fractures,

such as immobility and drowsiness, or bedridden state, due to cerebrovascular events, etc. We have highlighted

these possibilities in the discussion section.

Revision:

(DISCUSSION, Unanswered Questions and Future Research)

While the reason for the elevated risk during the pre-exposure period may lie in the relatively unstable disease

status of patients, the actual reasons will need to be elucidated by further studies. The risks during the exposure to

29

the treatment could reflect a composite consequence of patients’ unstable disease status, effects of medications on

relieving neuropsychiatric symptoms, and side effects of medications. For example, some side effects of

antipsychotic medications such as orthostatic hypotension, sedation, and extrapyramidal symptoms could

increase the risks of falls and fractures, while others such as immobility, drowsiness or being bedridden could

reduce these risks. However, these explanations are based mainly on clinical observations, and could not be

exhaustively tested in the current study. Future studies considering the severity of major NCD (e.g., Mini Mental

State Examination scale or Clinical Dementia Rating scale) and patient reported information might provide a

better platform to address these issues.


I’d avoid the term ‘show interactions with’ to describe the co-exposure of anti-psychotics and ChEIs in

preference of the term ‘show additive effects in combination with’. It is not clear if there was an effect that it

would be classed a drug-drug interaction.

Thank you for the suggestion. In response to your Comment 17, we have removed the analysis for synergistic

effects and the related descriptions from the revised manuscript. We agree with you that this is another topic that

requires deeper rationale and more detailed definitions.


The E value should not appear for the first time in the discussion and should be explained for general readers in

the methods section and why considerable unmeasured confounders would be required to explain away the

estimate.

Thank you for the suggestion. We have added the explanation for the use of the E-value in the method section.

Revision:

(METHODS, Statistical Analysis and Covariates, 2nd paragraph)

Because some potentially important confounders such as patients’ body weight, daily life activities, and the use

of walking aids were not available in the database, we adopted a quantitative bias assessment tool, the E-value

approach, to evaluate the minimum effect from an unmeasured confounder that would suffice to obviate the

association found between the exposure and the outcome [36]. For example, an E-value of 5 indicates that the

unmeasured confounder would need to be associated with both exposure and outcome by a factor of more than

five times in order to render the observed association irrelevant. Computation of the E-value was based on the

work of Mathur et al [37].

(DISCUSSION, Strengths and weaknesses)

Sixth, our study may be subject to unmeasured confounders such as patients’ daily living activities. The

quantitative bias assessment (i.e., E-value) showed that the potentially unmeasured confounders need to have

very large effect size so as to refute the observed high risks of falls and fractures during the pre-exposure period.

According to the literature [38], potential unmeasured confounders have not been shown to have such a large

effect size, and we thus concluded that the results were not affected significantly by these confounders.


36. VanderWeele, T.J. and P. Ding, Sensitivity analysis in observational research: introducing the E-value. Annals of internal

medicine, 2017. 167(4): p. 268-274.

30

37. Mathur, M.B., et al., Web Site and R Package for Computing E-values. Epidemiology (Cambridge, Mass.), 2018. 29(5): p.

e45-e47.

38. Fernando E, Fraser M, Hendriksen J, Kim CH, Muir-Hunter SW. Risk factors associated with falls in older adults with

dementia: a systematic review. Physiotherapy Canada. 2017;69(2):161-70.


The authors mention that they compared their findings to data from other countries – this needs to be expanded

upon. It is too brief to draw out any comparisons or conclusions.

Thank you for the comment. The reason we selected these guidelines/studies was to confirm that the “dosage” of

medication used in our study population was similar to that in the other studies. The results showed our dosages

were within the ranges reported by previous studies. We have revised the manuscript to make this clearer.

Daily doses (mg) Masopust, et al [39] Haw, et al [40] Lin, Yi-Ting, et al [41] Our study

Haloperidol 0.5-5 0.5-2 - Median: 0.8

Risperidone 0.25-1 0.5-2 Mean: 1 Median: 1.0

Olanzapine 2.5-7.5 2.5-10 - Median: 2.0

Quetiapine 12.5-150 25-150 Mean: 43.8 Median: 40.0

Revision:


Lastly, our study did not evaluate the dose-response relationship between the drugs and the risk of falls and

fractures. However, we did compare the dosages of antipsychotic medications from our study population with

those reported in guidelines and previous studies, and the dosages were within the suggested ranges [39-41].

Moreover, the stratification analysis by PDD/DDD ratio indicated the dosages of antipsychotic medications did

not have a differential impact on the risks of falls and fractures.


39. Masopust, J., et al., Treatment of behavioral and psychological symptoms of dementias with psychopharmaceuticals: a

review. Neuropsychiatric disease and treatment, 2018. 14: p. 1211.

40. Haw, C., G. Yorston, and J. Stubbs, Guidelines on antipsychotics for dementia: are we losing our minds? Psychiatric Bulletin,

2009. 33(2): p. 57-60.

41. Lin, Y.-T., et al., Dosage and duration of antipsychotic treatment in demented outpatients with agitation or psychosis. Journal

of the Formosan Medical Association, 2015. 114(2): p. 147-153.


P9 Line 12 – typo: Cognitive impairment should not be pleural ‘cognitive impairments’

We have corrected it. Thank you.


P9 Line 13 - ChEIs are not disease modifying medications that ‘slow’ progression – please rephrase and please

put ‘respectively’ after ‘control BSPD’ to make clear to readers you are not suggesting ‘antipsychotics can slow

the progress of cognitive impairment’.

We have revised it accordingly. Thank you.

Revision:

(DISCUSSION, Meaning of the Study, 1st paragraph)

31

ChEIs can improve cognitive function, and antipsychotic medications can control neuropsychiatric symptoms.

Therefore, they are commonly prescribed for patients with major NCD.


The ‘Unanswered questions and future research’ section is ineloquent – please rephrase this – I get the idea but it

is very poorly phrased.

Thank you for the comment. We have rephrased the ‘Unanswered questions and future research’ section.

Revision:

(DISCUSSION, Unanswered questions and future research)


status of patients, the actual reasons will need to be elucidated by further studies. The risks during the exposure to

the treatment could reflect a composite consequence of patients’ unstable disease status, effects of medications on

relieving neuropsychiatric symptoms, and side effects of medications. For example, some side effects of







Comment 27: Conclusion

In the conclusion, I would again avoid the implication that somehow anti-psychotic medications represent some

sort of potential treatment for falls associated with dementia – it is as the authors hint at, much more likely to be

due to reverse causation and as I have indicated an element of indication bias that could not be identified in the

study (related to dementia severity and other confounders related to the diagnosis and dementia subtypes, as well

as the doses of anti-psychotics, which were not available/provided).

Thank you for the suggestion. We have revised the conclusion section accordingly. We greatly appreciate all of

your comments, which have substantially improved our manuscript.

Revision:

(Conclusion)





period, although the magnitude was much lower than during the pre-exposure periods. Prevention strategies and


status.

32

Reviewer 3: Professor Jeremy M Jacobs

Overall Comment

This is a very interesting study, which addresses an increasingly common clinical dilemma- whether or not to

treat the behavioral and psychological symptoms of dementia (BPSD) with antipsychotics, despite the increased

risk of falls associated with treatment. Indeed it is widely accepted that the use of neuroleptics among people

with dementia is associated with increased falls and fractures, morbidity (particularly cardiovascular and

cerebrovascular), and is considered as off label "black label" in the USA. Nonetheless, among clinicians this

consensus is often somewhat counterintuitive, particularly when faced with extreme psychomotor agitation,

which often seems to place the patient at far greater immediate risk than the increased risks associated with

neuroleptics which have been repeatedly described in epidemiological data.

This novel study examines the risk of falls and fractures associated with antipsychotics (first and second

generation ) as well as Cholinesterase inhibitors among people aged over 65, using data from the National Health

Insurance database in Taiwan, which includes medical data covering 23 million people (99% of population). The

study includes 10, 750 subjects, average age at baseline 75, 65% females, free from psychiatric illness/previous

use of antipsychotics/cholinesterase inhibitors, who were admitted at least once during the years 2006-2017 for a

fall and or fracture. Subjects with early mortality during the 3 months post fall were excluded, and "washout”

periods were also taken into consideration. They used a self-controlled case series design to examine the risk for

falls and fractures according to medications in the "non-exposure" versus the (immediate) "pre exposure" versus

the "exposure" periods. The exposure period was examined according to either antipsychotics alone (either first

or second gen), or cholinesterase inhibitors or both. As the authors point out, the method was unable to adjust for

time sensitive patient factors, specifically disease progression during the different time-periods. Thus, take for

example, a subject aged 75 at entry in 2006 who was treated with medication from 2015-2017: fall risk in the

"non exposed” period aged 75 would be compared to fall risk aged 85 at the start of the "exposed "period. In

addition to the consequences of natural aging (both biological and chronological) affecting fall risk, the time at

which dementia developed remains unknown. The overall finding of the study is that fall risk was lowest in the

"non exposed" period, highest (by a factor of nearly five) in the "pre exposure" period, and much lower in the

"exposed "period (around 1.4 compared to non exposed). The explanation of the findings is that the fall risk of

untreated BPSD carries a far higher risk of falls and fractures than treated BPSD.

Thank you very much for the supportive comment and the summary of our study. We appreciate the example and

have addressed the issues in detail in Comment 1-4.

33

This is a novel and very interesting research finding, resonates with clinical real world practice, and may

contribute to the ongoing debate and dilemma concerning the use of neuroleptics in BPSD. Overall the article is

concise, fairly clearly written, logical, clearly presented. The hypothesis and rationale are clearly spelled out, and

appropriate references provided (however, they might provide greater depth on the literature concerning

increased risk/morbidity/mechanism/ mortality associated with neuroleptics; and where possible also for

acetylcholinesterase inhibs). The methodology requires greater detail and depth. Results are presented clearly,

easy on the reader, figures and tables intuitive and easy to understand (except fig 2). The discussion, is short, to

the point, clear, and covers the studies limitations. In my opinion the paper is interesting relevant to a wide

readership and covers a very common clinical entity. The data presented, which seems to explain a common

dilemma in daily practice, is in contrast to much of the existing research which currently shapes consensus

opinion. My suggestions for improvement surround the need for far greater explanation concerning methodology.

We appreciate the positive comments and the constructive suggestions. We have revised the manuscript

accordingly, especially the method section, with more details as you suggested. Please review our point-by-point

responses below.

Comment 1-1:

Since aimed at a wide readership, the methodology requires to be explained in greater detail. Particularly, it

remains unclear to me after several readings how the different time periods (non/pre/yes-exposure) were dealt

with. Specifically, it seems very important to know how the "pre exposure" period was defined –since this is the

area which provides the novel research finding of higher risk with no treatment compared to treatment.

Thank you for the suggestion. We have revised the method section and explained the definition of the "pre-

exposure" period in detail. Moreover, we have revised the Figure 2 to make it clearer and easier to understand.

Revision:


We classified the study time into five discrete periods: 1) 14-day pre-exposure period prior to the exposure to

drugs, 2) exposure to ChEIs alone, 3) exposure to antipsychotic medications alone, 4) exposure to a combination

of ChEIs and antipsychotic medications, and 5) non-exposure when neither ChEIs nor antipsychotic medications

were given. The 14-day pre-exposure period was designed to evaluate the increased incidence of falls and

fractures due to neuropsychiatric symptoms of major NCD before the initiation of medications, and to benchmark

the risk magnitude during the exposure periods.

(Figure 2, Study scheme and definitions of exposure periods)

34

Abbreviation: ChEI, cholinesterase inhibitor; AP, antipsychotic medication

We defined a 3-year washout period before the index date for exclusion of patients with a prior history of falls and

fractures. We defined a 1-year baseline period to assess patients’ baseline covariates. We classified the study time into

five discrete periods: 1) 14-day pre-exposure period prior to the exposure to drugs, 2) exposure to ChEIs alone, 3)

exposure to antipsychotic medications alone, 4) exposure to a combination of ChEIs and antipsychotic medications,

and 5) non-exposure when neither ChEIs nor antipsychotic medications were given. If a refill was noted within 14

days after the end date of the last prescription (i.e., 14-day grace period), it was considered as a continuation of a

previous prescription. The remaining periods without prescription of ChEIs or antipsychotic medications were defined

as non-exposure.

Comment 1-2:

Page 6 for example (line 39) mentions that the pre exposure period was adjusted (typo adjested) from 1-4 weeks.

The logic here remains unclear, and requires to be fully spelled out.

Thank you for the question and the correction of a typo.

There is no clear evidence from the literature on the exact duration that is adequate for assessing baseline risk in

the pre-exposure period. We assumed 14 days because many elderly patients with chronic conditions were

followed up every two weeks, based on our experience. We have redefined pre-exposure periods with various

lengths of 7, 21, and 28 days, beyond the original 14-day period, and repeated the analysis to examine the

adequacy of the original definition of 14-day pre-exposure and the robustness of our results. Please review the

results below and the full version in Supplementary Table 3-10.

Revision:


We further redefined the pre-exposure period to include various lengths of 7, 21 and 28 days to test the adequacy

of a 14-day pre-exposure period in the main analysis (Supplementary Table 3-10).


When we redefined multiple lengths of pre-exposure periods to 7, 21 and 28 days, the adjusted IRRs of these pre-

exposure periods were 9.49 (8.64, 10.43), 4.91 (4.56, 5.30) and 4.43 (4.14, 4.75), respectively.


35

Furthermore, we have tested various lengths of pre-exposure period from 7 days to 28 days. The IRR was the

highest when the length of the pre-exposure period was defined as 7 days (IRR: 9.49), and it decreased as the

duration of the pre-exposure period lengthened - 14 days (IRR: 6.17), 21 days (IRR: 4.91), and 28 days (IRR: 4.43).

Based on these results, it can be concluded that a 7-day pre-exposure duration probably represents a period of rapid

deterioration. On the other hand, a duration of more than 21 days possibly captures a relatively stable status.

Therefore, our decision to use 14 days appears to be adequate.



Comment 1-3:

As regards the Non-exposed period-was the whole period taken and compared to the relatively short and the

extremely short exposed or pre exposed periods.

Thank you for the question. Although the different periods were of different durations, they were comparable

because the SCCS was based on the comparison of incidence “rates”; that is, the numbers of cases divided by the

observed duration.

Comment 1-4:

Here again, as in the example above, for a 75 year old subject enrolled at 2006, with treatment started in 2015,

then the comparison would be the average of 10 years at ages 75-85, compared to the average of 1-2 years at age

85-86. Again, how was the bias of aging and dementia progression adjusted for?

Events, n

Follow-up time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)




Pre-exposure 474 580.21 81.69 (74.58, 89.30) 9.49 (8.64,10.43)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.32 (1.23, 1.42)

***Main analysis

14-day prior to exposure (n=15,278)


Pre-exposure 657 1,254.90 52.35 (48.46, 56.47) 6.17 (5.69, 6.69)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.35 (1.26, 1.45)




Pre-exposure 792 1,924.81 41.15 (38.35, 44.09) 4.91 (4.56, 5.30)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.37 (1.27, 1.47)



Pre-exposure 949 2,591.75 36.62 (34.34, 39.00) 4.43 (4.14, 4.75)



Combination 1,270 12,037.35 10.55 (9.98, 11.14) 1.39 (1.29, 1.49)

36

Considering aging and dementia progression of patients, we used a time-varying manner to adjust for the age-

related effects of patients. To minimize the effects from age and age-related factors, we used one-year boundaries

for the analysis as suggested by a study published by BMJ [42]. This approach is to break down the study periods

not only by “exposure status” but also by “one-year boundary”, as presented below (for example, the time

periods were classified as 12 different periods). The concept is to compare the risk of exposure and non-exposure

within the same year, and an overall estimate is generated by Poisson regression model, which incorporates the

estimates from each year band. We have revised the manuscript and explained this in detail.

Revision:

(Method, Statistical Analysis and Covariates, 2nd paragraph)

Therefore, we adopted a one-year boundary to split the study duration in order to take into account possible age-

related effects when estimating the effects. That is, we partitioned the observation period not only by the

exposure status, but also by one-year intervals (e.g., the first, second and third years etc.) This time-varying

method to adjust for age effects has been described in detail elsewhere [42].


42. Petersen I, Douglas I, Whitaker H. Self-controlled case series methods: an alternative to standard epidemiological study

designs. BMJ. 2016;354:i4515.

Comment 2:

The study outcomes were falls and fractures. It is somewhat unclear throughout the text whether or not the actual

outcome is a fall resulting in a fracture, or a fall (on its own with no fracture) or any fracture (??not resulting

from a fall??). Sometimes the text is falls and fractures, sometimes falls or fractures. Only in figure 3 does it

become apparent. Similarly only on page 7 is it specifically stated, line 39-42: states that subgroup sensitivity

analyses were performed for falls and fractures separately, as well as falls and fractures together (all events, in fig

3) with consistent results.

Thank you for the clarification question. Our primary analysis included a composite of falls and fractures, after

which we analysed falls and fractures separately in the secondary analyses.

From the database, we were not able to differentiate whether the fractures were caused by falls; therefore, we

selected only traumatic fractures (excluding pathological fractures and stress fractures) as the study outcomes to

improve the validity. We also restricted the outcomes to specific falls and hip fractures in one of the sensitivity

analyses because previous studies have reported that 95% of hip fractures were caused by falls [43]. The results

from the sensitivity analysis were consistent. We have revised the manuscript to avoid confusion. Please review

the results below and the full version in Supplementary Table 3-8.

Revision:


37

The primary outcome of interest was a composite of falls (ICD-9 code: E880-E888; ICD-10 code: W00-W19)

and traumatic fractures (ICD-9 code: 800-829; ICD-10 code: S02, S12, S22, S32, S42, S52, S62, S72, S82, S92),

whereby we also analysed falls and fractures separately in secondary analyses.


Moreover, we carried out a sensitivity analysis and selected specific diagnostic codes for falls and fractures (i.e.,

falls from a different level [ICD-9: E880-E884; ICD-10: W00.1, W00.2, W05-W17], falls from the same level

[ICD-9: E885-E886; ICD-10: W00.0, W01-W04, W18], and hip fractures [ICD-9: 820; ICD-10: S72]) to

examine the validity of the outcomes. The reason for selecting hip fractures was that more than 95% of hip

fractures are related to falls [43] (Supplementary Table 3-8).


43. Parkkari J, Kannus P, Palvanen M, Natri A, Vainio J, Aho H, et al. Majority of hip fractures occur as a result of a fall and

impact on the greater trochanter of the femur: a prospective controlled hip fracture study with 206 consecutive patients.

Calcified tissue international. 1999;65(3):183-7.



Comment 3:

Figure 2 was very unclear to me and needs to be improved in my opinion. Where is the Non Exposure period? It

seems to be lacking a legend.

Thank you for the comment. We have revised our Figure 2 accordingly, and we have also added on a legend for

the non-exposure period.

(Figure 2, Study scheme and definitions of exposure periods)

Events, n

Follow-up time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)

Redefine outcome by specific codes, all

events (n=5,458)


Pre-exposure 379 453.89 83.50 (75.40, 92.23) 10.16 (9.10,11.35)



Combination 537 4,321.32 12.43 (11.41, 13.51) 1.67 (1.49, 1.87)

Falls (n=621)


Pre-exposure 73 51.14 142.70 (112.70, 178.50) 10.77 (8.19,14.16)



Combination 92 404.43 22.75 (18.44, 27.77) 1.49 (1.09, 2.04)

Fractures (n=5,149)


Pre-exposure 357 428.79 83.26 (74.95, 92.24) 10.41 (9.29,11.66)



Combination 496 4,115.68 12.05 (11.03, 13.15) 1.70 (1.51, 1.91)

38


We defined a 3-year washout period before the index date for exclusion of patients with a prior history of falls and

fractures. We defined a 1-year baseline period to assess patients’ baseline covariates. We classified the study time into

five discrete periods: 1) 14-day pre-exposure period prior to the exposure to drugs, 2) exposure to ChEIs alone, 3)

exposure to antipsychotic medications alone, 4) exposure to a combination of ChEIs and antipsychotic medications,

and 5) non-exposure when neither ChEIs nor antipsychotic medications were given. If a refill was noted within 14

days after the end date of the last prescription (i.e., 14-day grace period), it was considered as a continuation of a

previous prescription. The remaining periods without prescription of ChEIs or antipsychotic medications were defined

as non-exposure.

Comment 4:

Figure 1: need a bit of clarification: 22,665 subjects reduced down to 10, 750. Why/who were the 11,915

excluded? Were they different to the 2,288 and 4,613 for whom a reason was given? The numbers need to be

clarified

Thank you for the question. We have revised the numbers in the flowchart and added the reasons for each

exclusion. Please review the revised Figure 1.


39

Reviewer 4: Dr. Jon Brännström

Overall Comment

Thank you for the opportunity to review this interesting manuscript. Wang et al has put the spotlight on a very

important and delicate matter in observational research - that of the difficulties in drawing conclusions about

causality from associations. The method is elegant, but there is some room for improvement. This manuscript has

potential of being transformed into an article worthy of publication, if improved further. I have two major points,

and several minor points, that need to be addressed.

Thank you very much for the very supportive comment. Please review our point-by-point response and

corresponding revisions below.

Comment 1:

The principle finding of the highest risk of falls and fractures during the 14-day pre-treatment period relies

heavily on the timing of dates set for initiation of treatment as well as the dates of diagnoses of falls and

fractures. Hence, the methods section (page 6, lines 3-23) needs to be more detailed and able to answer the

following questions:

‐ Outcome date. Diagnoses of falls and fractures were collected from inpatient claims, but how were the dates

of the diagnoses registered? Were diagnoses registered at the date of the actual occurrence of the fall or

fracture, at admission to inpatient care, or at release from inpatient care? Depending on how the database

(NHID) is organized and on how diagnoses are dated in the register, it is possible that the outcome event is

registered later than the actual event. This needs to be addressed.

‐ Index (exposure) date. Are all medications administered during inpatient care taken into account in the

analyses? It is not uncommon for a pharmacological treatment to be initiated during inpatient care, and

continued after release from hospital as a prescription. This could result in the index/exposure date being

registered days or weeks after the actual initiation of treatment. This needs to be clarified.

Thank you for the critical questions. Yes, we agree with you that the accuracy of dates for exposure and outcome

is especially important for self-controlled case series. We have the exact dates of the prescriptions for both

inpatient and outpatient claims. However, for those who had falls or fractures during hospitalization, the outcome

date may be a few days or weeks later than the actual date of event occurrence because the diagnoses may only

have been registered at release from inpatient care. We may thus have underestimated the risk during exposure

period for those who had falls or fractures during hospitalization and then discontinued the medication (the event

was misclassified to non-exposure period).

Revision:


Seventh, we did not have the exact outcome dates for those who had falls or fractures during hospitalization

because the diagnoses were registered at the discharge date, and thus the outcome dates we analysed may have

been later than the dates of actual event occurrences. Therefore, we may have underestimated the risk during the

exposure period for those who had falls or fractures during hospitalization and then discontinued the medications.

40

Comment 2:

I can understand why you have excluded patients with schizophrenia, but to exclude patients with depression

does not make sense to me. The prodromal phase of major neurocognitive disorder (NCD; dementia) often

includes depression, and depression and major NCD comorbidity is common. Excluding people with depression

makes the conclusions of your results only valid for a subgroup of patients with major NCD. I strongly suggest

that you redo the selection of participants to include those with psychiatric illness other than primary psychotic

conditions, such as schizophrenia. If not, this must be addressed as a major limitation.

We appreciate the comment. Originally we excluded patients with these underlying conditions (schizophrenia,

bipolar disorder, and depression) to ensure the antipsychotic medications were used to treat neuropsychiatric

symptoms of major neurocognitive disorder (NCD). However, we agree with your opinion and therefore have

removed “depression” from the exclusion criteria. Please review Figure 1 for the study selection flowchart.

Revision:

(METHODS, Study population)


medications were used to treat the neuropsychiatric symptoms of major NCD.


41

Comment 3:

Throughout the manuscript the terms “dementia” and “behavioural and psychological symptoms of dementia

(BPSD)” are used. Consider using the more up-to-date terminology of “neurocognitive disorder (NCD)” and

“neuropsychiatric symptoms of NCD” instead.

Thank you for the suggestion. We have updated the terminology throughout the manuscript.

Comment 4:

Consider analysing more time frames than pre-exposure (14 days), non-exposure and during exposure (3

different). What about 1-14 days following exposure? Several previous studies have found strong associations

between antipsychotics and falls/fracture soon after initiation of treatment. This would certainly be an interesting

addition to your study, and probably easy to conduct. A more difficult period to analyse is 1-14 days following

discontinuation of treatment. Does discontinuation lower or raise the incidence rate? Knowing when someone

actually discontinues treatment is quite difficult in a registry study however, so this might not be feasible.

Thank you for the inspiring suggestion. We have added a sensitivity analysis which further specifies the periods

within 1-14 days following exposure, and also 1-14 days following discontinuation of treatment. We found the

risk during the 1-14 day period following discontinuation of treatment was higher than during the non-exposure

period. Please review the results below and the full version in Supplementary Table 3-12.

Revision:

(Supplementary Figure 1. Study scheme and definitions of exposure periods for sensitivity analysis)


We defined a 3-year washout period before the starting date to exclude patients with a prior history of falls and

fractures. We defined a 1-year baseline period to assess patients’ baseline covariates. 1) 14-day pre-exposure period

prior to the exposure to drugs, 2) exposure to ChEIs alone, 3) exposure to antipsychotic medications alone, 4)

exposure to a combination of ChEIs and antipsychotic medications, and 5) non-exposure when neither ChEIs nor

antipsychotic medications were given. If a refill was noted within 14 days after the end date of the last prescription

(i.e., 14-day grace period), it was considered as a continuation of a previous prescription. We further classified the

periods 6) within 1-14 days after exposure, and 7) within 1-14 days after discontinuation of treatment. The remaining

time periods without prescription of ChEIs or antipsychotic medications were defined as non-exposure.

(METHODS, Sensitivity Analyses, 3rd paragraph)

42

Some studies have found strong associations between antipsychotic medications and falls and fractures

immediately after treatment initiation or shortly after discontinuation of treatment [44-45]. We therefore defined

two more study periods of 14 days: one after treatment initiation, the other following treatment discontinuation

(Supplementary Table 3-12).


Finally, the adjusted IRRs were 3.31 (2.96, 3.70) within 14 days after treatment initiation and 1.24 (1.05, 1.47)

within 14 days after discontinuation of treatment.


Moreover, the IRR within 14 days after treatment initiation was higher than during other exposure periods,

suggesting that a minimum duration of exposure might be required to stabilize these patients. The IRR within 14

days after discontinuation of treatment was higher than during the non-exposure period, suggesting that clinical

attention is still necessary in the initial stage after patients discontinue their treatments.


44. Pouwels, S., et al., Antipsychotic use and the risk of hip/femur fracture: a population-based case-control study. Osteoporosis

international : a journal established as result of cooperation between the European Foundation for Osteoporosis and the

National Osteoporosis Foundation of the USA, 2009. 20(9): p. 1499-1506.

45. Torstensson, M., et al., Danish register-based study on the association between specific antipsychotic drugs and fractures in

elderly individuals. Age and Ageing, 2016. 46(2): p. 258-264.



Comment 5:

The method of the current manuscript, as well as the findings of the study, have obvious similarities to a study by

Brannstrom et al, published in JAMDA (DOI: 10.1016/j.jamda.2020.06.047). This paper should be cited and your

results compared to those of that study. The results of that study would be in support of your conclusions.

Thank you very much for the recommendation. We have learnt a lot from the study by Brannstrom et al. We have

summarized and compared the findings from Brannstrom et al with ours in the revised manuscript.

Revision:

(DISCUSSION, Principal Findings, 2nd paragraph)

Our results showed that patients may have already been at high risk before receiving the treatment, implying that

the interconnection among patients’ underlying conditions, drug effects on relieving neuropsychiatric symptoms,

and side effects of ChEIs and antipsychotic medications has increased the difficulty of delineating the exact

Events, n

Follow-up time,

person-years,

sum

Incidence rate Adjusted IRR

(95%CI)

Redefine study periods (n=15,278)


Pre-exposure (day -14 days of exposure) 657 1,254.90 52.35 (48.46, 56.47) 6.12 (5.64, 6.64)

Exposure (day 1 to day 14 of exposure) 325 1,136.03 28.61 (25.62, 31.85) 3.31 (2.96, 3.70)

Cholinesterase inhibitor (day 14+ of exposure) 1,746 18,613.18 9.38 (8.95, 9.83) 1.16 (1.09, 1.24)

Antipsychotic medication (day 14+ of exposure) 1,164 12,666.02 9.19 (8.67, 9.73) 1.13 (1.05, 1.22)

Combination (day 14+ of exposure) 1,178 11,652.41 10.11 (9.55, 10.70) 1.26 (1.17, 1.36)

Day 1 to day 14 after medication discontinuation 142 1,362.32 10.42 (8.81, 12.25) 1.24 (1.05, 1.47)

43

contributing factors and quantifying the magnitude of the risk of falls and fractures that each factor poses. Similar

findings have been reported in previous studies. For example, Brännström et al. [46] reported that the highest risk

of hip fractures occurred before initiation of antipsychotic medications (OR: 9.09; 95% CI: 7.00 to 11.81 within -

16 to -30 days and OR: 5.84; 95% CI: 4.42 to 7.71 within -1 to -15 days), rather than after receiving treatment

(OR: 4.31; 95% CI: 3.05 to 6.10 within 1 to 15 days).



the Initiation of Treatment. Journal of the American Medical Directors Association. 2020;21(11):1636-42.e6.Pratt N,

Roughead EE, Ramsay E, Salter A, Ryan P. Risk of hospitalization for hip fracture and pneumonia associated with


Comment 6:

You only count the first incidence of a fall or fracture, as subsequent falls are not independent of previous

falls/fractures. As correct as this may be, this could lead an overestimation of falls/fractures in the first specified

time frame, the pre-exposure phase. Consider performing sensitivity analyses including all counts of falls and

fractures, to minimize this risk.

Thank you for the suggestion. We have added a sensitivity analysis including all counts of falls and fractures to

address the issue. Please review the results below and the full version in Supplementary Table 3-9.

Revision:


In the main analysis, we only included the first incidence of falls or fractures since subsequent events might not

be independent of previous falls or fractures. However, we performed a sensitivity analysis including all episodes

of falls and fractures in order to evaluate whether the exclusion of subsequent outcomes had significantly

impacted the results (Supplementary Table 3-9).



Comment 7:

You claim to have performed sensitivity analyses excluding those who died, and with different lengths of the pre-

exposure period. No results of the sensitivity analyses were included in the review material, however, and they

should be presented in full for transparency.

Thank you for the comment. We have added all the results from sensitivity analyses to the revised manuscript.

Table 3 presents a summary of results for adjusted IRR only. Detailed results are presented in Supplementary

Table 3-1 to 3-12.

Events, n

Follow-up time,

person-years,

sum

IR (95%CI),

/100 person-years

Adjusted IRR

(95%CI)

All counts of falls and fractures (n=15,278)


Pre-exposure 1,159 1,254.90 92.36 (87.16, 97.79) 5.53 (5.21, 5.88)



Combination 2,726 12,037.35 22.65 (21.81, 23.51) 1.31 (1.25, 1.38)

44

Comment 8:

Seeing how you excluded 8 participants due to missing sex information, I was expecting the results to be sex-

disaggregated. You should consider presenting results separately for women and men, to be able to detect

possible sex differences in the pattern of incidence. Several studies have shown sex differences in the

presentation of neuropsychiatric symptoms, as well as drug prescription patterns, so at least the main analyses

should be available as sex-disaggregated.

Thank you for the suggestion. We have added the suggested analysis accordingly. The results were consistent

with the main analysis. Please review the results below and the full version in Supplementary Table 3-1.

Revision:


We stratified the patients by sex (male or female) and age group (65 to 74 years, 75 to 84 years, and 85 years or

above) to minimize the effects of sex and age (Supplementary Table 3-1 and 3-2).

(DISCUSSION, Meaning of the study, 1st paragraph)

From our sensitivity analyses, we identified some groups who might have a higher risk of outcomes at the

baseline. For example, patients with advanced age or male sex had higher risks of falls and fractures.



Comment 9:

Exclusions. In Figure 1 you present the flowchart and exclusions are presented in two boxes. In the second box

the specified exclusions do not add up to the total number of exclusions. Please specify the reason for all

exclusions.

We have revised the Figure 1 and added the justification for each exclusion criterion.

Revision:


Events, n

Follow-up

time,

person-years,

sum

IR (95%CI),


Stratification by sex

Male (n=5,088)


Pre-exposure 285 421.32 67.64 (60.13, 75.85) 7.63 (6.73, 8.65)



Combination 443 3,591.78 12.33 (11.22, 13.52) 1.48 (1.31, 1.67)

Female (n=10,190)


Pre-exposure 372 833.58 44.63 (40.26, 49.34) 5.35 (4.80, 5.95)



Combination 827 8,445.56 9.79 (9.14, 10.48) 1.29 (1.18, 1.41)

45






Comment 10:

Throughout the manuscript minor errors occur (e.g. spelling, abbreviations not spelled out, figure legends not

being fully informative). Consider have the manuscript professionally edited.

Thank you for your suggestion. We have completed professional editing by an English native speaker.

Comment 11:

In your discussion you draw the conclusion the drugs mitigate the risk of falling and that the slightly higher

incidence rate ratios during exposure means that the treated underlying conditions are responsible for this

increase. This is possible of course, but another possibility (and just as likely, if not more likely) is that the drugs

themselves pose a small increase in risk. Your discussion should be more nuanced and humble towards the

possibility of an actual risk associated with psychotropic drug use. Although your results, and those of some

previous studies, point towards residual confounding and confounding by indication being highly prevalent in the

often found association between psychotropic drug use and falls/fractures, this does not mean that psychotropics

are protective in this regards, and surely this has not been proven.

We appreciate the critical comment. We have revised our discussion in both depth and scope. We have also

revised our conclusion according to your comment.

Revision:

(DISCUSSION, Principal findings, 1st paragraph)




46



exposure period, may be due to the neuropsychiatric symptoms in addition to the medication use. Our results

suggest that the increased risk of falls and fractures was very likely resulted from the underlying BPSD, rather

than the exposure of ChEIs and antipsychotics. The use of the study drugs may in fact mitigate the risk. This

implied that the patients’ condition might not have been fully restored to a steady status, despite receiving

treatment. This conclusion remained robust throughout a series of subgroup and sensitivity analyses.

(DISCUSSION, Unanswered questions and future research)


status of patients, the actual reasons will need to be elucidated by further studies. The risks during the exposure

to the treatment could reflect a composite consequence of patients’ unstable disease status, effects of medications

on relieving neuropsychiatric symptoms, and side effects of medications. For example, some side effects of







(Conclusion)





period, although the magnitude was much lower than during the pre-exposure periods. Prevention strategies and


status.

may 26 , 2021

Documents