Cost-effectiveness of a pharmacist-led information technology intervention for reducing rates of clinically important errors in medicines management in general practices (PINCER) : Technical Appendix

1. Summary of the PINCER trial hypothesis, methods, clinical results, initial economic analysis and costs of the intervention.

This is a summarised version of material that has already appeared in Avery et al 2009[1] and Avery et al 2012.[2]

1.1 PINCER interventions

Intervention practices received a 12-week multi-faceted pharmacist-led IT-enabled intervention (henceforth referred to as ‘PINCER’). As we were proactively identifying patients at high risk of potentially serious errors, we considered it unethical to have a no-intervention control arm; the control arm practices therefore received computer-generated feedback on at-risk patients (henceforth referred to as ‘Simple feedback’). It should be noted that all general practices had access to a limited amount of computerised decision support for prescribing, such as drug interaction checking.

Simple feedback (control arm): Following baseline data collection, control practices received computerized feedback on patients identified as at risk from potentially hazardous prescribing and medicines management plus brief written educational materials explaining the importance of each type of error. Practices were asked to institute changes they considered necessary within 12 weeks following the baseline data collection.

PINCER (intervention arm): This comprised of simple feedback, plus a pharmacist-led IT-enabled complex intervention lasting 12 weeks. First, the pharmacist arranged to meet with the practice team to discuss the computer-generated feedback on patients with errors. All doctors were encouraged to attend along with at least one member of the nursing staff, the practice manager and at least one member of the reception staff.

Before the meeting members of staff were provided with a brief summary of the objectives of the PINCER intervention and a summary of the findings from the computer search. At the meeting the pharmacists were asked to use the principles of educational outreach while also taking account of human error theory, in their discussions. Then, pharmacists used techniques to help correct medication errors that had been identified and prevent future medication errors. Interventions included: review of patients’ medical records; discussions with GPs to decide on actions; inviting patients to be reviewed or to have blood tests, and working with the practice team to improve local safety systems.

1.2 PINCER trial hypothesis

The trial aimed to determine whether PINCER was more effective than Simple feedback in reducing the proportions of patients at risk of the three primary outcome measures relating to prescribing and monitoring errors at six months post-intervention. A range of other prescribing and monitoring-related errors were also included.

1.3 PINCER methods

An overview of the trial design (Registration number: ISRCTN21785299) is outlined in this appendix. Further details are available in the published trial protocol.[1]

1

Trial design and eligibility: We conducted a two-arm pragmatic cluster randomised trial. A cluster design was chosen because the intervention was at the level of the general practice. General practices were eligible to participate if they were computerised with electronic prescribing.

Randomization and masking: The general practice was the unit of allocation. Eligible practices were stratified by centre (two strata: Manchester and Nottingham) and list size (three strata: <2500, 2500-6000, >6000 patients) and randomly allocated within strata (1:1 ratio) to the PINCER or Simple feedback arms using block randomisation with varying block sizes of 2 or 4. General practices, pharmacists and researchers visiting practices were not masked to treatment arm allocation. All outcome data for the trial were extracted from patient records using pre-specified electronic searches.

Outcome measures: Our primary and secondary outcome measures were chosen based on errors considered important in terms of burden and severity of iatrogenic harm in primary care, and those that were detectable from GP computer systems:

Primary:

1. Patients with a history of peptic ulcer who have been prescribed a non-selective NSAID2. Patients with asthma who have been prescribed a beta-blocker3. Patients aged 75 years and older who have been prescribed an ACE inhibitor or a loop diuretic long-

term who have not had a computer-recorded check of their renal function and electrolytes in the previous 15 months

Secondary:

2a Patients with asthma (and no history of CHD) who had been prescribed a beta-blocker

4. Proportions of women with a past medical history of venous or arterial thrombosis who have been prescribed the combined oral contraceptive pill

5. Patients receiving methotrexate for at least three months who have not had a recorded full blood count (Outcome 5a) and/or liver function test (Outcome 5b) within the previous three months

6. Patients receiving warfarin for at least three months who have not had a recorded check of their INR within the previous 12 weeks

7. Patients receiving lithium for at least three months who have not had a recorded check of their lithium levels within the previous three months

8. Patients receiving amiodarone for at least six months who have not had a thyroid function test within the previous six months

9. Patients receiving prescriptions of methotrexate without instructions that the drug should be taken weekly

10. Patients receiving prescriptions of amiodarone for at least one month who are receiving a dose of more than 200mg per day

Composite secondary outcome measures:

Patients with at least one prescribing problem (a combination of outcome measures 1, 2 & 4) Patients with at least one monitoring problem (a combination of outcome measures 3, 5, 6, 7 & 8)

2

Statistical analysis: Characteristics of practices and participants were described at baseline and compared informally between treatment arms. The prevalence of primary and secondary outcome measures was described at each time point by allocation arm, at the level of the patient, using the numerator, denominator and percentage. An intention-to-treat (ITT) analysis was used such that practices were analysed in the arms they were allocated to regardless of whether they received the intervention or not.[3, 4] All outcome measures were compared between arms using random effects logistic regression, with patient at level one and practice at level two, to estimate odds ratios and 95% confidence intervals (CI). The main analysis adjusted for stratum as a fixed effect (practice level), baseline medication-related errors (patient level), deprivation[5] and training status (practice level). Intra-class correlation coefficients (ICCs) (and 95% CI) were estimated from these models.

Significance was assessed based on likelihood ratio tests with a p value of <0 05 taken as significant. All∙ analyses were undertaken using Stata Version 10.[6] Outcome data were obtained for all participating practices at both the six- (primary) and 12-month assessment points, hence there were no missing data. No adjustments were made for multiple endpoints. Models were checked by examining plots of standardised empirical Bayes estimates for the random effects and sensitivity analyses undertaken excluding practices with estimates above or below two standard deviations (SDs).

Sample size calculations: Separate sample size calculations were performed for each primary outcome measure. Sample sizes unadjusted for clustering were calculated using nQuery Advisor® Version 6.0[7] and were inflated to adjust for clustering[8] using ICCs and average cluster sizes estimated from 43 general practices contributing anonymous clinical data to the QRESEARCH database (www.qresearch.org).[2] A total of 66 practices were required to detect a difference between an 11% reduction in error rate in the Simple feedback arm and a 50% reduction in the PINCER arm for the primary outcome measures, with 80% power and two-tailed alpha of 0 05. ∙

3

1.4 Clinical results

Table 1 Prevalence of prescribing and monitoring problems at main six month follow-up by allocation arm (from Avery et al 2012 [2])

OM Outcome Simple feedback (%) PINCER (%) Adjusted odds ratio* (95% CI)

ICC

Primary outcome measures

1 Patients with a history of peptic ulcer prescribed an NSAID without a PPI / Patients with a history of peptic ulcer without a PPI

78/2 035 (3·8) 61/1 852 (3·3) 0·91 (0·59 to 1·39) 6·54x10-7

2 Patients with asthma prescribed a beta-blocker / Patients with asthma

692/23 520 (2·9) 545/21 359 (2·6) 0·78 (0·63 to 0·97) 0·008

3 Patients aged ≥75 on long term ACEIs or diuretics without urea and electrolyte monitoring in the previous 15 months / Patients aged ≥75 on long term ACEIs or diuretics

452/5 813 (7·8) 306/5 242 (5·8) 0·63 (0·41 to 0·95) 0·13

Secondary outcome measures

2a Patients with asthma and not CHD prescribed a beta-blocker / Patients with asthma and not CHD

414/22 294 (1·9) 326/20 283 (1·6) 0·79 (0·62 to 1·02) 0·009

4 Female patients with a history of venous or arterial thromboembolism and arterial thrombosis prescribed combined oral contraceptives / Female patients with a history of venous or arterial thromboembolism and arterial thrombosis

15/2 987 (0·5) 4/2 640 (0·2) 0·57 (0·05 to 6·17) 0·24

5a Patients prescribed methotrexate for ≥3 months without a full blood count in last three months / Patients prescribed methotrexate for ≥ 3 months

194/552 (35·1) 130/531 (24·5) 0·51 (0·27 to 0·99) 0·22

5b Patients prescribed methotrexate for ≥ 3 months without a liver 186/552 (33·7) 134/531 (25·2) 0·50 (0·28 to 0·91)† 0·16

4

function test in last three months / Patients prescribed methotrexate for ≥ three months

6 Patients prescribed warfarin for ≥ 3 months without an INR in last three months / Patients prescribed warfarin for ≥ three months

69/1 752 (3·9) 76/1 877 (4·1) 0·98 (0·52 to 1·85) 0·10

7 Patients prescribed lithium for ≥ 3 months without a lithium level in last three months / Patients prescribed lithium for ≥ three months

88/213 (41·3) 56/176 (31·8) 0·50 (0·29, 0·85) 0·02

8 Patients prescribed amiodarone for ≥ 6 months without a thyroid function test in the last six months / Patients prescribed amiodarone for ≥ 6 months

92/247 (37·3) 80/233 (34·3) 0·77 (0·41 to 1·43) 0·11

9 Patients prescribed methotrexate without instructions to take weekly / Patients prescribed methotrexate

13/309 (4·2) 0/271 (0·0) Not calculable

10 Patients prescribed amiodarone for ≥ one month at a dose >200mg/day / Patients prescribed amiodarone for ≥ one month

1/231 (0·4) 1/232 (0·4) 0·95 (0·06 to 15·45)‡§ 1·07x10-5

Patients with at least one prescribing problem / Patients at risk of at least one prescribing problem

785/27 808 (2·8) 610/25 246 (2·4) 0·78 (0·64 to 0·94) 0·01

Patients with at least one monitoring problem / Patients at risk of at least one monitoring problem

901/8 011 (11·3) 652/7 449 (8·8) 0·64 (0·51 to 0·82)† 0·05

**Adjusted for randomisation stratum, baseline prevalence of errors, deprivation and training status unless otherwise stated. Number does not equal the sum of the denominators in each arm, as this only includes those with baseline and follow-up data

†Includes interaction between treatment arm and covariate dichotomised at the median value (≤ median vs. > median) ‡§ adjustment for other variables not calculable

5

1.5 Intervention costs

Costs were obtained from the perspective of the English NHS. The study was not powered to detect differences in costs because there is no prior study upon which to base a power calculation. The time horizon was six months in the base case. As all costs were incurred at practice level, correction for clustering was not required. The only cost associated with the Simple feedback arm required the researchers to go back into the practices at set time periods to generate error reports from GP systems. These costs reflect the equivalent resource that would be consumed in practice to generate these error reports. The PINCER intervention comprised report generation costs, plus a training session; facilitated meetings; monthly meetings; practice feedback meetings; time spent in each practice outside meetings following up errors:

1. Error report generation: Researchers involved with the study estimated that running computer queries took two hours and report printing took 15 minutes.

2. Training session: The initial pharmacist training session costs reflect set-up costs of the intervention and would be incurred if the intervention were rolled out into clinical practice.

3. Facilitated meetings: Five quarterly facilitated meetings were held for the pharmacists running the interventions across the 36 intervention practices to provide a strategic overview of the initiative and to maximise homogeneity of the intervention. In clinical practice, a facilitated meeting would equate to a strategic practice meeting.

4. Monthly meetings: The aim of the monthly meeting between practice pharmacists and the Trial Manager was to deal with operational issues within individual practices. Twelve monthly meetings were held. These meetings would equate to operational meetings and, in practice, would be added onto other Primary Care Trust (PCT) team pharmacist meetings with general practices.

5. Practice feedback sessions: The aim of the feedback session was to provide each practice with feedback and support on management of errors, using root cause analysis to identify how systems could be improved. Between one and three practice feedback sessions were held for each of the 36 intervention practices. The length of time spent per practice was not recorded, but was estimated by the pharmacists to be about one hour.

6. Time spent dealing with errors: PINCER pharmacists also spent time working on the intervention outside the meetings listed above to deal with errors identified. On the basis of information recorded by the pharmacists, the mean time spent dealing with each error was 23 3 minutes (median 18 4 minutes,∙ ∙ range 0 – 180 minutes). The time spent on dealing with these errors was calculated for each practice. Where data were missing for time spent dealing with an error, it was assumed that time taken equated to the mean time taken for that error. The mean time spent in a PINCER intervention practice on the errors included in the economic analysis was 1 106 minutes (median 873 minutes, range 155 – 3 585 minutes).The general practices involved in the Simple feedback arm and the PINCER intervention arm may have spent time correcting errors and improving safety systems, but these data were not collected in either arm.

Table 2 provides a summary of the cost components for the Simple feedback and PINCER arms.

Cost component Method of cost allocation

Error report generation Flat rate per practice

6

Pharmacist training

Preparation for meetings

Pharmacist facilitated meetings

Monthly meetings

Allocated to intervention practices according to list-size, calculated by deriving the total cost for the 36 practices, and then allocating a portion of that cost to a practice based on the list-size.

Practice feedback meetings Allocated to the PINCER intervention practices according to how many were carried out

Time spent dealing with errors Costs of time spent per error detected were calculated and added to the cost for each practice

Table 2: Summary of cost components for Simple feedback and PINCER intervention arms

1.6 Initial economic evaluation

Overall approach: An indicative prospective economic evaluation was undertaken of PINCER compared with Simple feedback in reducing proportions of patients at risk from prescribing errors in general practice, from the perspective of the English NHS. The outcome for the economic analysis was the number of errors detected by the report generation process in both the PINCER and Simple feedback arms at six months and 12 months after the intervention. For the economic analysis, outcome measures 1, 2, 3, 5(a and b), 7 and 8 were used. We did not include outcome measure 4 because the number of patients with errors was very small; we did not include outcome measures 6, 9 and 10 because there were difficulties obtaining full data in all practices, as described in our trial protocol.[1]

Methods: Regression analysis was planned to assess the effect of base list-size and at-risk list-size, as well as the following: i) number of GPs, in order to capture scale effects (this included the square of base list-size and at-risk list-size to also capture non-linear economies of scale); ii) Quality and Outcomes Framework (QOF - http://www.qof.ic.nhs.uk) and medicine-related QOF score (both were tested but QOF score was more informative), in order to capture efficiency; iii) Strategic Health Authority, in order to capture any potential regional fixed effects, and iv) demographic information on area-level deprivation, average ages and gender proportions. The negative binomial model was used for regression analysis of errors.[9] Costs were estimated via generalised linear modelling (GLM) assuming a gamma distribution. Incremental cost-effectiveness ratios (ICERs) were calculated for differences in error rates between the Simple feedback and PINCER interventions. Bootstrapping with replacement was employed, to identify the magnitude of uncertainty around the ICERs, utilising Microsoft Excel, using a minimum of 10 000 iterations to obtain 2 5%∙ and 97 5% percentiles of the ICER distribution.[10] ∙

Findings: Adjustment of outcomes: Negative binomial regression determined that only intervention and list-size were important predictors of error rates. An interaction term was also included between intervention and scale variables. It was not significant, however, and was subsequently removed. The model was estimated using the robust sandwich-estimator of the variance-covariance matrix.5 Marginal effects from the final model are in Avery et al (Web-table 6).[2] Base list-size was scaled downwards for regression; the coefficient reflects

7

the marginal increase in errors from a predicted mean of 33 69 per practice of an additional 100 patients.∙ The variables for number of GPs and at-risk list-size were not statistically significant when base list-size was included, suggesting that they are all representing the overall catchment of the practice. Neither area-level deprivation nor QOF scores were statistically significant. Besides using information criteria to select specifications and to choose between Poisson and negative binomial regression models, standard approaches to testing for independence in the errors were used, including fitting covariates to the residuals and fitting residuals to the fitted values. This supported the models above as appropriate approaches.

Adjustment of costs: Adjusted costs were estimated via GLM assuming a gamma distribution. Only the PINCER group was used in this analysis, since intervention costs in the Simple feedback group were constant (see Web-table 7 in Avery et al[11]). Only base list-size was significant.

Probabilistic incremental economic analysis: A probabilistic incremental cost-effectiveness analysis was completed using the adjusted cost and outcome data outlined above. The predicted errors and costs following the negative binomial regression for errors and the GLM regression for cost were used to characterise the distributions of incremental cost and effect. This allowed for bootstrapping with broader probabilistic sensitivity analysis since the values of the covariates were allowed to vary in the sample. Figure 2 illustrates the ICER distribution at six and 12 months.

-200

0

200

400

600

800

1000

1200

-14-12-10-8-6-4-202

Diff

eren

ce in

cos

t per

pra

ctic

e be

twee

n PI

NC

ER in

terv

entio

n an

d Si

mpl

e fe

edba

ck/£

Difference in number of errors per practice between PINCER intervention and Simple feedback

12 months 6 months

Figure 1 Cost effectiveness plane (cost per error avoided at six and 12 months)

8

2. Results from outcome measure-specific modelsA brief summary of the rationale, literature sources and search strategy, model structure, probabilities, effect of error, costs and utilities for each of the outcome measure-specific models is presented here. Illustrative representations of each model are followed by tables which detail transition probabilities, HRQL utility weights for each health state and the associated costs. The models are provided in detail in Appendices 1 to 6 of the full economics report.[12]

2.1 Patients with a medical history of peptic ulcer who have been prescribed a non-selective NSAID and no PPI

NSAIDs are associated with upper gastrointestinal complications.[13] Each year in the UK, NSAIDs cause about 3,500 hospitalisations for, and 400 deaths from, ulcer bleeding in people aged 60 years or above.[14] Risks of bleeding are increased when the drug is prescribed long-term.[15] When offering treatment with an oral nonselective NSAID, the drug selected should be co-prescribed with a PPI.[15]

The search strategy built on the systematic review and economic evaluation by Brown et al. (2006).[16, 17] Additionally, an inclusive search strategy was implemented using the key terms “peptic ulcer” and ‘bleeding$’ in combination with NSAIDs. As the patient group under study is defined as “patients with a past medical history of peptic ulcer who have been prescribed a non-selective NSAID and no PPI”, only data on the defined subgroup were evaluated.

There are many published decision-analytic models in relating to the use of NSAIDs (see Brown et al.[16]). These models primarily concentrate on two areas: switching between NSAIDs and the point at which gastro-protective agents or COX-2 inhibitors need to be added to therapy. The ACCES Model (Arthritis Cost Consequence Evaluation System)[18] is a model that examines the use of different gastro-protective agents and COX-2 inhibitors. In Brown et al, four pathways were defined in using NSAIDs: no adverse event; discomfort; symptomatic ulcer and serious gastrointestinal events.[16] We used these models to inform the design of our model.

We were able to access data from relevant RCTs and observational studies to provide relatively reliable estimates of probabilities in the appropriate patient group. Identifying the effect of the error on outcomes was straightforward, where in the study by Pettit et al (2000),[18] risk reductions were documented for patients who were prescribed an NSAID with PPI compared with patients with NSAIDs alone. Observational data then provided relevant, detailed and up-to-date information on death rates associated with hospitalisation for a gastric bleed in the UK.[16]

Previous work by Elliott and colleagues provided relevant disaggregated resource use data for each of the specified health states[17], allowing combination with UK unit costs. Relevant health status valuation was derived for each of the health states.[19] No GI adverse event was taken from UK Euroqol values for United Kingdom age 40-49.[20] For the other states, we used decrements from Spiegel et al[19] starting from the No GI adverse event state. So, utilities for Discomfort are: Utility(No GI adverse event) – (1-utility discomfort Spiegel), assuming a SE of 2% around these numbers and applying a gamma distribution. The same method was used for the other states. The model structure is presented in Figure 2.

9

Figure 2 Markov model for patients with a past medical history of peptic ulcer who have been prescribed a non-selective NSAID and no PPI

2.2 Patients with a history of asthma who have been prescribed a beta-blocker

Following case reports of bronchoconstriction in asthmatics caused by β-blockers, some resulting in death, the Committee on Safety of Medicines issued the following advice:

“…β-blockers, even those with apparent cardioselectivity, should not be used in patients with asthma or a history of obstructive airways disease, unless no alternative treatment is available. In such cases the risk of inducing bronchospasm should be appreciated and appropriate precautions taken.”[21]

Small-scale safety studies confirm that non-cardioselective β-blockers do cause bronchoconstriction, which can be severe in some asthmatics.[22-27] A number of studies have shown that topical timolol eye drops cause bronchoconstriction, and reduce the efficacy of bronchodilator therapy.[28-30] Betaxolol eye drops do not appear to have this effect.[29, 30]

The final search strategy was necessarily broad due to the lack of published data in this area, combining all asthma-related codes with all beta-blocker-related codes. This search produced 524 references. Studies were included if they examined issues on the incidence and/or prevalence of respiratory problems caused by taking β-blockers. Most of the papers assessed short term administration of ß-blockers. Studies with a treatment period of less than a week were excluded. Many other studies lacked information on the specific patient subgroups discussed here (patients with asthma prescribed ß-blockers). After this selection two references remained eligible for this research question. Further handsearching of reference lists, finally provided another four relevant studies. Data from the following studies were included in the final modelling: one clinical trial[31] three observational studies,[32-34] and two economic studies.[35, 36]

Very little clinical or economic data were available to populate this decision model. Therefore the model is quite simple, uses data from a range of sources and oral and ocular exposure to non-selective ß–blockers were combined within the model. The defined population of interest was adult patients with asthma prescribed the selective ß-blocker atenolol, as the representative agent for the group.

10

Five acute exacerbation levels in asthma can be identified from the literature: brittle asthma, moderate asthma exacerbation, acute severe asthma and life threatening asthma and near-fatal asthma.[37] Discussion with PINCER clinical colleagues suggests that these various levels are difficult to distinguish in practice, do not reflect decision-making and are therefore less useful in economic modelling. From a pragmatic point of view, the five levels above were collapsed to three levels for this model: no exacerbation; brittle asthma was combined with moderate exacerbation and life-threatening asthma was combined with near-fatal asthma. The model developed by Price et al[36] and also used by Steuten et al[35] also collapse acute asthma exacerbations into two levels of severity, with a severe exacerbation being treated in secondary care, and a moderate exacerbation being treated in primary care.

Identifying the effect of the error on outcomes was limited to moderate[34] and severe[32] exacerbations. The Price and Steuten model divided asthma further into “well controlled” and “suboptimal control”, the latter indicating mild exacerbations managed by the patient without health service contact. In our model the health state “no symptoms” reflected both these health states, due to the lack of data available regarding the effect of selective or non-selective ß–blockers on inducing these mild exacerbations.

No data on resource use associated with management of people with asthma who are prescribed ß-blockers could be retrieved so the estimated patient-level resource use associated with each health state was obtained from a range of published sources on management of asthma[35, 36, 38] and consultation with PINCER team clinical experts.

Utility weights for all health states in the model were obtained from a model developed by Steuten et al.[35] The utilities were derived from trial data (658 Dutch adults with asthma managed with daily inhaled corticosteroids), were measured using EQ-5D,[20] and standard error, alpha and beta parameters were reported and could be used in our model. The model structure is presented in Figure 3.

Figure 3 Markov model for patients with asthma and a ß-blocker prescription

2.3 Patients aged 75 years and older who have been prescribed an ACEI long-term who have not had a recorded check of their renal function and electrolytes in the previous 15 months

11

ACEIs, by altering glomerular perfusion, may result in a decrease in creatinine clearance, and in an increase in serum creatinine and serum potassium.[39] Hence, the risk of developing renal dysfunction is ongoing and regular monitoring helps to reduce the incidence of adverse events such as acute renal failure (ARF) and hyperkalaemia by adjusting doses or stopping the ACEI. The proportion of ambulatory patients with ACEI therapy who received monitoring in a one-year period was only 67.5%.[40]

A literature search was conducted using the search string ‘(Angiotensin-Converting Enzyme Inhibitors OR ACE Inhibitors OR Enalapril OR Captopril OR Lisinopril OR Perindopril OR Ramipril OR Fosinopril) AND (Hyperkalemia OR Hyperkalaemia OR Kidney Failure, Acute OR Acute Renal Failure) and produced 1218 references. 21 studies examined issues on the incidence and/or prevalence of hyperkalaemia and ARF in patients treated with ACEI. But examination of the full text of the retrieved references found only two references eligible for this research question. Further hand-searching produced another two references.

No models were found representing the events here, so our model was developed through discussion with clinical experts on the PINCER team.

Probabilities were available for developing ARF in the presence of [41] and absence of [42, 43] monitoring and death due to ARF[44], although causes of ARF were not necessarily ACEI-related. However,

Probabilities of developing hyperkalaemia in the present of monitoring were taken from a retrospective analysis of the results of the Studies of Left Ventricular Dysfunction (SOLVD) the incidence of hyperkalaemia in 3364 patients, treated with enalapril, was evaluated.[45] The probability of developing hyperkalaemia from ACEIs when not monitored was not well investigated and after discussion with clinical experts regarding the biological logic of this we had to assume a similar rate as for development of ARF.

No resource use data were available, so clinical opinion was sought to populate the model and reference cost were used (“electrolyte disorders in general” and ARF) to estimate hospital admission costs.

The utility weight for the state of ‘no symptoms’ was obtained from patients with hypertension.[46] No utility weight for “hyperkalaemia” or “electrolyte disorders in general” was found in the literature. As clinical hyperkalaemia requires urgent medical management,[47, 48] the utility weight for ‘hyperkalaemia’ was from a study of patients with coronary heart disease and heart failure admitted to hospital.[49] The post-hyperkalaemia state was obtained from a study of utility loss following cardiovascular events.[50] The utility weight for “ARF” was obtained with renal failure while receiving haemodialysis.[51] Post-ARF utility was obtained from two year follow-up of 703 patients that had received renal replacement therapy for ARF.[52] The model structure is presented in Figure 3.

Figure 4 Markov model for patients aged 75 years and older who have been prescribed an ACEI long-term who have not had a recorded check of their renal function and electrolytes in the previous 15 months

12

2.4 Patients receiving methotrexate for at least three months who have not had a recorded full blood count and/or liver function test within the previous three months

Methotrexate, commonly used in the treatment of psoriasis and rheumatoid arthritis (RA) is associated with adverse incidents and deaths so full blood counts (FBC) and liver function tests (LFT) are recommended at 1-3 monthly intervals.[53] Research suggests inadequate monitoring[54] and lack of effectiveness of interventions to improve monitoring.[55] Little data are available to assess the level of monitoring of long term methotrexate therapy. There are no studies examining the economic impact of monitoring or not monitoring methotrexate.

A literature search was conducted using the search string ‘(Methotrexate) AND (Drug Monitoring OR Blood Cell Count OR Liver Function Tests)’, producing 263 references. Thirteen studies examined the incidence of abnormal liver function tests and abnormal full blood counts in patients treated with methotrexate. Finally, reference lists of the retrieved references of the electronic search were hand-searched. This search produced another two references.

Due to the lack of published models, we have developed our model informed by the literature on methotrexate toxicity,[56-58] bone marrow suppression[59] and liver failure,[60] and ratified with our PINCER clinical experts. In this model we assume that, once clinical manifestations of liver toxicity or bone marrow suppression appear, therapy and monitoring will be installed or methotrexate therapy will be stopped. This is defined as the states ‘no liver toxicity post liver toxicity’ and ‘no bone marrow suppression post bone marrow suppression’.

Probabilities were available for the development of bone marrow suppression and acute liver failure in the presence of[58] and absence of[57, 60] monitoring, although we had to assume these probabilities with disparate sources and patient groups could be applied to one hypothetical population.

13

No resource use data were available, so clinical opinion was sought to populate the model and reference cost were used (“liver function disorders” and “pyrexia of unknown origin” ) to estimate hospital admission costs.

Utilities were available for ‘No symptoms’ from patients with psoriasis taking methotrexate.[61] The utility weight for liver toxicity and post-toxicity was derived from a systematic review of health-state utilities in liver disease.[62] No utility weights were found in the literature for patients experiencing or post- bone marrow suppression taking methotrexate. Estimated utility weights of 0.75 and 0.80 was assigned. The model structure is presented in Figure 5.

Figure 5 Markov model for patients receiving methotrexate for at least three months who have not had a recorded full blood count and/or liver function test within the previous three months

2.5 Patients receiving lithium for at least three months who have not had a recorded check of their lithium levels within the previous three months

Lithium is used in the prophylaxis and treatment of mania, in the prophylaxis of bipolar disorder (manic depressive disorder),[63] and is an effective adjunctive treatment in resistant recurrent depression.[64] Lithium is a drug associated with many adverse effects and 75-90% patients treated with lithium will show signs or symptoms of toxicity during their treatment.[65] Long term use of lithium therapy has been associated with thyroid disorders, renal impairment and mild cognitive and memory impairment.[63] Poor patient adherence (18-52%) has been reported due to side effects and perceived lack of efficacy against depressive episodes.[66] Sub-therapeutic levels, and associated increased risk of relapse, are usually associated with poor adherence.[67-69] Lithium has a small therapeutic/toxic ratio[21] and doses are adjusted to achieve serum lithium concentration of 0.6-1.2 mmol/litre. Over-dosage, usually with a serum-lithium concentration of over 1.5 mmol/litre, may be fatal and toxic effects include tremor, ataxia, dysarthria, nystagmus, renal impairment and convulsions. Serum lithium concentrations above 2 mmol/litre

14

require emergency poisoning treatment. To keep serum lithium concentration levels within a therapeutic range, therapeutic drug monitoring (TDM) should be carried out every three months.[21]

The key words ‘bipolar’ and ‘lithium’, ‘TDM’ OR ‘therapeutic drug monitoring’ OR ‘monitoring’ OR ‘therapy’ OR ‘therapeutic’ OR ‘manic’ OR ‘depressive’ OR ‘depressed’ OR ‘depression’ OR ‘suicide’ OR ‘suicide rates’ were used. During the literature search two systematic reviews [70, 71] of decision analytic models based on bipolar disorder, and a bipolar disorder model designed by NICE [72], looked at the effect lithium (among other medications) has on reducing relapse rates as well as suicide in this population group. NICE guideline number 38 [72]introduces a model for the medical and pharmacological management of bipolar disorder. No published model was found that examined the effect of medication monitoring on the disease. Therefore we had to develop a new model in discussion with a neurologist (William Whitehouse, University of Nottingham).

Lithium has two main benefits when used in patients with bipolar disorder: reducing the probability of relapse into a manic or depressive state and reducing the probability of suicide. In our model, we assume that patients in the therapeutic and supra-therapeutic range have the same relapse and suicide incidence, in the absence of evidence to the contrary. Patients who are sub-therapeutic do not realise these benefits of lithium. Previously published lithium models do not take into account the effects of lithium intoxication, [70, 71] possibly because the literature around lithium intoxication is complex and sometimes contradictory. Adverse effects occur at therapeutic and sub-therapeutic levels, and relapse or non-response can occur when the drug is within therapeutic range.[67] Waring et al. (2007) describes three patterns of lithium toxicity:[73] Acute intoxication: in patients not receiving lithium previously; Acute-on-therapeutic intoxication: acute ingestion whilst on lithium therapy; Chronic intoxication: arising insidiously over time due to lithium accumulation. Waring and other authors suggest there is a limited link between serum levels of lithium and toxicity severity.[67, 73, 74] This means that the adverse effects of being supra-therapeutic are not clearly different from those in therapeutic or sub-therapeutic levels as chronic toxicity occurs due to long term accumulation, so can present at any serum lithium concentration.[67] The use of lithium TDM may reduce the chance of this occurring over the long term, but this is not clear. Monitoring renal and thyroid function may be more useful. Acute intoxication (as in an overdose) would present clinically, so TDM has limited value here. As there were no clear primary data to populate the model for lithium acute or chronic toxicity, it was excluded from the model on the assumption that it would probably occur equally between monitored and un-monitored patients. Scott and Pope[75] and Schumann et al[76] report low adherence levels of 53% and 59.2% respectively in general samples of people taking lithium. Poor adherence to lithium is attributed to lack of acceptance of prophylaxis in general, the effectiveness of lithium and the severity of illness.[76] Published evidence suggests that regular monitoring of lithium levels encourages increased patient adherence to the medication. [66]

In our model, a pooled mean adherence of 56.1% was used for patients that were not receiving TDM, from Scott and Pope[75] and Schumann et al [76]. Rosa et al [66] reported that attendance at a regular outpatient mood disorder clinic resulted in 85.6% adherence to lithium treatment leading to therapeutic lithium levels. For this model we assumed the Rosa et al[66] sample represented patients that receive TDM every 3 months. This means that patients being appropriately monitored will be more likely to be in the therapeutic range. Patients without monitoring are less likely to be in the therapeutic range and are more likely to present with relapse (or suicide).

15

Rosa et al[66] reports 14.4% non-adherence in regularly monitored patients, and the pooled results report 43.9% non-adherence in non-monitored patients. If we hold the assumption that patients that are sub-therapeutic are so because of non-adherence (compared to therapeutic and supra-therapeutic patients being adherent to their medication) then appropriate lithium monitoring (i.e. TDM every 3 months) decreases non-adherence by 32.8% [14.4/43.9]; increasing patients within a therapeutic range by 28.2% (from 70.23% within therapeutic range in the non-monitored group to 90.04% in the monitored group).

It is assumed a person will not move from sub-therapeutic to supra/therapeutic until an adverse event occurs and they begin to be monitored again, i.e. a person will only adhere to their medication, after making the decision to be non-adherent, if they are monitored or an adverse event occurs to change their mind.

Costs of monitoring and treatment were derived from NICE Guideline for bipolar disorder.[72, 77]

Utilities were taken from a study of bipolar patients receiving lithium therapy.[78] It is assumed in the model that patients that do not receive lithium therapy have a better health state than those patients receiving lithium therapy when in a stable state, due to lack of adverse effects relating to lithium. As no published utility data were available for bipolar patients in a depressive relapse, data were used from patients with major depression.[79] The model structure is presented in Figure 6.

Figure 6 Markov model for patients receiving lithium for at least three months who have not had a recorded check of their lithium levels within the previous three months

2.6 Patients receiving amiodarone for at least six months who have not had a thyroid function test within the previous six months

Amiodarone is highly effective in suppressing recurrent atrial fibrillation.[80, 81] However, it is associated with serious thyroid dysfunction. [81, 82] A recently developed less toxic agent, dronedarone, has not replaced amiodarone, as was originally hoped, due to its lesser efficacy in cardioversion.[83] Amiodarone usage can result in amiodarone-induced thyrotoxicosis (AIT), and amiodarone-induced-hypothyroidism (AIH).[82, 84-87] AIT and AIH can cause significant patient morbidity [87, 88]. The clinical presentation of AIH is more subtle than that of AIT, which can be more dramatic with life-threatening cardiac instability. AIH can result in a slower heart rate, weight gain, shivers, and emotional instability.[89] Diagnosis and management

16

Relapse: manic Relapse: depressed

Stable: sub-therapeutic Death/suicide

Stable: supra-therapeutic or

therapeutic

of AIT in particular is quite complex, with wide variation in practice, including variation in opinions about whether patients should remain on amiodarone in the presence of AIT.[90]

The following search was used with the key words ‘amiodarone’ AND ‘induced’ AND ‘thyroid’ were used which resulted in 373 articles, of which 25 were used to inform model design. No published model was found that examined the effect of medication monitoring on the disease and there are no UK guidelines for treatment of AIT and thus, there is some lack of consensus around approach to management.[91]. Therefore we had to develop a new model in discussion with an international expert on amiodarone (Jayne Franklin, University of Birmingham). The UK and the USA have a high dietary intake of iodine, such that AIT is less common than AIH.[89, 92] Therefore UK/US-based evidence was used wherever possible. Users of amiodarone have well documented side effects other that of the thyroid, such as lung, central nervous system, and skin-related side effects. These other side effects are considered equal in both the intervention and the treatment arm.

No studies were found that reported the effects of monitoring thyroid function on patient outcomes. In this model, AIT and AIH are assumed to occur whether monitoring takes place or not. If patients are monitored, it is assumed that they will have a lower probability of staying in the state ‘Untreated AIH’ or ‘Untreated AIT’, with the associated increased risk of morbidity and mortality. If a patient is being monitored regularly, AIT will be picked up and treated within one cycle. The probability of surgical management via thyroidectomy in AIT is 0.081.[90] There were no studies reporting probability of surgical management of AIT if patients are un-monitored. It was assumed that the probability will be higher than zero as patients may be picked up by chance, at a rate of 10% of the rate in the monitored group.

Resource use was obtained from studies reporting management of AF, AIT and AIH.[89, 93, 94]

Health status valuations were difficult to obtain for users of amiodarone with thyroid complications. Patient-level data from a UK population with atrial fibrillation were used to provide a baseline utility for people with no symptoms (EQ-5D: 0.78 (SD: 0.21).[95] Utility decrements for thyroid toxic events was based on the Quality of Wellbeing scale from Nolan et al[96] which was the only paper that measured the quality of life on a scale of 0 to 1 post-treatment for AIT and AIH. For utility decrement following a thyroidectomy, a clinical expert value was taken from Esnaola et al.[97] The model structure is presented in Figure 7.

Figure 7 Markov model for patients receiving amiodarone for at least six months who have not had a thyroid function test within the previous six months

17

References1. Avery A, Rodgers S, Cantrill J, Armstrong S, Elliott R, Howard R, et al. Protocol for the PINCER trial: a cluster

randomised trial comparing the effectiveness of a pharmacist-led IT-based intervention with simple feedback in reducing rates of clinically important errors in medicines management in general practices. Trials 2009,10:28.

2. Avery AJ, Rodgers S, Cantrill JA, Armstrong S, Cresswell K, Eden M, et al. A pharmacist-led information technology intervention for medication errors (PINCER): a multicentre, cluster randomised, controlled trial and cost-effectiveness analysis. The Lancet 2012,379:1310-1319.

3. Hollis S, Campbell F. What is meant by intention to treat analysis? Survey of published randomised controlled trials. BMJ 1999,319:670-674.

4. Shih WJ. Problems in dealing with missing data and informative censoring in clinical trials. Curr Control Trials Cardiovasc Med 2002,3:4.

5. Department for Communities and Local Government. Indices of Deprivation 2004 In; 2004.6. StataCorp LP. Stata data analysis and statistical Software. Special Edition Release 10.1 edition 2008.7. Elashoff JD. nQuery Advisor®. Version 6.0 Software. In. Los Angeles: CA; 2005.8. Kerry SM, Bland JM. Statistics notes: The intracluster correlation coefficient in cluster randomisation. BMJ

1998,316:1455-1460.9. Hilbe JM. Negative binomial regression. New York: Cambridge University Press; 2007.10. Briggs A, Klaxton K, Sculpher M. Decision Modelling For Health Economic Evaluation: Oxford University

Press; 2006.11. Assmann SF, Pocock SJ, Enos LE, Kasten LE. Subgroup analysis and other (mis)uses of baseline data in clinical

trials. Lancet 2000,355:1064-1069.12. Elliott RA, Putman K, Franklin M, Verhaeghe N, Annemans L, Eden M, et al. Economic evaluation of a

pharmacist-led IT-based intervention with simple feedback in reducing rates of clinically important errors in medicines management in general practices (PINCER). In. Edited by Department of Health Patient Safety Research Portfolio. Birmingham, UK; 2013.

13. Hernandez-Diaz S, Rodriguez L. Association between nonsteroidal anti-inflammatory drugs and upper gastrointestinal tract bleeding/perforation: an overview of epidemiologic studies published in the 1990s. Archives of Internal Medicine 2000,160:2093-2099.

14. Hawkey CJ, Langman MJ. Non-steroidal anti-inflammatory drugs: overall risks and management. Complementary roles for COX-2 inhibitors and proton pump inhibitors. [Review] [113 refs]. Gut 2003,52:600-608.

18

AIT surgicalmanagement

Death

AIH medicalmanagment

No symptoms

AIT medicalmanagementAIT untreated Post treated

AIT

AIH untreated

15. National Institute for Health and Clinical Excellence. Guidance on the use of cyclo-oxygenase (Cox) II selective inhibitors, celecoxib, rofecoxib, meloxicam and

etodolac for osteoarthritis and rheumatoid arthritis. In: Technology Appraisal 27. London; 2001.16. Brown TJ, Hooper L, Elliott RA, Payne K, Webb R, Roberts C, et al. A comparison of the cost-effectiveness of

five strategies for the prevention of non-steroidal anti-inflammatory drug-induced gastrointestinal toxicity: A systematic review with economic modelling. Health Technology Assessment 2006,10:iii-183.

17. Elliott RA, Hooper L, Payne K, Brown TJ, Roberts C, Symmons D. Preventing non-steroidal anti-inflammatory drug-induced gastrointestinal toxicity: are older strategies more cost-effective in the general population? Rheumatology 2006,45:606-613.

18. Pettitt D, Goldstein JL, McGuire A, Schwartz JS, Burke T, Maniadakis N. Overview of the arthritis cost consequence evaluation system (ACCES): A pharmacoeconomic model for celecoxib. Rheumatology 2000,39:33-42.

19. Spiegel BMR, Chiou CF, Ofman JJ. Minimizing complications from nonsteroidal antiinflammatory drugs: cost effectiveness of competing strategies in varying risk groups. Arthritis and Rheumatism 2005,53:185-197.

20. Euroqol Group. Measuring Self-Reported Population Health-An International Perspective based on EQ-5D. In; 2008.

21. British Medical Association, Royal Pharmaceutical Society of Great Britain. British National Formulary. 56 ed. London: BMJ Group and RPS Publishing; 2011.

22. Kotlyar E, Keogh AM, Macdonald PS, Arnold RH, McCaffrey DJ, Glanville AR. Tolerability of carvedilol in patients with heart failure and concomitant chronic obstructive pulmonary disease or asthma. Journal of Heart and Lung Transplantation 2002,21:1290-1295.

23. Sue DY, Hansen JE, Wasserman K. Beta-adrenergic blockade with pindolol (LB-46) in mild to moderate asthma. Chest 1981,80:537-542.

24. Addis GJ, Thorp JM. Effects of oxprenolol on the airways of normal and bronchitic subjects. European Journal of Clinical Pharmacology 1976,9:259-263.

25. Giulekas D, Georgopoulos D, Papakosta D, Antoniadou H, Sotiropoulou E, Vamvalis C. Influence of pindolol on asthmatics and effect of bronchodilators. Respiration 1986,50:158-166.

26. Nordstrom LA, MacDonald F, Gobel FL. Effect of propranolol on respiratory function and exercise tolerance in patients with chronic obstructive lung disease. Chest 1975,67:287-292.

27. Devereux G, Fishwick K, Aiken TC, Bourke SJ, Hendrick DJ. Adverse effects of a single dose of (+)-sotalol in patients with mild stable asthma. British Journal of Clinical Pharmacology 1998,46:79-82.

28. Bohm E, Fabel H. [Changes in lung function following administration of eyedrops containing timolol, metipranolol, pindolol and pilocarpine in healthy probands and patients with mild bronchial asthma]. [German]. Klinische Wochenschrift 1987,65:920-924.

29. Schoene RB, Abuan T, Ward RL, Beasley CH. Effects of topical betaxolol, timolol, and placebo on pulmonary function in asthmatic bronchitis. American Journal of Ophthalmology 1984,97:86-92.

30. Dunn TL, Gerber MJ, Shen AS, Fernandez E, Iseman MD, Cherniack RM. The effect of topical ophthalmic instillation of timolol and betaxolol on lung function in asthmatic subjects. American Review of Respiratory Disease 1986,133:264-268.

31. Cowper PA, DeLong ER, Whellan DJ, Allen LaPointe NM, Califf RM. Economic effects of beta-blocker therapy in patients with heart failure. American Journal of Medicine 2004,116:104-111.

32. Brooks TW, Creekmore FM, Young DC, Asche CV, Oberg B, Samuelson WM. Rates of hospitalizations and emergency department visits in patients with asthma and chronic obstructive pulmonary disease taking beta-blockers. Pharmacotherapy 2007,27:684-690.

33. Chen J, Radford MJ, Wang Y, Marciniak TA, Krumholz HM. Effectiveness of beta-blocker therapy after acute myocardial infarction in elderly patients with chronic obstructive pulmonary disease or asthma. Journal of the American College of Cardiology 2001,37:1950-1956.

34. Chen YF, Avery AJ, Neil KE, Johson C, Dewey ME, Stockley IH. Incidence and possible causes of prescribing potentially hazardous / contraindicated drug combinations in general practice. Drug Safety 2005,28:67-80.

35. Steuten L, Palmer S, Vrijhoef B, van Merode F, Spreeuwenberg C, Severens H. Cost-utility of a disease management program for patients with asthma. International Journal of Technology Assessment in Health Care 2007,23:184-191.

36. Price MJ, Briggs AH. Development of an economic model to assess the cost effectiveness of asthma management strategies. Pharmacoeconomics 2002,20:183-194.

37. British Thoracic Society. 2008 British Guideline on the Management of Asthma - updated June 2009. In; 2009.38. British Thoracic Society S. 2008 British Guideline on the Management of Asthma. Thorax 2008,63:iv1-iv121.39. Rastegar A, Soleimani M. Hypokalaemia and hyperkalaemia. Postgrad.Med.J. 2001,77:759-764.

19

40. Amir O, Hassan Y, Sarriff A, Awaisu A, Abd AN, Ismail O. Incidence of risk factors for developing hyperkalemia when using ACE inhibitors in cardiovascular diseases. Pharm.World Sci. 2009,31:387-393.

41. Baraldi A, Ballestri M, Rapana R, Lucchi L, Borella P, Leonelli M, et al. Acute renal failure of medical type in an elderly population. Nephrol.Dial.Transplant. 1998,13 Suppl 7:25-29.

42. Mittalhenkle A, Stehman-Breen CO, Shlipak MG, Fried LF, Katz R, Young BA, et al. Cardiovascular risk factors and incident acute renal failure in older adults: the cardiovascular health study. Clin.J.Am.Soc.Nephrol. 2008,3:450-456.

43. Knight EL, Glynn RJ, McIntyre KM, Mogun H, Avorn J. Predictors of decreased renal function in patients with heart failure during angiotensin-converting enzyme inhibitor therapy: results from the studies of left ventricular dysfunction (SOLVD). Am.Heart J. 1999,138:849-855.

44. Wynckel A, Ebikili B, Melin JP, Randoux C, Lavaud S, Chanard J. Long-term follow-up of acute renal failure caused by angiotensin converting enzyme inhibitors. Am.J.Hypertens. 1998,11:1080-1086.

45. de Denus S, Tardif JC, White M, Bourassa MG, Racine N, Levesque S, et al. Quantification of the risk and predictors of hyperkalemia in patients with left ventricular dysfunction: a retrospective analysis of the Studies of Left Ventricular Dysfunction (SOLVD) trials. Am.Heart J. 2006,152:705-712.

46. Burstrom K, Johannesson M, Diderichsen F. Swedish population health-related quality of life results using the EQ-5D. Qual.Life Res. 2001,10:621-635.

47. Alfonzo AV, Isles C, Geddes C, Deighan C. Potassium disorders--clinical spectrum and emergency management. Resuscitation 2006,70:10-25.

48. Kim HJ, Han SW. Therapeutic approach to hyperkalemia. Nephron 2002,92 Suppl 1:33-40.49. Spiraki C, Kaitelidou D, Papakonstantinou V, Prezerakos P, Maniadakis N. Health-related quality of life

measurement in patients admitted with coronary heart disease and heart failure to a cardiology department of a secondary urban hospital in Greece. Hellenic.J.Cardiol. 2008,49:241-247.

50. Lindgren P, Kahan T, Poulter N, Buxton M, Svarvar P, Dahlof B, et al. Utility loss and indirect costs following cardiovascular events in hypertensive patients: the ASCOT health economic substudy. Eur.J.Health Econ. 2007,8:25-30.

51. Lee AJ, Morgan CL, Conway P, Currie CJ. Characterisation and comparison of health-related quality of life for patients with renal failure. Curr.Med.Res.Opin. 2005,21:1777-1783.

52. Ahlstrom A, Tallgren M, Peltonen S, Rasanen P, Pettila V. Survival and quality of life of patients requiring acute renal replacement therapy. Intensive Care Med. 2005,31:1222-1228.

53. The National Patient Safety Agency. Towards the safer use of oral methotrexate. In. Edited by NHS National Patient Safety Agency. London; 2004.

54. Collin B, Srinathan SK, Finch TM. Methotrexate: prescribing and monitoring practices among the consultant membership of the British Association of Dermatologists. British Journal of Dermatology 2008,158:793-800.

55. Raebel MA, Lyons EE, Chester EA, Bodily MA, Kelleher JA, Long CL, et al. Improving laboratory monitoring at initiation of drug therapy in ambulatory care: a randomized trial. Archives of Internal Medicine 2005,165:2395-2401.

56. Choi HK, Hernan MA, Seeger JD, Robins JM, Wolfe F. Methotrexate and mortality in patients with rheumatoid arthritis: a prospective study. Lancet 2002,359:1173-1177.

57. Bologna C, Viu P, Picot MC, Jorgensen C, Sany J. Long-term follow-up of 453 rheumatoid arthritis patients treated with methotrexate: an open, retrospective, observational study. British Journal of Rheumatology 1997,36:535-540.

58. Haustein UF, Rytter M. Methotrexate in psoriasis: 26 years' experience with low-dose long-term treatment. Journal of the European Academy of Dermatology & Venereology 2000,14:382-388.

59. Lim AY, Gaffney K, Scott DG. Methotrexate-induced pancytopenia: serious and under-reported? Our experience of 25 cases in 5 years. Rheumatology 1051,44:1051-1055.

60. Malatjalian DA, Ross JB, Williams CN, Colwell SJ, Eastwood BJ. Methotrexate hepatotoxicity in psoriatics: report of 104 patients from Nova Scotia, with analysis of risks from obesity, diabetes and alcohol consumption during long term follow-up. Canadian Journal of Gastroenterology 1996,10:369-375.

61. Revicki D, Willian MK, Saurat JH, Papp KA, Ortonne JP, Sexton C, et al. Impact of adalimumab treatment on health-related quality of life and other patient-reported outcomes: results from a 16-week randomized controlled trial in patients with moderate to severe plaque psoriasis. British Journal of Dermatology 2008,158:549-557.

62. McLernon DJ, Dillon J, Donnan PT. Health-state utilities in liver disease: a systematic review. [Review] [53 refs]. Medical Decision Making 2008,28:582-592.

63. Young AH, Newham JI. Lithium in maintenance therapy for bipolar disorder. Journal of Psychopharmacology 2006,20:17-22.

64. Bauer M, Whybrow PC, Angst J, Versiani M, Moller HJ, World Federation of Societies of Biological Psychiatry Task Force on Treatment Guidelines for Unipolar Depressive D. World Federation of Societies of Biological Psychiatry (WFSBP) Guidelines for Biological Treatment of Unipolar Depressive Disorders, Part 2:

20

Maintenance treatment of major depressive disorder and treatment of chronic depressive disorders and subthreshold depressions.[see comment]. [Review] [160 refs]. World Journal of Biological Psychiatry.3(2):69-86 2002.

65. Groleau G. Lithium toxicity. Emergency Medicine Clinics of North America 1994,12:511-531.66. Rosa AR, Marco M, Fachel JMG, Kapczinski F, Stein AT, Barros HMT. Correlation between drug treatment

adherence and lithium treatment attitudes and knowledge by bipolar patients. Progress in Neuro-Psychopharmacology and Biological Psychiatry 2007,31:217-224.

67. Sharma S, Joshi S, Chadda RK. Therapeutic Drug Monitoring of Lithium in Patients with Bipolar Affective Disorder: experiences from a tertiary care hospital in India. American Journal of Therapeutics 2009,16:393-397.

68. Goodwin GM. Recurrence of mania after lithium withdrawal. British Journal of Psychiatry 1994,164:149-152.69. Tondo L, Baldessarini RJ. Long-term lithium treatment in the prevention of suicidal behavior in bipolar disorder

patients. Epidemiologia e psichiatria sociale 2009,18:179-183.70. Bridle C, Palmer S, Bagnall A, Darba J, Duffy S, Sculpher M, et al. A rapid and systematic review and

economic evaluation of the clinical and cost-effectiveness of newer drugs for treatment of mania associated with bipolar affective disorder. Health technology assessment (Winchester, England) 2004,8:iii.

71. Soares-Weiser K, Bravo Vergel Y, Beynon S, Dunn G, Barbieri M, Duffy S , et al. A systematic review and economic model of the clinical effectiveness and cost-effectiveness of interventions for preventing relapse in people with bipolar disorder. Health Technol Assess 2007,11:1-226.

72. NICE. The management of bipolar disorder in adults, children and adolescents, in primary and secondary care: national clinical practice guideline number 38: The British Psychological Society & The Royal College of Psychiatrists; 2006.

73. Waring WS, Laing WJ, Good AM, Bateman DN. Pattern of lithium exposure predicts poisoning severity: Evaluation of referrals to a regional poisons unit. Quarterly Journal of Medicine 2007,100:271-276.

74. Chen KP, Shen WW, Lu ML. Implication of serum concentration monitoring in patients with lithium intoxication. Psychiatry and Clinical Neurosciences 2004,58:25-29.

75. Scott J, Pope M. Nonadherence with mood stabilizers: Prevalence and predictors. Journal of Clinical Psychiatry 2002.

76. Schumann C, Lenz G, Berghöfer A, Müller-Oerlinghausen B. Non-adherence with long-term prophylaxis: a 6-year naturalistic follow-up study of affectively ill patients. Psychiatry research 1999,89:247-257.

77. British Medical Association, Royal Pharmaceutical Society of Great Britain. British National Formulary. 56 ed. London: BMJ Group and RPS Publishing; 2008.

78. Revicki DA, Hanlon J, Martin S, Gyulai L, Nassir Ghaemi S, Lynch F, et al. Patient-based utilities for bipolar disorder-related health states. Journal of affective disorders 2005,87:203-210.

79. Revicki DA, Wood M. Patient-assigned health state utilities for depression-related outcomes: differences by depression severity and antidepressant medications. Journal of affective disorders 1998,48:25-36.

80. Robinson G, Antoniou S. Dronedarone. Clinical Pharmacist 2010,2:334-336.81. Thorne SA, Barnes I, Cullinan P, Somerville J. Amiodarone-Associated Thyroid Dysfunction : Risk Factors in

Adults With Congenital Heart Disease. Circulation 1999,100:149-154.82. Hofmann A, Nawara C, Ofluoglu S, Holzmannhofer J, Strohmer B, Pirich C. Incidence and predictability of

amiodarone-induced thyrotoxicosis and hypothyroidism. Wiener Klinische Wochenschrift 2008,120:493-498.83. Le Heuzey J-Y, De Ferrari GM, Radzik D, Santini M, Zhu J, Davy J-M. A short-term, randomized, double-

blind, parallel-group study to evaluate the efficacy and safety of dronedarone versus amiodarone in patients with persistent atrial fibrillation: the DIONYSOS study. Journal of Cardiovascular Electrophysiology 2010,21:597-605.

84. Martino E, Safran M, Aghini-Lombardi F, Rajatanavin R, Lenziardi M, Fay M , et al. Environmental Iodine Intake and Thyroid Dysfunction During Chronic Amiodarone Therapy. Ann Intern Med 1984,101:28-34.

85. Loh K-C. Amiodarone-induced thyroid disorders: a clinical review. Postgrad Medical Journal 76 2000:133-140.

86. Daniels G. Amiodarone-Induced Thyrotoxicosis. The Journal of clinical Endocrinology & Metabolism vol. 86 no.1 2001:3-8.

87. Osman F, Franklyn JA, Sheppard MC, Gammage MD. Successful Treatment of Amiodarone-Induced Thyrotoxicosis. Circulation 105 2002:1275-1277.

88. Klein I, Ojamaa K. Mechanisms of Disease: Thyroid Hormone and the Cardiovascular System. New England Journal of Medicine 2001,344:501-509.

89. Cardenas G, Cabral J, Leslie C. Amiodarone-induced thyrotoxicosis: Diagnosis and Therapeutic Strategies. Cleveland Clinical Journal of Medicine 2003,70:624-631.

90. Bartalena L, Wiersinga WM, Tanda ML, Bogazzi F, Piantanida E, Lai A , et al. Diagnosis and management of amiodarone-induced thyrotoxicosis in Europe: results of an international survey among members of the European Thyroid Association. Clinical endocrinology 2004,61:494-502.

21

91. Tanda ML, Piantanida E, Lai A, Liparulo L, Sassi L, Bogazzi F , et al. Diagnosis and management of amiodarone-induced thyrotoxicosis: similarities and differences between North American and European thyroidologists. Clinical endocrinology 2008,69:812-818.

92. Lee SM, Lewis J, Buss DH, Holcombe GD, Lawrance PR. Iodine in British foods and diets. British Journal of Nutrition 1994,72:435-446.

93. Beastall GH, Beckett GJ, Franklyn J, Fraser WD, Hickey J, John R, et al. UK Guidelines for the Use of Thyroid Function Tests. In. London: The Association for Clinical Biochemistry, British Thyroid Association, British Thyroid Foundation; 2006.

94. Bogazzi F, Bartalena L, Dell'Unto E, Tomisti L, Rossi G, Pepe P, et al. Proportion of type 1 and type 2 amiodarone-induced thyrotoxicosis has changed over a 27-year period in Italy. Clinical endocrinology 2007,67:533-537.

95. Buxton M, Caine N, Chase D, Connelly D, Grace A, Jackson C, et al. A review of the evidence on the effects and costs of implantable cardioverter defibrillator therapy in different patient groups, and modelling of cost-effectiveness and cost-utility for these groups in a UK context. NCCHTA Health Technol Assess 2006,10:1 - 132.

96. Nolan JP, Tarsa NJ, DiBenedetto G. Case-finding for unsuspected thyroid disease: costs and health benefits. American journal of clinical pathology 1985,83:346.

97. Esnaola NF, Cantor SB, Sherman SI, Lee JE, Evans DB. Optimal treatment strategy in patients with papillary thyroid cancer: A decision analysis* 1. Surgery 2001,130:921-930.

22