general methods of investigation
DESCRIPTION
General Methods of Investigation. 1.Chance observations 2.Case histories individual cases case series 3.Uncontrolled trials of an intervention 4.Cross-sectional (naturalistic) studies 5. Case-control studies 6.Prospective follow-up studies 7.Randomized clinical trial. No planned - PowerPoint PPT PresentationTRANSCRIPT
General Methods of Investigation
1. Chance observations
2. Case histories– individual cases– case series
3. Uncontrolled trials of an intervention
4. Cross-sectional (naturalistic) studies
5. Case-control studies
6. Prospective follow-up studies
7. Randomized clinical trial
No plannedconcurrentcomparisongroup
Examples of Case-Control Studies
1. Multi-Center study of SIDS (Hoffman H, Ann NY Acad Sci, 1988)
2. Influenza vaccine effectiveness (Treanor JJ et al, CID, 2012)
3. Coffee drinking and CHD
4. Soluble biomarkers and all-cause mortality (nested case-control study)
Example
Treanor JJ et al. Effectiveness of seasonal influenza vaccines in the United States during a season with circulation of all three vaccine strains. CID 55:951-959, 2012.
Purpose: To assess vaccine effectiveness during the 2010-2011 season.
Design: Case-control study
Cases: Persons in 5 communities (4 states) seeking care for an acute respiratory illness (ARI) with positive respiratory specimens.
Controls:Persons in same communities with ARI with negative respiratory specimen
Factor: At least 1 dose of seasonal influenza vaccine at least 14 days before symptom onset.
Vaccine
Test Pos.Cases
Test Neg.Controls
317 1958 2275
Vaccine use among cases = 317 / 1028 = 31%Vaccine use among controls = 1958 / 3684 = 53%
Odds ratio (OR) = (317x1726)/(1958x711)= 0.39 (95% CI: 0.34-0.45)Adj VE = 60% (95% CI: 54-66%); adj. for site, demographics, Insurance, and high risk conditions.
No Vaccine 711 1726 2437
1028 3684 4,712
Effectiveness of Seasonal Influenza Vaccine
Example
Hennekens et al. Coffee drinking and death due to coronary heart disease. NEJM 294:633-36, 1976.
Purpose: To investigate the relation between coffee drinking and death due to CHD.
Design: Individually matched case-control study
Cases: Married, white men, aged 30-70 who died from CHD within 24 hours of symptom onset according to death certificate
Controls:Age, sex, neighborhood matched
Agent: Coffee consumption as assessed by interview with wife 2-8 weeks after death
Consumption 3 months prior to death or interview
Case-Control Study Hennekens et al. Coffee drinking and death due to coronary
heart disease. NEJM, 1976.
1+ cups/day
CHDCases
NeighborhoodControls
500 485 985
Prevalence of coffee drinking among cases = 500 / 649 = 77%
Prevalence of coffee drinking among controls = 485 / 649 = 75%
None 149 164 313
649 649 1,298
Matched Analysisfor Coffee Study
1+ cups/day
1+ cups/day None
359 126 485
Odds Ratio (OR) = = 1.12
None 141 23 164
500 149 649
Cases
141126
^
Controls
Other Considerations in Interpreting Findings from Observational Studies
Bias (def.)
A systematic error usually introduced by investigator and/or patient which leads to incorrect estimates of the association between a risk factor and a disease endpoint.
– Case and control selection and recall bias are common problems in case-control studies
Possible Sources of Bias in Vaccine and Coffee Case-Control Studies
1. Identification of cases and controls
2. Interviewer
3. Vaccine receipt, and wife’s report or memory of spouse’s coffee consumption
Nested Case-Control Study: SMART Study
Baseline plasma samples were identified for patients who died (85 patients) and for two matched controls for each death (170 patients). Matching was on country, age (+/- 5years), gender and approximate date of randomization (+/- 3 months).
Conditional logistic for matched studies used to estimate odds ratios (OR) for mortality with participants in lowest quartile as reference.
Adjusted OR consider covariates corresponding to age, race, ART, HIV RNA, CD4+ count, BMI, and total/HDL cholesterol at baseline, smoking, diabetes, hep B/C co-infection, use of lipid and BP lowering medication
PLoS Medicine 2008; 5(10) e203
Nested Case Control Design
Time Axis0
Two matched controls for every case were chosen. Follow-up for all members of the cohort (horizontal white lines) begins at randomization (zero-time axis).
Biomarker and All-Cause Mortality Associations
Baseline Level OR (4th/1st QRT)Univariate P-value
D-dimer 12.4 <0.0001
IL-6 8.3 <0.0001
hsCRP 2.0 0.05
Cohort Study Example: Framingham Heart Study
Goal: 6,000 men and women aged 30-59 estimated to yield 2,000 new cases by the end of the 20th century
Selection of Sample• Annual publication by town of Framingham• Stratified by family size and location of
residence• Sample unit - family (cluster)• Systematic sampling within stratum
Result
• Acceptance rate = 69%• Eventual “starting” sample
4469 respondents+ 740 volunteers
5209
Potential for bias?• Prevalence data• Association of risk factors with disease incidence
Another Cohort Study Example
Shekelle et al. MRFIT behavior pattern study: Type A behavior and incidence of coronary heart disease. Am J Epid 122:559, 1985.
Question: Is type A behavior associated with an increased incidence of CHD.
Design: Prospective follow-up study – cohort study within a randomized clinical trial
Risk factor:Behavior pattern assessed by interview (4 point scale)
– Each interview taped and reviewed– Quality assurance (J Chronic Dis 1978; 32:293-305)
Endpoint:CHD death or non-fatal MI in 7 years
– Mortality review committee– Blinding assessments
Study subjects:MRFIT men (aged 35-37 with risk factors for CVD)
Overview of MRFIT Behavior Pattern Study
Prospective Observational Study
Type A
CHDEvent
No CHDEvent
94 2,220 2,314
Odds Ratio (OR) = = 0.92
Type B 35 761 769
129 2,981 3,110
94 (761)35 (2,220)
^
Shekelle RB et al, Amer J Epid 1985; 122: 559-570.
Prospective Observational Study
No.CHD Events
No. Person Years ofFollow-up
94 15,973
Probability of CHD in 7 years among Type A participants = (94 / 15,973) x 1,000 = 5.9
For Type B = (35 / 5,514) x 1,000 = 6.3
35 5,514
129 21,487
Type A
Type B
RR = = 0.94;5.96.3
^ Adjusted RR = 0.87 (95% CI: 0.59-1.28)^
Conclusion
“Type A behavior was not associated with
CHD in MRFIT … further study is needed.”
The MRFIT BehaviorPattern Study
“Employed ad-hoc, poorly chosen, inadequately trained, non-professional clerks to administer the structured interview?”
“The widely disseminated negative findings of the MRFIT study have delayed the introduction of a procedure that could have prolonged the lives of hundreds of thousands of patients with CHD.”
Source: Friedman M. Am Heart J, 115:950-35, 1988.
Cohort Study Considerations
• Representativeness of sample
• Confounding
• Bias due to incomplete endpoint ascertainment (e.g., differential lost-to-follow-up rates) – missing data
• Protocol for risk factor (predictor) assessment
• Number of participants with outcome of interest (sampling variability)
Randomized Clinical Trial
• A prospective study in which the investigators determine who receives the intervention or agent and who does not by a random process
• A carefully planned manipulation of a “natural” state
• “A prospective study comparing the effect and value of intervention(s) against a control in human beings”
Some Resources for Finding Trials
• Registered trials– www.clinicaltrials.gov
• Trial registry developed in England– www.controlled-trials.com
• Published trials– www.ncbi.nlm.nih.gov/sites/entrez?db=pubmed\
Essential Components of a Randomized Clinical Trial
1. Clearly stated question/hypothesis
2. Statistical design (sample size, power) to address hypothesis
3. Definition of target population
4. A control group
5. Random method of treatment assignment after informed consent
6. Excellent follow-up and unbiased endpoint ascertainment
7. Monitoring plan
Aspirin Myocardial Infarction Study (AMIS)
• Randomized double-blind placebo-controlled study
Purpose: To determine whether regular use of aspirin results in a reduction in 3-year mortality among patients with at least one documented MI
(secondary prevention trial)
Risk Indicators According to Aspirin Use in Women Aged 34-59 Years
0 1 - 3 4 - 6 7 - 14 ≥ 15
No. 52,630 15,540 7,518 7,352 4,998
Age (years) 45.9 45.7 45.4 46.8 47.8
Hypertension (%) 15.0 13.2 16.3 17.9 17.7
Smokers (%) 28.2 28.9 30.6 30.2 30.8
High cholesterol (%) 5.0 4.6 4.9 6.1 5.9
Aspirin / Week
Manson et al. A prospective study of aspirin use and primary prevention of cardiovascular disease in women. JAMA, 1991.
More aspirin/week associated with greater prevalence of risk factors.
Study subjects:– men and women 30-69– no contraindication to aspirin– with a previous documented MI (secondary prevention)
Treatments: - aspirin (0.5 grams twice daily) - placebo
Follow-up - for at least 3 years
Endpoint: - total mortality (most deaths expected to be due
to cardiovascular disease)
AMIS Overview
AMIS: Characteristics at Baseline
Men (%) 88.4 89.4White (%) 91.7 91.5SBP (mmHg) 127.9 128.2Cholesterol (µ mol/l) 6.1 6.1Age (years) 54.8 54.8Cigarette smoker (%) 27.5 27.2No. MI’s 1.2 1.2
Aspirin(N=2267)
Placebo(N=2257)
Note difference from study by Manson.
AMIS: Total Mortality Findings
Aspirin
Deaths Survivors
245 2,022 2,267
(RR) = 1.11, P = 0.20
Placebo 219 2,038 2,257
4,524
^
Adj. (RR) = 1.05, P = 0.50^
4,060464
Conclusion
No beneficial effect of aspirin– at this stage of Rx– at this dose– when given for 3 years– on total mortality (some reduction in non-fatal MI
and stroke was found)
“Aspirin is not recommended for routine use in patients who have survived an MI”
Other Notable Findings
1. Vital status of all but 9 patients ascertained
2. Average missed visit rate approximately 6% for both groups
3. Average 1.6 capsules per day taken; platelet aggregation and urine tests for compliance consistent with capsule counts
4. Side effects more common with aspirin
5. Multiple outcomes assessed
Antiplatelet Regimensand CVD Morbidity and Mortality
Cardiff-IAspirin 58/615 76/624 25% ± 16Cardiff-II Aspirin 129/847 185/878 32% ± 10Paris-I A or A+Dip 244/1620 4(77/406)* 25% ± 13Paris-II A+Dip 154/1563 218/1565 32% ± 9AMIS Aspirin 395/2267 427/2257 10% ± 7CDP-A Aspirin 88/758 110/77121% ± 14GAMIS Aspirin 39/317 49/309 25% ± 20ART Sulphin 102/813 130/816 24% ± 12ARIS Sulphin 38/365 57/362 37% ± 18Micristin Aspirin 65/672 106/666 43% ± 13Rome Diphrid 9/40 19/40 66% ± 28
Adjusted* total for 1321/9877 1685/9914 25% ± 4all prior MI trials
Trialsanalysed
AntiplateletRegimen Antiplatelet
Adjustedcontrols*
Oddsreduction (SD)
* The actual PARIS-I control result (which is used for calculation of O-E) is 77/406, but to match the PARIS-I treatment group size this control contributes fourfold (308/1624) to the adjusted total number of events and of patients. This adjustment has no effect on the calculations of statistics.
Reference: Peto R, et al., J Clin Epid, 1:12-40, 1995.
Hierarchy of Evidence
Coherence of evidence from multiple sources
Systematic review of well-designed, large randomized trials
Strong evidence from one large randomized trial
Systematic review of small trials (e.g., surrogate outcome studies)
Systematic review of from well-designed cohort studies
Strong evidence from one cohort study
Unsystematic observations (expert opinions)
Adapted from Devereaux PJ et al, Evidence-Based Cardiology, 2nd Edition,
BMJ Books, 2003.
Recommendations on Aspirin
• Aspirin is beneficial for secondary prevention (benefits clearly outweigh risks)
• For primary prevention, the picture is not so clear based on several large trials:– British Doctor’s Study– Physicians’ Health Study– Thrombosis Prevention Trial– Hypertension Optimal Treatment Trial– Primary Prevention Project– Women’s Health Study
Recent Recommendations
• Antithrombotic Trialists’ Collaboration (Lancet 2009)– “In primary prevention without disease, aspirin is of uncertain value
as the reduction in occlusive events needs to be weighed against any increase in major bleeds.”
• U.S. Preventive Task Force (Ann Intern Med 2009)– Encourage men age 45-79 to use aspirin if potential benefits of MI
reduction outweigh bleeding harm; encourage women aged 45-79 to use aspirin if potential ischemic stroke reduction outweigh bleeding harm; do not use < 45 and insufficient evidence > 80 years.
Important Steps in Any Study
Minimize Random Errors.– Solutions: sample size, replication, standardization
of measurement protocols
Minimize Systematic Errors– Solutions: measurement of key confounders,
excellent follow-up, randomization, blinding
Observation / Experiment
“If important alternative hypotheses are compatible with available evidence, then the question is unsettled, even if the evidence is experimental. But, if only one hypothesis can explain all the evidence, then the question is settled, even if the evidence is observational.”
Cornfield, J. Principles of research.Johns Hopkins University,Department of Biostatistics, PaperNo. 325.
Summary• Consideration of bias and confounding variables is
central in the interpretation of findings from observational studies
• Since uncontrolled confounding threatens the validity of findings from observational studies, it is essential that in the design possible confounders be identified and measured
• Data analyses should be aimed at quantifying the influence of confounding factors
• Not all possible confounders are known and not all can be measured, thus designs which eliminate/minimize confounding are particularly important in studying small/moderate effects