Cohort study Analysis Akbar soltani Assistant Professor of

Cohort study: Analysis Akbar soltani Assistant Professor of Medicine and Endocrinology Evidence-Based Medicine Working Team (EBMWT) Tehran University of Medical Sciences (TUMS) Shariati Hospital EBMWT A. SOLTANI TUMS

Critical appraisal: Cohort Studies • Definition and Purpose • Types • Strengths & Weaknesses • PECO • Biases in Cohort Studies • Analysis EBMWT A. SOLTANI TUMS

Analysis : Measures of Association & Impact

Self evaluation I. Several studies have found that approximately 85% of cases of lung cancer are due to cigarette smoking. This measure is an example of: a. An incidence rate b. An attributable risk c. A relative risk d. A prevalence risk e. A proportionate mortality ratio EBMWT A. SOLTANI TUMS

10 years cohort of smoking and CHD At beginning of Developed study CHD NOT Developed CHD 2, 000 Healthy smokers 65 1935 4, 000 Healthy nonsmokers 20 3980 EBMWT A. SOLTANI TUMS

Self evaluation 2. The incidence of CHD in smokers that can be attributed to smoking is=? 3. The proportion of the total incidence of CHD in smokers that is attributable to smoking is=? EBMWT A. SOLTANI TUMS

Self evaluation 2. The incidence of CHD in smokers that can be attributed to smoking is=27. 5/1000 3. The proportion of the total incidence of CHD in smokers that is attributable to smoking is=85% EBMWT A. SOLTANI TUMS

Self evaluation In a cohort study of smoking and lung cancer, the incidence of lung cancer among smokers was found to be 9/1, 000 and the incidence among nonsmokers was 1/1, 000. From another source we know that 45% of the total population were smokers. 4. The incidence of lung cancer attributable to smoking in the total population is: 5. The proportion of the risk in the total population that is attributable to smoking: EBMWT A. SOLTANI TUMS

Self evaluation In a cohort study of smoking and lung cancer, the incidence of lung cancer among smokers was found to be 9/1, 000 and the incidence among nonsmokers was 1/1, 000. From another source we know that 45% of the total population were smokers. 4. The incidence of lung cancer attributable to smoking in the total population is: 3. 6/1000 5. The proportion of the risk in the total population that is attributable to smoking: 78. 3% EBMWT A. SOLTANI TUMS

Risk for Disease n Epidemiologic Measures of Association n Strength of association n Effect of a factor (exposure) in exposed group n n Attributable Risk (AR) Importance of a factor (exposure) in the population n EBMWT Relative Risk (RR) – cohort study Relative Odds (Odds Ratio) (OR) – cohort or casecontrol Population Attributable Risk (PAR) A. SOLTANI TUMS

Level of Risk Ie In In exposed group In non-exposed group Absolute risks EBMWT A. SOLTANI TUMS

Level of Risk Ie incidence due to the exposure In Attributable risk (AR) AR = Ie- In incidence not due to the exposure In exposed group Background Risk In non-exposed group AR: To find how much of the total risk in exposed persons is due to the exposure, subtract the background risk (In) from the total risk (Ie) EBMWT A. SOLTANI TUMS

EXAMPLE Develop CHD SMOKE 84 R NOT 87 Do not develop CHD Total INCIDENCE PER 1000 PER YEAR 2196 3000 28. 0 4913 5000 17. 4 SMOKER EBMWT A. SOLTANI TUMS

Attributable risk for exposed: n n Attributable risk for exposed: 28 -17. 4/1000=10. 6/1000 Proportion: 28 -17. 4/28=37. 9% EBMWT A. SOLTANI TUMS

PAR=Incidence in total population- incidence in nonexposed Smokers=44% Nonsmokers =56% Develop CHD SMOKE 84 R NOT 87 Do not develop CHD Total INCIDENCE PER 1000 PER YEAR 2196 3000 28. 0 4913 5000 17. 4 SMOKER EBMWT A. SOLTANI TUMS

Attributable risk for total population n n n n Incidence in total population- incidence in nonexposed Smokers=44% Nonsmokers =56% 28/1000*44%+17. 4/1000*56%=22. 1/1000 -17. 4/1000=4. 7/1000=PAR Proportion 22. 1 -17. 4/22. 1=21. 3% Elimination of smoking=21. 3% reduction in incidence of CAD EBMWT A. SOLTANI TUMS

Attributable risk for total population? n Incidence in total population-incidence in nonexposed n We need: n n n The incidence among smokers The incidence among nonsmokers The proportion of the total population that smokes EBMWT A. SOLTANI TUMS

PAR% in Case-Control Studies n Cohort study where P = % population exposed n Case-control study If % controls exposed % population exposed: where Pcon = % controls exposed EBMWT A. SOLTANI TUMS

Attributable Risk n EBMWT The interpretation of the AR is dependent on the assumption that a cause-effect relationship exists between exposure and disease. A. SOLTANI TUMS

Age adjusted death rate/100000 LUNG CANCER CORONARY HEART MORTALITY DIS. MORTALITY CIGARETTE SMOKERS NONSMOKERS RELATIVE RISK ATTRIBUTABLE RISK %ATTRIBUTABLE RISK EBMWT 140 669 10 413 14. 0 1. 6 130 per 100, 000 per year 92. 9% 256 per 100, 000 per year A. SOLTANI 38. 3% TUMS

Doll & Peto, BMJ 1976 LUNG CANCER CORONARY HEART MORTALITY DIS. MORTALITY CIGARETTE SMOKERS NONSMOKERS RELATIVE RISK ATTRIBUTABLE RISK %ATTRIBUTABLE RISK EBMWT 140 669 10 413 14. 0 1. 6 130 per 100, 000 per year 256 per 100, 000 per year 130/140=92. 9% 256/669=38. 3% A. SOLTANI TUMS

Mortality in Male Physicians n n EBMWT Smoking much stronger risk factor for mortality from lung ca than for CHD However, if smoking is causally related to both diseases, the elimination of cigarettes would prevent far more deaths among smokers from CHD than from lung ca as shown by the ARs A. SOLTANI TUMS

Mortality in Male Physicians n EBMWT Explanation for this is that while death from lung cancer is a relatively rare occurrence among nonsmokers, the annual death rate for CHD in nonsmokers is high: 413 per 100, 000. A. SOLTANI TUMS

Mortality in Male Physicians n EBMWT So, even a relatively small (60%, RR=1. 6) increased risk of CHD mortality associated with smoking will affect a much larger number of people than a 14 -fold increased risk of death from lung ca. A. SOLTANI TUMS

Relative Risk vs. Attributable Risk n Strength of association n n RR and OR n n Etiologic studies of disease: is the factor a cause? n n n EBMWT how strong is the association? A. SOLTANI How much risk is due to exposure? AR and AF (PAR and PAF) Addresses issues of clinical and public health practice significance how much excess risk? TUMS

ANALYSIS OF AGE, BIRTH COHORT AND PERIOD EFFECTS EBMWT A. SOLTANI TUMS

Age Effects n n EBMWT Change in rate of a condition according to age, irrespective of birth cohort and calendar time. Shows strong age effect with little or no cohort effect (data overlap from cohort to cohort that is there is the same age pattern for each birth cohort). A. SOLTANI TUMS

Cross sectional study: prevalence of disease-1995 EBMWT 30 -39 45 40 -49 40 50 -59 36 60 -69 31 70 -79 27 Does prevalence decrease with age? Not necessarily TUMS A. SOLTANI

Birth Cohort Effects n n EBMWT Change in rate of a condition according to year of birth, irrespective of age and calendar time. Result from lifetime experience not just experiences related to birth (e. g. diet; smoking habits). A. SOLTANI TUMS

Examples Smoking and lung cancer n Fat intake and CAD n Radiation and thyroid cancer n Health status may be partially dependent on past exposure n (Prevalence is determined by year of birth) EBMWT A. SOLTANI TUMS

Period Effects n n EBMWT Change in the rate of a condition affecting an entire population at some point in time, irrespective of age and birth cohort. Examples: new vaccine; disease outbreak; war; natural disaster, insulin, massive migration A. SOLTANI TUMS

SUMMARY EBMWT A. SOLTANI TUMS

Measures of disease occurrence and exposure effect in analytic study designs Type of Analytical Study Measure of disease (outcome) occurrence Measure of exposure effect Ecological Rate, Risk, Prevalence Correlation or Regression Coefficient Cross-sectional Prevalence ratio, Prevalence difference, Odds ratio Cohort Rate, Risk, Odds, Mean or Median Rate Ratio, Risk Ratio, Odds Ratio, Rate Difference, Vaccine efficacy, Difference between Means or Medians Case-Control None 1 Odds Ratio, Vaccine efficacy Intervention Rate, Risk, Odds, Mean or Median Rate Ratio, Risk Ratio, Odds Ratio, Rate Difference, Vaccine efficacy, Difference between Means or Medians 1 Unless the sampling fraction is known for both cases and controls i. e. proportion of cases TUMS A. SOLTANI and proportion of controls sampled from the population is known EBMWT

Applications of different observational and analytic study designs (number of crosses indicates usefulness) Ecological Crosssectional Case-control Cohort Investigation of rare diseases ++++ - Investigation of rare exposures ++ - - ++++ Studying multiple outcomes + ++ - ++++ Studying multiple exposures ++ ++ +++ Measurement of time relationship between exposure and outcome + - + ++++ Direct measurement of incidence - - + ++++ Investigation of long latent periods - - +++ EBMWT A. SOLTANI (if TUMS retrospective)

attribute cohort Case. c Cross. s Classification of population Free from condition Case. control Disease+exposure+- Sample represented Non-diseased Uncertain Source ? Survivors in time Temporal sequence Prospective retrospective function Compare incidence rates compare prevalence of EX Retrospective relation revealed at a point or period Describe association of EX & disease simultaneously EBMWT A. SOLTANI TUMS

attribute cohort Case. c Cross. s Outcome comparison Incidence of disease Prevalence of EX Prevalence of disease in EX &UNEX Risk measurement RR, AR (precise) Odds ratio Prevalence ratio(inexact estimate of RR)also Odds ratio Evidence for causality strong Need more Only suggestive careful analysis bias Easy to manage Need more effort to manage EBMWT A. SOLTANI May be very difficult to manage TUMS

basis cohort Case. c Cross. s Rare condition Not practical best NA To determine precise risk best Only estimate possible Gives relative prevalence Temporality best NA NA For administrative purpose NA NA best If attrition is a serious problem (if >30% conclusion will be suspected NA Attrition is usually minimal Attrition may have occurred before the study If selective survival is a problem best NA NA If all factors are unknown best NA Less appropriate Time and money Most expensive Least expensive In between EBMWT A. SOLTANI TUMS

Strengths and Weaknesses of Observational Analytic Study Designs Ecological Crosssectional Casecontrol Cohort Selection bias NA Medium High Low Information bias ( Recall) NA High Low Loss to follow up NA NA Low High Medium Low Time required Low Medium High Cost Low Medium High Probability of: n n Confounding NA = Not EBMWT applicable A. SOLTANI TUMS

THANK YOU EBMWT A. SOLTANI TUMS