Epidemiology GPH-GU 2106 Fall 2024 Lecture 9 - Bias PDF

8/31/24 Epidemiology GPH-GU 2106 - Fall 2024 Lecture 9 – Bias 1 Lecture 9 agenda To understand the difference between random error and systematic bias in epidemiologic studies. To understand how to describe the direction of bias...

8/31/24 Epidemiology GPH-GU 2106 - Fall 2024 Lecture 9 – Bias 1 Lecture 9 agenda To understand the difference between random error and systematic bias in epidemiologic studies. To understand how to describe the direction of bias To learn about the main types of bias that can impact study results obtained from epidemiologic studies. 2 1  8/31/24 Error vs. Bias 3 Observe an association Is it real? Real Spurious Is it due to presence Bias/Chance of 3rd variable? Not due to Due to 3rd variable 3rd variable Could it be due to chance? Confounding Unlikely Likel y CAUSAL Evidence for causation not as strong 4 2  8/31/24 Error & Bias Error – arises from two different types of processes: Random processes, such as sampling, where results can change unpredictably For example, lets say you do a draw of 1,000 random samples from population with (known!) 50% women Estimate of % Women SAMPLE SET 1 SAMPLE SET 2 Distribution of women N = 100 N = 1,000 Highest 68 54.9 10% above 56 above 51.9 25% above 53 above 51.0 25% below 47 below 48.9 10% below 44 below 48.0 Lowest 33 45.0 From V. Schoenbach, Epid160 5 Random Error & Sample Size Systematic error/bias Error Random error Sample Size 6 3  8/31/24 Another example of random error Reality % Study results Size of induration (mm) WHO 7 Error & Bias Error – arises from two different types of processes: Random processes, such as sampling, where results can change unpredictably Systematic error – also known as bias, stems from the design, conduct or analysis of a study that results in a mistaken estimate of an exposure’s effect on disease (Schlesselman, 1982) 8 4  8/31/24 How this would look as systematic error or BIAS Study results Reality % Size of induration (mm) WHO 9 Directions of bias 10 5  8/31/24 Direction of bias Positive Bias Negative Bias Bias AWAY from the null Bias TOWARD the null Observed value is greater Observed value is smaller (stronger) than true value (weaker) than true value OBSERVED TRUE OBSERVED TRUE 3.5 2.5 2.5 3.0 2.0 1.0 1.0 2.0 0.50 0.75 0.50 0.25 0.25 0.5 0.75 1.5 2.0 2.5 3.0 3.5 0.25 0.5 0.75 1.5 2.0 2.5 3.0 3.5 1.0 1.0 11 Types of bias 12 6  8/31/24 Classifying types of bias Two major types: 1. Selection bias – flawed selection of study participants Confounding….coming up next week! 2. Information bias – flaws or inaccuracies in measurement or classification of relevant information 13 Selection Bias Systematic error in selecting subjects into 1 or more of the study groups, such as cases and controls, or exposed and unexposed 14 7  8/31/24 Selection Bias: in a case-control study Cases: patients hospitalized with a diagnosis of pancreatic cancer Controls: patients hospitalized for other reasons by the same gastroenterologist who had hospitalized the case Results: found a strong relationship between coffee drinking and pancreatic cancer Question: did the controls have a comparable exposure probability to the cases? 15 What happened? % Coffee Drinking Pancreatic Controls Non-cases in the general Cancer population Cases 16 8  8/31/24 What happened? POPULATION Pancreatic Cancer Yes No CONTROLS who DID NOT drink coffee are Yes Coffee over-represented No Pancreatic Cancer Yes No Yes Coffee No STUDY SAMPLE 17 Selection Bias: in a cohort study Exposed: workers who handle asbestos (100% participation) Unexposed: workers in other areas of the factory who agree to participate (50% participation) Results: found NO relationship between asbestos and lung cancer 18 9  8/31/24 What happened? POPULATION Lung Cancer UNEXPOSED workers who Yes No participate are those at high risk for lung cancer, so Asbestos Yes unexposed with disease are No over-represented Lung Cancer Yes No Yes Asbestos No STUDY SAMPLE 19 Selection bias due to differential loss to follow-up in a cohort study 80% follow-up Complete cohort Observed cohort Disease No Disease Disease No Disease Exposed 40 160 32 144 Unexposed 20 180 18 162 Total 60 340 50 306 Risk ratio 2.0 1.8 90% follow-up 20 10  8/31/24 Selection bias due to non-response bias in a case-control study Target population Study population Cases Controls Cases Controls Exposed 200 20,000 180 200 Unexposed 400 40,000 300 400 Total 600 60,000 480 600 Odds ratio 1.0 1.2 21 How would selective survival impact recruitment in the following cohort studies: Study of the natural history of HIV infection Study 1: Recruit participants at time they seroconvert Study 2: Recruit participants by means of a serologic survey to detect prevalent infections. Study of the effects of hypertension in an elderly cohort Study 1: Recruit a sample of elderly adults living in a nursing home. Study of environmental tobacco smoke (ETS) exposure in newborns Study 1: Enroll women who have recently given birth and assess ETS exposure levels during pregnancy 22 11  8/31/24 Information Bias Systematic error in obtaining information regarding subjects in the study cases may be misclassified as controls and vice versa exposed may be misclassified as unexposed and vice versa 23 Potential sources of information bias Respondent: inability to understand, recall, articulate; unwillingness to disclose or social desirability Data collector: unclear or ambiguous questions, lack of a neutral demeanor, insufficiently conscientious, inaccurate transcription, fraud Data managers: inaccurate transcription, misreading, miscoding, programming errors Data analysts: variable coding and programming errors Study investigator: inadequate appreciation of the characteristics of the measure or of the relations being studied 24 12  8/31/24 Types of Information Bias Bias from surrogate interviews: spouses/friends/etc. may not have accurate information Surveillance bias: disease ascertainment in monitored population is better than in general population. Recall bias: differential recall of exposure between cases and controls Reporting bias: socially desirable responding Wish bias 25 IMPACT of information bias: Differential vs. non-differential misclassification bias Important question: is information error occurring differently for the groups being compared? If yes, then “differential” misclassification of exposure or outcome If no, then “non-differential” misclassification of exposure or outcome 26 13  8/31/24 Information Bias in a case-control study Cases: newborns with congenital malformations Controls: healthy newborns Results: found a strong relationship between mother’s recall of infection during pregnancy and malformation RECALL BIAS: Parents of children with congenital malformations were more likely to report infection during pregnancy than parents of children without congenital malformations 27 What happened? POPULATION Misclassification of unexposed as Case Control exposed is more common in cases pregnancy Infection Yes than in controls DIFFERENTIAL during MISCLASSIFICATION No Case Control pregnancy Infection Yes during No STUDY SAMPLE 28 14  8/31/24 What is the impact on our measure of association? Differential misclassification resulting in bias away from the null POPULATION STUDY SAMPLE Case Control Case Control pregnancy Infection Yes 50 25 pregnancy 75 during Yes 25 Infection during No 50 100 100 No 25 Real OR BIASED OR = (50*100)/(25*50) =(75*100)/(25*25) =4 = 12 29 What if exposure misclassification is similar in both cases and controls? Case Control Yes Exposure No Non-differential misclassification Usually biases estimate of association towards 1 (the null) 30 15  8/31/24 Information Bias in a case-control study: Example #2 Cases: hospitalized cases of myocardial infarction (MI) in elderly adults Controls: elderly adults who have not been hospitalized for MI Results: found a weak relationship between smoking and MI 31 What happened? POPULATION Many true cases of MI are misclassified as non- cases, and are included as controls (they were not hospitalized and had no major symptoms) Case Control Yes Smoke Misclassification of cases No as controls is similar in smokers and non- Case Control smokers NON- DIFFERENTIAL Yes MISCLASSIFICATION Smoke No STUDY SAMPLE 32 16  8/31/24 What is the impact on our measure of association? Non-differential misclassification resulting in bias towards the null STUDY SAMPLE POPULATION Case Control Case Control Yes 50 25 Yes 25 50 Smoke Smoke No 50 100 No 25 125 REAL OR BIASED OR =(100*50)/(50*25) =(125*25)/(25*50) =4 =2.5 33 Information Bias: in a cohort study Exposed: women who use oral contraceptives (OC) Unexposed: women who do not use OC Results: found a strong relationship between OC use and thrombophlebitis SURVEILLANCE BIAS: also called detection bias Women who are on oral contraceptives are more likely to receive a diagnosis of thrombophlebitis 34 17  8/31/24 What happened? POPULATION Misclassification of disease as Thrombophlebitis non-diseased is different in exposed Yes No and unexposed persons DIFFERENTIAL MISCLASSIFICATION Yes OC Use No Thrombophlebitis Yes No Yes OC Use No STUDY SAMPLE 35 What is the impact on our measure of association? Differential misclassification resulting in bias away from the null POPULATION STUDY SAMPLE Thrombophlebitis Thrombophlebitis Yes No Yes No Yes 50 25 Yes 50 25 OC Use OC Use No 50 100 No 30 120 REAL RR BIASED RR = (50/75)/(50/150) = (50/75)/(30/150) = 2.0 = 3.33 36 18  8/31/24 How to reduce information bias Precise operational definitions of variables Detailed measurement protocols Repeated measurements on key variables Training, certification, and re-certification of study staff Data audits (of interviewers, of data centers) Data cleaning – visual, computer Re-running all analyses prior to publication 37 Evaluating selection & information bias Why did it occur? What effect does it have on the observed association? What could have been done to control for bias in this study? to prevent it in future studies? ALWAYS REPORT POTENTIAL SOURCES OF BIAS IN YOUR STUDY REPORTS! 38 19  8/31/24 Examples of reporting on bias 39 20