Bias and Confounding Epidemiology PDF

‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Epidemiology Bias, Confounding, Prof. Liat Lerner Research errors Random errors: Systematic – Sampling errors (bias) – Lack of – Selection of precision subjects – Variability in...

‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Epidemiology Bias, Confounding, Prof. Liat Lerner Research errors Random errors: Systematic – Sampling errors (bias) – Lack of – Selection of precision subjects – Variability in – Collection of measurement information about exposure and outcome – confounding 1  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Bias Any systematic error in a study, in the design, conduct or analysis, that results in an incorrect estimate of an association between exposure and risk of outcome (disease). Bias Any systematic error in a study that results in an incorrect estimate of an association between exposure and risk of outcome (disease). A complete elimination of research bias is often unachievable 2  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Bias Any systematic error in a study that results in an incorrect estimate of an association between exposure and risk of outcome (disease). A complete elimination of research bias is often unachievable We try to avoid it as much as we can in the design, and minimize it’s effect during analysis. Bias Any systematic error in a study that results in an incorrect estimate of an association between exposure and risk of outcome (disease). A complete elimination of research bias is often unachievable We try to avoid it as much as we can in the design, and minimize it’s effect during analysis. Bias is a result of an error in the design or conduct of a study 3  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Types of Bias: Selection Selection bias – If the way in which cases and controls, or exposed and nonexposed individuals, were selected is such that an apparent association is observed, when in reality exposure and disease are unrelated. Types of Bias: Selection Selection bias – Subjects included in a study are not representative of the population Case-Control study: when cases are more likely to be selected if exposed or controls are more likely to be selected if unexposed - Coffee drinking and pancreatic cancer Mitigation strategy: stratification according to level of drinking 4  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Non-response Bias: Selection If the response rate of people with the disease (cases) who were exposed is higher than that of cases who were not exposed, an apparent association is observed, when in reality exposure and disease are unrelated. – Asthma study via mailed questionnaires – lower response rate of smokers. Mitigation: collect data on non-respondents Types of Bias: Information Information bias – When the means for obtaining information about the subjects in the study are inadequate so that as a result some of the information gathered regarding exposures and/or disease outcome is incorrect – When the level of information is different in exposed and in unexposed, or in cases and in controls. 5  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Types of Bias: Information Information bias – Recall bias – Interviewer bias (mitigation: blinding to the outcome) Misclassification bias Misclassification bias Case control: some cases will be misclassified as controls, and some controls will be misclassified as cases. Misclassify exposure status When exposure data are based on interview without objective ascertainment… Selby JV et al, NEJM 1992. 6  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Types of Misclassification bias Differential misclassification: the rate of misclassification differs between study groups Cases are misclassified as exposed more often than controls are… – Biases towards the H1 or the H0 Selby JV et al, NEJM 1992. Types of Misclassification bias Non-differential misclassification: a problem inherent in the data collection methods, unrelated to exposure status or to study group (cases or controls) – Biases towards the H0 (null): RR 1 Selby JV et al, NEJM 1992. 7  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Some types and sources of information bias Bias in abstracting records Bias in interviewing Bias from surrogate interviews Surveillance bias – monitored population Recall bias Reporting bias – Wish bias “why me?” Estimating the impact of Bias What would the effect be on the RR/OR? When RR is underestimated we are more confident of our findings Biases can work in different directions: Does screening sigmoidoscopy change mortality rates from colorectal cancer: a case control study. How might those screened be different than those not screened? Selby JV et al, NEJM 1992. 8  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Evaluating Epidemiologic Associations 1. Could the association have been observed by chance? 2. Could the association be due to bias? 3. Could other confounding variables have accounted for the observed relationship? Observed Association Causal? Maternal Coffee smoking Drinking Low birth Pancreatic weight Cancer 9  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Distribution of Pancreatic Cancer Cases and Controls by Coffee- Drinking Habits and Estimates of Risk Ratios Sex Category Total Male Number of cases 216 Number of controls 307 Adjusted relative risk* 2.6 95% Confidence interval 1.2-5.4 Female Number of cases 151 Number of controls 336 Adjusted relative risk* 2.3 95% Confidence interval 1.2-4.6 MacMahon et al. 1981 Types of Association Real and Spurious Associations: A. Causal B. Due to Confounding Maternal Coffee smoking Drinking Smoking Low birth Pancreatic weight Cancer 10  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Confounding 1. Smoking is a known risk factor for pancreatic cancer 2. Smoking is associated with coffee drinking, but is not a result of coffee drinking 3. The observed association between coffee drinking and pancreatic cancer is the result of confounding by cigarette smoking Confounding Confounding variable (C): – interferes with the observed association – Is associated with the putative causal factor (A) – Is causally associated with the outcome (B) C A B ? 11  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Controlling for Confounders In the design: – Randomization: not always possible – Restriction: exclude subjects positive to confounder – Matching: exposed & unexposed by confounder In the analysis – Stratification – Adjustment: Multivariate analysis Distribution of Pancreatic Cancer Cases and Controls by Sex and by Coffee- Drinking Habits and Estimates of Risk Ratios Sex Category Coffee Consumption (Cups/Day) 0 1-2 3-4 ≥5 Total Male Number of cases 9 94 53 60 216 Number of 32 119 74 82 307 controls Adjusted relative risk* 1.0 2.6 2.3 2.6 2.6 95% Confidence interval - 1.2-5.5 1.0-5.3 1.2-5.8 1.2-5.4 Female Number of cases 11 59 53 28 151 Number of 56 152 80 48 336 controls Adjusted relative risk* 1.0 1.6 3.3 3.1 2.3 95% Confidence interval - 0.8-3.4 1.6-7.0 1.4-7.0 1.2-4.6 MacMahon et al. 1981 12  ‫תשפ"ה‬/‫טבת‬/‫י"ג‬ Estimates of Relative Risk of Cancer of the Pancreas Associated with Use of Coffee and Cigarettes (sex and age adjusted) Coffee Drinking (Cups/Day) Cigarette Smoking Status 0 1-2 ≥3 Total Never smoked 1.0 2.1 3.1 1.0 Ex-smokers 1.3 4.0 3.0 1.3 Current smokers 1.2 2.2 4.6 1.2(0.9-1.8) Total* 1.0 1.8 2.7 (1.0-3.0) (1.6-4.7) * Reference category is the group that uses neither cigarettes nor coffee McMahon B, Yen S, Trichopoulos D, et al. Coffee and cancer of the pancreas. N Engl J Med. 1981, 13