Bias Confounding Association Causation Dr Asif Rehman Bias

Bias, Confounding Association & Causation Dr. Asif Rehman

Bias Definition Any systematic error that results in an incorrect estimation of the association between an exposure and outcome is called bias

Types of Bias More that 50 types of bias are identified in epidemiological studies, but for simplicity, they are broadly grouped in to two categories: 1. Selection bias 2. Information bias

Types of Bias 1. Selection bias It occurs when the inclusion of subjects in a study depends in some way on the outcome of interest. It occurs mainly in case control studies and not in prospective cohort study as the outcome of interest has not yet occurred.

Selection bias Another means of selection bias could be due to inappropriate source of selection e. g Case selected from hospital and controls from household surveys. In this case it is possible that a number of demographic and lifestyle variables could be different amongst the cases & controls leading to non comparability between groups & incorrect results with respect to association between exposure and outcome.

Selection bias examples A classical example of selection bias is a study conducted to see the association between oral contraceptive and thrombo-embolism. There was a concern in this study that as physicians were already aware of the possible relationship b/w OC & thrombo-embolism, hence proportion of women that had been hospitalized for evaluation of thromboembolism was all current users of Ocs. So any increased frequency of thromboembolism in oral contraceptive users could be in part due to the fact that hospitalization and the determination of the diagnosis were both influenced by the history of OC use.

Selection bias examples In a clinical trail a selection bias can occur if there is no randomization. Suppose that the principal investigator is taking a decision as to which pts are going to be included in the standard drug group and which patient is going to be included in the new drug group. If the principal investigator is allowed to do so, he might include all the healthy patients in the new drug group and all the patients who are sick and (having multiple co-morbid conditions) in the standard drug group. Thus he can show better outcomes among the new drug group ( who are healthy pts) compared to the standard treatment group (who are sicker) and present results which are not true. The process of randomization ensures that selection bias can not take place, by ensuring that the principal investigator and his team members are not aware which pt is going to control group and which pt is going to intervention group.

Information bias It includes any systematic error in the measurement of information on exposure or outcome. Its is further classified in to different categories. .

Information bias Different types: 1. Recall bias 2. Interviewer bias 3. Reporting bias 4. Loss to follow up bias 5. Misclassification bias

Information bias Recall bias: It occurs when individuals with previous adverse health outcomes remember and report their previous exposure differently or with different degree of completeness and accuracy than those who are unexposed. It can lead to an over or underestimate of the association between exposure and disease, depending on whether the cases recall their exposure to a greater or lesser extent than the control.

Information bias Recall bias: For example, in a case control study mothers whose recent pregnancies had ended in fetal death (case) may report their exposure experience (drug history) differently than a group whose pregnancy had ended normally (control). That is, cases may have a better recall on past exposure than controls. It can be reduced by: 1. Collecting exposure data from work or medical record. 2. Blinding participants to study hypothesis.

Information bias Interviewer bias: It refers to any systematic difference in the soliciting, recording or interpretation of information by interviewer from study participants and can effect every type of epidemiological study.

Information bias Loss to follow-up bias: It is a major concern in a cohort or any prospective study. When persons lost to follow up differ from those who remain in the study with respect to both the exposure and outcome, any observed association will be biased. Even very small loss to follow up can be a potential for bias as long as such loss is related to both exposure an disease.

Information bias Misclassification Bias: It occurs when the sensitivity/ and or specificity of procedure to detect exposure and/ or outcome is not perfect, that is exposed/disease subjects can be classified as non-exposed and vice versa, based on the means of determination which may be unclear or non standardized. It is inevitable in every study and always a potential for concern and therefore should be care fully evaluated.

Control of bias is mostly done at the design phase of the study. For the control of the selection bias: 1. Correct choice of the study population (sampling procedure) 2. Randomization

Control of bias For the control of information bias: 1. Correct training of interviewers and use of clearly written protocols ensuring uniform methodology of obtaining information. 2. Use of standardized, tested instruments for data collection, and utilizing uniform source of data on all study subjects. 3. Maintaining of a complete records and having definite means of contact with respondents to prevent loss to follow up 4. Use of clearly defined means of determination of both exposure and outcome variables. 5. Blinding of interviewees and interviewers to study objective.

Confounding A confounder is an external variable that changes the effect of a dependent and independent variable. Confounding can lead to an over or an under estimation of the true association between exposure and outcome. Universal confounders: 1. Age 2. Gender 3. Socio economic status

Confounding example Example 1 In a study conducted to determine the association between smoking and myocardial infarction, Age can be a confounder as it is associated with both exposure and outcome independently.

Confounding example Example 2 A study was conducted to see the association between coffee consumption and pancreatic cancer. Smoking can be a confounder because most of the coffee users are smokers.

Control of Confounding can be controlled in study design by matching and randomization. In analysis it can be controlled through stratification and multivariate analysis.

Association and Causation A correlation/Association between variables, however, does not automatically mean that the change in one variable is the cause of the change in the values of the other variable. Causation indicates that one event is the result of the occurrence of the other event; i. e. there is a causal relationship between the two events

Association and Causation Guidelines for judging whether an association is Causal 1. Temporal association 2. Strength of association 3. Dose response relationship 4. Replication of the finding 5. Cessation of exposure 6. Consistency with other knowledge 7. Specificity of the association

Association and Causation Temporal Association: If a factor is believed to be the cause of a disease, exposure to the factor must have occurred before the disease developed. It is often easier to establish a temporal relationship in a prospective cohort study than in a case-control study or a retrospective cohort study

Association and Causation Strength of the Association: The strength of the association is measured by the relative risk (or odds ratio). The stronger the association, the more likely it is that the relation is causal.

Association and Causation Dose-Response Relationship: As the dose of exposure increases, the risk of disease also increases. For example of the dose-response relationship for cigarette smoking and lung cancer. If a dose-response relationship is present, it is strong evidence for a causal relationship.

Association and Causation Replication of the Findings: If the relationship is causal, we would expect to find it consistently in different studies and in different populations. Replication of findings is particularly important in epidemiology. If an association is observed, we would also expect it to be seen consistently within subgroups of the population and in different populations, unless there is a clear reason to expect different results.

Association and Causation Cessation of Exposure. If a factor is a cause of a disease, we would expect the risk of the disease to decline when exposure to the factor is reduced or eliminated.

Association and Causation Consistency with Other Knowledge: If a relationship is causal, we would expect the findings to be consistent with other data.

Association and Causation Specificity of the Association. An association is specific when a certain exposure is associated with only one disease; this is the weakest of all the guidelines and should probably be deleted from the list. Cigarette manufacturers have pointed out that the diseases attributed to cigarette smoking do not meet the requirements of this guideline, because cigarette smoking has been linked to lung cancer, pancreatic cancer, bladder cancer, heart disease, emphysema, and other conditions.

THANK YOU