Biases in Epidemiological study

Epidemiological Study Aims To know the cause of disease and risk factor Epidemiological Study Aims know the extent of disease in the community, to study the natural history and prognosis of disease evaluate the existing and new preventive and therapeutic measures and modes of health delivery, provide foundation for development of policy.

Error Error- A false or mistaken result obtained in a study or experiment Sources of error- Random error and Systematic error Random error -Random error is when a value of the sample measurement diverges from that of true population value -due to chance alone. Systematic error- Any trend in the collection, analysis, interpretation, publication, or review of data that can lead to conclusions that are systematically different from the truth

Bias Any systematic error in the design, conduct or analysis of a study that results in a mistaken estimate of an exposure’s effect on the risk of the disease.” -Leon Gordis Bias in research denotes deviation from the truth

Impact of Bias Our ability to apply the results obtained in our study population to a Broader population is called the generalizability, or external validity, A randomized trial Study is internally valid if the randomization is proper and the study is free of other biases, no major methodologic Problems

Classification of Bias Selection bias 2. Information bias/Measurement bias 3. Bias due to confounding

Selection Bias Selection bias is a systematic error resulting from the way the subjects are either selected in a study or else are selectively lost to follow up. This occurs when the subjects included in the study are not truly representative of the target population. Selection bias can cause an overestimate or underestimate of association.

Example Study of smoking and lung cancer, the exposure is distributed among cases and controls in target population: Lung cancer + Lung cancer - Smoking + 100 200 Smoking - 400 True OR in target population = (100*400)/ (100*200) =2

If the selection probabilities for all the cells in the table are equal at 90%, the 2*2 table of selection probabilities would be: If selection probabilities are unequal, and non proportional, then selection bias will occur Lung cancer + Lung cancer - Smoking + 100*0.90= 90 200*0.90 = 180 Smoking - 100*0.90 = 90 400*0.70= 280 Lung cancer + Lung cancer - Smoking + 100*0.90= 90 200*0.90 = 180 Smoking - 100*0.90 = 90 400*0.90= 360 OR= (90*360)/ (90*180) =2 OR= (90*280)/ (90*180) = 1.6

Types of selection bias Self-selection Bias Berkson’s Bias Incidence – prevalence bias/Neyman’s bias Healthy Worker Effect Bias due to loss to follow up

1a. Self-selection Bias Volunteers were recruited. Common source of selection bias Volunteers induced bias Individuals who volunteer for study possess different characteristics than average general population. Example- A case control study was planned to explore association of family history of Diabetes and presence of diabetes in subjects. Volunteers were recruited. Diabetic Volunteers with family history of diabetes may be more likely to participate in the study- leading to bias and over estimation of results

Example Self selection bias Diabetes + Diabetes - F/H /0 Diabetes+ 300 200 F/H /O Diabetes + Diabetes - F/H/0 Diabetes + 240(80%) 120(60%) F/H /O 180(60%) OR=3.0 True OR= 2.25 Self selection bias

1b. Berkson’s Bias Hospital selective bias. Hospital cases when compared to hospital controls can have bias, if the exposure increases the chance of admission. The cases in a hospital will have disproportionately higher number of subjects with that exposure. Persons with two or more diseases have a higher probability to be hospitalized than persons with only one disease.

Example Example: People who have both peptic ulcer and also smoke, are more likely to be admitted in the hospital than people who have either of them. A Case-Control study trying to evaluate relationship between smoking and peptic ulcer may find a stronger association between the two.

1c. Incidence – Prevalence bias /Neyman's bias Bias occurs when we estimate the risk of disease on basis of data collected at a given point in a series of survivors rather than on data gathered during a certain time period in a group of incident cases. It creates a case group not representative of cases in the community. Major issue in Case-control and cross-sectional study.

Many cases and controls gave history of regular physical exercise; Example Case control study to assess the protective effect of physical exercise on MI- was undertaken by taking cases of MI and healthy controls and asking them about the history of regular physical exercise Many cases and controls gave history of regular physical exercise; So, the study concluded that regular physical exercise does not protect against MI 25% to 33% of the cases of acute MI die within the first 3 hrs. Only those who survive get admitted to the hospital and are available as cases. Now regular physical exercise may be an important factor in helping the person to overcome the acute myocardial episode Thus ,the cases who did exercise were the ones who lived to give such a history.

1d. Healthy Worker Effect Form of selection bias The basic rule of selection and comparisons in research should be to ‘compare likes with likes’ Example: A comparative study between the health status of military and civilian population may show - a better health status of soldiers , as soldiers undergo initial medical examination in which ,unfit people are excluded and only ‘healthy workers’ are included in the army.

1e. Bias due to loss to follow up If the subjects drop out/are withdrawn/ die before assessment of study outcome, the study results would be different, than if they would have continued. Differential loss to follow up in a prospective cohort study on oral contraceptives and thromboembolism After 40% loss to follow up TE Normal OC + 20 9980 OC - 10 9990 TE Normal OC + 8 5980 OC - 5990 RR = 1 (biased) RR = 2 (truth)

2.Information bias/ Measurement bias Information bias can occur 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 outcomes is incorrect. Due to inaccuracies in methods of data acquisition, subjects may be misclassified and thereby introducing a misclassification bias. For example, in a case-control study, some people who have the disease (cases) may be misclassified as controls, and some without the disease (controls) may be misclassified as cases. TYPES of misclassification bias - 2a. Non differential Misclassification bias 2b. Differential Misclassification bias

2a. Non differential mis-classification bias When the misclassification is the same across the groups to be compared, for example, exposure is equally misclassified in cases and controls - NDM It is just a problem inherent in the data collection methods. The amount and direction of misclassification is same in cases and controls. The usual effect NDM is that the relative risk or odds ratio tends to be diluted, and it is shifted towards 1.0. So it is less likely to detect an association ,even if one really exists.

2a. Non differential misclassification bias…. . A Case- Control study comparing CAD cases & controls for history of diabetes. With non-differential Misclassification CAD + CAD - Diabetes+ 40 10 Diabetes- 60 90 CAD + CAD- Diabetes + 20 5 Diabetes - 80 95 OR = (40x90)/(10x60) = 6 (True) OR = (20x95)/(5x80) = 4.75 With non-differential Misclassification (only half of the diabetics are correctly recorded as such in case and controls)

2b. Differential misclassification When errors in classification of exposure or outcome are more frequent in one group. - DM For example, misclassification of exposure may occur such that unexposed cases are misclassified as being exposed more often than the unexposed controls are misclassified as being exposed. This can occur if -

Differential misclassification….. 1. Differences in accurately remembering exposures (unequal) Mothers of malformed infants tend to remember more mild infections that occurred during perinatal period than did mothers of normal infants. There was a tendency for DM in regard to prenatal infection, in that more unexposed cases ( mothers of malformed infants) were misclassified as exposed ( perinatal infection)than were unexposed controls. The result was an apparent association of malformations with infections, even though none existed.

2b. Differential misclassification… 2.Interviewer or Recorder bias. Example: interviewer knows better about hypothesis 3. More accurate information in one of the groups. Example: Case-Control study with cases from one Hospital and controls from another ,with differences in record keeping.

Recall Bias Type of information bias People with disease may remember exposures differently (more or less accurately) than those without disease. To minimize: 1. Use a control group that has a different disease 2. Use questionnaires that are constructed to maximize accuracy and completeness 3. For socially sensitive questions, such as alcohol and drug use, use self administered questionnaire instead of an interviewer 4. If possible, assess past exposures from pre-existing records

Interviewer bias Systematic differences in soliciting, recording, or interpreting information Minimized by- 1. Blinding the interviewers if possible 2. Using standardized questionnaires consisting of closed ended, easy to understand questions 3. Training all interviewer to adhere to the question and answer format strictly 4. Obtaining data or verifying data by examining pre existing records (eg- medical records or employment records)

Biases specific to the studies

Biases in Case- Control study Selection bias Information bias Bias due to confounding Biases in Case- Control study

Biases in Cohort study Selection bias Follow up bias Information bias Bias due to confounding Biases in Cohort study

Biases in Clinical Trial Selection Bias Ascertainment bias Consent bias Dilution bias Attrition bias Analytical bias Publication bias Choice of question bias Choice of population bias Technical bias Chance bias Biases in Clinical Trial

Ascertainment Bias Consent Bias This occurs when consent to take part in the trial occurs AFTER randomisation. Most frequent danger in Cluster trials Consent Bias This occurs when the person reporting the outcome can be biased. Example, of homeopathy study of histamine, showed an effect when researchers were not blind to the allocation but no effect when they were. Multiple sclerosis treatment appeared to be effective when clinicians were unblinded but ineffective when blinded Ascertainment Bias

Dilution bias Attrition Bias This occurs when the intervention or control group get the opposite treatment. This affects all trials where there is non-adherence to the intervention. For example, in a trial of probiotics, about 4% of the controls are getting the treatment and 35% of the intervention group stop taking their treatment. This will ‘dilute’ any apparent treatment effect. Usually most trials lose participants after randomisation. This can cause bias, particularly if attrition differs between groups. If a treatment has side-effects this may make drop outs higher among the less well participants, which can make a treatment appear to be effective when it is not We can avoid this by using Intention to Treat Analysis, where we keep as many of the patients in the study as possible even if they are no longer in treatment

Biases in screening programmes Volunteer bias Lead time bias Length time bias Overdiagnosis bias Biases in screening programmes

Lead time bias Lead time is the period of time between the detection of a medical condition by screening and when it ordinarily would be diagnosed because a pt. experiences symptoms and seeks medical care The earlier detection of disease by screening gave him an additional 3 years of postoperative and 3 years of normal life

Length time bias Over diagnosis Bias Persons who initiate screening program have almost unlimited enthusiasm for the program. Even cytologists reading pap smears may become so enthusiastic that they may tend to over read the smears (false positive readings). Consequently the abnormal group will be diluted with women who are free of disease Over diagnosis Bias Form of selection bias- a statistical distortion of results which can lead to incorrect conclusions about the data Length time bias can occur when lengths of intervals are analysed by selecting intervals that occupy randomly chosen points in time or space. Example: the impression that detecting tumors through screening causes cancers to be less dangerous, when the reality is that less dangerous cancers are simply more likely to be detected by screening.

How to control Selection bias Sampling the cases and controls in the same way. Matching Randomization Using a population based sample.

How to control measurement bias Development of explicit, objective criteria for measuring environmental characteristics and health outcomes. Careful consistent data collection- for example, through use of standardized instruments ;objectives, closed ended questionnaires; valid instruments. Careful consistent use of data instruments- for example, through use of standardized training and instruction manuals, blinding to the extent possible Development and application of quality control/ quality assurance procedures Use of multiple sources of data / Data cleaning and coding Analysis and adjustment, if necessary, to take account of measurement bias

3.Confounding Mixing or blurring of effects Researcher attempts to relate exposure to outcome but actually measure effect of 3rd factor, termed as confounding variable. A confounding variable is associated with exposure, affects outcome, but not an intermediate link in chain of causation between exposure and outcome

Examples … confounding HEART DISEASE COFFEE DRINKING (Smoking increases the risk of heart ds) SMOKING (Coffee drinkers are more likely to smoke)

For a variable to be a confounder It should be a known risk factor for the disease or the outcome It should be associated with the exposure It should not be in direct chain or linked between the exposure and outcome It should be differentially distributed in the two group

Control of confounding During design of epidemiological study: Randomization Restriction-Subjects chosen for study are restricted to only those possessing a narrow range of characteristics, to equalize important extraneous factors Matching - For each patient in one group there are one or more patients in the comparison group with same characteristics, except for the factor of interest During analysis of study: Stratification- The process of or the result of separating a sample into several sub-samples according to specified criteria such as age groups, socio-economic status etc. multivariate analysis

References Gordis L, Epidemiology,, Elsevier Saunders. Library of Congress Cataloging-in-Publication Data. Fifth Edition .2014 pg no 262-269 Beaglehole R, Bonita R, Kjellstrom T, Basic Epidemiology, 2nd Edition(2006)WHO. Linienfeld, Stolley. Foundations of epidemiology. Third edition. Miguel Delgado-Rodrı´guez, Javier Llorca ; Bias; J Epidemiol Community Health ;2004 pg 1-7.