Lecture 10 Bias and Confounders (2024-2025)

Errors, Bias & Confounders Basic Infection Control Measures 1st semester / Year 3 2024-2025 Dr. Sanabel Barakat DDMS. MSc. PhD. Week 10 JCD.  ILOs 1. Understand how errors occur in epide...

Errors, Bias & Confounders Basic Infection Control Measures 1st semester / Year 3 2024-2025 Dr. Sanabel Barakat DDMS. MSc. PhD. Week 10 JCD.  ILOs 1. Understand how errors occur in epidemiological studies 2. Identify type1 and type2 errors 3. Define bias and confounders 4. Explore different types of bias 5. Learn how to avoid bias in epidemiological studies 2  Introduction The two central goals in epidemiological studies are 1- Obtaining a precise estimate of the true value of the population parameter 2- Establishing causal association between exposures and outcomes, both may be threatened by different errors that may arise in almost any part or stage of epidemiological studies 3  Introduction Every study is subject to error. These errors may arise from Random variations Biological variations Errors in measurement Errors in study design and planning lack of perfect knowledge about the phenomenon under study that could contribute to the various types of errors. 4  Errors in epidemiological studies In epidemiological studies, errors can be Random. Systematic Confounders Type 1 & Type 2 errors 5  Random Error and Bias 6  Random Error and Bias Errors and biases may impact any part of a study. In every step of study design, analysis, and interpretation (solid arrows), there is scope for errors and biases (dotted arrows) resulting from investigator bias, random error, and various study biases. A research question may be set up in such a way that it does not directly answer the intended question and as a result of various errors and biases it obfuscates the truth. Maximizing strengths (+) and minimizing threats (–) at different stages impact making correct inferences. In this trade-off, we try to minimize errors so that they are not large enough to change the conclusions in important ways. Only if the study is able to arrive at a correct inference can the truth about the issue be unearthed. 7  1. Random Errors Are those that affect equally the participants, irrespective of being exposed to or affected by any other condition. These errors may occur roughly equally between exposed and unexposed (or case/control) groups (non-differential errors) They occurs as random variation or due to chance Its potential effect of modifying the analysis is lower than that of systematic errors. 8  1. Random Errors Random errors affect the accuracy of the studies, and overcoming it eventually demands to increase the sample size and the commitment to improving the quality of the measurements. These are unpredictable errors and therefore it will be difficult to find a method to adjust for such errors Random Errors should be avoided and duly considered in the research procedure 10  Management of Random Error Random error in a study may be corrected by improving 1-sampling and 2-study design, 3- increasing sample size, 4- reducing measurement variability in instruments, 5- by using strict measurement criteria (e.g., use of regularly calibrated accurate instruments, or using averages of multiple measurement). 6- During study implementation, it may be minimized using stringent quality control measures. 7- Use of appropriate and more efficient analytical methods can also help minimize random errors. 11  Management of Random Error Random error can also be perceived as the inverse of precision. In epidemiology, most random error is generated from sample selection (sampling error); the overall goal of studies is to minimize random errors by a sampling method that would tend to equalize (at least in theory) the distribution of unknown factors in cases and controls or among the exposed and unexposed groups. 12 2- Systematic Errors Bias Occur when there is some factor that modifies the results, and this factor is more prevalent among some group of participants in the study, for instance, those affected by the disease or those exposed to some other relevant factor. i.e. These errors may occur differently between comparison groups (differential error).  2- Systematic Errors Bias When systematic errors affect the collected data, the magnitude of estimated associations can change significantly, and the researcher has no control over this process. For example, if in a case-control study the information regarding past exposures is obtained in a face-to-face interview with the cases, but through the phone with the controls, it is possible that the recollection of these exposures is more accurate in a group than in the other 14  2- Systematic Errors Bias These are predictable or systematic (bias). “Bias” can be defined as any pattern in the gathering, description, analysis, interpretation, publication, or review of the data, whose potential effect is to induce different conclusions of the reality The most frequent systematic errors that can reduce the validity of epidemiological studies are: Selection bias, Observation bias (information or measurement), 15  Systematic Errors Bias There are three important, somewhat related bias “effects” that may impact epidemiological studies. Careful attention must be paid to avoid these sources of errors. The Pygmalion Effect The Placebo Effect The Hawthorne Effect 18  The Pygmalion Effect This “self-fulfilling prophecy,” also called the Rosenthal effect, occurs when some participants perform much better than others in certain situations just because they were expected to perform better in those situations (Rosenthal & Jacobson, 1992). 19  The Placebo Effect A placebo is a noneffective drug. Some patients may respond to a placebo even if no active drug is given to them, perhaps because the patient believes that the given drug works. 20  The Hawthorne effect This effect implies that individuals will change their behavior if they are aware they are being observed. The “observer effect” refers to changes that the mere act of observation induces on the phenomenon being observed. 21  A- Selection Bias Berkson’s bias Occurs especially in case control studies Selection biases occur when (i) participants are erroneously classified with respect to some characteristic of interest (ii) participants are preferably enrolled in the study according to some characteristics (iii) unequal allocation of individuals in the sample; It can be controlled by using careful control selection criteria. 22  A- Selection Bias Loss to follow-up Occurs especially in cohort studies. Those missing follow-up appointments may be sicker, or have greater exposure to one or more risk factors compared to those who attend regularly. For example, in examining oral and systemic disease linkages in ambulatory settings, those who develop serious conditions may be admitted to the hospital and may not report for their appointments. 23  A- Selection bias Nonresponse bias Occurs when study participants with certain characteristics selectively do not participate in the study. The bias may be directly proportional to response rate and can be reduced by increasing the response rate. 24  A- Selection bias Membership bias Involves those people who choose to be members of a group with shared attributes (e.g., gym users) and they might differ from others in important ways. For example, such people are more health conscious, have better oral hygiene habits, and eat a less sugary diet. 25  B- Information bias Recall bias Occurs due to inaccurate recall of past exposures or disease status details. When people try to recall remote exposure histories, cases may search their memories more deeply and intently compared to controls leads to better recall among cases. Occurs in case control studies 27  B- Information bias Interviewer bias Occurs when interviewers are not blinded to case status of the participant, and try to seek replies more deeply among cases by “clarifying” questions or seeking clearer explanations from cases as compared to controls. Blinding of interviewers to case status (to the best extent possible) can minimize interviewerbias. 28  B- Information bias Observer bias Similar to interviewer bias, occurs especially in cohort studies and clinical trials when the observers may seek outcomes more deeply among the exposed compared to the unexposed group. Blinding of observers to outcome`s definition (or hypothesis) (to the best extent possible) can minimize bias. 29  B- Information Bias Respondent bias Occurs when the outcome information is dependent upon participants’ assessment. For example, ascertaining diagnoses of pain, This is dependent on participants’ observed report and may be subject to inaccuracies compared to an examiner- defined clinical outcome. 30  Temporal bias Occurs when conclusions are based on inaccurate temporal sequences of cause and effect. This occurs mostly in cross-sectional and case-control studies when it is difficult to ascertain the temporal sequence of cause and effect. 35  Length bias Occurs in diseases that have a long detectable preclinical phase. Active screening may detect more cases in earlier preclinical phases compared to non-screened diagnosed cases (which might be at later phases). Effectiveness of screening programs should consider the potential effect of length bias. 36  Lead time bias Lead time is the time by which diagnosis can be advanced by early detection and screening compared to the usual time when diagnosis is made. Early detection and treatment may impose a false sense of increased survival rate, when in fact the survival time increase would be mainly due to the extra time accrued through early diagnosis and not a better posttreatment success. 37 38  Compliance bias Occurs due to differential compliance to treatment; for example, taking a once a day dose vs multiple dosages a day. 39  Remember In contrast to random errors, biases are errors that are systematic deviations from the truth. Therefore, essentially, biases are predictable. Although relatively easy to intellectualize, estimation and quantification of biases can be challenging. Bias can be described as inaccuracies that occur in one group and not the other. For example, unpredictable error in sampling would be random error, but selectively choosing a sample of a certain type may lead to a biased sample. Biases threaten the validity of the study. 42  Control of Bias Bias is mostly controlled at the design stage before the study starts and/or at the implementation stage of the study. Bias can be minimized by using various techniques such as valid data collection, disease definition, and information collection tools; blinding of data-analyst, interviewer, and clinical investigator to the case/exposure status of the participant; validation of case definition, diagnosis, and exposure data; minimizing time between exposure/event and data collection so that participants do not have to recall events far off in time; 43  Control of Bias Using ancillary questions to check responses to important questions that may be subject to recall bias; verifying answers to questions with objective measures or clinical records; measuring exposure and outcomes at the same level using standardized calibration methods to improve within- and inter- examiner agreement; setting up periodic procedure reviews; Well designed well conducted monitoring procedures and study protocols. 44  confounding variable A confounding variable is defined as a Characteristic of the observation units, when it is associated with both of the exposure and health outcomes, but it is not an intermediary path between the possible cause and effect. When the estimates of the association between two factors can be attributed, entirely or partially, to a third factor not taken into consideration, this third factor is considered a confounder. 47  confounding variable A confounding variable is defined as a There are some strategies to control confounding in epidemiological studies, such as restriction and matching, and statistical techniques, such as stratification and multivariable analysis. In case these strategies have not been applied or the study has not been planned adequately, the effect of confounding may affect the conclusion of the study. 48  Effects of confounding variables Positive confounding occurs when the confounder leads to overestimation of the real strength of the association Negative confounding occurs when confounder leads to underestimation of the effect. The inversion of the association, which can occur, when the crude analysis indicates risk and the adjusted assessment indicate protection (or vice versa) 49  Closing remarks In general, biases, errors and confounders may be minimized through improved design, better accuracy, and strict quality control procedures. Bias cannot be corrected by sample size adjustment. An epidemiologic study should try to balance the threats and strengths within feasible limits while recognizing that “perfect” studies are utopias. In this trade-off, one tries to minimize errors so that they are not large enough to change the conclusions in important or meaningful ways. 50  Type 1 & Type 2 Errors Sometimes, by chance alone, a sample is not representative of the population. Thus the results in the sample do not reflect reality in the population, and the random error leads to an erroneous inference. A type I error (false-positive) occurs if an investigator rejects a null hypothesis that is actually true in the population; a type II error (false-negative) occurs if the investigator fails to reject a null hypothesis that is actually false in the population. 51  Type 1 & Type 2 Errors Although type I and type II errors can never be avoided entirely, the investigator can reduce their likelihood by increasing the sample size (the larger the sample, the less is the likelihood that it will differ substantially from the population). False-positive and false-negative results can also occur because of bias (observer, instrument, recall, etc.). (Errors due to bias, however, are not referred to as type I and type II errors.) 52  Making a statistical decision always involves uncertainties, so the risks of making these errors are unavoidable in hypothesis testing. 53  In statistics, a Type I error is a false positive conclusion, while a Type II error is a false negative conclusion. The probability of making a Type I error is the significance level, or alpha (α), while the probability of making a Type II error is beta (β). These risks can be minimized through careful planning in your study design. 54