Bias in health research

Classification and common biases encountered in health research

1. R. VENKITACHALAM

2. Contents Introduction Classification of bias Bias in clinical trials Bias in qualitative research Case scenarios Summary and conclusion References

3. Introduction Bias is a fundamental concept in epidemiology It is defined as ‘deviation of results or inferences from the truth , or processes leading to such deviation’ - Grimes and Schulz, 2002 It is the result of errors  not random but systematic  results invalid Results mainly from faulty design

4. Scientifically speaking, bias can be explained as: lack of internal validity or incorrect assessment of the association between an exposure and an effect in the target population in which the statistic estimated has an expectation that does not equal the true value. Delgado-Rodríguez M, Llorca J. Bias. J Epidemiol Community Health. 2004 Aug 1;58(8):635–41.

5. Etymology mid 16th century (in the sense ‘oblique line’; also as an adjective meaning ‘oblique’): from French  biais , from Provençal, perhaps based on Greek  epikarsios  ‘oblique’.

6. Classification of bias Several classifications of bias exist in literature. Prominent among them are: Sackett (19 types) and Choi (65 types) – based on stages of research Maclure and Schneeweiss – causal diagram theory Kleinbaum et al – three main groups (selection, information and confounding) Steineck and Ahlbom – misclassification, misrepresentation and analysis deviation

7. Maclure and Schneeweiss episcope of bias

8. WHO classificatio n Selection bias Occurs from the manner in which study population is selected Most common type of bias in health research Seen in observational and analytical studies Ascertainment or information bias Occurs due to measurement error or misclassification of subjects according to one or more variables

9. WHO classification Health research methodology - A Guide for training in Research methods. Second edition. WHO Bias Selection bias Prevalence – incidence bias Admission rate bias Non-response bias Information bias Diagnostic bias Recall bias

10. How to reduce selection bias The study population should be clearly identified i.e. clear definition of study population. The choice of the right comparison/ reference group (unexposed or controls) is crucial In a cohort study : exposed and unexposed groups should be identical but for the exposure in a retrospective cohort study , the selection of exposed and unexposed groups should be done without knowing the outcome (disease status).

11. In a case-control study : the control group should reflect the exposure of the population which gave rise to the cases controls should be selected independently of the exposure status precise case definition and exposure definition should be used by all investigators. In a clinical trial: Randomization and allocation concealment from the investigator

12. How to reduce information bias Blinding Placebo arm in case of RCT’s

13. 1. SELECTION BIAS Error introduced when the study population does not include the target population Causes: Due to design Bad definition of eligible population Lack of accuracy of sampling frame Uneven diagnostic procedures in target population Due to implementation

14. 1.1 Inappropriate definition of eligible population Occurs when kind of patients gathered does not represent the cases originated in the population It is of the following types: 1.1.a – Competing risks bias When two or more outputs are mutually exclusive, any of them competes with each other in the same subject. Eg : death

15. 1.1.b – Healthcare access bias When patients admitted to an institution does not represented the cases originated in the community Popularity bias Centripetal bias Referral filter bias Diagnostic/treatment access bias 1.1.c – Length bias Cases with disease with longer duration are more easily included in surveys

16. 1.1.d - Neyman bias When a series of survivors is selected, if the exposure is related to prognostic factors, or the exposure itself is a prognostic determinant , the sample of cases offers a distorted frequency of exposure . 1.1.e – Spectrum bias In the assessment of validity of a diagnostic test, this bias is produced when researchers include only clear or definite cases, not representing the whole spectrum of disease. Also applicable for controls

17. 1.1.f – Survivor treatment selection bias In observational studies, patients who live longer have more probability to receive a certain treatment 1.1.g – Berkson bias When probability of hospitalization of cases and controls differ, and it is also influenced by exposure 1.1.h – Healthy worker effect Lower mortality observed in employed population compared to general population

18. 1.1.h – Inclusion bias When one or more conditions of controls are related with exposure. Frequency of exposure higher in control group. Seen commonly in hospital based studies 1.1.i – Exclusion bias When controls with conditions related to exposure are excluded where as cases with co-morbidities are included

19. 1.2 lack of accuracy of sampling frame 1.2.a – Non random sampling bias Results in non-representative sample 1.2.b – telephone random sampling bias Excludes some households from the sample . . Coverage issues

20. In systematic reviews and meta analysis, selection of samples (relevant studies) is most important 1.2.c – Citation bias More frequently cited, more easily found 1.2.d – Dissemination bias Biases in retrieval of information (language, reporting of results) 1.2.e – Post-hoc analysis Due to subgroup analysis which give misleading results 1.2.f – Publication bias When published reports do not represent the studies carried out on that association

21. 1.3 uneven diagnostic procedures In case control studies, if exposure influences diagnosis of disease, detection bias occurs 1.3.a – diagnostic suspicion bias Exposure is taken as a diagnostic criterion 1.3.b – Mimicry bias When benign conditions mimic clinically to the disease

22. 1.4 during study implementation 1.4.a – Loss to follow up Attrition/withdrawal is uneven in exposure and outcome categories  study results affected 1.4.b - Missing information bias Seen mostly in multivariate analysis Missing data affects study outcome 1.4.c – Non-response bias This type of bias is due to refusals to participate in a study. The individuals concerned are likely to be different from individuals who do participate . Non-respondents must be compared with respondents with regard to key exposure and outcome variables in order to ascertain the relative degree of non-response bias.

23. 2. INFORMATION BIAS It occurs during data collection. Three types of information bias are: Misclassification bias Ecological fallacy bias Regression to the mean bias

24. 2.1 Misclassification bias When sensitivity and/or specificity of the procedure to detect exposure and/or effect is not perfect, ie exposed/diseased subjects can be classified as non-exposed/non-diseased subjects and vice-versa Two types: Differential misclassification bias Non-differential misclassification bias

25. 2.1.a – Detection bias Seen in studies with follow-up (cohort, clinical trials) 2.1.b – Observer/Interviewer bias Knowledge of hypothesis, disease status, exposure status or intervention received can influence data recording Interviewers can influence errors into a questionnaire or guide the respondents to a particular answer.

26. 2.1.c – Recall bias If presence of disease influences the perception of its causes (rumination bias) or In a trial, if patients knows what they receive, it may influence their answers (participant expectation bias) 2.1.d – Reporting bias Participants can collaborate with researchers and give answers in the directions they perceive are of interest ( obsequiesness bias) Existence of a case triggers family information (family aggregation bias) Measures or sensitive questions that embarrass or hurt can be refused Reporting of socially undesirable behaviours (underreporting bias)

27. 2.2 Ecological fallacy It is produced when analysis realized in an ecological (group level) analysis are used to make inferences at the individual level. Eg : if exposure and disease are measured at group level, exposure disease relations can be biased from those obtained at the individual level

28. 2.3 Regression to the mean It is a phenomenon that a variable that shows an extreme value on its first assessment will tend to be closer to the centre of its distribution on a later measurement . Eg : high cholesterol level measurement

29. 2.4 Other information biases 2.4.a – Hawthorne effect Increase in outcome under study in participants who are aware of being observed. 2.4.b – Lead time bias The added time of illness produced by the diagnosis of a condition during its latency period.

30. 2.4.c – protopathic bias When an exposure is influenced by early (subclinical) stages of disease Sick quitter bias: People with risky behaviours ( eg : alcohol consumption) quit their habit as a consequence of disease . . Which will mention them as non-exposed 2.4.d – temporal ambiguity When it cannot be established that exposure precedes effect. Seen in cross-sectional and ecological studies

31. 2.4.e – Will Rogers phenomenon Improvement in diagnostic tests refines disease staging in diseases such as cancer It is seen when survival rates are measured across time and even among centres with different diagnostic capabilities 2.4.f – Work-up bias (verification bias) In the assessment of validity of a diagnostic test, it is produced when the execution of gold standard is influenced by the results of the assessed test, ie reference test is less frequently performed when the test result is negative

32. 3. Confounding It occurs when a variable is a risk factor for an effect among non-exposed persons and is associated with the exposure of interest in the population from which the effect derives, without being affected by the exposure or the disease Confounding can occur in every epidemiological study Susceptibility bias is a synonym

33. Confounding can be neutralised at the design stage of a research (for example, by matching or randomisation ) and/or at the analysis, given that the confounders have been measured properly 3.1 - Confounding by group: I t is produced in an ecological study, when the exposure prevalence of each community ( group) is correlated with the disease risk in non-exposed of the same community

34. 3.2 - Confounding by indication T his is produced when an intervention (treatment) is indicated by a perceived high risk , poor prognosis, or simply some symptoms. Here the confounder is the indication, as it is related to the intervention and is a risk indicator for the disease

35. 4. Specific biases in trials 4.1 – allocation of intervention bias Seen in non- randomised trials when sequence of allocation is known in advance or concealment is unclear/inadequate 4.2 – compliance bias In trials requiring adherence to intervention, the degree of adherence (compliance) influences efficacy assessment of the intervention

36. 4.3 – Contamination bias When intervention-like activities find their way into control group Seen in community based trials due to relationships among members and interference by mass media etc 4.4 – Lack of intention to treat analysis In RCT’s, analysis should be done keeping participants in the group originally assigned to. Otherwise, bias results . . .

37. Publication bias Scientific journals are most likely to accept studies that have ‘positive findings’ than those with ‘negative findings’. Creates false impression in the literature and may cause long-term consequences to the scientific community To overcome this bias, several journals have been launched which publish only negative findings. Eg : Journal of Pharmaceutical Negative Results, Journal of Negative Results in Biomedicine, Journal of Interesting Negative Results

38. BIAS in clinical trials

39. Other biases in RCT’s Planning phase During conduct During Reporting During dissemination of results During uptake of results

40. During planning phase Choice of question bias Regulation bias Wrong design bias

41. During planning phase Choice of question bias Hidden agenda bias Self-fulfilling prophecy bias Cost and convenience bias Funding availability bias Regulation bias Wrong design bias

42. During planning phase Choice of question bias Regulation bias IRB/Bureaucracy bias Complicated informed consent Wrong design bias

43. During conduct of trial Population choice bias Intervention choice bias Outcome choice bias

44. During conduct of trial Population choice bias Gender bias Age bias Pregnancy bias Special circumstances bias Informed consent / literary bias Intervention choice bias Outcome choice bias

45. During conduct of trial Population choice bias Intervention choice bias Too early bias Too late bias Learning curve bias Complexity bias Outcome choice trial

46. During conduct of trial Population choice bias Intervention choice bias Outcome choice trial Measurement bias Time term bias

47. During reporting of trial Withdrawal bias Selective reporting bias Fraud bias

48. During reporting of RCT Withdrawal bias Selective reporting bias Social desirability bias Data dredging bias Interesting data bias Fraud bias

49. During dissemination of results Publication bias Language bias Country of publication bias Time lag bias

50. During uptake of RCT Prominent author bias Trial design bias Complimentary medicine bias Flashy title bias Careless reading bias

51. During uptake of RCT Prominent author bias Esteemed author Esteemed professor Friendship Trial design bias Complimentary medicine bias Flashy title bias Careless reading bias

52. During uptake of RCT Prominent author bias Trial design bias Favoured design bias Large trial bias Small trial bias Multicentre trial bias Complimentary medicine bias Flashy title bias Careless reading bias

53. BIAS in qualitative research

54. BIAS in qualitative research In qualitative research, bias affects validity and reliability of findings and thus the results Categories of biases seen in qualitative research are: Moderator bias Biased questions Biased answers Biased sampling Biased reporting

55. Moderator bias The moderator’s facial expressions, body language, tone, manner of dress, and style of language may introduce bias . Similarly, the moderator’s age, social status, race, and gender can produce bias . Some conditions unavoidable, still some can be controlled

56. Biased questions Leading question bias Misunderstood question bias U nanswerable question bias Question order bias. Always ask: general questions before specific questions unaided before aided questions positive questions before negative questions behavior questions before attitude questions

57. Biased answers Consistency bias Dominant respondent bias Error bias Hostility bias Moderator acceptance bias Mood bias Overstatement bias Reference bias Sensitivity bias Social acceptance bias Sponsor bias

58. Biased sampling Due to errors in sampling. Professional respondents should be avoided Biased reporting Experiences , beliefs, feelings, wishes, attitudes, culture, views, state of mind, reference, error, and personality can bias analysis and reporting.

59. Case scenarios

60. Scenario 1 Investigators recruited both cases and controls from a defined catchment area in the general population. This is often difficult to do in the absence of a comprehensive registry. Suppose investigators had recruited all cases of oral cancer from a comprehensive national registry, between August 1, 2013 to July 31, 2014. What bias might be intoduced if controls were obtained from the catchment area? Selection Bias

61. Scenario 2 Suppose you are designing a case-control study on association between smoking and oral cancer. Which of the following methods of acquiring cases and controls is most practical to reduce bias? Scenario 1: Cases from AIMS (tertiary hospital) and controls from xyz dental clinic Scenario 2: Cases from AIMS and controls from admission office of AIMS with diagnosis of any other cancer other than oral cancer Scenario 3: Cases from AIMS and controls from Panangad satellite clinic

62. Scenario 3 What potential bias could have been introduced if you found out that those who interviewed cases took 30 minutes longer on average than those who interviewed controls? Selection bias Information bias Volunteer bias Loss to follow up bias

63. Scenario 4 A study reported a significant association between long-term use of smoking and oral cancer compared to no smoking and oral cancer. The duration of exposure varied among both cases and controls. What bias could arise when trying to measure exposures that happened over different time periods? Misclassification bias Measurement bias Recall bias

64. Scenario 5 What effect would you observe if the interviewers were aware of the disease status of the study subjects? It would benefit the validity of results since interviewer would understand more precisely of the disease and collect better data for cases The results would likely not change It could damage the results by introducing interviewer bias

65. SUMMARY Bias is any trend or deviation in truth or data collection, analysis, interpretation and publication which can cause false conclusions Occurs either intentionally or unintentionally Due to consequences of bias, it is unethical to conduct and publish a biased research even unintentionally

66. Confounding effect cannot be completely avoided. Every researcher should therefore be aware of all potential sources of bias and undertake all possible actions to reduce and minimize the deviation from the truth . If deviation is still present, authors should confess it in their articles by declaring the known limitations of their work.

67. References 1. 8.4 Introduction to sources of bias in clinical trials [Internet]. [cited 2014 Dec 22]. Available from: http:// handbook.cochrane.org/chapter_8/8_4_introduction_to_sources_of_bias_in_clinical_trials.htm 2. 9781405132664_4_003.indd - 9781405132664_4_003.pdf [Internet]. [cited 2014 Dec 22]. Available from: http:// www.blackwellpublishing.com/content/BPL_Images/Content_store/Sample_chapter/9781405132664/9781405132664_4_003.pdf 3. Delgado-Rodríguez M, Llorca J. Bias. J Epidemiol Community Health. 2004 Aug 1;58(8):635–41. 4. bias.aspx [Internet]. [cited 2014 Dec 22]. Available from: http:// www.ashpfoundation.org/mainmenucategories/researchresourcecenter/fosteringyounginvestigators/ajhpresearchfundamentalsseries/bias.aspx 5. Sackett DL. Bias in analytic research. J Chronic Dis. 1979;32(1-2):51–63.

68. 6. Kopec JA, Esdaile JM. Bias in case-control studies. A review. J Epidemiol Community Health. 1990 Sep;44(3):179–86. 7. FEM - Preventing bias [Internet]. [cited 2014 Dec 22]. Available from: https://wiki.ecdc.europa.eu/fem/w/fem/preventing-bias.aspx 8 . Pannucci CJ, Wilkins EG. Identifying and Avoiding Bias in Research. Plast Reconstr Surg. 2010 Aug;126(2):619–25. 9 . Qualitative Research Bias - How to Recognize It [Internet]. [cited 2014 Dec 22]. Available from: http://www.focusgrouptips.com/qualitative-research.html 10 . Types of bias.pdf [Internet]. [cited 2014 Dec 22]. Available from: http://www.medicalbiostatistics.com/Types%20of%20bias.pdf