The effects of joint misclassification of exposure and disease on epidemiologic measures of association exposure and disease on epidemiologic measures of association

doi:10.1016/0895-4356(93)90119-L

Journal of Clinical Epidemiology

Volume 46, Issue 10, October 1993, Pages 1195-1202

https://doi.org/10.1016/0895-4356(93)90119-L Get rights and content

Abstract

This paper addresses the effects of simultaneous misclassification of both exposure and disease on epidemiologic measures of association. If misclassification of a dichotomous exposure is independent of a dichotomous disease status and vice versa (non-differential misclassification), and misclassification of exposure is independent of misclassification misdisease, then the bias is always toward the null. In practice, however, errors in exposure and disease ascertainment may often be correlated. In this case, the observed exposure-disease association may be strongly biased in any direction even with non-differential misclassification. As an important corollary, the assertion commonly made in the discussion of epidemiologic study results that the observed measures of association can only be biased toward the null due to presumedly non-differential misclassification has to be viewed as inadequate unless the assertion that exposure and disease misclassification are independent is also justified. Inferences regarding the degree and direction of bias due to misclassification of exposure and disease should consider plausible degrees of correlation in classification errors in addition to the overall misclassification rates. Whenever possible, sensitivity analyses should be performed to provide a quantitative basis for such inferences.

References (21)

K.T. Copeland et al.
Bias due to misclassification in the estimate of relative risk
Am J Epidemiol
(1977)
K.M. Flegal et al.
The effects of exposure misclassification on estimates of relative risk
Am J Epidemiol
(1986)
K.J. Rothman
Modern Epidemiology
(1986)
S.D. Walter et al.
Estimation of test error rates, disease prevalence and relative risk from misclassified data: a review
J Clip Epidemiol
(1988)
D.G. Kleinbaum et al.
Epidemiologic Research. Principles and Quantitative Methods
(1982)
A.M. Walker et al.
Analysis of case-control data derived in part from proxy respondents
Am J Epidemiol
(1988)
R.E. Dales et al.
Respiratory health effects of home dampness and molds among Canadian children
Am J Epidemiol
(1991)
P. Kristensen
Bias from nondifferential but dependent misclassification of exposure and outcome
Epidemiology
(1992)
M. Chavange et al.
Correlated nondifferential misclassifications of disease and exposure: application to a cross-sectional study of the relation between handedness and immune disorders
Int J Epidemiol
(1992)
S. Greenland et al.
Confounding and misclassification
Am J Epidemiol
(1985)

There are more references available in the full text version of this article.

Cited by (40)

Awareness of and potential for dependent error in the observational epidemiologic literature: A review
2019, Annals of Epidemiology
Citation Excerpt :
Removed articles were replaced via random sampling of the remaining screened articles to yield a final 100 articles meeting inclusion criteria (Supplement Figure 1). We used Google scholar to identify studies published between January 1992 and July 2016 citing one of two prominent research articles by Kristensen [9] and Brenner [10] on dependent misclassification. After initial review to include only English language, peer-reviewed publications, full-text review using the same inclusion and exclusion criteria as the random sample, yielded a final sample of 39 articles (Supplement, Methods & Figure 1).
Measurement error discussions often assume classification errors of key variables are independent. Yet, small amounts of dependent error can create large biases in effect estimates. The purpose of this review was to evaluate frequency of measurement error discussions and potential for dependent error in the observational literature.
Two samples of articles analyzing exposure-outcome contrasts were collected: a random sample (n = 100) from high-impact epidemiology and medical journals (June 2015–July 2016), and a citation-based sample (n = 39) of studies citing one of two prominent dependent misclassification articles (through July 2016). We extracted study details, recorded measurement error mentions, and qualitatively assessed dependent error potential.
Measurement error was often discussed. No random sample articles explicitly mentioned dependent error, compared with 59% of the citation-based sample. The random sample was found to be at low risk of exposure-outcome (15% plausible/probable) but increased risk for exposure-confounder (38% plausible/probable) dependency. The citation-based sample was at higher risk for dependent error (exposure-outcome: 46% plausible/probable; exposure-confounder: 61% plausible/probable).
Although measurement error was frequently mentioned, potential impact on observed results was rarely discussed in-depth or quantified. Dependent error mentions were rare, even among studies deemed susceptible. Further education and steps to avoid dependent error are needed.
Adjusting for multiple-misclassified variables in a study using birth certificates
2013, Annals of Epidemiology
Citation Excerpt :
Adjustment for exposure-related selection bias is straightforward algebraically; assuming classification occurs before selection, one simply lets fC and fN vary with exposure (becoming fCj and fNj in the initial selection-bias adjustment), but reliable information on this variation would rarely be available. A number of methods for adjustment and sensitivity analysis can be applied when the exposure and outcome classification errors are dependent [16,18,32–36]. For example, the matrix method we used [16,18] could be applied, although the matrix entries (Table 2) would no longer be simply products of the separate exposure and outcome classification probabilities.
Birth certificates are a convenient source of population data for epidemiologic studies. It is well documented, however, that birth certificate data can be highly inaccurate. Nonetheless, studies based on birth certificates are routinely analyzed without accounting for sources of data errors. We focused on the association between maternal cigarette smoking and cleft lip and palate based on birth certificate data.
We adjusted odds ratio estimates simultaneously for exposure and outcome misclassification. We also calculated odds ratios adjusted for exposure misclassification only and outcome misclassification only.
Adjustment for both maternal smoking during pregnancy and clefting resulted in adjusted odds ratios that ranged from less than 1.0 to much greater than the unadjusted estimate of 1.16, with most adjusted estimates outside of the 95% confidence limits (1.01, 1.33).
Because of the potentially large impact of birth certificate classification errors, we suggest that inferences from these or similar records employ quantitative methods for incorporating uncertainties caused by data errors.
Tradeoffs between accuracy measures for electronic health care data algorithms
2012, Journal of Clinical Epidemiology
Citation Excerpt :
One of the primary reasons to prioritize one measure of algorithm accuracy over another is to reduce bias (or distortion) in the risk estimate (i.e., the magnitude of the association between the exposure and the health outcome). The relationship between misclassification and bias of risk estimates is complex, however [44–50]. We will not explore the literature in detail, although we note several factors that may make it difficult to determine how misclassification will affect the estimate of the association between the exposure and outcome:
We review the uses of electronic health care data algorithms, measures of their accuracy, and reasons for prioritizing one measure of accuracy over another.
We use real studies to illustrate the variety of uses of automated health care data in epidemiologic and health services research. Hypothetical examples show the impact of different types of misclassification when algorithms are used to ascertain exposure and outcome.
High algorithm sensitivity is important for reducing the costs and burdens associated with the use of a more accurate measurement tool, for enhancing study inclusiveness, and for ascertaining common exposures. High specificity is important for classifying outcomes. High positive predictive value is important for identifying a cohort of persons with a condition of interest but that need not be representative of or include everyone with that condition. Finally, a high negative predictive value is important for reducing the likelihood that study subjects have an exclusionary condition.
Epidemiologists must often prioritize one measure of accuracy over another when generating an algorithm for use in their study. We recommend researchers publish all tested algorithms—including those without acceptable accuracy levels—to help future studies refine and apply algorithms that are well suited to their objectives.
A review of five cardiology journals found that observer variability of measured variables was infrequently reported
2008, Journal of Clinical Epidemiology
Citation Excerpt :
Consequently, rigorous research design incorporates planned assessments of both interobserver and intra-observer variability where applicable [2], so that interpretation of research results may be understood in the context of the reliability of the measures used [3]. Observer variability may affect both the performance and the interpretation of a test, and contributes to participant misclassification, underestimation of sample size requirements, as well as to the magnitude of effect in cohort, case-control, method comparison, and diagnostic accuracy studies [4–8]. The clinical consequences of such disagreement have been recognized for more than 50 years [9].
To investigate the reporting of the analysis of interobserver and intra-observer variability within clinical research studies from five high-impact cardiology journals published in 2005.
A cross-sectional study using a combined electronic and manual search identified 180 of 511 eligible articles that reported the assessment of observer variability. Sixty of these were randomly selected for detailed review.
The proportion of the 60 studies reporting interobserver variability, intra-observer variability, or both were 27%, 17%, and 53%, respectively. The reported methodological design of interobserver and intra-observer analyses included a specific protocol in 42% and 33%, identified observers as independent in 31% and 17%, as blinded in 50% and 31%, and identified a prior statistical plan in only 33% and 36%, respectively. Pearson correlation was the most reported measure for continuous variables, and the methods of Bland and Altman were reported in 15% of interobserver and 14% of intra-observer studies, respectively. For categorical variables, a kappa statistic was reported in 82% and 80%, respectively.
Reliability assessment is hampered by unclear and incomplete reporting of interobserver and intra-observer analysis. For continuous variables, inappropriate methods were most frequently reported as being done.
Exact inference for the risk ratio with an imperfect diagnostic test
2017, Epidemiology and Infection
Fundamentals of Epidemiology
2024, Fundamentals of Epidemiology

View all citing articles on Scopus

View full text

Original articleThe effects of joint misclassification of exposure and disease on epidemiologic measures of association exposure and disease on epidemiologic measures of association

Abstract

Bias due to misclassification in the estimate of relative risk

Am J Epidemiol

The effects of exposure misclassification on estimates of relative risk

Am J Epidemiol

Modern Epidemiology

Estimation of test error rates, disease prevalence and relative risk from misclassified data: a review

J Clip Epidemiol

Epidemiologic Research. Principles and Quantitative Methods

Analysis of case-control data derived in part from proxy respondents

Am J Epidemiol

Respiratory health effects of home dampness and molds among Canadian children

Am J Epidemiol

Bias from nondifferential but dependent misclassification of exposure and outcome

Epidemiology

Correlated nondifferential misclassifications of disease and exposure: application to a cross-sectional study of the relation between handedness and immune disorders

Int J Epidemiol

Confounding and misclassification

Am J Epidemiol

Original article
The effects of joint misclassification of exposure and disease on epidemiologic measures of association exposure and disease on epidemiologic measures of association