Adjustment for non-differential misclassification error in the generalized linear model

It is well known that estimates of association between an outcome variable and a set of categorical covariates, some of which are measured with misclassification, tend to be biased upon application of the usual methods of estimation that ignore the classification error. We propose a method to adjust for misclassification in covariates when one applies the generalized linear model. In the case where one can observe some true covariates only through surrogates, we combine a latent class analysis with the approach to incorporate multiple surrogates into the model. We include discussion on the efficacy of repeated measurements which one can view as a special case of multiple surrogates with identical distribution. We provide two examples to demonstrate the applicability of the method and the efficacy of multiple replicates for a covariate subject to misclassification in a regression framework.     

Proper interpretation of non-differential misclassification effects: expectations vs observations

BACKGROUND: Many investigators write as if non-differential exposure misclassification inevitably leads to a reduction in the strength of an estimated exposure-disease association. Unfortunately, non-differentiality alone is insufficient to guarantee bias towards the null. Furthermore, because bias refers to the average estimate across study repetitions rather than the result of a single study, bias towards the null is insufficient to guarantee that an observed estimate will be an underestimate. Thus, as noted before, exposure misclassification can spuriously increase the observed strength of an association even when the misclassification process is non-differential and the bias it produced is towards the null. METHODS: We present additional results on this topic, including a simulation study of how often an observed relative risk is an overestimate of the true relative risk when the bias is towards the null. RESULTS: The frequency of overestimation depends on many factors: the value of the true relative risk, exposure prevalence, baseline (unexposed) risk, misclassification rates, and other factors that influence bias and random error. CONCLUSIONS: Non-differentiality of exposure misclassification does not justify claims that the observed estimate must be an underestimate; further conditions must hold to get bias towards the null, and even when they do hold the observed estimate may by chance be an overestimate.     

Quantifying bias due to allele misclassification in case-control studies of haplotypes

OBJECTIVES: Genotyping errors can induce biases in frequency estimates for haplotypes of single nucleotide polymorphisms (SNPs). Here, we considered the impact of SNP allele misclassification on haplotype odds ratio estimates from case-control studies of unrelated individuals. METHODS: We calculated bias analytically, using the haplotype counts expected in cases and controls under genotype misclassification. We evaluated the bias due to allele misclassification across a range of haplotype distributions using empirical haplotype frequencies within blocks of limited haplotype diversity. We also considered simple two- and three-locus haplotype distributions to understand the impact of haplotype frequency and number of SNPs on misclassification bias. RESULTS: We found that for common haplotypes (>5% frequency), realistic genotyping error rates (0.1-1% chance of miscalling an allele), and moderate relative risks (2-4), the bias was always towards the null and increases in magnitude with increasing error rate, increasing odds ratio. For common haplotypes, bias generally increased with increasing haplotype frequency, while for rare haplotypes, bias generally increased with decreasing frequency. When the chance of miscalling an allele is 0.5%, the median bias in haplotype-specific odds ratios for common haplotypes was generally small (<4% on the log odds ratio scale), but the bias for some individual haplotypes was larger (10-20%). Bias towards the null leads to a loss in power; the relative efficiency using a test statistic based upon misclassified haplotype data compared to a test based on the unobserved true haplotypes ranged from roughly 60% to 80%, and worsened with increasing haplotype frequency. CONCLUSIONS: The cumulative effect of small allele-calling errors across multiple loci can induce noticeable bias and reduce power in realistic scenarios. This has implications for the design of candidate gene association studies that utilize multi-marker haplotypes.     

Full-likelihood approaches to misclassification of a binary exposure in matched case-control studies

We consider analysis of matched case-control studies where a binary exposure is potentially misclassified, and there may be a variety of matching ratios. The parameter of interest is the ratio of odds of case exposure to control exposure. By extending the conditional model for perfectly classified data via a random effects or Bayesian formulation, we obtain estimates and confidence intervals for the misclassified case which reduce back to standard analytic forms as the error probabilities reduce to zero. Several examples are given, highlighting different analytic phenomena. In a simulation study, using mixed matching ratios, the coverage of the intervals are found to be good, although point estimates are slightly biased on the log scale. Extensions of the basic model are given allowing for uncertainty in the knowledge of misclassification rates, and the inclusion of prior information about the parameter of interest.     

The effect of non-differential outcome misclassification on estimates of the attributable and prevented fraction

This paper considers the effect of non-differential outcome misclassification on the population attributable fraction and the population prevented fraction. I examine the bias in the attributable and the prevented fraction derived from a risk ratio estimate as a function of the sensitivity and specificity of the outcome classification, the true risk ratio, the prevalence of the exposure, and the baseline disease frequency. With outcome misclassified, disease frequency is an important determinant of the magnitude of the bias; the rarer the disease, the more severe is the bias. For both the attributable and the prevented fraction, the specificity of the outcome classification has a greater influence on the magnitude of the bias than the sensitivity; this is in contrast to the dominant effect of sensitivity in situations of exposure misclassification. Also, unlike the findings in the exposure misclassification, the bias due to outcome misclassification does not increase monotonically with increased prevalence of exposure. For the attributable and prevented fraction derived from an odds ratio estimate, the specificity of the outcome classification does not have a greater influence on bias than the sensitivity, and a perfect specificity alone does not lead to unbiased effect estimates if the sensitivity of the outcome classification is imperfect.     

Adjusting for differential misclassification in matched case-control studies utilizing health administrative data

In epidemiological studies of secondary data sources, lack of accurate disease classifications often requires investigators to rely on diagnostic codes generated by physicians or hospital systems to identify case and control groups, resulting in a less-than-perfect assessment of the disease under investigation. Moreover, because of differences in coding practices by physicians, it is hard to determine the factors that affect the chance of an incorrectly assigned disease status. What results is a dilemma where assumptions of non-differential misclassification are questionable but, at the same time, necessary to proceed with statistical analyses. This paper develops an approach to adjust exposure-disease association estimates for disease misclassification, without the need of simplifying non-differentiality assumptions, or prior information about a complicated classification mechanism. We propose to leverage rich temporal information on disease-specific healthcare utilization to estimate each participant's probability of being a true case and to use these estimates as weights in a Bayesian analysis of matched case-control data. The approach is applied to data from a recent observational study into the early symptoms of multiple sclerosis (MS), where MS cases were identified from Canadian health administrative databases and matched to population controls that are assumed to be correctly classified. A comparison of our results with those from non-differentially adjusted analyses reveals conflicting inferences and highlights that ill-suited assumptions of non-differential misclassification can exacerbate biases in association estimates.     

Accounting for error due to misclassification of exposures in case-control studies of gene-environment interaction

We consider analysis of data from an unmatched case-control study design with a binary genetic factor and a binary environmental exposure when both genetic and environmental exposures could be potentially misclassified. We devise an estimation strategy that corrects for misclassification errors and also exploits the gene-environment independence assumption. The proposed corrected point estimates and confidence intervals for misclassified data reduce back to standard analytical forms as the misclassification error rates go to zero. We illustrate the methods by simulating unmatched case-control data sets under varying levels of disease-exposure association and with different degrees of misclassification. A real data set on a case-control study of colorectal cancer where a validation subsample is available for assessing genotyping error is used to illustrate our methods.     

The effect of non-differential exposure misclassification on estimates of the attributable and prevented fraction

This paper considers the effect of non-differential exposure misclassification on the population attributable fraction and the population prevented fraction as a function of the sensitivity and specificity of the exposure classification, the true relative risk, and the true prevalence of the exposure. Given a certain set of sensitivity, specificity, and prevalence of the exposure, the apparent attributable fraction is a constant proportion of the true attributable fraction regardless of the true relative risk. This observation does not hold for the apparent prevented fraction and the apparent relative risk, both of which vary with the true relative risk. For both the attributable and the prevented fraction, the sensitivity of the exposure classification has a greater influence on the magnitude of the bias than the specificity; also, the higher the prevalence of the exposure, the larger is the bias caused by the imperfect exposure classification.     

Impact of severity of illness bias and control group misclassification bias in case-control studies of antimicrobial-resistant organisms.

BACKGROUND: Case-control studies often analyze risk factors for antibiotic resistance. Recently published articles have illustrated that randomly selected control-patients may be preferable to those with the susceptible phenotype of the organism. A possible methodologic problem with randomly selected control-patients is potential bias due to control group misclassification. This occurs if some control-patients did not have clinical cultures performed and thus might have been unidentified case-patients. If this bias exists, these studies might be expected to report lower odds ratios (ORs) because control-patients would be more like case-patients. OBJECTIVE: To analyze potential biases that might arise due to control group misclassification and potentially larger selection biases that may be introduced if control-patients are required to have at least one clinical culture. PATIENTS: One hundred twenty case-patients, 770 control-patients in group 1, and 510 control-patients in group 2. METHODS: Two case-control studies. Case-patients had clinical cultures positive for imipenem-resistant Pseudomonas aeruginosa. The first group of control-patients were random. The second group of control-patients were identical to those in group 1 except being required to have at least one clinical culture. RESULTS: Univariate analyses showed higher ORs for case-patients versus control-patients in group 1 (imipenem [OR, 12.5], piperacillin-tazobactam [OR, 3.7], and vancomycin [OR, 4.7]) as compared with case-patients versus control-patients in group 2 (imipenem [OR, 8.0], piperacillin-tazobactam [OR, 2.5], and vancomycin [OR, 3.0]). CONCLUSION: Requiring control-patients to have at least one clinical culture introduces a selection bias likely because it eliminates patients with less severe illness.     

Methods to adjust for misclassification in the quantiles for the generalized linear model with measurement error in continuous exposures

In many biomedical studies, covariates of interest may be measured with errors. However, frequently in a regression analysis, the quantiles of the exposure variable are often used as the covariates in the regression analysis. Because of measurement errors in the continuous exposure variable, there could be misclassification in the quantiles for the exposure variable. Misclassification in the quantiles could lead to bias estimation in the association between the exposure variable and the outcome variable. Adjustment for misclassification will be challenging when the gold standard variables are not available. In this paper, we develop two regression calibration estimators to reduce bias in effect estimation. The first estimator is normal likelihood‐based. The second estimator is linearization‐based, and it provides a simple and practical correction. Finite sample performance is examined via a simulation study. We apply the methods to a four‐arm randomized clinical trial that tested exercise and weight loss interventions in women aged 50–75years. Copyright © 2015 John Wiley & Sons, Ltd.     

Differential misclassification of exposure in case-crossover studies

The aim of this study is to examine 2 types of differential misclassification of exposure in case-crossover studies. The first is the outcome-dependent misclassification of exposure, meaning that if an event has occurred, it could affect the reporting of exposure. The second is differential misclassification of exposure as a result of fading memory over time, which arises if the length of the recall period for case and control windows differs. We use empirical data from a case-crossover study of triggers of attacks in Ménière's disease. The study applied the matched-pair interval control window sampling approach. We examined misclassification in relation to 2 different types of exposures: emotional stress and salty food intake. The study covered repeated events reported by the same patients and involved the sampling of many control windows. Because some of these windows were related to case events and some unrelated, we were able to conduct both case-crossover and control-crossover analyses. Although this group of Ménière patients are well aware of their disease, and many of them have definite ideas regarding what triggers attacks, neither outcome-dependent misclassification nor differential misclassification of exposure resulting from fading memory over time seemed to be a major problem.     

On correcting for misclassification in twin studies and other matched-pair studies

In twin studies (and other matched-pair studies) of the effect of a K-level risk factor on disease risk, one must estimate the proportion of pairs in each of K2 possible pair categories, of which K(K-1) categories represent discordant pairs. In particular, for a binary factor, one must estimate proportions within two discordant-pair categories and the variances of functions of these estimates. This paper shows how to do so when misclassification is present and stable estimates of the classification rates are available. Unlike methods that estimate only the discordance ratio, one can use the methods presented here to improve estimates of epidemiologic effects.     

Effects of nondifferential exposure misclassification in ecologic studies.

Although many authors have argued against inferring individual-level exposure-disease relations from ecologic data because of the potential "ecological fallacy." the availability of data from diverse populations promotes the continued use of this rapid and inexpensive study design. In ecologic studies, the exposure status of groups is often defined by the proportion of individuals exposed. In these studies, nondifferential exposure misclassification of individuals is shown to produce overestimation of exposure-disease associations that may be extreme when the ecologically derived rate ratios are applied to individuals. This overestimation contrasts with the bias toward the null resulting from nondifferential misclassification of a binary exposure in epidemiologic studies conducted at the individual level. Given the magnitude of the potential bias from nondifferential exposure misclassification and other sources, quantitative estimates of individual-level rate ratios from ecologic data should be interpreted with extreme caution.     

Quantifying misclassification bias in cohort studies of vaccine efficacy

A simple mathematical model is presented to quantify the bias due to misclassification in prospective cohort studies of vaccine efficacy. Limitations of methods based on quantifying misclassification in 2 x 2 tables are discussed. The model is applied to data from three examples of the design and analysis of studies of pertussis vaccine efficacy.     

Bias in case-control studies

Case-control studies are largely used to explore differences between groups of individuals. They can identify potential risk factors associated with disease, or they can investigate patient behaviour, such as why some people do not attend for services. As such, case-control studies are often used to generate or test hypotheses about causal factors. Nonetheless, bias is always a danger in case-control studies, arising especially from the way in which study samples are selected or from the collection of retrospective data. Confounding also remains a problem. This short paper explores ways in which such flaws can be uncovered in published studies.     

Assessment of alcohol consumption by mailed questionnaire in epidemiological studies: evaluation of misclassification using a dietary history interview and biochemical markers

Background: Self-reported information on alcohol from questionnaires is generally assumed to introduce misclassification of consumption, mainly in the direction of underestimation. The aim of this study was to evaluate self-reported information on alcohol consumption from a mailed questionnaire by comparing to a dietary history interview and biochemical markers of alcohol intake. Subjects and Methods: For 76 male twin pairs of the Finnish Twin Cohort Study aged 40–70 years information on self-reported alcohol consumption was collected through mailed questionnaire and dietary history interview. Carbohydrate-deficient transferrin (CDT), Gamma-glutamyltransferase (Gamma-GT) and mean corpuscular volume (MCV) were determined from blood samples. Results: Mean levels of CDT, gamma-GT and MCV showed a rise with increased self-reported alcohol consumption already at low levels of reported consumption (<20 g alcohol/day). There was a positive correlation between reported amount alcohol intake per day and levels of CDT (r = 0.46), gamma-GT (r = 0.32) and MCV (r = 0.36) but within the high consumption group ( 30 g/day) there was no such correlation. The questionnaire had sensitivity of 28–43% and specificity of 89% for identification of high consumers of alcohol using the biochemical markers as reference and sensitivity 41% and specificity 94% using the dietary history interview as reference. Sensitivity was improved when information on binge drinking (82%) or possible drinking problems (73%) was considered. Conclusion: Comparison to dietary history interview as well as to biochemical markers indicate that self-reported information on alcohol consumption from a mailed questionnaire may be used to distinguish between groups with different levels of alcohol consumption. The suggested misclassification of high consumers implies that only strong associations between high alcohol intake and disease are likely to be detected in studies based on questionnaire data.     

Correcting for misclassification in two-way tables and matched-pair studies

Barron has presented a matrix formula that allows correction for nondifferential misclassification in two-by-two tables. We show how extensions of his result can be applied to situations involving differential misclassification, matched data, and arbitrary two-way tables. We also discuss some important limitations in the application of algebraic corrections for misclassification. In particular, we show that study misclassification rates need not reflect population misclassification rates if the distribution of the misclassified variable is determined by aspects of the study design.