Adjustment For Misclassification Research Articles

Bayesian Adjustment for Misclassification in Mortality Data

Bayesian Adjustment for Misclassification in Mortality Data

Potential sensitivity of bias analysis results to incorrect assumptions of nondifferential or differential binary exposure misclassification.

Results of bias analyses for exposure misclassification are dependent on assumptions made during analysis. We describe how adjustment for misclassification is affected by incorrect assumptions about whether sensitivity and specificity are the same (nondifferential) or different (differential) for cases and noncases. We adjusted for exposure misclassification using probabilistic bias analysis, under correct and incorrect assumptions about whether exposure misclassification was differential or not. First, we used simulated data sets in which nondifferential and differential misclassification were introduced. Then, we used data on obesity and diabetes from the National Health and Nutrition Examination Survey (NHANES) in which both self-reported (misclassified) and measured (true) obesity were available, using literature estimates of sensitivity and specificity to adjust for bias. The ratio of odds ratio (ROR; observed odds ratio divided by true odds ratio) was used to quantify magnitude of bias, with ROR = 1 signifying no bias. In the simulated data sets, under incorrect assumptions (eg, assuming nondifferential misclassification when it was truly differential), results were biased, with RORs ranging from 0.18 to 2.46. In NHANES, results adjusted based on incorrect assumptions also produced biased results, with RORs ranging from 1.26 to 1.55; results were more biased when making these adjustments than when using the misclassified exposure values (ROR = 0.91). Making an incorrect assumption about nondifferential or differential exposure misclassification in bias analyses can lead to more biased results than if no adjustment is performed. In our analyses, incorporating uncertainty using probabilistic bias analysis was not sufficient to overcome this problem.

Spatial correlation in Bayesian logistic regression with misclassification

Standard logistic regression assumes that the outcome is measured perfectly. In practice, this is often not the case, which could lead to biased estimates if not accounted for.This study presents Bayesian logistic regression with adjustment for misclassification of the outcome applied to data with spatial correlation. The models assessed include a fixed effects model, an independent random effects model, and models with spatially correlated random effects modelled using conditional autoregressive prior distributions (ICAR and ICAR(ρ)). Performance of these models was evaluated in a simulation study.Parameters were estimated by Markov Chain Monte Carlo methods, using slice sampling to improve convergence.The results demonstrated that adjustment for misclassification must be included to produce unbiased regression estimates. With strong correlation the ICAR model performed best. With weak or moderate correlation the ICAR(ρ) performed best. With unknown spatial correlation the recommended model would be the ICAR(ρ), assuming convergence can be obtained.

Marginal analysis of longitudinal ordinal data with misclassification in both response and covariates

Marginal methods have been widely used for the analysis of longitudinal ordinal and categorical data. These models do not require full parametric assumptions on the joint distribution of repeated response measurements but only specify the marginal or even association structures. However, inference results obtained from these methods often incur serious bias when variables are subject to error. In this paper, we tackle the problem that misclassification exists in both response and categorical covariate variables. We develop a marginal method for misclassification adjustment, which utilizes second-order estimating functions and a functional modeling approach, and can yield consistent estimates and valid inference for mean and association parameters. We propose a two-stage estimation approach for cases in which validation data are available. Our simulation studies show good performance of the proposed method under a variety of settings. Although the proposed method is phrased to data with a longitudinal design, it also applies to correlated data arising from clustered and family studies, in which association parameters may be of scientific interest. The proposed method is applied to analyze a dataset from the Framingham Heart Study as an illustration.

Bayesian Sample Size Determination for Case-Control Studies When Exposure May be Misclassified

Odds ratios are frequently used for estimating the effect of an exposure on the probability of disease in case-control studies. In planning such studies, methods for sample size determination are required to ensure sufficient accuracy in estimating odds ratios once the data are collected. Often, the exposure used in epidemiologic studies is not perfectly ascertained. This can arise from recall bias, the use of a proxy exposure measurement, uncertain work exposure history, and laboratory or other errors. The resulting misclassification can have large impacts on the accuracy and precision of estimators, and specialized estimation techniques have been developed to adjust for these biases. However, much less work has been done to account for the anticipated decrease in the precision of estimators at the design stage. Here, we develop methods for sample size determination for odds ratios in the presence of exposure misclassification by using several interval-based Bayesian criteria. By using a series of prototypical examples, we compare sample size requirements after adjustment for misclassification with those required when this problem is ignored. We illustrate the methods by planning a case-control study of the effect of late introduction of peanut to the diet of children to the subsequent development of peanut allergy.

Extended Matrix and Inverse Matrix Methods Utilizing Internal Validation Data When Both Disease and Exposure Status Are Misclassified.

The problem of misclassification is common in epidemiological and clinical research. In some cases, misclassification may be incurred when measuring both exposure and outcome variables. It is well known that validity of analytic results (e.g. point and confidence interval estimates for odds ratios of interest) can be forfeited when no correction effort is made. Therefore, valid and accessible methods with which to deal with these issues remain in high demand. Here, we elucidate extensions of well-studied methods in order to facilitate misclassification adjustment when a binary outcome and binary exposure variable are both subject to misclassification. By formulating generalizations of assumptions underlying well-studied "matrix" and "inverse matrix" methods into the framework of maximum likelihood, our approach allows the flexible modeling of a richer set of misclassification mechanisms when adequate internal validation data are available. The value of our extensions and a strong case for the internal validation design are demonstrated by means of simulations and analysis of bacterial vaginosis and trichomoniasis data from the HIV Epidemiology Research Study.

Expected estimating equations via EM for proportional hazards regression with covariate misclassification

In epidemiological and medical studies, covariate misclassification may occur when the observed categorical variables are not perfect measurements for an unobserved categorical latent predictor. It is well known that covariate measurement error in Cox regression may lead to biased estimation. Misclassification in covariates will cause bias, and adjustment for misclassification will be challenging when the gold standard variables are not available. In general, statistical modeling for misclassification is very different from that of the measurement error. In this paper, we investigate an approximate induced hazard estimator and propose an expected estimating equation estimator via an expectation-maximization algorithm to accommodate covariate misclassification when multiple surrogate variables are available. Finite sample performance is examined via simulation studies. The proposed method and other methods are applied to a human immunodeficiency virus clinical trial in which a few behavior variables from questionnaires are used as surrogates for a latent behavior variable.

Misclassification and the use of register‐based indicators for depression

To study the degree to which depression indicators based on register data on hospital and antidepressant treatment suffer from differential misclassification with respect to gender, age and social group. Data on 7378 persons were obtained by linking a cross-sectional survey of Danish adults aged 40 and 50 years with population-based registers. Misclassification was analysed by comparing survey data to register data on major depression using the method proposed by Rothman and Greenland. Differential misclassification was found. Adjustment for misclassification reduced women's odds ratios from 2.18 to 1.00 for hospital treatment and from 1.70 to 1.10 for antidepressants. For the lower social group, the corresponding odds ratios increased from 1.18 to 3.52, and from 1.35 to 2.32 respectively, whereas odds ratios with respect to age remained almost unchanged. Differential misclassification should be considered when register-based information about hospital and antidepressant treatment are used as depression indicators.

A Bayesian Adjustment for Covariate Misclassification with Correlated Binary Outcome Data

Estimated associations between an outcome variable and misclassified covariates tend to be biased when the methods of estimation that ignore the classification error are applied. Available methods to account for misclassification often require the use of a validation sample (i.e. a gold standard). In practice, however, such a gold standard may be unavailable or impractical. We propose a Bayesian approach to adjust for misclassification in a binary covariate in the random effect logistic model when a gold standard is not available. This Markov Chain Monte Carlo (MCMC) approach uses two imperfect measures of a dichotomous exposure under the assumptions of conditional independence and non-differential misclassification. A simulated numerical example and a real clinical example are given to illustrate the proposed approach. Our results suggest that the estimated log odds of inpatient care and the corresponding standard deviation are much larger in our proposed method compared with the models ignoring misclassification. Ignoring misclassification produces downwardly biased estimates and underestimate uncertainty.

A Bayesian approach for analyzing a cluster-randomized trial with adjustment for risk misclassification

Bayesian hierarchical modelling techniques have some advantages over classic methods for the analysis of cluster-randomized trial. Bayesian approach is also becoming more popular to deal with measurement error and misclassification problems. We propose a Bayesian approach to analyze a cluster-randomized trial with adjusting for misclassification in a binary covariate in the random effect logistic model when a gold standard is not available. This Markov chain Monte Carlo (MCMC) approach uses two imperfect measures of a dichotomous exposure under the assumptions of conditional independence and non-differential misclassification. Both simulated numerical example and real clinical example are given to illustrate the proposed approach. The Bayesian approach has great potential to be used in misclassification problem in generalized linear mixed model (GLMM) since it allow us to fit complex models and identify all the parameters. Our results suggest that Bayesian approach for analyzing cluster-randomized trial and adjusting for misclassification in GLMM is flexible and powerful.

The effects of radiotherapy for the treatment of contralateral breast cancer

Radiotherapy following breast-conserving surgery for the treatment of first primary breast cancer is the standard of care and is widely used despite its small survival benefit. The effects of radiotherapy in metachronous contralateral breast cancer are unknown. We examined the use of radiotherapy and its effect on cause-specific and all-cause mortality among women with metachronous contralateral breast cancer treated with breast-conserving surgery in community settings. Using data from the 1985-2000 Surveillance, Epidemiology, and End Results program, we identified women with stage 0-III metachronous contralateral breast cancer that occurred at least six months after stage 0-III first primary breast cancer. Cause-specific and all-cause mortality of women age 40-69 who did and who did not receive radiotherapy following breast-conserving surgery for metachronous contralateral breast cancer were compared in proportional hazard models using propensity scores to balance covariates by radiotherapy use. We adjusted for misclassification of radiotherapy use. Based on misclassification-corrected analyses, 43.2 percent of 1,083 women with metachronous contralateral breast cancer did not receive radiotherapy after BCS. After adjustment for propensity scores and radiotherapy misclassification, women who did not receive radiotherapy had 2.2 times greater risk of cause-specific and 1.7 times greater risk of all-cause mortality. In community settings, a high percentage of women with stage 0-III metachronous contralateral breast cancer did not receive radiotherapy following breast-conserving surgery. Unlike the small survival benefit of radiotherapy after first primary breast cancer, omission of radiotherapy after metachronous contralateral breast cancer significantly increased the risk of cause-specific and all-cause mortality.

Comparison of HIV Type 1 Incidence Observed during Longitudinal Follow-Up with Incidence Estimated by Cross-Sectional Analysis Using the BED Capture Enzyme Immunoassay

The BED capture enzyme immunoassay (BED CEIA) for recent infection was developed for the estimation of HIV-1 incidence in a population from a single cross-sectional survey. To evaluate performance, we applied the assay to specimen sets obtained from a longitudinal cohort study, the AIDSVAX B/B vaccine trial, in which there was an independent and conventional measure of observed incidence. The BED CEIA was performed on specimens obtained during follow-up for seroconversion conducted every 6 months for 3 years. There was excellent agreement between the observed and BED-estimated incidence for all the intervals. The cumulative, annualized incidence observed in the cohort was 3.10 new infections per 100 person-years (95% CI, 2.57-3.63). The corresponding BED-estimated incidence was 2.91 (2.30-3.53). We also estimated the effect of varied prevalence on a fixed incidence. Because some specimens from persons with longer-term infection are classified as recent by the assay, this can inflate the incidence estimate. We quantify this effect and discuss potential mitigation by excluding certain specimens on clinical grounds, by relying on trend differences rather than absolute incidence estimates, by secondary confirmatory testing, or by analytic adjustments for misclassification. Cross-sectional HIV incidence estimation circumvents many of the drawbacks associated with longitudinal cohort studies, but there are test-specific limitations that should be considered in the design of population surveys.

Statistical adjustment for misclassification of seat belt and alcohol use in the analysis of motor vehicle accident data

The use of seat belts and alcohol is often misreported in police motor vehicle accident reports for a number of reasons. To avoid penalties, occupants often over report seat belt use and under report alcohol use. Police officers sometimes fail to account for evidence such as presence of belt burn, condition of belts, odor of alcohol, crash patterns, etc. Biased conclusions result when using misclassified accident data to estimate the effectiveness of seat belts in preventing injuries and reducing medical costs. We investigated the effects of misclassification of seat belt and alcohol use on the odds ratio of injury as well as medical costs. A statistical method and a SAS program were developed to adjust odds ratios of injury and medical cost estimates to account for misclassification of seat belts and alcohol use. The method allowed for incorporation of variables that could affect misclassification of seat belt and alcohol use. We conducted a Monte Carlo simulation and found that when there were large differences between the misclassification rates for major and minor injury, the unadjusted odds ratio could have up to a 90% bias while our adjusted odds ratio was effectively unbiased. To illustrate the method, we estimated the misclassification rates of seat belt and alcohol use by comparing merged police and hospital reports from Nebraska motor vehicle accident data sets (1996–1997) and then evaluated the bias of the odds ratio of injury and medical costs estimates due to misclassification. Our results showed that the bias of the odds ratio of injury and medical costs due to misclassification of seat belts and alcohol use depended both on the amount of misclassification and the reported frequencies. Misclassification about seat belt and alcohol use only slightly biased the unadjusted odds ratio estimates and mean hospital charge, while misclassification resulted in approximately a 69% underestimate of the total medical costs savings due to seatbelts. However, due to the small size of the merged Nebraska police and hospital data set used to estimate misclassification rates, these results are likely somewhat imprecise.

Curious phenomena in Bayesian adjustment for exposure misclassification

Many epidemiologic investigations involve some discussion of exposure misclassification, but rarely is there an attempt to adjust for misclassification formally in the statistical analysis. Rather, investigators tend to rely on intuition to comment qualitatively on how misclassification might impact their findings. We point out several ways in which intuition might fail, in the context of unmatched case-control analysis with non-differential exposure misclassification. Particularly, we focus on how intuition can conflict with the results of a Bayesian analysis that accounts for the various uncertainties at hand. First, the Bayesian adjustment for misclassification can weaken the evidence about the direction of an exposure-disease association. Second, admitting uncertainty about the misclassification parameters can lead to narrower interval estimates concerning the association. We focus on the simple setting of unmatched case-control analysis with binary exposure and without adjustment for confounders, though much of our discussion should be relevant more generally.

Racial Misclassification and Disparities in Cardiovascular Disease Among American Indians and Alaska Natives

National vital event data suggest that cardiovascular disease (CVD) mortality rates are lower for American Indians and Alaska Natives (AIAN) than for the general US population, but these data are disproportionately flawed for AIAN because of racial misclassification. Vital event data adjusted for racial misclassification and published by the Indian Health Service were used to compare trends in CVD mortality from 1989 to 1991 to 1996 to 1998 between AIAN, US all-races, and US white populations. Without misclassification accounted for, AIAN initially had the lowest mortality rates from major CVD, but by the end of the study, their rates were the highest. Adjustment for misclassification revealed an early and rapidly growing disparity between CVD mortality rates among AIAN compared with rates in the US all-races and white populations. By 1996 to 1998, the age- and misclassification-adjusted number of CVD deaths per 100,000 among AIAN was 195.9 compared with age-adjusted rates of 166.1 and 159.1 for US all races and whites, respectively. The annual percent change in CVD mortality for AIAN was 0.5 compared with -1.8 in the other groups. Regardless of racial misclassification, the most striking and widening disparities were found for middle-aged AIAN, but CVD mortality among AIAN > or =65 years of age was lower than in the other populations. A previously underrecognized disparity in CVD mortality exists for AIAN, particularly among middle-aged adults. Moreover, these disparities are increasing. Efforts to reduce CVD mortality in AIAN must begin before the onset of middle age.

Adjustment for misclassification in studies of familial aggregation of disease using routine register data.

This paper discusses the misclassification that occurs when relying solely on routine register data in family studies of disease clustering. A register study of familial aggregation of schizophrenia is used as an example. The familial aggregation is studied using a regression model for the disease in the child including the disease status of the parents as a risk factor. If all the information is found in the routine registers then the disease status of the parents is only known from the time when the register started and if this information is used unquestioningly the parents who have had the disease before this time are misclassified as disease-free. Two methods are presented to adjust for this misclassification: regression calibration and an EM-type algorithm. These methods are used in the schizophrenia example where the large effect of having a schizophrenic mother hardly shows any signs of bias due to misclassification. The methods are also studied in simulations showing that the misclassification problem increases with the disease frequency.

Revisiting the Association between Environmental Tobacco Smoke Exposure and Lung Cancer Risk

Hackshaw et al. estimated that adjustment for bias due to misclassification of smoking habits reduces the observed relative risk of lung cancer in non-smoking women associated with smoking by the husband from 1.24 to 1.18. This bias estimate is little affected by using an alternative method for misclassification adjustment or by using updated data from 47 studies rather than data from the 37 studies used by Hackshaw et al. The bias is increased if strong evidence of much higher misclassification rates in Asian women is taken into account and could then explain about half the observed association. Misclassification correction has not previously been attempted for dose-response data. We describe a suggested approach and apply it to data relating risk to amount smoked by the husband. As shown in paper II of this series the unadjusted increase in risk per 10 cigarettes/ day smoked by the husband is 10% (95% CI 5 to 15%), reducing to either 6% (95% CI 1 to 11%) or 9% (95% CI 5 to 14%) after adjustment for confounding by fruit, vegetable and fat consumption and education using respectively unweighted or weighted means to combine evidence on the ETS/confounder association from different studies. Further adjustment for plausible levels of misclassification similarly reduced these two estimates to, respectively, 2% (95% CI –3 to 8%) or 5.5% (95% CI 0 to 11%). Difficulties in applying misclassification bias corrections are discussed. Other sources of bias will be considered in later papers in this series.

Revisiting the Association between Environmental Tobacco Smoke Exposure and Lung Cancer Risk

Hackshaw et al. estimated that adjustment for bias due to misclassification of smoking habits reduces the ob served relative risk of lung cancer in non-smoking wom en associated with smoking by the husband from 1.24 to 1.18. This bias estimate is little affected by using an alter native method for misclassification adjustment or by using updated data from 47 studies rather than data from the 37 studies used by Hackshaw et al. The bias is increased if strong evidence of much higher misclassifi cation rates in Asian women is taken into account and could then explain about half the observed association. Misclassification correction has not previously been at tempted for dose-response data. We describe a suggest ed approach and apply it to data relating risk to amount smoked by the husband. As shown in paper II of this series the unadjusted increase in risk per 10 cigarettes/ day smoked by the husband is 10% (95% Cl 5 to 15%), reducing to either 6% (95% Cl 1 to 11 %) or 9% (95% Cl 5 to 14%) after adjustment for confounding by fruit, vege table and fat consumption and education using respec tively unweighted or weighted means to combine evi dence on the ETS/confounder association from different studies. Further adjustment for plausible levels of mis classification similarly reduced these two estimates to, respectively, 2% (95% Cl -3 to 8%) or 5.5% (95% Cl 0 to 11 %). Difficulties in applying misclassification bias cor rections are discussed. Other sources of bias will be con sidered in later papers in this series.

Sociodemographic and Health Care System Factors Influencing Cervical Cancer Screening among Young American Indian Women

Age-adjusted cervical cancer mortality rates (Indian Health Service 1997) are higher for women in the Oklahoma Indian Health Service (IHS) area (3.3 per 100,000) than for the general U.S. female population (2.5 per 100,000). But reported rates may not provide a complete picture of the problem due to underreporting of Indian race on death certificates (ibid.). With an adjustment for misclassification, the rate increases to 4.9 per 100,000, almost twice that of the general U.S. female population.

The Effect of Diagnostic Misclassification on Non-Cancer and Cancer Mortality Dose Response in A-Bomb Survivors

We used the EM algorithm in the context of a joint Poisson regression analysis of cancer and non-cancer mortality in the Radiation Effects Research Foundation (RERF) Life Span Study (LSS) to assess whether the observed increased risk of non-cancer death due to radiation exposure (Shimizu et al., RERF Technical Report 02-91, 1991) can be attributed solely to misclassification of cancer as non-cancer on death certificates. We show that greater levels of dose-independent misclassification than are indicated by a series of autopsies conducted on a subset of LSS members would be required to explain the non-cancer dose response, but that a relatively small amount of dose-dependence in the misclassification of cancer would explain the result. The adjustment for misclassification also results in higher risk estimates for cancer mortality. We review applications of similar statistical methods in other contexts and discuss extensions of the methods to more than two causes of death.