Estimation and inference for case-control studies with multiple non-gold standard exposure assessments: with an occupational health application

Our plain-English summary of Estimation and inference for case-control studies with multiple non-gold standard exposure assessments: with an occupational health application, by H. Chu, S.R. Cole, Y. Wei, et al (Biostatistics, 2009).

The study

• Authors: H. Chu, S.R. Cole, Y. Wei, J.G. Ibrahim
• Journal: Biostatistics
• Published: 2009;10(4):591-602
• Cited by: 7 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

In occupational case-control studies, work-related exposure assessments are often fallible measures of the true underlying exposure. In lieu of a gold standard, often more than 2 imperfect measurements (e.g. triads) are used to assess exposure. While methods exist to assess the diagnostic accuracy in the absence of a gold standard, these methods are infrequently used to correct for measurement error in exposure-disease associations in occupational case-control studies. Here, we present a likelihood-based approach that (a) provides evidence regarding whether the misclassification of tests is differential or nondifferential; (b) provides evidence whether the misclassification of tests is independent or dependent conditional on latent exposure status, and (c) estimates the measurement error-corrected exposure-disease association. These approaches use information from all imperfect assessments simultaneously in a unified manner, which in turn can provide a more accurate estimate of exposure-disease association than that based on individual assessments. The performance of this method is investigated through simulation studies and applied to the National Occupational Hazard Survey, a case-control study assessing the association between asbestos exposure and mesothelioma.

In plain English

This was a statistics methods paper, not a new patient study. The authors developed a way to correct exposure-disease estimates when exposure is judged by several imperfect methods. They tested it in computer simulations and on existing case-control data on asbestos exposure and mesothelioma from the National Occupational Hazard Survey (208 cases, 533 controls). Exposure had been judged by next-of-kin reports, an occupational hygienist, and a job exposure matrix. Alone, next-of-kin reports gave an odds ratio of 10.74 (95% confidence limits 7.27 to 15.90). The other methods gave 4.65 (3.19 to 6.77) and 2.06 (1.48 to 2.86). The article's table counts tie 4.65 to the hygienist and 2.06 to the matrix, though its text implies the reverse. After correction, the odds ratio was 7.47 under one model (limits 3.68 to 11.30 in the text, 3.962 to 10.977 in a table) and 16.12 (5.22 to 27.03) under the authors' final model. In simulations, single assessments underestimated the true value. The authors concluded that wrong assumptions about measurement error can noticeably affect results and that models lacking identification need cautious interpretation.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Statistical methods paper: a likelihood-based (latent class, probit random-effects) approach to correct exposure misclassification when several imperfect exposure assessments are available. It was evaluated by simulation studies and applied to data from a previously published case-control study (Nam and others, 2005).
• Who was studied: Applied example: 208 mesothelioma cases and 533 controls from the National Occupational Hazard Survey. Cases were selected between 1975 and 1980 from the New York State Health Department Cancer Registry, the Los Angeles County Cancer Surveillance Program, and 39 Veterans Administration hospitals. Controls came from the same geographic area or the same hospital. Simulations used 2000 replications, each with 250 cases and 500 controls.
• Exposure: Asbestos exposure (no fiber type specified), classified as yes or no by three methods: next-of-kin reports of occupational exposure, a job exposure matrix applied to work histories, and an occupational hygienist's classification based on work histories
• Main results:
  • Uncorrected odds ratios of mesothelioma for those assessed as exposed versus not exposed, by each single method, were 10.74 (95% CL 7.27 to 15.90) for next-of-kin reports. The other two were 4.65 (95% CL 3.19 to 6.77) and 2.06 (95% CL 1.48 to 2.86). The counts in Table 1 give 4.65 for the expert (hygienist) assessment and 2.06 for the job exposure matrix. The article's text lists these two methods in the opposite order.
  • Under the model with partially differential misclassification and conditional independence (Model III), the corrected odds ratio was 7.47. The text gives its 95% CL as 3.68 to 11.30; Table 2 prints 3.962 to 10.977.
  • Under Model III, next-of-kin reports were more sensitive for cases than for controls (0.77 vs. 0.36). The job exposure matrix was less sensitive for cases than for controls (0.59 vs. 0.75).
  • Under the authors' 'final' model, which allowed partially differential misclassification and constrained conditional dependence (Model IV), the corrected odds ratio was 16.12 (95% CL 5.22 to 27.03). The model assuming nondifferential misclassification and conditional independence gave 14.58 (95% CL 7.26 to 21.90).
  • In simulations with a true log odds ratio of 2.0, data came from a partially differential, dependent misclassification model. Using a single assessment gave average log odds ratio estimates of only 0.288, 0.926 and 0.727.
• The authors' conclusion: The authors concluded that wrongly specifying whether misclassification is differential or nondifferential, and whether it is conditionally independent or dependent, can noticeably affect estimates of exposure-disease associations. They said different misclassification models should be routinely explored, and that the corrected estimates are potentially more accurate because they combine all three assessments.
• Limitations the authors state:
  • With only 3 assessments, a model with both full differential and full conditional dependent misclassification cannot be identified, so the two cannot be told apart
  • The final model in the case study was chosen by an ad hoc approach
  • The method assumes the latent variable is independent of disease status
  • In simulations, the likelihood ratio test often failed to select the true dependent model (probabilities of 0.607, and 0.011 in another scenario)
  • Further theoretical research is needed on whether the order of testing the assumptions matters, and results from models lacking identification need cautious interpretation

Topics

Medical terms in this study, explained in our Lexicon: Case-Control Study, Asbestos Exposure, Mesothelioma.

Related studies

• A Bayesian approach to strengthen inference for case-control studies with multiple error-prone exposure assessments — J. Zhang, S.R. Cole, et al. (2013)
• Hierarchical regression for multiple comparisons in a case-control study of occupational risks for lung cancer — M. Corbin, L. Richiardi, et al. (2012)
• Analysis of the effect of occupational exposure to asbestos based on threshold regression modeling of case-control data — A. Chambaz, D. Choudat, et al. (2014)
• Application of a method to evaluate the quality of work histories and document the exposure assessment process — C. Rice, E.F. Heineman (2003)
• Application of Pattern Mining Methods to Assess Exposures to Multiple Airborne Chemical Agents in Two Large Occupational Exposure Databases from France — J.F. Sauvé, A. Emili, et al. (2022)
• Global corporate policies and international "double standards" in occupational and environmental health — B. Castleman (1999)

Browse return to medical research.