Sources of uncertainty in dose-response modeling of epidemiological data for cancer risk assessment

Our plain-English summary of Sources of uncertainty in dose-response modeling of epidemiological data for cancer risk assessment, by L. Stayner, A.J. Bailer, R. Smith, et al (Annals of the New York Academy of Sciences, 1999).

The study

• Authors: L. Stayner, A.J. Bailer, R. Smith, S. Gilbert, F. Rice, E. Kuempel
• Journal: Annals of the New York Academy of Sciences
• Published: 1999;895:212-22
• Cited by: 15 articles, as counted by NIH iCite
• Read it: PubMed · Publisher

Abstract

Epidemiologic data is increasingly being used for dose-response analysis in risk assessment. The Environmental Protection Agency (EPA) and other U.S. agencies have expressed a preference for using epidemiologic data rather than toxicologic data when possible. However, there are a number of important sources of uncertainty in using epidemiologic data for this purpose that need to be clearly recognized and, when possible, quantified. This paper presents a critical review of the major sources of uncertainty in the use of epidemiologic data for cancer risk assessment. These may include: (1) study design issues such as potential confounding and other biases, inadequate sample size, and followup, (2) the choice of the data set, (3) specification of the dose-response model, (4) estimation of exposure and dose, and (5) unrecognized variability in susceptibility. Examples from risk assessments for cadmium, asbestos, and diesel exhaust are used to illustrate the potential magnitude of some of these sources of uncertainty. It is shown that the overall uncertainty from these various sources combined may often result in highly uncertain risk estimates from dose-response modeling of epidemiologic data. For this reason, we believe it is best to present a range of possible risk estimates, which, to the extent possible, reflects the variability and uncertainty inherent in the dose-response evaluation of epidemiologic data.

In plain English

This paper was a review, not a new study. The authors discussed sources of uncertainty in using studies of people to estimate cancer risk. They illustrated these with analyses of chrysotile asbestos and cadmium, several their own, and preliminary results from an ongoing diesel exhaust analysis. A hypothetical lung cancer study of workers showed that such studies would generally be unable to detect an excess risk below 1 per 1,000 workers. Lung cancer risk slopes for chrysotile asbestos from a study of textile workers and a study of Quebec chrysotile miners differed by more than one order of magnitude (more than tenfold). For cadmium-exposed workers, lung cancer risk estimates varied by nearly an order of magnitude depending on the statistical model used. In the chrysotile textile worker analysis, the statistical estimate of a threshold exposure was zero. Preliminary simulations of errors in exposure duration only, for railroad workers exposed to diesel exhaust, showed a model slope varying about fourfold from lowest to highest, with 95% of results within a factor of two. The authors concluded that it is best to present a range of possible risk estimates.

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: Review of sources of uncertainty in dose-response modeling of epidemiologic data for cancer risk assessment. It is illustrated with examples from risk analyses, several of them the authors' own, and with preliminary results from an ongoing analysis of diesel exhaust.
• Who was studied: No study population of its own. Examples were drawn from chrysotile asbestos textile workers, Quebec chrysotile miners, cadmium-exposed workers, and railroad workers exposed to diesel exhaust (an ongoing analysis). A hypothetical occupational lung cancer cohort was used for sample-size calculations.
• Exposure: Chrysotile asbestos (lung cancer), occupational cadmium (lung cancer), and diesel exhaust particles (railroad workers)
• Main results:
  • In a hypothetical retrospective cohort study of lung cancer in workers, detecting an excess risk of 1 per 1,000 (relative risk 1.02) would need 15,605 expected deaths and 399,605 workers. The authors stated that cohort studies would generally be incapable of detecting an excess risk of less than 1 per 1,000 workers.
  • Lung cancer risk slopes for chrysotile asbestos from the authors' study of textile workers and from a study of Quebec chrysotile miners differed by more than one order of magnitude
  • In the authors' earlier analysis of occupational cadmium exposure and lung cancer, risk estimates varied by nearly an order of magnitude depending on the dose-response model used
  • In the chrysotile asbestos textile worker analysis, a restricted cubic spline model fitted the lung cancer data very similarly to the additive relative rate model
  • Applying a formal threshold-estimation method to the chrysotile asbestos lung cancer analysis, the authors found that the maximum likelihood estimate of the threshold parameter was zero
  • Preliminary Monte Carlo simulations (1,000 runs) of errors affecting estimates of duration of diesel exposure in railroad workers were run as part of an ongoing analysis. The slope of the linear relative rate model varied by a factor of approximately four (minimum versus maximum), with 95% of results within a factor of two.
  • This is not a finding of this paper; the authors cited it from other studies. Most case-control studies of mesothelioma found a strong association with lung burdens of amphibole asbestos, but not of chrysotile asbestos.
• The authors' conclusion: The authors concluded that the combined uncertainty from these sources may often make risk estimates from epidemiologic data highly uncertain. They believed it is best to present a range of possible risk estimates reflecting that variability and uncertainty.
• Limitations the authors state:
  • The authors described their discussion as brief, covering some of the major sources of uncertainty
  • The diesel exhaust simulation results were preliminary and reflected only part of the overall uncertainty. The authors expected that uncertainty to increase as other sources were considered.
  • The authors stated that the extent of uncertainty from relying on external exposure rather than tissue dose was difficult to judge at the time

Topics

Medical terms in this study, explained in our Lexicon: Dose-response relationship, Confounding.

Related studies

• Risk assessment and source analysis of heavy metals in soil around an asbestos mine in an arid plateau region, China — X. Li, D. Ding, et al. (2024)
• An air quality data analysis system for interrelating effects, standards, and needed source reductions: Part 13--Applying the EPA Proposed Guidelines for Carcinogen Risk Assessment to a set of asbestos lung cancer mortality data — R.I. Larsen (2003)
• Cancer incidence within a cohort occupationally exposed to asbestos: a study of dose--response relationships — B. Clin, F. Morlais, et al. (2011)
• Dose-response modeling of NLRP3 inflammasome-mediated diseases: asbestos, lung cancer, and malignant mesothelioma as examples — L.A. Cox (2019)
• Banding carcinogenic risks in developed countries: A procedural basis for qualitative assessment — B.W. Stewart (2008)
• The contribution of environmental monitoring in the epidemiological assessment of exogenous risk. The experience of ARPA in the Emilia-Romagna Region of Italy — A. Zavatti, P. Lauriola (1999)

Browse more research in Annals of the New York Academy of Sciences, or return to medical research.