Lung cancer risk prediction models and asbestos exposure: a validation study on the Western Australia Asbestos Review Program

Our plain-English summary of Lung cancer risk prediction models and asbestos exposure: a validation study on the Western Australia Asbestos Review Program, by C. Kumarasamy, K. Betts, R. Norman, et al (Occupational and Environmental Medicine, 2026).

The study

• Authors: C. Kumarasamy, K. Betts, R. Norman, K. Bennett, P. Franklin, M. Tammemägi, F. Brims
• Journal: Occupational and Environmental Medicine
• Published: 2026;83(3):154-161
• Read it: PubMed · Publisher · Free full text

Abstract

Objectives: Asbestos exposure raises the lung cancer risk and has supra-additive synergy alongside tobacco exposure. Lung cancer screening (LCS) is effective when high-risk populations are targeted. This study examined the utility of various LCS eligibility and risk prediction models in an asbestos-exposed population.

Methods: The Western Australia Asbestos Review Program (ARP) consists of individuals with known exposure to asbestos. All participants underwent annual review with low-dose CT screening. The performance of the Prostate, Lung, Colorectal and Ovarian (PLCO)m2012, PLCOm2014 and PLCOocc models, Liverpool Lung Project V.2 (LLPv2) and Bach models, together with the United States Preventive Services Task Force (USPSTF)2021 and Australian LCS eligibility criteria were validated on the ARP.

Results: The cohort consisted of 2126 participants of which 85.4% were male with a median (IQR) age of 70 (63-75) years old. Former smokers comprised 55.1% (n=1172) and never smokers 36.2% (n=769) of the cohort. Median smoking and cessation duration were 24 years (IQR: 13-72) and 32 years (IQR: 22-41), respectively. Lung cancer was diagnosed in 51 (2.4%) participants.When applying the risk models to the ARP cohort, the area under the curve for all models was modest, ranging from 0.602 to 0.675. All models underestimated risk in this cohort during calibration assessment, with the exception of the LLP V.2, which overestimated risk.

Conclusions: In an asbestos exposed population, current LCS eligibility criteria and risk models mostly underestimate the risk of lung cancer, reflecting the need for improved risk prediction models that adequately account for asbestos exposure.

© Author(s) (or their employer(s)) 2026. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ Group.

In plain English

This validation study tested tools for selecting people for lung cancer screening in the Western Australia Asbestos Review Program. Its members have known asbestos exposure and have yearly low-dose CT scans. Exposure came mostly from mixed-fibre construction and manufacturing work, or from crocidolite mining and residence at Wittenoom. Of 2126 participants aged 45 to 90, 51 (2.4%) were diagnosed with lung cancer between September 2012 and September 2023. Five risk models showed modest ability to separate who did and did not develop lung cancer (areas under the curve 0.602 to 0.675). Four models predicted fewer cancers than occurred, and the LLPv2 model predicted more. Only PLCOm2012 and LLPv2 differed significantly in calibration, the authors stated, yet the printed PLCOm2014 interval (0.46 to 0.79) also lay below 1, where predicted and observed totals match. The Australian eligibility criteria identified only 3 of 42 cancers among people who had smoked. The authors concluded that current criteria underperform in this asbestos-exposed group. They noted that regular screening may itself raise detected cancer rates, so the poor calibration could not be firmly linked to asbestos.

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: Validation study testing five existing lung cancer risk prediction models (PLCOm2012, PLCOm2014, PLCOocc, LLPv2, Bach) and two eligibility criteria (USPSTF2021 and Australian MSAC) in an asbestos-exposed CT screening cohort. The outcome was new lung cancers diagnosed between September 2012 and September 2023.
• Who was studied: 2126 of the 2195 participants in the Western Australia Asbestos Review Program, aged 45 to 90 years, with sufficient data for analysis.
• Exposure: Program members qualified through any of: at least 3 months of occupational asbestos exposure (mostly construction and manufacturing industries with mixed-asbestos fibre exposure); pleural plaque on imaging; or being former crocidolite (blue asbestos) mine workers or township residents at Wittenoom, Western Australia.
• Main results:
  • 51 lung cancers (2.4% of the cohort) were diagnosed. Of these, 36 were in former smokers and nine in never smokers.
  • Discrimination (area under the curve) across the models ranged from 0.602 (95% CI 0.505 to 0.698) for PLCOm2012 to 0.675 (95% CI 0.590 to 0.761) for LLPv2.
  • The PLCOocc model, which adds an asbestos term, had the same discrimination as the standard PLCOm2012 model.
  • The authors reported that differences in performance between the risk models in this cohort were not significant.
  • Mean calibration compares the total of predicted cancers with the number observed; a value below 1 means the model underestimated risk overall. The authors noted that a value of 1 does not necessarily mean good calibration across the whole range of predicted risk. LLPv2 overestimated overall lung cancer risk, with a mean calibration value of 2.54 (95% CI 1.9 to 3.65).
  • The other four models underestimated risk: PLCOm2014 0.58 (95% CI 0.46 to 0.79), PLCOm2012 0.7 (95% CI 0.54 to 0.98), Bach 0.76 (95% CI 0.59 to 1.04) and PLCOocc 0.83 (95% CI 0.63 to 1.18).
  • The authors stated that only the PLCOm2012 and LLPv2 calibrations were significantly different from the null. This is inconsistent with the PLCOm2014 interval the article prints (0.46 to 0.79), which also lies entirely below 1, the value at which predicted and observed totals match.
  • All risk models had higher sensitivity than the USPSTF2021 and Australian MSAC eligibility criteria.
  • The Australian MSAC criteria had a sensitivity of 0.07, identifying only 3 of the 42 lung cancers among participants with tobacco smoking exposure.
  • F1 scores (a combined measure of accuracy, from 0 to 1) were poor. They ranged from 0.068 for PLCOm2012 to 0.0937 for Bach, and were 0.0395 for both the USPSTF2021 and MSAC criteria.
• The authors' conclusion: The authors concluded that proposed eligibility criteria for lung cancer screening underperform in a cohort with a history of asbestos exposure, compared with their performance in other standard populations. In their view, this highlights the need for tailored risk prediction models for high-risk cohorts exposed to occupational carcinogens.
• Limitations the authors state:
  • Because the cohort is regularly screened with CT, detected lung cancer rates are higher than in the general population. The authors said this surveillance bias limits their ability to link the models' miscalibration to asbestos exposure.
  • The cohort is unusual, and other asbestos-exposed populations may not share its exposure levels or patterns, which limits generalisability.
  • The relatively small cohort limited the number of lung cancer cases, which led to overlapping confidence intervals between models.
  • Some predictor variables, including the binary asbestos variable and the COPD variable, were defined differently across the models.

Topics

The National Library of Medicine indexes this study under subjects our site covers: Lung cancer.

Medical terms in this study, explained in our Lexicon: Lung Cancer, Asbestos Exposure, Low-Dose CT Screening.

Related studies

• Development and validation of a lung cancer risk prediction model for African-Americans — C.J. Etzel, S. Kachroo, et al. (2008)
• Predictive accuracy of the Liverpool Lung Project risk model for stratifying patients for computed tomography screening for lung cancer: a case-control and cohort validation study — O.Y. Raji, S.W. Duffy, et al. (2012)
• Predictions of future cases of asbestos-related disease among former miners and millers of crocidolite in Western Australia — N.H. de Klerk, B.K. Armstrong, et al. (1989)
• Ultra-low-dose chest computer tomography screening of an asbestos-exposed population in Western Australia — F.J. Brims, C.P. Murray, et al. (2015)
• Cancer Risk Prediction Using Machine Learning for Supporting Early Cancer Diagnosis in Symptomatic Patients: A Systematic Review of Model Types — F. Pennisi, S. Borlini, et al. (2025)
• Autoimmune antibodies and asbestos exposure: Evidence from Wittenoom, Western Australia — A. Reid, P. Franklin, et al. (2018)

Browse more research in Occupational and Environmental Medicine, or return to medical research.