Utilizing serial screening data to estimate dwell time of cancers and colorectal adenomas and develop new cancer stage progression models
Principal Investigator
Name
Joachim Worthington
Degrees
Ph.D.
Institution
The University of Sydney
Position Title
Senior Research Fellow
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-2032
Initial CDAS Request Approval
Mar 16, 2026
Title
Utilizing serial screening data to estimate dwell time of cancers and colorectal adenomas and develop new cancer stage progression models
Summary
Reliable estimates of the benefits, harms, and costs of cancer screening require an understanding of the cancer’s dwell time, the interval between cancer onset and diagnosis in an unscreened population. Understanding this dwell time allows us to assess the potential impact of screening via estimated prevalence of undetected cancers and potential downstaging effects of early detection through screening in larger cancer progression models. Observed population-level data alone is insufficient to estimate dwell time; repeat observations including negative screening tests are required.
As part of our program of work on the health and economic implications of cancer screening, we plan to develop new estimates and improve our existing estimates of cancer dwell time and stage progression across prostate, lung, colorectal, and ovarian cancers, using data from the PLCO screening trial. Large databases of patient-level screening data are not available in Australia, where our research team is based, and so we plan to augment the PLCO data with local population-level estimates from the Australian Institute of Health and Welfare.
Our analysis will include designing, testing, and validating mathematical approaches to modelling cancer dwell time and stage progression on the PLCO database. This will include novel methodology like shared frailty scores to capture correlated disease progression times, biologically realistic non-Markov models of stage progression risk, and fast Bayesian uncertainty estimates for differential equation-based models. Successfully generating and validating estimates for the PLCO cancers will allow us to use these methods to generate estimates for uncommon cancers where data is sparse or unavailable once validated.
These findings and new methodologies will be used to improve our existing Policy1 cancer modelling platform and inform our planned future research on screening for uncommon and rare cancers, including the potential health-economic impact of multi-cancer early detection (MCED) tests. Analysing a database with patient data across multiple cancers will also allow us to incorporate the impact of competing risks in our modelling and assess the potential impact of multiple cancer diagnoses.
For cancers like colorectal cancer where detection and removal of precancerous adenoma/lesions is part of the benefits of routine screening, understanding adenoma recurrence and the impact of surveillance after a positive screen is crucial to understanding the overall benefits and costs of screening. To achieve this we are also requesting the Study of Colonoscopy Utilization (SCU) and Recurrent Adenoma databases to supplement the core PLCO data.
This research will build on and inform our existing program of work on cancer prevention, including development of the Policy1 platform. See https://www.sydney.edu.au/medicine-health/our-research/research-centres/cancer-elimination-collaboration.html for details.
Aims
Estimate cancer dwell time distributions and stage progression rates using PLCO screening data: Define PDE-based time-to-event models using the methodology described in Worthington et al, Medical Decision Making (2025) incorporating cancer onset and stage progression. Calibrate these models to the PLCO database by calculating maximum likelihood estimates based on repeated screening observations, following the statistical method applied by Wu et al (Statistics and its Interface, 2022). Estimate age-specific cancer onset rates and dwell-time dependent stage progression rates based on the model parameters. Adapt these findings for the Australian population where appropriate data is available.
Develop and validate novel modelling methods: examine the viability of extensions to the time-to-event models, such as a shared frailty model of stage progression to capture correlations between transition times in more vs less aggressive cancers (see Gorfine et al, Annual Review of Statistics and its Applications, 2023). Develop fast Bayesian methods to assess parameter uncertainty in PDE-based models. Apply these methods, tested on the large PLCO database, to other cancers where data is sparse, and develop a multi-cancer model to assess the impact of multi-cancer early detection tests.
Estimate adenoma recurrent risk and the impact of routine surveillance in colorectal cancer: Extend the Policy1-Bowel model (Lew et al, Lancet Public Health, 2017) to capture more details in adenoma recurrence using the Recurrent Adenoma and Study of Colonoscopy Utilization (SCU) databases, to assess the impact of routine surveillance on long-term cancer outcomes.
Collaborators
Joachim Worthington University of Sydney
Marianne Weber University of Sydney
Yue He University of Sydney
Samuel Davis University of Sydney