Estimating Overdiagnosis in Cancer Screening Studies
Principal Investigator
Name
Ruth Etzioni
Degrees
PhD
Institution
Fred Hutchinson Cancer Research Center
Position Title
Full Member
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-205
Initial CDAS Request Approval
Apr 11, 2016
Title
Estimating Overdiagnosis in Cancer Screening Studies
Summary
Millions of men undergo screening for early detection of prostate cancer every year. Overdiagnosis, the detection by screening of cancer that would never become clinically diagnosed, is now recognized as the main driver of harms due to screening. This presents an especially difficult clinical dilemma in prostate cancer, where latent cancers are highly prevalent in older age groups. Overdiagnosed cases cannot be helped by treatment, and this overtreatment carries a high price in terms of patient morbidity and economic costs. Knowledge about overdiagnosis is critical for sound screening policies and for informed patient decision making. However, overdiagnosis depends on screening practices and personal factors, and many published studies are biased or do not apply to populations that differ from those used for estimation.
One objective of our ongoing research is to advance knowledge about how to validly estimate overdiagnosis and to provide useful information about overdiagnosis associated with specific cancer screening settings to inform screening policy and clinical decision making. As a critical component of this work, we will adapt an established model of prostate cancer natural history and detection to estimate overdiagnosis rates associated with screening under different assumptions about disease natural history, under different screening policies, and for different population subgroups defined by patient and tumor characteristics. These estimates will be made available to policy makers and clinicians via online calculators.
To achieve this work, we are requesting access to the Prostate dataset of the Prostate, Lung, Colorectal, and Ovarian cancer screening trial. Using these data, we will calibrate our model of natural history and practice patterns to reproduce incidence patterns in each arm by patient age and year of diagnosis and tumor stage and grade. In principle, because the intervention and control arms received different intensities of screening and because incidence data includes tumor stage and grade, the model can identify the risks of onset, progression from lower grade/stage to higher grade/stage states, and clinical detection (in the absence of screening) most consistent with incidence counts in both arms. We will implement different model specifications including: (1) all cancers begin as low grade and early stage and can progress to later grades and more advanced stages, (2) a fraction of cancers can progress to later grades and more advanced stages as in (1) and the remainder cannot progress, and (3) as in (2) but allowing only progression to later grades or only progression to more advanced stages.
Using the estimated models, we will estimate overdiagnosis frequencies for subgroups of individuals that vary in terms of age at diagnosis, severity of comorbid conditions, screening/biopsy histories, and tumor stage and grade. We previously estimate this model under a single set of natural history assumptions using PLCO data as part of an ongoing PLCO project (PLCO-9). The knowledge generated by this application will improve screening policies and informed clinical decisions around overdiagnosis risks and should provide a valuable tool in the ongoing battle to improve the balance of benefit and harm associated with cancer screening.
One objective of our ongoing research is to advance knowledge about how to validly estimate overdiagnosis and to provide useful information about overdiagnosis associated with specific cancer screening settings to inform screening policy and clinical decision making. As a critical component of this work, we will adapt an established model of prostate cancer natural history and detection to estimate overdiagnosis rates associated with screening under different assumptions about disease natural history, under different screening policies, and for different population subgroups defined by patient and tumor characteristics. These estimates will be made available to policy makers and clinicians via online calculators.
To achieve this work, we are requesting access to the Prostate dataset of the Prostate, Lung, Colorectal, and Ovarian cancer screening trial. Using these data, we will calibrate our model of natural history and practice patterns to reproduce incidence patterns in each arm by patient age and year of diagnosis and tumor stage and grade. In principle, because the intervention and control arms received different intensities of screening and because incidence data includes tumor stage and grade, the model can identify the risks of onset, progression from lower grade/stage to higher grade/stage states, and clinical detection (in the absence of screening) most consistent with incidence counts in both arms. We will implement different model specifications including: (1) all cancers begin as low grade and early stage and can progress to later grades and more advanced stages, (2) a fraction of cancers can progress to later grades and more advanced stages as in (1) and the remainder cannot progress, and (3) as in (2) but allowing only progression to later grades or only progression to more advanced stages.
Using the estimated models, we will estimate overdiagnosis frequencies for subgroups of individuals that vary in terms of age at diagnosis, severity of comorbid conditions, screening/biopsy histories, and tumor stage and grade. We previously estimate this model under a single set of natural history assumptions using PLCO data as part of an ongoing PLCO project (PLCO-9). The knowledge generated by this application will improve screening policies and informed clinical decisions around overdiagnosis risks and should provide a valuable tool in the ongoing battle to improve the balance of benefit and harm associated with cancer screening.
Aims
Estimate an existing cancer natural history model using detailed incidence data from the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial under varying assumptions about disease natural history, including whether all cancers are progressive or whether a fraction may be non-progressive. The estimated models will be used to estimate personalized (by patient age and comorbidity and tumor grade and stage) frequencies of overdiagnosis among prostate cancers detected by screening in the US and to examine sensitivity and bias associated with the natural history assumptions.
Collaborators
Roman Gulati, Fred Hutchinson Cancer Research Center