Quantifying cancer screening overdiagnosis using synthetic excess incidence
An important concern in cancer screening is overdiagnosis, detecting preclinical cancer on screening that would not have developed into symptomatic cancer in the absence of screening. New liquid biopsy cancer screening technologies are being developed which require quantification of overdiagnosis. Because the technology is developing rapidly, this quantification should involve only short-term observational data. Unfortunately estimating the fraction overdiagnosed is “notoriously difficult,” even for approximations. Estimation based on previous modeling methods strongly relies on assumptions about the natural history of cancer. My objective was to develop a method for estimating the fraction overdiagnosed based on short-term observational data with less reliance on assumptions about the natural history of cancer.
Methods and Results.
I framed the problem as computing synthetic excess incidence for periodic screening versus no screening based on data from at least two screens. This allowed me to apply previous methodology to estimate the age-specific incidence of cancer in the absence of screening. I also modeled cancer incidence after the oldest age screened which yielded estimated lower and upper bounds with a sensitivity analysis. Applying the methodology to short-term observational data from lung (MLP), breast (HIP) and colon cancer screening trials (MCSS), I obtained estimates of the fraction overdiagnosed that were consistent with estimates from other methods.
Project Goal: Apply methodology to PLCO data
This request applies to PLCO lung, prostate, and ovarian cancer.
Part 1. (Synthetic excess incidence method)
Separately for lung, prostate, and ovarian cancer, I would like a table in the following form.
The rows are age (in years at the time of the screen).
The columns are (1) number who received first screen, (2) number positive on the first screen and positive for cancer on workup, (3) number who received a subsequent screen, (4) number positive on the subsequent screen and positive for cancer on work-up.
(a) For prostate and ovarian cancer, a screen refers to either screening test.
(b) A subsequent screen is a screen occurring 1 year after a negative screen
(c) For PLCO lung, first screen is T0 and subsequent screens are T1 and T2 (if following a negative screen)
(d) For PLCO prostate, first screen is T0 and subsequent screens are T1 , T2, T3 (if following a negative screen)
(e) For PLCO ovarian, first screen is T0 and subsequent screens are T1 , T2, T3 (if following a negative screen)
Part 2 (alternative method of estimation for comparison)
Separately for lung, prostate, and ovarian cancer, for each arm (control and screened), I would like a table where the rows are year since randomization and the columns are (1) number of cancers cases in control group diagnosed in that year (either screening or with symptoms), (2) number of cancers cases in screened group diagnosed in that year (either screening or with symptoms) (3) the number of cancer cases detected on screening in that year (so zeros after the time of the last screen),
Hormuzd Katki NCI
Ping Hu NCI
Philip Prorok NCI (contractor)