Assessing overdiagnosis of fecal immunological test screening for colorectal cancer with a digital twin approach

Abstract

Evaluating the magnitude of overdiagnosis associated with stool-based service screening for colorectal cancer (CRC) beyond a randomized controlled trial is often intractable and understudied. We aim to estimate the proportion of overdiagnosis in population-based service screening programs for CRC with the fecal immunochemical test (FIT). The natural process of overdiagnosis-embedded disease was first built up to learn transition parameters that quantify the pathway of non-progressive and progressive screen-detected cases calibrated with sensitivity, while also taking competing mortality into account. The Markov algorithms were then developed for estimating these transition parameters based on Taiwan FIT service CRC screening data on 5,417,699 residents aged 50-69 years from 2004 to 2014. Following the digital twin design with the parallel universe structure for emulating the randomized controlled trial, the screened twin, mirroring the control group without screening, was virtually recreated by the application of the above-mentioned trained parameters to predict CRC cases containing overdiagnosis. The ratio of the predicted CRCs derived from the screened twin to the observed CRCs of the control group minus 1 was imputed to measure the extent of overdiagnosis. The extent of overdiagnosis for invasive CRCs resulting from FIT screening is 4.16% (95% CI: 2.61-5.78%). The corresponding figure is increased to 9.90% (95% CI: 8.41-11.42%) for including high grade dysplasia (HGD) and further inflated to 15.83% (95% CI: 15.23-16.46%) when the removal adenoma is considered. The modest proportion of overdiagnosis modelled by the digital twin method, dispensing with the randomized controlled trial design, suggests the harm done to population-based FIT service screening is negligible.

Fig. 1. Randomized controlled trial (RCT) and digital twin design for overdiagnosis in population-based screening.

a The design of randomized controlled trial for evaluation of overdiagnosis; b the hidden disease natural history process from normal, PCDP, until the clinical phase among the control group in the absence of screening; c the overdiagnosis-embedded multi-state Markov model and algorithms for learning parameters with adjustment for sensitivity and competing mortality from the screened group; d the overdiagnosis-embedded multi-state Markov model and algorithms with adenoma for learning parameters with adjustment for sensitivity and competing mortality from the screened group.