Design of randomized controlled trials to estimate cancer-mortality reductions from multicancer detection screening.

Abstract

Determining whether screening with multicancer detection (MCD) tests saves lives requires randomized controlled trials (RCTs). To inform RCT design, we estimated cancer-mortality outcomes from hypothetical MCD RCTs. We used the Hu-Zelen model, previously used to plan the National Lung Screening Trial (NLST), to estimate mortality reductions, sample size, and power for 9 cancers for different RCT design parameters and MCD test parameters. Our base-case RCT with 5 yearly screens and 100 000 people ages 60-74 in each arm, who also undergo standard-of-care screens, has 87%-89% power to detect a 9%-10% mortality reduction (Number Needed to Screen [NNS] = 578-724) over 7-9 years. The majority of prevented deaths were from lung cancers (base-case [64%-66%] and all sensitivity analyses), 8%-10% from colorectal cancer, and 26% from the other 7 cancers, mostly from stomach or ovary or esophagus (due to excellent stage 1 survival) and less from liver or pancreas (poor stage 1 survival) or head and neck or lymphoma (excellent stage 4 survival). There was limited power for mortality reductions at most individual cancer sites. Base-case findings were sensitive to test sensitivity, stage shifts, and mean sojourn times in the preclinical state (especially for lung cancer), but 90% power could be recovered by recruiting a substantially higher risk population. Large-scale MCD RCTs would have 89% power to detect an approximate 10% cancer mortality reduction over a relatively short 7-9 year timeframe from trial entry. The estimated NNS for MCD RCTs compares favorably with mammographic screening. Most prevented cancer deaths in a well-powered MCD RCT would likely be from lung cancer. Non-lung and non-colorectal cancer sites could be a meaningful proportion of prevented cancer deaths, but power is insufficient to isolate non-lung-cancer mortality reductions.