Cancer Causes & Control · Nov 6, 2026
Proxy-mediated bias in non-invasive colorectal cancer risk predictors: an evidence appraisal for multiethnic Southeast Asian population
By Suhaimei US, Rahman HA, Kassim N et al.
Purpose Colorectal cancer (CRC) is rising across Southeast Asia (SEA) and leads by incidence in Brunei Darussalam. Machine learning (ML) applications for CRC in ASEAN remain endoscopy-based and diagnostic; no risk-stratification model using routinely collected non-invasive data has been reported. This review asks not which predictors associate with CRC risk, but which can serve as valid, equitable ML inputs in multiethnic SEA populations, where lifestyle variables may encode ethnic or religious identity rather than biological risk (proxy-mediated bias). Methods This narrative review used qualitative synthesis rather than meta-analysis. PubMed, Scopus and Google Scholar were searched to 9 July 2026 (English language). Six predictor categories were appraised for Asian evidence strength, SEA-specific data, and proxy-bias risk: dietary and lifestyle factors, anthropometric measures, faecal immunochemical test (FIT), complete blood count (CBC), medication history, and emerging predictors; 22 studies were prioritised for appraisal. Results FIT and BMI (Asian threshold ≥23.0 kg/m²) showed the most consistent Asian evidence; FIT, with age, sex and CBC, carried the lowest proxy-bias risk. CBC-based ML performs well in Western populations (AUC 0.736–0.82) but is unvalidated in any Asian population. The central paradox: the most available predictor has the weakest regional evidence, whilst those with strongest Asian evidence are most entangled with identity. Lifestyle predictors require explicit proxy-bias management; medication-history evidence was inconsistent. Conclusion A four-criterion framework (effect size, biological plausibility, proxy-bias risk, and routine data availability) and a two-pathway delivery model provide an evidence-grounded basis for equitable, risk-stratified CRC early detection in Brunei and comparable SEA systems.