Objective To compare the performance of existing sigmoidoscopy-based strategies in predicting advanced proximal neoplasia (APN) in an asymptomatic Chinese cohort.

Design We included all screening participants aged 50–70 years who received colonoscopy between 2008 and 2014 in Hong Kong. Sigmoidoscopy yield was estimated from the colonoscopic findings based on the: (1) UK flexible sigmoidoscopy; (2) Screening for COlon REctum (SCORE); (3) NORwegian Colorectal Cancer Prevention (NORCCAP) trials and (4) US clinical index based on age, gender and distal findings. The sensitivity, specificity, the number of subjects needed to screen (NNS) and the number of subjects needed to refer (NNR) for colonoscopy to detect one APN were evaluated. Binary logistic regression modelling identified the distal findings associated with APN.

Results From 5879 eligible subjects, 132 (2.2%) had APN. The US strategy achieved the highest sensitivity for APN detection (42.0%) and the UK criteria attained the highest specificity (96%). The US criteria led to the lowest NNS (92 vs 103–267) and the UK criteria required the least NNR (12 vs 16–21). Using the US strategy, the rates of APN detected were 1.4% (low-risk group), 2.2% (intermediate risk) and 5.9% (high risk). The c-statistics of the UK, SCORE, NORCCAP and the US criteria were 0.55±0.03; 0.59±0.03; 0.59±0.03 and 0.62±0.05 respectively.

Conclusions The US criteria had the highest sensitivity for detection of APN and lowest NNS and the UK score had the highest specificity and the lowest NNR. The performance of all these four criteria to predict APN is limited, highlighting an urgent need to devise a novel APN prediction system for Asian subjects.