To the Editor,

Risk-stratification screening efficiently reduces the incidence and mortality rate of colorectal cancer (CRC) [1], but no risk-stratification model has been extensively validated in China where the limited colonoscopy resources are mainly occupied by low-risk individuals with non-specific gastrointestinal symptoms (NSGS) [2, 3], who are considered equivalent to average-risk population for the risk of advanced colorectal neoplasia (ACN) [4, 5]. Current guidelines struggle to recommend to start colonoscopy screening at the age of 45 or 50 [6, 7]. However, any “one-size-fits-all” standard for age may prevent the detection of many early-onset CRCs [8, 9]. Herein, under the dilemma of inefficient detection, limited resources, and increasing early-onset CRC screening faced by colonoscopy practice, we developed and validated a risk-stratification model  for colorectal neoplasia (CN).

From 2018 to 2020, the National Colorectal Polyp Care (NCPC) program was implemented in 175 centers nationwide (Fig. 1A), where consecutive adult individuals who had no alarming symptoms or signs of CRC were enrolled, regardless of NSGS [4]. All participants completed questionnaires regarding baseline information and life risk factors and received fecal immunochemical tests (FITs) and colonoscopies. A central database was established to manage the uploaded data from all centers (accessed at www.ncrcgastro.org). The primary outcome was the CN [10]. The details of methods, including exclusion criteria, outcome measures, sample size calculation, and statistical analysis, are illustrated in Additional file 1: Supplementary Methods.

Fig. 1
figure 1

A The distribution of 175 participating centers in the provincial-level administrative regions of China. B Flowchart of enrollment, allocation, and study design. C Independent risk factors for colorectal neoplasia in the multivariate logistic regression model and points assigned to the NCPC score. * Points were assigned by dividing the Log-Odds coefficients by the absolute value of the smallest coefficient (BMI 0.163) and rounding up to the nearest integer. D Predicting performance of NCPC score in the derivation cohort, validation cohort, NSGS cohort, and asymptomatic cohort. * No significant differences were found for AUC between derivation and validation cohort (P = 0.80) or between NSGS and asymptomatic cohort (P = 0.31). NCPC, national colorectal polyp care; CEA, carcinoembryonic antigen; FIT, fecal immunochemical test; CRC, colorectal cancer; CN, colorectal neoplasia; OR, odds ratio; CI, confidence interval; BMI, body mass index; FDR, first-degree relative; PNC, previous negative colonoscopy; NSGS, non-specific gastrointestinal symptom; and AUC, area under the receiver operating characteristic curve

Full size image

A total of 10,164 participants were enrolled (Fig. 1B), whose clinical characteristics were comparable between the derivation and validation cohort (Additional file 2: Tables S1–2). The univariate analysis identified 11 potential risk factors, and eight variables (sex, age, body mass index, smoking, drinking, diabetes, first-degree relative of CRC, history of previous negative colonoscopy) were identified as independent predicting factors for developing NCPC score (Additional file 2: Table S3, Fig. 1C), while the other variables were excluded (Additional file 2: Table S4). The NCPC score was divided into three levels according to the mean CN prevalence: low risk (LR 0–14, 0–17.4%), intermediate risk (IR 15–17, 18.8–24.0%), and high risk (HR 18–28, ≥ 25.9%) (Additional file 2: Table S5). Compared with FIT- individuals, FIT+ individuals showed higher risks for CN, ACN, and CRC in all subgroups of NCPC score (all P < 0.001) (Additional file 2: Table S6). Therefore, the risk-stratification model (Changhai Li’s Model) triaged individuals with IR or HR NCPC scores or FIT+ as increased-risk individuals to receive colonoscopy.

The model showed good calibrations no significant difference of area under curve (AUC) between the derivation and validation cohort (0.68 vs. 0.68, P = 0.80); consistent predicting performance in risk-stratification ability and individuals’ distribution were confirmed in the deviation and validation cohort (Figs. 1D, 2A). No significant difference of AUC was also found between NSGS and asymptomatic population (0.68 vs. 0.67, P = 0.31), where the predicting performances were demonstrated to be similar; individuals’ distribution and prevalence of CN and ACN were also found to be consistent between NSGS and asymptomatic individuals (Figs. 1D, 2B).

Fig. 2
figure 2

A–B Risk-stratification based comparisons of CN, ACN, and CRC prevalence between derivation and validation cohort or between NSGS and asymptomatic cohort. & Low risk represents participants with FIT- and low-risk score, and high risk represents participants with FIT+ or intermediate/high-risk score; * P value for intermediate risk vs. low risk; # P value for high risk vs. low risk. C–E Comparison of AUCs for CN, ACN, and CRC between the NCPC score and selected risk models or FIT for overall cohort. F Performance of NCPC score or risk-stratification model guided colonoscopy and estimated reduction of colonoscopy burden. & Low risk represents participants with FIT- and low-risk score, and high risk represents participants with FIT + or intermediate/high-risk score; reduction of NNS = (NNS by primary colonoscopy – NNS by NCPC (+FIT)-based algorithm)/(NNS by primary colonoscopy). AUC, area under the receiver operating characteristic curve; CN, colorectal neoplasia; CI, confidence interval; ACN, advanced colorectal neoplasia; CRC, colorectal cancer; NCPC, National Colorectal Polyp Care; BMI, body mass index; FDR, first-degree relative; PNC, previous negative colonoscopy; APCS, Asia–Pacific Colorectal Screening score; FIT, fecal immunochemical test; RR, relative risk. NNS, number needed for screening colonoscopy to detect one lesion; and ROC, receiver operating characteristic

Full size image

Compared with FIT or other Asian models, the NCPC score showed the best discriminative ability for CN [0.67, P < 0.001] and ACN [0.70, P < 0.001 or = 0.002] [1, 11, 12] (Fig. 2C-E). The NCPC score could identify 70.7% CN, 77.7% ACN, and 78.7% CRC when reducing 29.2%, 35.5%, and 36.4% number needed for screening colonoscopies to detect one lesion (NNS), respectively (Fig. 2F). The risk-stratification model could identify 73.5% CN, 82.6% ACN, and 93.6% CRC when recommending 52.7% individuals to receive colonoscopy (Fig. 2F). By using risk-stratification model, only 25.6% young individuals will be recommended for colonoscopy, and 55.8% ACN and 72.7% CRC of young population could be identified when reducing 54.2% and 64.8% corresponding NNS, respectively (Fig. 2F).

In summary, a risk-stratification model (Changhai Li’s Model) for CN, consisting of FIT and NCPC score, was developed and validated to improve the efficiency of CRC screening. The model was able to save almost a half colonoscopy resources when maintaining a high sensitivity for CN, ACN, and CRC. Notably, 55.8% early-onset ACN and 72.7% early-onset CRC were identified with only 25.6% young individuals receiving colonoscopy. Consistent risk-stratifying performance was demonstrated between NSGS and asymptomatic population, which could rationally promote scope of CRC screening to cover the previously “ignored” NSGS population and avoid “indication gaming.” This model holds the promise as a feasible risk-stratification approach to improve the colonoscopy efficiency and CRC-screening scope in China and other countries with limited resources.