The CRC screening in community allied third-grade class-A hospital (21) has been launched in Guangzhou of China since 2015 (22). The screening protocol (see below) was conducted in Guangzhou primary care units and their allied hospitals qualified for colonoscopy. Short message service (SMS) reminders of free CRC screening were sent to all permanent residents aged 40–74 years old (22). The primary care units were required to sensitize eligible people within their jurisdictions to first complete the HRFQ then undergo FIT twice. Participants with a positive result in any of the above mentioned tests were referred to undergo colonoscopy at designated allied hospitals through advice notes issued by the screening physicians. Additionally, participants with negative HRFQ and negative FIT results were also allowed for colonoscopy after their physicians’ approval. The primary care units were responsible for the follow-up of all these participants and a specialized online CRC screening system was established for data collection and management (23).

For the present study, data collected between 2015 and 2021 from Yuexiu District of Guangzhou ( Supplementary Figure S1 ) were included. We focused on participants having colonoscopy outcomes but with negative results for both the HRFQ and FIT screens. All colonoscopy examinations were performed in the allied medical institutions including Sun Yat-sen University Cancer Center (SYSUCC), the largest oncology base in southern China.

The screening strategy, maintained in this 7-year study, has been documented in previous studies, including ours (22–24). In brief, after providing a written informed consent, all participants were asked to complete the primary screening, that is, a HRFQ questionnaire followed by double FIT tests within two consecutive weeks. The HRFQ collected basic information of the participants (e.g., sex and age) and their medical history and health-related behavior associated with CRC cancer (e.g., alcohol drinking and smoking history). Participants were determined to be HRFQ positive if they had 1) personal history of cancer, 2) personal history of intestinal polyps, 3) first-degree relative(s) with CRC, or 4) two or more of the following risk factors: 4a) chronic diarrhea (diarrhea lasting for more than 3 months in the past two years, and each episode lasting for more than 1 week), 4b) chronic constipation (over the past two years, constipation lasting for more than 2 months), 4c) mucoid or bloody feces (over the past two years), 4d) history of chronic appendicitis or appendectomy, 4e) history of chronic cholecystitis or cholecystectomy, and/or 4f) major trauma or painful event in the past 20 years. In addition, a second test of FIT was required for every participant regardless of the result of the first. And a FIT test was considered positive when the hemoglobin concentration in the sample was ≥ 100 ng/ml, which corresponds to ≥ 20 μ g Hb/g feces. Eventually, a primary screening was defined as positive if any of the HRFQ or FIT test was positive. During the colonoscopy, polyps with a diameter < 5 mm were resected, if possible; any neoplasm with a diameter ≥ 5 mm was biopsied first and proceeded with polypectomy or colectomy depending on pathological report and feasibility of endoscopic surgery (23, 24). Abnormal findings in colonoscopy included benign lesions (i.e., intestinal polyp, enterocolitis, and non-adenomatous lesions), non-advanced adenoma, advanced adenoma, and CRC. Non-advanced adenoma was defined as adenomas with the following features: size < 10 mm, tubular histology and low-grade dysplasia. Advanced adenoma was defined as an adenoma of  ≥ 10 mm or with a histological examination showing ≥ 25% villous component or high-grade dysplasia. If multiple lesions presented, the participant was classified by the most advanced one. When summarizing, advanced colorectal neoplasms (ACRN) were defined as a cancer or advanced adenoma, and overall colorectal neoplasms (CRN) were defined as a cancer or any adenoma.

The HRFQ is a self-administered questionnaire, in which data on medical history and health-related behavior were filled by the participants while physical measurements and laboratory tests were performed by trained staffs. In the questionnaire, the smoking status was defined as “yes” (current/former) if they smoked at least one cigarette a day for more than 6 months, while the alcohol drinking status was defined as “yes” (current/former) if he/she had a drink at least once a week for more than 6 months. Additionally, a history of night work was defined as working three or more hours between 12 pm and 5 am at least once a month which lasted for more than half a year during their entire career, and sedentary work was referred as working in a sitting, reclining or lying posture in office, driving, or consoles etc. for more than half of the working time during their whole career. Moreover, history of diabetes was self-reported.

Some indicators for analysis were transformed from the collected data. Among them, age was calculated as (date of informed consent - date of birth)/365.25. The marriage status was collected according to four categories (married, unmarried, divorced, or windowed), but summarized into two (married or others). The education background was retrieved from five categories (illiteracy, primary school, secondary or middle school, college or undergraduate, or graduate student), but aggregated into three (primary school or below, secondary or middle school, or college or above). The Body mass index (BMI) was calculated by dividing measured weight (kg) by the square of the height (m²) and categorized according to the recommendation of Working Group on Obesity in China (25), where overweight or obesity was defined by BMI ≥ 24 kg/m².

A total of 79,722 records during our study period were retrieved from the CRC screening system. Quality control was first performed on these data. Repeated screenings (n=8,829) were identified in advance by comparing their names and identification numbers, and only their earliest screening records were included, which ensures every single participant has their earliest CRC screening in their residential areas. Next, the following exclusion criteria were applied: a previous history of CRC (n=149), a previous history to undergo a colonoscopy, sigmoidoscopy, or barium contrast enema (n=101), and subjects with age<40 or >=75 (n=834). Finally, 69,809 participants were eligible and involved in our downstream analysis ( Figure 1 ).

In the analysis, continuous variables were presented as mean ± SD (standard deviation), while categorical variables were described as the frequency with percentage. Our analysis focused on a special sub-population who underwent colonoscopy with negative primary screening results and the discovery of potential reasons for false negatives. To detect differences between groups, a t-test for continuous variables, and a Chi-square test or Fisher’s exact test for categorical variables were used. Additionally, to explore the reasons for colonoscopy uptake within this special sub-population, a retrospective telephone survey was conducted with an initial cohort of 199 participants from three communities, in late 2024.

Binary logistic regression models were conducted to compare the predictive ability of different screening methods. To assess robustness, we conducted sensitivity analyses. First, we repeated the analyses after excluding HRFQ-identified symptomatic individuals among screening-negative participants who underwent colonoscopy (e.g., chronic diarrhea, constipation, mucous/bloody stools, appendicitis or appendectomy, and cholecystitis or cholecystectomy). Second, to account for key confounders that might affect the performance of FIT and HRFQ, multivariable logistic regression model adjusting for age, sex, BMI, education, and marital status was applied. All analyses were performed in R with a version of 4.2.2 (26). A two-sided P value<0.05 was considered statistically significant. When multiple comparison occurred, a Bonferroni correction was applied.

Between 2015 and 2021, a total of 69,809 eligible residents aged 40 to 74 in Yuexiu District of Guangzhou ( Supplementary Figure S1 ) attended the CRC screening. Only their first screening records were included in this study. All of them completed the HRFQ questionnaire with a positive rate of 19.9% and 80.0% of them underwent at least once FIT test with a positive rate of 11.4% ( Figure 1 ). A summary of these results showed 18,883 (27.0%) participants were identified as primary screening positives, defined by positive HRFQ or at least one positive FIT result, and the other 50,926 (73.0%) were categorized as negatives. Of these participants, 7,466 with positive primary screening and 527 with negative primary screening underwent colonoscopy. Surprisingly, a comparable prevalence of benign lesion(s) (18.7% vs. 17.5%), non-advanced adenoma (17.8% vs. 15.7%), advanced adenoma (10.5% vs. 9.7%) and CRC (2.1% vs. 1.5%) was found between these two sub-populations, with a P value of 0.287 showing no between-group difference ( Table 1 ).

To investigate the differences in the prevalence of dysplasia at various levels between primary screening positive and negative participants who underwent colonoscopy, additional statistical comparisons were made ( Table 1 ). Likewise, similar prevalence was found for CRN (30.4% vs. 26.9%, P = 0.101), ACRN (12.6% vs. 11.2%, P = 0.376), and CRC (2.1% vs. 1.5%, P = 0.418). These patterns persisted in a sensitivity analysis excluding HRFQ-identified symptomatic individuals from the screening-negative group, which yielded similar prevalence estimates ( Supplementary Table S1 ). This motivated us to further explore the demographic and clinical characteristics of these primary-screening-negative participants to identify and distinguish them from other asymptomatic populations for improvement in CRC screening programs.

Basic characteristics of subjects who underwent colonoscopy were first elucidated and compared between participants with negative and positive primary screening results ( Table 2 ). Of the 7,993 participants who did colonoscopy, 41.7% were men and their mean age was 59.9 ± 8.1 years. The comparison analysis demonstrated that among the population with colonoscopy for CRC diagnosis, the proportions of participants who were current or past smokers (4.7% vs. 10.4%, P < 0.001), had a history of night work (13.3% vs. 23.2%, P < 0.001), and had a sedentary job (26.9% vs. 44.2%, P < 0.001) were lower in subjects who got negative primary screening results than in those with positive primary screening results. Except for these, no significant between-group differences were found for other characteristics. The CRC risk assessment under the current screening protocol showed all the primary screening negative participants who did colonoscopy reached the low CRC risk criteria whereas 29.0% of them had one of the risk conditions ( Supplementary Table S2 ). Similar results were found for comparison between primary screening negative participants who did colonoscopy and all with positive primary screening ( Supplementary Table S3 ).

To investigate why participants who were negative in primary screening might benefit from a colonoscopy, baseline characteristics of subjects who did and did not undergo colonoscopy were compared ( Table 3 ). The results demonstrated that among participants with primary screening negative results, the proportion of middle-aged adults less than 60 (45.9% vs. 37.3%, P < 0.001) and people with college or above education background (24.5% vs. 17.2%, P < 0.001) was higher in subjects who did colonoscopy than those who did not. Moreover, participants that opted for a colonoscopy showed higher rates for choosing healthy behaviors, including less smoking (4.7% vs. 8.5%, P=0.002), having a history of night work (13.3% vs. 18.1%, P = 0.005), having sedentary work (26.9% vs. 35.4%, P < 0.001), and having a lower rate of diabetes (7.2% vs. 11.5%, P = 0.003); but more of them had one risk factor for CRC (29.0% vs. 22.1%, P<0.001), with significant differences in a history of chronic diarrhea (5.1% vs. 2.0%, P < 0.001), chronic constipation (9.3% vs. 4.7%, P < 0.001), or mucoid or bloody feces (5.7% vs. 3.7%, P = 0.020) in the past two year. To further explore the reasons for colonoscopy uptake among participants with negative primary screening, a retrospective telephone survey was conducted, but yielded a low response rate (13.1%, 26/199). Over half of the respondents (53.8%) could not recollect their reasons for pursuing the colonoscopy, while 26.9% reported self-perceived abnormalities in bowel movements ( Supplementary Table S4 ).

Further, we evaluated the detection ability of different screening methods for all grades of colorectal dysplasia to discover their contribution in lesion identification. Among participants undergoing colonoscopy, the overall detection rates for CRN, ACRN and CRC under current screening protocol were 30.2%, 12.5% and 2.1%, respectively ( Supplementary Table S5 ). Relative to negative FIT, positive FIT displayed higher detection rates for CRN and ACRN regardless of HRFQ results ( Figures 2A, B ), which were largely attributed to enhanced detection rates for AA and CRC but not non-AA ( Figures 2C–E ). This suggested FIT positive had a better risk predictive ability for AA and CRC. However, positive HRFQ showed an opposite trend. That is, negative HRFQ showed higher or close detection rates for CRN and ACRN whenever FIT was negative or not ( Figures 2A, B ). This might be mainly ascribed to the declining predictive ability of HRFQ for non-AA and AA but not CRC ( Figure 2C–E ). All these suggested that positive HRFQ does not have discriminatory ability for CRC, as expected.

The logistic regression demonstrated consistent results ( Figure 3 ). That is, FIT positives had significantly higher predictive ability for AA (OR: 2.87; 95% CI: [2.46, 3.34]) and CRC (OR: 8.33; 95% CI: [5.59, 12.88]), CRN (OR: 1.71; 95% CI: [1.54, 1.89]) and ACRN (OR: 3.56; 95% CI: [3.08 4.11]). Nevertheless, HRFQ positives showed significantly lower predictive ability for AA (OR: 0.44; 95% CI: [0.38, 0.50]) and CRC (OR: 0.39; 95% CI: 0.29, 0.53]), and CRN (OR: 0.66; 95% CI: [0.59, 0.73]), and ACRN (OR: 0.41; 95% CI: [0.36, 0.47]). Multivariable logistic models adjusting for age, sex, BMI, education, and marital status showed similar estimates, reinforcing the robustness of these findings ( Supplementary Figure S2 ).