Derivation and validation of a predictive model for advanced colorectal neoplasia in asymptomatic adults

Application of machine-learning model to optimize colonic adenoma detection in India

AIMS

There is limited data on the prevalence and risk factors of colonic adenoma from the Indian sub-continent. We aimed at developing a machine-learning model to optimize colonic adenoma detection in a prospective cohort.

METHODS

All consecutive adult patients undergoing diagnostic colonoscopy were enrolled between October 2020 and November 2022. Patients with a high risk of colonic adenoma were excluded. The predictive model was developed using the gradient-boosting machine (GBM)-learning method. The GBM model was optimized further by adjusting the learning rate and the number of trees and 10-fold cross-validation.

RESULTS

Total 10,320 patients (mean age 45.18 ± 14.82 years; 69% men) were included in the study. In the overall population, 1152 (11.2%) patients had at least one adenoma. In patients with age > 50 years, hospital-based adenoma prevalence was 19.5% (808/4144). The area under the receiver operating curve (AUC) (SD) of the logistic regression model was 72.55% (4.91), while the AUCs for deep learning, decision tree, random forest and gradient-boosted tree model were 76.25% (4.22%), 65.95% (4.01%), 79.38% (4.91%) and 84.76% (2.86%), respectively. After model optimization and cross-validation, the AUC of the gradient-boosted tree model has increased to 92.2% (1.1%).

CONCLUSIONS

Machine-learning models may predict colorectal adenoma more accurately than logistic regression. A machine-learning model may help optimize the use of colonoscopy to prevent colorectal cancers.

TRIAL REGISTRATION

ClinicalTrials.gov (ID: NCT04512729).

Colorectal Cancer Risk Assessment and Precision Approaches to Screening: Brave New World or Worlds Apart?

Current colorectal cancer (CRC) screening recommendations take a "one-size-fits-all" approach using age as the major criterion to initiate screening. Precision screening that incorporates factors beyond age to risk stratify individuals could improve on current approaches and optimally use available resources with benefits for patients, providers, and health care systems. Prediction models could identify high-risk groups who would benefit from more intensive screening, while low-risk groups could be recommended less intensive screening incorporating noninvasive screening modalities. In addition to age, prediction models incorporate well-established risk factors such as genetics (eg, family CRC history, germline, and polygenic risk scores), lifestyle (eg, smoking, alcohol, diet, and physical inactivity), sex, and race and ethnicity among others. Although several risk prediction models have been validated, few have been systematically studied for risk-adapted population CRC screening. In order to envisage clinical implementation of precision screening in the future, it will be critical to develop reliable and accurate prediction models that apply to all individuals in a population; prospectively study risk-adapted CRC screening on the population level; garner acceptance from patients and providers; and assess feasibility, resources, cost, and cost-effectiveness of these new paradigms. This review evaluates the current state of risk prediction modeling and provides a roadmap for future implementation of precision CRC screening.

A risk scoring system for advanced colorectal neoplasia in high-risk participants to improve current colorectal cancer screening in Tianjin, China

Background

Given the limited effectiveness of the current Chinese colorectal cancer (CRC) screening procedure, adherence to colonoscopy remains low. We aim to develop and validate a scoring system based on individuals who were identified as having a high risk in initial CRC screening to achieve more efficient risk stratification and improve adherence to colonoscopy.

Methods

A total of 29,504 screening participants with positive High-Risk Factor Questionnaire (HRFQ) or faecal immunochemical test (FIT) who underwent colonoscopy in Tianjin from 2012–2020 were enrolled in this study. Binary regression analysis was used to evaluate the association between risk factors and advanced colorectal neoplasia. Internal validation was also used to assess the performance of the scoring system.

Results

Male sex, older age (age ≥ 50 years), high body mass index (BMI ≥ 28 kg/m2), current or past smoking and weekly alcohol intake were identified as risk factors for advanced colorectal neoplasm. The odds ratios (ORs) for significant variables were applied to construct the risk score ranging from 0–11: LR, low risk (score 0–3); MR, moderate risk (score 4–6); and HR, high risk (score 7–11). Compared with subjects with LR, those with MR and HR had ORs of 2.47 (95% confidence interval, 2.09–2.93) and 4.59 (95% confidence interval, 3.86–5.44), respectively. The scoring model showed an outstanding discriminatory capacity with a c-statistic of 0.64 (95% confidence interval, 0.63–0.65).

Conclusions

Our results showed that the established scoring system could identify very high-risk populations with colorectal neoplasia. Combining this risk score with current Chinese screening methods may improve the effectiveness of CRC screening and adherence to colonoscopy.

Risk Scoring Systems for Predicting the Presence of Colorectal Neoplasia by Fecal Immunochemical Test Results in Chinese Population

INTRODUCTION

Adherence to colonoscopy screening for colorectal cancer (CRC) is low in general populations, including those tested positive in the fecal immunochemical test (FIT). Developing tailored risk scoring systems by FIT results may allow for more accurate identification of individuals for colonoscopy.

METHODS

Among 807,109 participants who completed the primary tests in the first-round Shanghai CRC screening program, 71,023 attended recommended colonoscopy. Predictors for colorectal neoplasia were used to develop respective scoring systems for FIT-positive or FIT-negative populations using logistic regression and artificial neural network methods.

RESULTS

Age, sex, area of residence, history of mucus or bloody stool, and CRC in first-degree relatives were identified as predictors for CRC in FIT-positive subjects, while a history of chronic diarrhea and prior cancer were additionally included for FIT-negative subjects. With an area under the receiver operating characteristic curve of more than 0.800 in predicting CRC, the logistic regression-based systems outperformed the artificial neural network-based ones and had a sensitivity of 68.9%, a specificity of 82.6%, and a detection rate of 0.24% by identifying 17.6% subjects at high risk. We also reported an area under the receiver operating characteristic curve of about 0.660 for the systems predicting CRC and adenoma, with a sensitivity of 57.8%, a specificity of 64.6%, and a detection rate of 6.87% through classifying 38.1% subjects as high-risk individuals. The performance of the scoring systems for CRC was superior to the currently used method in Mainland, China, and comparable with the scoring systems incorporating the FIT results.

DISCUSSION

The tailored risk scoring systems may better identify high-risk individuals of colorectal neoplasia and facilitate colonoscopy follow-up. External validation is warranted for widespread use of the scoring systems.

External validation of models for predicting risk of colorectal cancer using the China Kadoorie Biobank

BACKGROUND

In China, colorectal cancer (CRC) incidence and mortality have been steadily increasing over the last decades. Risk models to predict incident CRC have been developed in various populations, but they have not been systematically externally validated in a Chinese population.  This study aimed to assess the performance of risk scores in predicting CRC using the China Kadoorie Biobank (CKB), one of the largest and geographically diverse prospective cohort studies in China.

METHODS

Nine models were externally validated in 512,415 participants in CKB and included 2976 cases of CRC. Model discrimination was assessed, overall and by sex, age, site, and geographic location, using the area under the receiver operating characteristic curve (AUC). Model discrimination of these nine models was compared to a model using age alone. Calibration was assessed for five models, and they were re-calibrated in CKB.

RESULTS

The three models with the highest discrimination (Ma (Cox model) AUC 0.70 [95% CI 0.69-0.71]; Aleksandrova 0.70 [0.69-0.71]; Hong 0.69 [0.67-0.71]) included the variables age, smoking, and alcohol. These models performed significantly better than using a model based on age alone (AUC of 0.65 [95% CI 0.64-0.66]). Model discrimination was generally higher in younger participants, males, urban environments, and for colon cancer. The two models (Guo and Chen) developed in Chinese populations did not perform better than the others. Among the 10% of participants with the highest risk, the three best performing models identified 24-26% of participants that went on to develop CRC.

CONCLUSIONS

Several risk models based on easily obtainable demographic and modifiable lifestyle factor have good discrimination in a Chinese population. The three best performing models have a higher discrimination than using a model based on age alone.

A scoping review of risk-stratified bowel screening: current evidence, future directions

PURPOSE

In this scoping review, we examined the international literature on risk-stratified bowel screening to develop recommendations for future research, practice and policy.

METHODS

Six electronic databases were searched from inception to 18 October 2021: Medline, Embase, PsycINFO, CINAHL, Cochrane Database of Systematic Reviews and Cochrane Central Register of Controlled Trials. Forward and backwards citation searches were also undertaken. All relevant literature were included.

RESULTS

After de-deduplication, 3,629 records remained. 3,416 were excluded at the title/abstract screening stage. A further 111 were excluded at full-text screening stage. In total, 102 unique studies were included. Results showed that risk-stratified bowel screening programmes can potentially improve diagnostic performance, but there is a lack of information on longer-term outcomes. Risk models do appear to show promise in refining existing risk stratification guidelines but most were not externally validated and less than half achieved good discriminatory power. Risk assessment tools in primary care have the potential for high levels of acceptability and uptake, and therefore, could form an important component of future risk-stratified bowel screening programmes, but sometimes the screening recommendations were not adhered to by the patient or healthcare provider. The review identified important knowledge gaps, most notably in the area of organisation of screening services due to few pilots, and what risk stratification might mean for inequalities.

CONCLUSION

We recommend that future research focuses on what organisational challenges risk-stratified bowel screening may face and a consideration of inequalities in any changes to organised bowel screening programmes.

Improved risk scoring systems for colorectal cancer screening in Shanghai, China

Background

An optimal risk‐scoring system enables more targeted offers for colonoscopy in colorectal cancer (CRC) screening. This analysis aims to develop and validate scoring systems using parametric and non‐parametric methods for average‐risk populations.

Methods

Screening data of 807,695 subjects and 2806 detected cases in the first‐round CRC screening program in Shanghai were used to develop risk‐predictive models and scoring systems using logistic‐regression (LR) and artificial‐neural‐network (ANN) methods. Performance of established scoring systems was evaluated using area under the receiver operating characteristic curve (AUC), calibration, sensitivity, specificity, number of high‐risk individuals and potential detection rates of CRC.

Results

Age, sex, CRC in first‐degree relatives, chronic diarrhoea, mucus or bloody stool, history of any cancer and faecal‐immunochemical‐test (FIT) results were identified as predictors for the presence of CRC. The AUC of LR‐based system was 0.642 when using risk factors only in derivation set, and increased to 0.774 by further incorporating one‐sample FIT results, and to 0.808 by including two‐sample FIT results, while those for ANN‐based systems were 0.639, 0.763 and 0.805, respectively. Better calibrations were observed for the LR‐based systems than the ANN‐based ones. Compared with the currently used initial tests, parallel use of FIT with LR‐based systems resulted in improved specificities, less demands for colonoscopy and higher detection rates of CRC, while parallel use of FIT with ANN‐based systems had higher sensitivities; incorporating FIT in the scoring systems further increased specificities, decreased colonoscopy demands and improved detection rates of CRC.

Conclusions

Our results indicate the potentials of LR‐based scoring systems incorporating one‐ or two‐sample FIT results for CRC mass screening. External validation is warranted for scaling‐up implementation in the Chinese population.

The established scoring systems derived from the logistic regression (LR) models, incorporating one‐ or two‐sample faecal immunochemical test (FIT) results as a predictor, have the potential to triage high‐risk individuals for colonoscopy in mass screening of colorectal cancer (CRC). More importantly, the cut‐off points of the scoring systems can be adjusted flexibly, facilitating the choices of cut‐off values for populations with abundant or limited resources.

Utility of machine learning in developing a predictive model for early-age-onset colorectal neoplasia using electronic health records

BACKGROUND AND AIMS

The incidence of colorectal cancer (CRC) is increasing in adults younger than 50, and early screening remains challenging due to cost and under-utilization. To identify individuals aged 35-50 years who may benefit from early screening, we developed a prediction model using machine learning and electronic health record (EHR)-derived factors.

METHODS

We enrolled 3,116 adults aged 35-50 at average-risk for CRC and underwent colonoscopy between 2017-2020 at a single center. Prediction outcomes were (1) CRC and (2) CRC or high-risk polyps. We derived our predictors from EHRs (e.g., demographics, obesity, laboratory values, medications, and zip code-derived factors). We constructed four machine learning-based models using a training set (random sample of 70% of participants): regularized discriminant analysis, random forest, neural network, and gradient boosting decision tree. In the testing set (remaining 30% of participants), we measured predictive performance by comparing C-statistics to a reference model (logistic regression).

RESULTS

The study sample was 55.1% female, 32.8% non-white, and included 16 (0.05%) CRC cases and 478 (15.3%) cases of CRC or high-risk polyps. All machine learning models predicted CRC with higher discriminative ability compared to the reference model [e.g., C-statistics (95%CI); neural network: 0.75 (0.48-1.00) vs. reference: 0.43 (0.18-0.67); P = 0.07] Furthermore, all machine learning approaches, except for gradient boosting, predicted CRC or high-risk polyps significantly better than the reference model [e.g., C-statistics (95%CI); regularized discriminant analysis: 0.64 (0.59-0.69) vs. reference: 0.55 (0.50-0.59); P<0.0015]. The most important predictive variables in the regularized discriminant analysis model for CRC or high-risk polyps were income per zip code, the colonoscopy indication, and body mass index quartiles.

DISCUSSION

Machine learning can predict CRC risk in adults aged 35-50 using EHR with improved discrimination. Further development of our model is needed, followed by validation in a primary-care setting, before clinical application.