Development and validation of a prediction model for early screening of people at high risk for colorectal cancer

Abstract

Background: Colorectal cancer (CRC) is a serious threat worldwide. Although early screening is suggested to be the most effective method to prevent and control CRC, the current situation of early screening for CRC is still not optimistic. In China, the incidence of CRC in the Yangtze River Delta region is increasing dramatically, but few studies have been conducted. Therefore, it is necessary to develop a simple and efficient early screening model for CRC.

Aim: To develop and validate an early-screening nomogram model to identify individuals at high risk of CRC.

Methods: Data of 64448 participants obtained from Ningbo Hospital, China between 2014 and 2017 were retrospectively analyzed. The cohort comprised 64448 individuals, of which, 530 were excluded due to missing or incorrect data. Of 63918, 7607 (11.9%) individuals were considered to be high risk for CRC, and 56311 (88.1%) were not. The participants were randomly allocated to a training set (44743) or validation set (19175). The discriminatory ability, predictive accuracy, and clinical utility of the model were evaluated by constructing and analyzing receiver operating characteristic (ROC) curves and calibration curves and by decision curve analysis. Finally, the model was validated internally using a bootstrap resampling technique.

Results: Seven variables, including demographic, lifestyle, and family history information, were examined. Multifactorial logistic regression analysis revealed that age [odds ratio (OR): 1.03, 95% confidence interval (CI): 1.02-1.03, P < 0.001], body mass index (BMI) (OR: 1.07, 95%CI: 1.06-1.08, P < 0.001), waist circumference (WC) (OR: 1.03, 95%CI: 1.02-1.03 P < 0.001), lifestyle (OR: 0.45, 95%CI: 0.42-0.48, P < 0.001), and family history (OR: 4.28, 95%CI: 4.04-4.54, P < 0.001) were the most significant predictors of high-risk CRC. Healthy lifestyle was a protective factor, whereas family history was the most significant risk factor. The area under the curve was 0.734 (95%CI: 0.723-0.745) for the final validation set ROC curve and 0.735 (95%CI: 0.728-0.742) for the training set ROC curve. The calibration curve demonstrated a high correlation between the CRC high-risk population predicted by the nomogram model and the actual CRC high-risk population.

Conclusion: The early-screening nomogram model for CRC prediction in high-risk populations developed in this study based on age, BMI, WC, lifestyle, and family history exhibited high accuracy.

Keywords: Colorectal cancer; Dietary habit; Early screening model; High-risk population; Living habit; Nomogram model; Questionnaire survey.

Figure 1

Nomogram for predicting early screening of individuals at high risk of colorectal cancer. The value of each variable was scored on a point scale from 0 to 100, after which the scores for each variable were added together. The total sum was located on the total points axis, which enabled us to predict the probability of early screening of individuals at high risk of colorectal cancer. Age, body mass index, and waist circumference were used as continuous variables. The family history group 0 = no and 1 = yes, and lifestyle group 1 = unhealthy lifestyle and 2 = healthy lifestyle. BMI: Body mass index; WC: Waist circumference.

Figure 2

Receiver operating characteristic curves curve for predicting early screening of individuals at high risk of colorectal cancer. A: Validation set: Receiver operating characteristic curves (ROC) curve for the nomogram generated using bootstrap resampling (500 times); B: Training set: ROC curve for the nomogram generated using bootstrap resampling (500 times). AUC: Area under the curve.

Figure 3

Calibration pot for predicting early screening of individuals at high risk of colorectal cancer. A: Validation set nomogram calibration plot; B: Training set nomogram calibration plot. When the solid line (performance nomogram) is closer to the dotted line (ideal model), the prediction accuracy of the nomogram is better.

Figure 4

Decision curve analysis for predicting early screening of individuals at high risk of colorectal cancer. A: Decision curve analysis (DCA) of the validation set prediction model; B: DCA of the training set prediction model. Red solid lines indicate the prediction models, gray solid lines indicate all populations at high risk for colorectal cancer (CRC), and solid horizontal lines indicate non-high-risk populations for CRC. The graph depicts the expected net gain for each individual relative to the high-risk CRC Nuo plot forecast. Net gain increases with the model curve.