Risk prediction of advanced colorectal neoplasia varies by race and neighbourhood socioeconomic status.

OBJECTIVE

Neighbourhood deprivation increases the risk of colorectal neoplasia and contributes to racial disparities observed in this disease. Developing race-specific advanced colorectal neoplasia (ACN) prediction models that include neighbourhood socioeconomic status has the potential to improve the accuracy of prediction.

METHODS

The study includes 1457 European Americans (EAs) and 936 African Americans (AAs) aged 50-80 years undergoing screening colonoscopy. Race-specific ACN risk prediction models were developed for EAs and AAs, respectively. Area Deprivation Index (ADI), derived from 17 variables of neighbourhood socioeconomic status, was evaluated by adding it to the ACN risk prediction models. Prediction accuracy was evaluated by concordance statistic (C-statistic) for discrimination and Hosmer-Lemeshow goodness-of-fit test for calibration.

RESULTS

With fewer predictors, the EA-specific and AA-specific prediction models had better prediction accuracy in the corresponding race/ethnic subpopulation than the overall model. Compared with the overall model which had poor calibration ( =0.053 in the whole population and =0.011 in AAs), the EA model had C-statistic of 0.655 (95% CI 0.594 to 0.717) and =0.663; and the AA model had C-statistic of 0.637 ((95% CI 0.572 to 0.702) and =0.810. ADI was a significant predictor of ACN in EAs (OR=1.24 ((95% CI 1.03 to 1.50), =0.029), but not in AAs (OR=1.07 ((95% CI 0.89 to 1.28), =0.487). Adding ADI to the EA-specific ACN prediction model substantially improved ACN calibration accuracy of the prediction across area deprivation groups ( =0.924 with ADI vs =0.140 without ADI) in EAs.

CONCLUSIONS

Neighbourhood socioeconomic status is an important factor to consider in ACN risk prediction modeling. Moreover, non-race-specific prediction models have poor generalisability. Race-specific prediction models incorporating neighbourhood socioeconomic factors are needed to improve ACN prediction accuracy.