Development and internal validation of a prediction model for suboptimal bowel preparation in individuals aged ≤50 years undergoing colonoscopy.

BACKGROUND

The incidence of early-onset colorectal cancer (EOCRC) has increased in recent decades, leading to more frequent colonoscopy use in younger adults. Adequate bowel preparation is essential for colonoscopy quality, but predictors of suboptimal bowel preparation and individualized risk assessment tools remain insufficiently defined in this population.

METHODS

We conducted a retrospective cohort study of 2,283 consecutive individuals aged ≤50 years who underwent complete colonoscopy between June 2017 and December 2025. Bowel preparation quality was assessed using the Boston Bowel Preparation Scale (BBPS). Suboptimal bowel preparation was defined as a total BBPS score <6 or any segmental score <2. Colonoscopy indication was evaluated separately. Two multivariable logistic regression models were defined according to predictor availability relative to selection of the bowel-preparation regimen: a baseline model containing patient characteristics available before selection of the bowel-preparation regimen and an extended model that additionally included preparation and treatment variables. Performance was assessed using receiver operating characteristic analysis, 1,000-resample bootstrap internal validation, calibration, Brier score, and decision curve analysis (DCA). Exploratory analyses examined age-group-by-predictor interactions in a same-centre older comparison cohort, calendar-period adjustment, and internal temporal validation.

RESULTS

Suboptimal bowel preparation occurred in 426 participants (18.7%). Colonoscopy indication was not associated with suboptimal preparation after adjustment (global likelihood-ratio P = 0.358). In the extended model, diabetes mellitus (OR: 10.701, 95% CI: 7.312-15.660) and constipation (OR: 6.250, 95% CI: 4.619-8.458) were strongly associated with suboptimal preparation. Split-dose preparation (OR: 0.459, 95% CI: 0.362-0.583) and linaclotide use (OR: 0.375, 95% CI: 0.196-0.718) were associated with lower odds. Apparent AUCs were 0.722 for the baseline model and 0.761 for the extended model; corresponding optimism-corrected AUCs were 0.714 and 0.752. The extended model improved discrimination modestly (ΔAUC = 0.039; likelihood-ratio P < 0.001). In internal temporal validation (development, 2017-2022; validation, 2023-2025), the extended model had an AUC of 0.769 but underestimated later risk (calibration intercept 0.389; slope 0.830).

CONCLUSION

Patient characteristics available before regimen selection provided moderate discrimination for suboptimal bowel preparation. Adding preparation and treatment variables modestly improved performance and estimated residual risk after the selected regimen. Because these variables were not randomly assigned, their associations should not be interpreted as causal treatment effects. The model was developed for individuals aged ≤50 years who completed standard preparation without rescue cleansing and underwent complete colonoscopy. Its absolute risk estimates reflect the pooled 2017-2025 case mix; recalibration or model updating and external validation are required before clinical use.