Objective To integrate multi-dimensional and multimodal data to develop a tool for predicting the risk of chronic kidney disease (CKD). Methods Data from the UK Biobank were utilized, involving 6561 participants recruited between 2006 and 2010, with a follow-up window from April 17, 2007 to November 30, 2022. In the development cohort (n=5248), a multimodal random survival forest (RSF) model was constructed, integrating conventional risk factors [variables from the CKD Prognosis Consortium (CKD-PC) equation], social determinants of health, Life’s Essential 8 data, and retinal optical coherence tomography imaging features. Comparative models included a base model (based solely on estimated glomerular filtration rate and urine albumin-to-creatinine ratio), the CKD-PC equation, a unimodal RSF model (conventional risk factors + social determinants of health + Life’s Essential 8 data), and an extended model (the multimodal RSF model plus a polygenic risk score). The performance of these models was compared in the validation cohort (n=1313). Results After a median follow-up of 12.7 years, 3.57% (234/6561) of the participants developed CKD. In the validation cohort, the 5-year concordance index (C-index) of the multimodal RSF model was 0.73 [95% confidence interval (CI) (0.68, 0.77)], which was significantly higher than that of the base model [C-index=0.67, 95%CI (0.65, 0.71)], the CKD-PC equation [C-index=0.64, 95%CI (0.58, 0.70)], and the unimodal RSF model [C-index=0.69, 95%CI (0.64, 0.73)], and the extended model with the inclusion of the polygenic risk score did not significantly improve predictive performance [C-index=0.72, 95%CI (0.67, 0.76)]. The results of the time-dependent area under the receiver operating characteristic curve analysis were consistent with these findings. Based on the predicted CKD risk derived from the multimodal RSF model for risk stratification, the actual proportions of individuals who developed CKD in the high-, medium-, and low-risk strata were 19.7%, 4.1%, and 1.5%, respectively. In addition to established risk factors, retinal imaging information, healthcare accessibility, and financial status were among the top-ranked predictors for CKD risk. Conclusion Integrating multi-dimensional and multimodal data—including conventional risk factors, social factors, lifestyle factors, and retinal imaging—can improve the performance of CKD risk prediction and stratification.
