Risk of onset of chronic kidney disease in type 2 diabetes mellitus (ROCK-DM): Development and validation of a 4-variable prediction model.

AIMS

The aim of this study was to develop and validate a prediction model for incident chronic kidney disease (CKD) in type 2 diabetes mellitus (T2DM), defined as eGFR < 60 ml/min/1.73m2 and/or urine albumin:creatinine ratio (UACR) > 3 mg/mmol in ≥ 2 consecutive readings ≥ 3 months apart.

METHODS

Model derivation was performed in the SingHealth Diabetes Registry, including patients aged ≥ 21 years diagnosed with T2DM without pre-existing CKD. External validation was performed in a single-center prospective observational cohort. Cox Proportional Hazard model was created to evaluate predictors associated with time-to-onset of incident CKD. Increasingly parsimonious models were assessed for discrimination and calibration. Models underwent external validation, benchmarking against existing models, and decision curve analysis.

RESULTS

25,142 (59 %) of 42,552 patients in the derivation cohort developed CKD over a median 4.0 years (IQR 2.1-7.7) follow up. An 18-variable model, 12-variable model, and 4-variable model (including age, duration of T2DM, eGFR, and previous non-persistent albuminuria) was developed. The 4-variable model had a C-statistic of 0.78 and good calibration on plots of observed-versus-predicted risk. The 12-variable and 18-variable models performed similarly. In the external validation cohort of 2249 patients, of whom 1035 (46 %) developed incident CKD, the 4-variable model had a C-statistic of 0.87. All models had better discrimination than existing benchmarks. Decision curve analysis of the 4-variable model showed positive net benefit for any threshold probability above 16 % for 2-year and 28 % for 5-year CKD risk.

CONCLUSION

The 4-variable model for prediction of incident CKD in T2DM demonstrates good performance, predicts both eGFR and albuminuria endpoints, and is simple-to-use. This may guide personalized care, resource allocation and population health.