Predicting 12- and 24-month documented CKD progression from routine laboratory data: a single-center temporal evaluation

Routine laboratory data may support CKD follow-up. We evaluated 12- and 24-month documented progression prediction and exploratory review outputs.

This retrospective single-center study used 2012-2025 hospital records. CKD required objective persistence for ≥90 days. Development used 2012-2021 indices; temporal evaluation used unseen patients indexed in 2022-2023. Among horizon-evaluable indices, the endpoint was documented sustained ≥40% estimated glomerular filtration rate (eGFR) decline or kidney failure. Development selected 12-month and conditional second-year models combined for non-decreasing cumulative risk. Laboratory and documentation outputs were assessed separately.

Temporal evaluation of horizon-evaluable patients included 143 patients (10 event patients; 21/364 positive indices) at 12 months and 87 patients (22 event patients; 50/247 positive indices) at 24 months. Area under the precision-recall curve (AUPRC) was 0.510 (95% confidence interval [CI] 0.217-0.789) and 0.677 (0.459-0.853), respectively. Incremental 12-month AUPRC over age, sex and current eGFR was uncertain (difference -0.009, 95% CI -0.171 to 0.123). At 24 months, area under the receiver operating characteristic curve (AUROC) was higher than the core comparator (difference 0.057, 95% CI 0.007-0.121); clinical utility was not evaluated. Calibration slopes were 0.996 (0.689-1.640) and 1.028 (0.676-1.661); wide CIs precluded reliable individual-risk calibration. Documentation ranking did not improve on prior-mention strategies.

This study is proof of concept. Small event counts, follow-up selection and uncertain incremental value limit interpretation. Multisite validation is required before implementation.