Real-World Validation and Optimization of CKD Progression Prediction Models Using US Mayo Clinic Data
Bibliographic record
Abstract
Background: Accurate chronic-kidney-disease (CKD) risk prediction tools stall in practice because data elements are often missing. Two deep-learning models, RRT Onset and Rapid Progression (RP), originally developed in Taiwan (China Medical University Hospital, CMUH) and granted FDA Breakthrough Device Designation were tested and optimized using a large, racially diverse U.S. cohort from the Mayo Clinic Platform (MCP). Methods: We analyzed 232,613 stage 3–5 CKD patients (25,214 under nephrology care) treated at Mayo Clinic sites (AZ, FL, MN) between 2018 and 2024. Outcomes were (1) RRT initiation and (2) eGFR decline (≥40%, slope ≤-5 mL/min/1.73m2/yr, or eGFR <15) within 2 and 5 years. Model variants assessed were: original, fine-tuned, and de novo (trained exclusively on MCP data). Discrimination (AUC), sensitivity, and specificity were compared with 4-variable Kidney Failure Risk Equation (KFRE). Results: The fine-tuned AI models delivered robust discrimination (RRT-AUC 0.85–0.86; RP-AUC 0.78–0.81). uACR was absent in 46% of records, resulting in an inability to calculate KFRE. Against KFRE (AUC 0.81), fine-tuned RRT predictions improved event reclassification by 12.9–41.4%, while maintaining sensitivity >78% and specificity >75%. Notably, de novo model did not offer significant performance advantages over the fine-tuned model (Table 1). Event rates increased with higher risk scores supporting effective risk stratification (Figure 1). Conclusion: FDA-designated AI models preserved accuracy despite nearly 50% data missingness and outperformed KFRE across U.S. centers. A rapid fine-tuning step, rather than de novo training, enables seamless local deployment, facilitating prospective trials and accelerating clinical adoption.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.009 | 0.015 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".