Development of a Predictive Tool for Maintenance Dialysis Initiation: A Retrospective Cohort Analysis of 589,284 Participants From the Shizuoka Kokuho Database Study

Rationale & Objective

Maintenance dialysis initiation is rare in early-stage chronic kidney disease (CKD), making accurate risk stratification difficult. We sought to build a simple prediction score based on routine health checkup data.

Study Design

Retrospective cohort study.

Setting & Participants

Adults (≥40 years) attending government-mandated health checkups in Shizuoka Prefecture, Japan, 2012-2020 (N = 589,284).

Predictors

Baseline demographics, vital signs, laboratory indices, medication use, and lifestyle factors routinely recorded at health checks.

Outcomes

Time from first health checkup to initiation of maintenance dialysis, ascertained from procedure codes; death treated as a competing risk.

Analytical Approach

Two-thirds of participants were randomly assigned to a training cohort (n = 392,856; events = 335) and one-third to a test cohort (n = 196,428; events = 179). Cause-specific Cox models generated hazard ratios that were converted to integer scores (maximum 31). Discrimination was evaluated with Harrell’s c-index and calibration with performed with cumulative incidence curves.

Results

During a median follow-up of 5.9 years, 514 participants (0.09%) initiated dialysis. Independent predictors included male sex, body mass index <18.5 kg/m², higher systolic blood pressure, hemoglobin A1c ≥8 %, lower estimated glomerular filtration rate, proteinuria, aspartate aminotransferase ≥50 IU/L, use of antihypertensive or antidiabetic drugs, and habitual smoking. The score showed excellent discrimination in both training (c-index, 0.916; 95% confidence interval [CI], 0.898-0.934) and test (c-index, 0.916; 95% CI, 0.889-0.943) cohorts. High-risk individuals (score ≥16; 0.3% of the cohort) had a 5-year dialysis incidence of 4.6%, yet 95% remained dialysis-free, underscoring the challenge of predicting this rare outcome.

Limitations

CKD onset predated cohort entry, prescription data were not analyzed for causal effects, and external validation is pending.

Conclusions

Our model, leveraging routine health checkup data, accurately identifies persons at elevated risk for future dialysis despite low event rates.

Plain Language Summary

Many early kidney problems go unnoticed until dialysis is suddenly required. We wondered whether the routine health-check data already collected each year in Japan could warn clinicians sooner. We followed a very large group of adults who had regular checkups, recording simple measures such as weight, blood pressure, blood sugar, urinary protein, kidney filtering rate, smoking status, and common prescriptions. Using these items, we built an easy point score that sorts people into low, moderate, or high risk of needing dialysis later on. Most people still avoided dialysis, showing how rare the outcome is, yet the score clearly highlighted those at greatest danger. Embedding this tool in electronic records could prompt earlier referrals and help protect kidney health and wellbeing.