JMIR Publications

ABSTRACT

Background:

Kidney Disease: Improving Global Outcomes (KDIGO) classifies chronic kidney disease using categories based on both glomerular filtration rate and albuminuria. Structured measures remain essential but may not capture contextual clinical concerns documented in narrative records.

Objective:

This study examined whether text-derived clinical-status language features in clinician notes were associated with deterioration in renal function in independent Greek- and English-language clinical cohorts.

Methods:

We retrospectively analyzed a Greek general-nephrology cohort (314 patients; 10,306 eGFR measurements) and an independent English-language MIMIC-IV replication cohort. The Greek primary analysis used a linear mixed-effects model over follow-up of up to 1,797 days. The MIMIC-IV primary analysis used a 90-day, patient-level fixed-horizon competing-outcome model (N=87,533). Separate analyses of Greek and MIMIC-IV data evaluated discrimination, calibration, and sensitivity to alternative feature specifications.

Results:

In the Greek cohort, a 1-SD higher negativity was associated with lower eGFR (β = -3.273 mL/min/1.73 m²; 95% CI, −5.049 to −1.498; P < .001) and a more negative time trajectory (time-by-score β = -1.324; 95% CI, −2.553 to −0.096; P = .035). In MIMIC-IV, a 1-SD higher negative-versus-reassuring polarity contrast was associated with renal deterioration (RRR=1.090; 95% CI 1.070–1.110; P<.001). Within eGFR G categories, polarity-score AUCs were 0.506–0.552 using discharge eGFR and 0.518–0.534 using the worst observed eGFR. Discrimination for renal deterioration remained modest (one-vs-rest AUC=0.6134; 95% CI 0.6092–0.6175), and adding both NLP features to the clinical model increased AUC by only 0.0018 (95% CI 0.0010–0.0025).

Conclusions:

Independently constructed clinical-language measures were associated with renal function deterioration across two languages and healthcare settings. The findings support cross-language replication of an association signal, not formal external validation, measurement equivalence, or clinically useful stand-alone prediction.