Dynamic hemoglobin trajectories and mortality in hemodialysis patients: a joint modeling study with iron and...

Mortality remains high among maintenance hemodialysis (MHD) patients, and laboratory biomarkers such as hemoglobin, ferritin, and phosphorus exhibit substantial temporal variability. Static measurements may fail to adequately reflect dynamic prognostic changes.

This study evaluated the longitudinal trajectories of hemoglobin, log-transformed ferritin, and phosphorus, and assessed their dynamic associations with all-cause mortality in MHD patients using joint modeling.

A total of 254 MHD patients with 814 repeated measurements were analyzed. Linear mixed-effects models were linked with Cox proportional hazards models. Three joint model structures were compared using information criteria, and dynamic predictive performance was evaluated using time-dependent AUC and Brier scores.

Over a median 18-month follow-up, 62 deaths (24.4%) occurred. Higher hemoglobin levels were independently associated with lower mortality (per 10 g/L increase: HR = 0.97; 95% CI: 0.95-1.00). Log-ferritin showed a negative but nonsignificant association, while phosphorus did not improve model fit. The hemoglobin + log-ferritin model achieved the best discrimination (AUC ≈ 0.70-0.75). Sensitivity analyses incorporating phosphorus or replacing ferritin with transferrin saturation produced consistent results, confirming robustness.

Dynamic hemoglobin changes show a modest but consistent inverse association with mortality risk in MHD patients and provide stable predictive value. Joint modeling may enhance risk stratification and support individualized anemia management in hemodialysis.