68/100

Moderate evidence

Cite
More
EmailLinkedInX

Model flags hemodialysis patients at risk of demoralization, a study of 592 finds

Study title: Demoralization syndrome in hemodialysis patients: Development and validation of an interpretable machine learning prediction model

Evidence type
Controlled or validated comparison
Sample
n = 592
Sources
1 sourcePeer-reviewed
Published
2 min read

Study title: Demoralization syndrome in hemodialysis patients: Development and validation of an interpretable machine learning prediction model

Researchers built a machine learning model to predict which maintenance hemodialysis patients will develop demoralization syndrome, a state of hopelessness and loss of meaning, within five months. The prospective multicenter cohort enrolled 592 patients, and in external validation on 245 patients the 12-variable model reached an AUC of 0.922.

Read the study abstract

Demoralization syndrome (DS)-an existential distress state marked by hopelessness, helplessness, and loss of meaning-is a prevalent yet underrecognized affective comorbidity among maintenance hemodialysis (MHD) patients. Unlike depression, it stems from impaired coping capacity and carries distinct suicidal ideation risk, yet no practical prediction tool exists for early identification in routine care. To develop and externally validate a parsimonious, interpretable machine learning model for predicting early DS…

The gain

In a multicenter cohort study of 592 hemodialysis patients, a 12-variable machine learning model predicted demoralization syndrome at 5 months with an AUC of 0.922 in external validation of 245 patients, matching a 28-variable model.

In brief

Demoralization differs from depression, stems from impaired coping and carries its own suicidal ideation risk, yet no practical tool exists for early identification in routine care. The authors say the model could support earlier spotting of distress in nephrology practice and prepare the ground for targeted psychosocial intervention.

At the chosen threshold the model caught 65.2% of cases (recall 0.652) with precision of 0.714, so roughly one in three flagged patients and a similar share of true cases were mishandled. The model has been tested on patient data but not shown to improve care.

Main points

  1. The cohort enrolled 592 maintenance hemodialysis patients without baseline demoralization, and new symptoms were assessed at a 5-month follow-up with the Demoralization Scale-II (cutoff 20 or higher).
  2. Twelve predictors were selected from 87 clinical, laboratory and psychological indicators using a funnel-like feature selection workflow.
  3. In external validation (N = 245) the Stacking-Ridge model had accuracy 0.886, recall 0.652, precision 0.714 and F1-score 0.682.
  4. SHAP analysis identified urinary microalbumin, PHQ-9 score and HALP score as the top contributors to predictions.
  5. Decision curve analysis indicated favorable clinical net benefit across a broad range of thresholds.

The full story

The team used a stacking ensemble with Bayesian-optimized base learners and interpreted it with SHapley Additive exPlanations. Sensitivity and ablation analyses tested how stable the predictors were and what each variable added.

The 12-variable version performed comparably to a 28-variable model, which the authors say lowers the burden of collecting clinical data. The top predictors combine a kidney marker (urinary microalbumin), a depression questionnaire score (PHQ-9) and the HALP score.

The authors present the tool as a way to identify affective distress early in nephrology practice and as groundwork for targeted psychosocial intervention, though the report describes prediction accuracy rather than patient outcomes.

Sources

1 on file
  1. Journal of Affective Disordersdoi:10.1016/j.jad.2026.122759
    Peer-reviewed2026-11-19

Talk it over

Three questions to start on, two more if you have time
The model caught 65.2% of demoralization cases, with precision of 0.714. How would you weigh the patients it missed against the patients it flagged who did not go on to develop it?

Start with a question

Model flags hemodialysis patients at risk of demoralization, a study of 592 findshttps://truvace.com/item/demoralization-syndrome-in-hemodialysis-patients-development-and-validation-of-a

  1. The model caught 65.2% of demoralization cases, with precision of 0.714. How would you weigh the patients it missed against the patients it flagged who did not go on to develop it?

    Answer in the thread
  2. In the multicenter cohort of 592 hemodialysis patients, 245 were used for external validation. Who might be missing from that group, and how could that shift the model's AUC of 0.922?

    Answer in the thread
  3. If you were a nephrology team member and the model flagged a patient as at risk of demoralization, what would you want to happen next, and who should decide it?

    Answer in the thread
If you have more time
  • Urinary microalbumin, PHQ-9 score and HALP score were the top predictors. What would you ask researchers about why these three stood out?

    Answer in the thread
  • What kind of trial would show whether flagging patients early leads to better psychosocial care, rather than only better prediction?

    Answer in the thread

What readers say