68/100

Moderate evidence

Cite
More
EmailLinkedInX

AI help lifted discharge instruction scores for 101 kidney disease patients, study finds

Study title: Large Language Model-Assisted Discharge Instructions for Diabetic Kidney Disease: Comparative Evaluation Study

Evidence type
Controlled or validated comparison
Sample
n = 101
Sources
1 sourcePeer-reviewed
Published
2 min read

Study title: Large Language Model-Assisted Discharge Instructions for Diabetic Kidney Disease: Comparative Evaluation Study

A study of 101 hospitalized patients with diabetic kidney disease compared discharge instructions drafted by junior residents alone with instructions written by a second group of residents of the same level using DeepSeek-V3.2. That produced 202 documents, which four clinical experts scored on a 5-point scale.

Read the study abstract

Background: Discharge instructions are an important component of the transition from hospital to home for patients with diabetic kidney disease (DKD), whose care often involves multiple comorbidities, medications, and stage-specific management needs. However, producing complete, accurate, clear, and personalized instructions can be challenging for junior physicians. Objective: This study aimed to evaluate whether large language model (LLM) assistance improves the quality of DKD discharge instructions produced by…

The gain

In a comparative study of 101 hospitalized diabetic kidney disease patients, four clinical experts rated discharge instructions written by junior residents with DeepSeek-V3.2 higher than resident-only versions on all five dimensions, with 100 of 101 cases improving by at least one point overall.

In brief

AI-assisted instructions scored significantly higher on completeness, accuracy, safety, clarity, personalization and overall quality, with large effect sizes (r=0.741-0.927). Overall quality improved by at least one point in 100 of 101 cases.

Resident-only quality fell as chronic kidney disease stage rose, while the AI-assisted group showed no significant stage correlation for most dimensions. The authors conclude the advantage was greatest in complex cases and see human-AI collaboration as a way to reduce documentation burden.

Main points

  1. The study covered 101 hospitalized patients with diabetic kidney disease and 202 discharge instruction documents in total.
  2. One group of junior residents drafted instructions alone; another group at the same level used DeepSeek-V3.2 for the same 101 patients.
  3. Two nephrologists and two endocrinologists independently scored each document on completeness, accuracy, safety, clarity and personalization.
  4. AI-assisted instructions scored higher on all five dimensions and overall (all Bonferroni-adjusted P<.001), with effect sizes of r=0.741-0.927.
  5. Among resident-only drafts, higher CKD stage correlated with lower accuracy, safety, clarity and overall quality (Spearman ρ=-0.33 to -0.37).
  6. In the AI-assisted group, correlations with CKD stage were mostly not significant (|ρ|<0.15).
  7. Overall quality improved by at least one point in 99% of cases (100/101; 95% CI 94.6%-100.0%).

The full story

Discharge instructions help patients with diabetic kidney disease move from hospital to home, and their care often involves several conditions, medications and stage-specific management. The authors note that writing complete, accurate, clear and personalized instructions can be hard for junior physicians.

Residents first drafted instructions for all 101 patients. A separate group of residents at the same level then wrote instructions for the same patients with help from DeepSeek-V3.2. Four specialists scored the documents, and agreement among them was high in both groups (ICC=0.817 for resident-only, 0.860 for AI-assisted).

Resident-only quality declined with more advanced CKD stage, while AI-assisted quality did not show a significant decline on most dimensions. The authors say LLM assistance improved completeness, clarity and personalization, especially in complex cases.

Sources

1 on file
  1. JMIR Medical Informaticsdoi:10.2196/100792
    Peer-reviewed2026-11-13

Talk it over

Three questions to start on, two more if you have time
Four clinical experts scored 202 discharge instructions, and 100 of 101 cases improved by at least one point with DeepSeek-V3.2. What can expert scores tell us about what happens to patients after they go home?

Start with a question

AI help lifted discharge instruction scores for 101 kidney disease patients, study findshttps://truvace.com/item/large-language-model-assisted-discharge-instructions-for-diabetic-kidney-disease

  1. Four clinical experts scored 202 discharge instructions, and 100 of 101 cases improved by at least one point with DeepSeek-V3.2. What can expert scores tell us about what happens to patients after they go home?

    Answer in the thread
  2. Resident-only quality fell as chronic kidney disease stage rose, while AI-assisted drafts mostly did not. Who benefits most from that pattern, and which patients might still be poorly served?

    Answer in the thread
  3. Two separate groups of junior residents wrote instructions for the same 101 patients, one alone and one with AI. How might that design have shaped the size of the gap?

    Answer in the thread
If you have more time
  • If the same residents wrote instructions with and without DeepSeek-V3.2, what result would convince you the AI made the difference?

    Answer in the thread
  • Which outcomes for the 101 patients, such as readmissions or questions to the clinic, would you want measured before a hospital adopted this routinely?

    Answer in the thread

What readers say