# On-premise medical AI agents for reliable clinical decision-making

- **DOI:** [10.1038/s41591-026-04609-x](https://doi.org/10.1038/s41591-026-04609-x)
- **Rating:** Sound (3.7 out of 5)
- **Verdict:** Cross-run diagnostic consistency improves error ranking in simulated clinical agents, but retrospective operating points do not establish deployment-ready autonomous triage.
- **Framework version:** v5.13
- **Pipeline version:** v0.4
- **Evaluation date:** 2026-11-22
- **Evaluation page:** https://www.nabu.science/paper/10.1038%2Fs41591-026-04609-x

## Key claims

1. Diagnostic behavioral consistency provided the strongest discrimination of correctness (area under the curve (AUC) = 0.860) and remained informative under stress testing (AUC = 0.875).
   - Support: Supported
   - Type: Primary finding
   - Basis: retrospective benchmark discrimination supports the ranking, with limited error counts and incompletely specified case-level correctness
   - Location: Abstract; Results Fig. 3a, Fig. 5c
2. At a consistency threshold of 0.90, 49.4% of cases were retained at 98.9% diagnostic accuracy.
   - Support: Supported
   - Type: Secondary
   - Basis: same-cohort retrospective counts support this operating point, not performance in a new clinical deployment
   - Location: Abstract; Results Fig. 6a,c
3. Across two Medical Information Mart for Intensive Care IV (MIMIC-IV)-derived benchmarks, the agent achieved 90.04% accuracy on a seven-disease task and 83.8% accuracy on a four-disease task, approaching a cloud baseline on the primary benchmark.
   - Support: Supported
   - Type: Secondary
   - Basis: simulated benchmark comparisons support the reported accuracies, with temperature selection performed on the evaluation set
   - Location: Abstract; Results Fig. 2a,b
4. These findings support a practical framework for institutionally governed clinical agents in which decision-time reliability signals identify a lower-risk subset for autonomous handling and defer the remainder for review.
   - Support: Tentative
   - Type: Secondary
   - Basis: retrospective simulations with configuration-dependent thresholds do not establish safe autonomous handling in clinical workflows
   - Location: Abstract, final sentence; Discussion
5. On VivaBench, ConsistencyDx again provided the strongest discrimination of correctness among the evaluated metrics (AUC = 0.719).
   - Support: Supported
   - Type: Secondary
   - Basis: curated case-report evaluation with approximate adjudication supports the metric ranking within this external benchmark
   - Location: Results, External benchmark evaluation on VivaBench

## Strengths

- **Consistency evaluated across agentic clinical settings** The evaluation compares eight diagnosis- and reasoning-stream signals in multi-turn clinical agents. Diagnostic consistency leads the evaluated metrics on MIRA-v2 and VivaBench and remains informative under reduced patient-history grounding.
  - Dimension: contribution
  - Location: Results, Figs. 3 and 5; Extended Data Fig. 4b
- **Automated scoring checked against physician judgments** Automated adjudication agrees with physician consensus on 167 of 181 reviewed cases, or 92.3%. A separate CDM comparison reports 98.6% agreement with deterministic fuzzy matching.
  - Dimension: methodological rigour
  - Location: Fig. 2e,f; Methods, Physician evaluation; Extended Data Fig. 6
- **Deployment limitations traced to concrete consequences** The Discussion connects encoder-dependent thresholds, approximately five-fold token use and age-related performance differences to calibration, infrastructure and fairness requirements. Persistent high-consistency errors are distinguished from benchmark-label disagreements.
  - Dimension: positioning
  - Location: Discussion; Results, Implementation sensitivity analyses; Extended Data Figs. 1 and 5

## Limitations

- **Practical autonomy framing exceeds retrospective validation** The abstract's practical-framework interpretation rests on an operating point assessed on the same MIRA-v2 cohort: 98.9% accuracy at 49.4% coverage. VivaBench reaches 89.9% at 32.0% coverage, and prospective clinical routing safety is not established.
  - Dimension: reporting
  - Location: Abstract, final sentence; Fig. 6; Extended Data Fig. 4c; Discussion, limitations
- **Case-level correctness construction remains insufficiently explicit** GLM-4.5-Air has an 88.4% five-run mean in the model comparison, while the confidence and triage baseline contains 499 correct cases out of 551, or 90.6%. These need not conflict, but the mapping from repeated runs to binary case-level correctness is not fully explained.
  - Dimension: methodological rigour
  - Location: Results, On-premise agent achieves competitive diagnostic performance; Fig. 3 legend; Fig. 5 legend; Methods, Multi-perspective confidence framework
- **External access boundary is not reconciled** The ethics statement says clinical text was inaccessible to external providers, yet the study reports a cloud GPT-5.2 baseline and an API-based Gemini evaluator. The text does not explain how these uses fit the stated governance boundary.
  - Dimension: positioning
  - Location: Methods, Ethics statement and Performance evaluation; Results, model comparison; Extended Data Table 2

## Reliability checks

Caveats (4 of 4 checks).

- **Results check:** Noted
  - "GLM-4.5-Air (88.4%) and GPT-OSS (85.3%)" (Results, On-premise agent paragraph (GLM-4.5-Air accuracy)) vs "Overall accuracy decreased from 90.6% (499/551) in the Baseline condition" (Fig. 5 legend). Compare the reported five-run mean of 88.4% with the case-level baseline of 499/551 = 90.6%, whose conversion from repeated-run outcomes is not specified. The effect on the result is unclear.
- **Reporting and transparency:** Noted. The statement that clinical text was inaccessible to external providers is not reconciled with the reported cloud baseline and API-based adjudication. Ethics approval, competing interests, data access and archived code are declared.
  - Conflict of interest: declared · Ethics approval: declared · Preregistration: not stated · Data availability: declared · Code availability: declared
- **Reference integrity:** Clean
  - 15 references in manuscript
  - 9 of 10 checkable references found in an index
  - 5 are books, websites or datasets — counted, but not index-checkable
  - References not verified automatically are resolved manually within 24 hours.

## Post-publication record

- **Status:** Clean
- **Summary:** No retraction or expression of concern found in Crossref (includes Retraction Watch).
- **Last checked:** 2026-11-22
