Skip to content
DraconiticResearch
Search Sources API
197,761 studies31 sourcesUpdated 58 min ago
PreprintCC BYmedRxiv

Revision Behavior and Explainability in Adaptive LLM Swarms for ICU Mortality Risk Prediction: A Two-Dataset Evaluation

  • C. Anderson1
  1. 1University of Ottawa
Read full textPDFDOICite

Abstract

PurposeTo evaluate how adaptive LLM swarm revision changes ICU mortality-risk outputs and to characterize evidence use, explanation indicators, auditability, and computational burden in final adaptive outputs relative to an independently executed fixed-voting (FV) architecture.

MethodsWe retrospectively analyzed 1,607 eICU encounters (1,500 stays) and 1,607 ICU-2012 encounters. Initial (ASI) and final (ASF) adaptive outputs were compared within runs for revision engagement and risk-score drift; final ASF and separately generated FV outputs were compared for evidence use, explanation indicators, auditability, and computation. Paired differences and 95% confidence intervals used 10,000 hospital-stay-clustered bootstrap replicates.

ResultsAt least one specialist revision trace occurred in 84.32% of eICU and 56.44% of ICU-2012 encounters. Mean ASF-minus-ASI risk-score changes were +0.0810 and +0.0539; 757 of 761 0.50-threshold crossings moved toward mortality, without clear AUROC or AUPRC improvement. In the independent benchmark, ASF explanations contained 1.26 and 0.55 more supporting-evidence items than FV, but counterevidence acknowledgement was 21.59 and 7.47 percentage points lower and unsupported-claim flags were 1.43 and 0.68 points higher. All final records met the reconstruction-completeness criterion, although ASF generated more warnings and required 1.97 and 1.52 times the FV runtime.

ConclusionAdaptive revision materially changed swarm operating behaviour. Independently, final ASF outputs showed greater supporting-evidence use but less balanced evidence engagement, more process warnings, and greater computational burden than FV. These automated artifact-level findings do not establish superior explanation quality or isolate revision as their cause.

Related

  1. On-Device ECG and Vital-Sign Prediction of 24-Hour Mortality and Unplanned ICU Admission in Emergency Department Patients: A Retrospective Evaluation of MEDI-AI
    Diagnostics (Basel, Switzerland)Peer-reviewedCC BYFull textGeneral medicineEurope PMC
  2. Competing Risk Prediction of COVID-19 Hospital Mortality with Model Diagnostics and Exploratory Evaluation in Mexico
    Zahra Sabzevari, Hamid Reza Marateb, Reza Rasti +3
    Diagnostics (Basel, Switzerland)Peer-reviewedCC BYFull textGeneral medicineEurope PMC
  3. Explainability-informed patient-disjoint evaluation of deep learning for plus disease in retinopathy of prematurity with an ophthalmologist reader study
    Rouhollah Kian Ara, Jan Kubíček, Sumit Kaushik +2
    Frontiers in medicinePeer-reviewedCC BYFull textGeneral medicineEurope PMC
  4. Comparison of Mortality and Revision Rates between Cemented and Cementless Fixation in Bipolar Hemiarthroplasty: A Nationwide Cohort Study
    Jung-Wee Park, Danbee Kang, Hyunsoo Kim +3
    Clinics in orthopedic surgeryPeer-reviewedCC BY-NCFull textSurgery & anaesthesiaEurope PMC
  5. A method for brain tumor detection and localization by means of deep learning and reliable prediction explainability
    Marcello Di Giammarco, Antonella Santone, Mario Cesarelli +2
    PLOS OnePeer-reviewedCC BYFull textNeuroscienceMedical imaging & radiologyPLOSEurope PMCPMC

Cite

BibTeXRIS
BibTeX
@misc{anderson2026revision,
  title = {{Revision Behavior and Explainability in Adaptive LLM Swarms for ICU Mortality Risk Prediction: A Two-Dataset Evaluation}},
  author = {C. Anderson},
  howpublished = {medRxiv},
  year = {2026},
  month = sep,
  doi = {10.64898/2026.09.01.26361960},
  url = {https://www.medrxiv.org/content/10.64898/2026.09.01.26361960v1},
  note = {CC BY},
}
RIS
TY  - UNPB
TI  - Revision Behavior and Explainability in Adaptive LLM Swarms for ICU Mortality Risk Prediction: A Two-Dataset Evaluation
AU  - C. Anderson
PY  - 2026
DA  - 2026/09/01
PB  - medRxiv
DO  - 10.64898/2026.09.01.26361960
UR  - https://www.medrxiv.org/content/10.64898/2026.09.01.26361960v1
AB  - PurposeTo evaluate how adaptive LLM swarm revision changes ICU mortality-risk outputs and to characterize evidence use, explanation indicators, auditability, and computational burden in final adaptive outputs relative to an independently executed fixed-voting (FV) architecture.  MethodsWe retrospectively analyzed 1,607 eICU encounters (1,500 stays) and 1,607 ICU-2012 encounters. Initial (ASI) and final (ASF) adaptive outputs were compared within runs for revision engagement and risk-score drift; final ASF and separately generated FV outputs were compared for evidence use, explanation indicators, auditability, and computation. Paired differences and 95% confidence intervals used 10,000 hospital-stay-clustered bootstrap replicates.  ResultsAt least one specialist revision trace occurred in 84.32% of eICU and 56.44% of ICU-2012 encounters. Mean ASF-minus-ASI risk-score changes were +0.0810 and +0.0539; 757 of 761 0.50-threshold crossings moved toward mortality, without clear AUROC or AUPRC improvement. In the independent benchmark, ASF explanations contained 1.26 and 0.55 more supporting-evidence items than FV, but counterevidence acknowledgement was 21.59 and 7.47 percentage points lower and unsupported-claim flags were 1.43 and 0.68 points higher. All final records met the reconstruction-completeness criterion, although ASF generated more warnings and required 1.97 and 1.52 times the FV runtime.  ConclusionAdaptive revision materially changed swarm operating behaviour. Independently, final ASF outputs showed greater supporting-evidence use but less balanced evidence engagement, more process warnings, and greater computational burden than FV. These automated artifact-level findings do not establish superior explanation quality or isolate revision as their cause.
ER  -