Language models flag spiritual pain in palliative nursing notes with many false positives
Source check: Source checked. Not independent scientific confirmation.
Finding: Gemini 2.5 Flash and Microsoft Copilot detected all 56 expert-labeled spiritual-pain notes in a selected 500-note palliative-care sample. Positive predictive values were only 34% and 25%, respectively. The retrospective study identifies substantial over-detection rather than validating an autonomous diagnostic or clinical screening workflow.
Technology: Gemini 2.5 Flash; Microsoft Copilot language-model service
Family Gemini: profile · models page
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Pretrained language models applied prompted spiritual-pain definitions to free-text nursing notes, producing classifications compared with independent expert annotations; error analysis characterized their misinterpretations.
- Test and result
- From 6,858 nursing records in the last three weeks of life of 109 patients, keyword filtering retained 1,808 records and 500 were sampled. Two researchers annotated spiritual pain with consensus resolution. Japanese few-shot prompts supplied definitions/examples to Gemini 2.5 Flash and Copilot. Both detected all 56 reference-positive notes; specificity was 0.75/0.61 and positive predictive value 0.34/0.25. Gemini false positives frequently confused declining physical function and delirium with spiritual pain.
- Limitation / still unknown
- Selected single-hospital Japanese records, multiple notes per patient and keyword enrichment limit generalization and apparent sensitivity. Reference is retrospective note interpretation, not direct patient assessment. Copilot underlying model version was not specified. No unfiltered-record screening, prospective clinical utility, improved care or autonomous diagnosis was established; human verification remains necessary.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC13719773/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 21 Nov 2026. Review scope: Read actual primary record sampling/filtering, consensus annotation, named language models and prompting Methods, result/error tables and limitations. Supplementary prompt appendices were not independently inspected. Exact primary identity/date and fresh primary PubMed correction/retraction relationships checked..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.