What makes a ‘good’ decision with artificial intelligence? A grounded theory study in paediatric care
Bibliographic record
Abstract
OBJECTIVE: To develop a framework for good clinical decision-making using machine learning (ML) models for interventional, patient-level decisions. DESIGN: Grounded theory qualitative interview study. SETTING: Primarily single-site at a major urban academic paediatric hospital, with external sampling. PARTICIPANTS: Sixteen participants representing physicians (n=10), nursing (n=3), respiratory therapists (n=2) and an ML specialist (n=1) with experience working in acute care environments were identified through purposive sampling. Individuals were recruited to represent a spectrum of ML knowledge (three expert, four knowledgeable and nine non-expert) and years of experience (median=12.9 years postgraduation). Recruitment proceeded through snowball sampling, with individuals approached to represent a diversity of fields, levels of experience and attitudes towards artificial intelligence (AI)/ML. A member check step and consultation with patients was undertaken to vet the framework, which resulted in some minor revisions to the wording and framing. INTERVENTIONS: A semi-structured virtual interview simulating an intensive care unit handover for a hypothetical patient case using a simulated ML model and seven visualisations using known methods addressing interpretability of models in healthcare. Participants were asked to make an initial care plan for the patient, then were presented with a model prediction followed by the seven visualisations to explore their judgement and potential influence and understanding of the visualisations. Two visualisations contained contradicting information to probe participants' resolution process for the contrasting information. The ethical justifiability and clinical reasoning process were explored. MAIN OUTCOME: A comprehensive framework was developed that is grounded in established medicolegal and ethical standards and accounts for the incorporation of inference from ML models. RESULTS: We found that for making good decisions, participants reflected across six main categories: evidence, facts and medical knowledge relevant to the patient's condition; how that knowledge may be applied to this particular patient; patient-level, family-specific and local factors; facts about the model, its development and testing; the patient-level knowledge sufficiently represented by the model; the model's incorporation of relevant contextual factors. This judgement was centred on and anchored most heavily on the overall balance of benefits and risks to the patient, framed by the goals of care. We found evidence of automation bias, with many participants assuming that if the model's explanation conflicted with their prior knowledge that their judgement was incorrect; others concluded the exact opposite, drawing from their medical knowledge base to reject the incorrect information provided in the explanation. Regarding knowledge about the model, we found that participants most consistently wanted to know about the model's historical performance in the cohort of patients in their local unit where the hypothetical patient was situated. CONCLUSION: Good decisions using AI tools require reflection across multiple domains. We provide an actionable framework and question guide to support clinical decision-making with AI.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Direct model labels (unvalidated)
Per-model category and study-design labels from the labeling rounds. They are machine output, unvalidated, and the disagreement between models ships as data. No study design here is MEDLINE-validated yet.
| Model arm | Categories | Study design | Confidence |
|---|---|---|---|
| gemma | Science and technology studies Domain: not available · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Qualitative | low |
| gpt | no category Domain: not available · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Qualitative | low |
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.005 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.003 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedLabeled directly by 2 models reading the full record.
The models disagree on parts of this classification; every voice is preserved in the section at the end of the page.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".