AI-assisted clinical summary and treatment planning for cancer care: A comparative study of human vs. AI-based approaches.
Bibliographic record
Abstract
1523 Background: Understanding a patient's clinical narrative, timeline, and history is critical for accurate treatment decision-making. However, reviewing and summarizing complex records is time-consuming and error-prone. Recent advancements in artificial intelligence (AI), specifically large language models (LLM), offer paths to improve quality and efficiency. Methods: A study was conducted on 50 breast cancer cases from an academic medical institution, utilizing all medical records—clinic, pathology, and radiology reports—up until the point of the initial treatment decision. All cases were processed using three different approaches: AI-assisted; full-AI; and human-only. In the AI-assisted method, two oncology physician assistants (PAs) revised AI-generated summaries to create clinical summaries. The full-AI method had AI independently produce clinical summaries, while the human-only method had the PAs compile summaries without AI. Eight board-certified international oncology specialists blindly evaluated summaries for faithfulness, completeness, and succinctness using a 3-point scale, ranked their preferences, and tried to predict which summaries were full-AI. Rankings were assessed using a Friedman test followed by a Wilcoxon signed-rank test, and full-AI prediction was assessed using a two-sided one-sample binomial test. After summarization, a distinct AI system with access to clinical guidelines provided treatment plans. These plans were then evaluated by a board-certified oncologist with access to the original treatment decision. Results: The study found specialists favored AI-assisted, followed by full-AI, and then human-only summaries, with average ranks of 1.73, 1.93, 2.34 respectively (lower is better, p<0.001). The difference between full-AI and AI-assisted was not significant (p=0.11). Evaluation scores (mean±95%CI, higher is better) showed AI-assisted, full-AI, and human-only scored 2.35±0.13, 2.14±0.14, 2.17±0.14 for faithfulness; 2.28±0.12, 2.01±0.12, 1.93±0.14 for completeness; and 2.33±0.12, 2.21±0.12, 1.99±0.13 for succinctness. The average summarization time was 19.71, 1.17, 26.03 minutes. Full-AI identification accuracy was 0.28 (not different from chance 0.33, p=0.46). With AI-assisted summaries, the treatment plans were accurate in 45 cases (90%) and partially accurate in 5 cases (10%). In the 5 partially accurate cases, the system was accurate with the provided input data, but there were inaccuracies with the input data, including incorrect formats or missing data. Conclusions: Incorporating LLMs into the creation of medical summaries has shown improvements in both quality and efficiency, achieving up to 22.2x speed up with full-AI, indicating that AI-assisted summarization tools can potentially enhance care quality. AI-assisted summaries yield accurate treatment plans when the input data is accurate.
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
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.017 | 0.079 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
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, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
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".