The landscape of drug sensitivity and resistance in sarcoma.

Sarcomas are rare malignancies with over 100 distinct histological subtypes. Their rarity and heterogeneity pose significant challenges to identifying effective therapies, and approved regimens show varied responses. Novel, personalized approaches to therapy are needed to improve patient outcomes. Patient-derived tumor organoids (PDTOs) model tumor behavior across an array of malignancies. We leverage PDTOs to characterize the landscape of drug resistance and sensitivity in sarcoma, collecting 194 specimens from 126 patients spanning 24 distinct sarcoma subtypes. Our high-throughput organoid screening pipeline tested single agents and combinations, with results available within a week from surgery. Drug sensitivity correlated with clinical features such as tumor subtype, treatment history, and disease trajectory. PDTO screening can facilitate optimal drug selection and mirror patient outcomes in sarcoma. We could identify at least one FDA-approved or NCCN-recommended effective regimen for 59% of the specimens, demonstrating the potential of our pipeline to provide actionable treatment information.

Comparison of thoracic epidural catheter and continuous peripheral infusion for management of traumatic rib fracture pain.

BACKGROUND: Thoracic epidural catheters (TECs) are useful adjuncts to multimodal pain regimens in traumatic rib fractures. However, TEC placement is limited by contraindications, patient risk profile, and provider availability. Continuous peripheral infusion of ketamine and/or lidocaine is an alternative that has a modest risk profile and few contraindications. We hypothesized that patients with multiple traumatic rib fractures receiving TECs would have better pain control, in terms of daily morphine milligram equivalents (MMEs) and mean pain scores (MPSs) when compared with continuous peripheral infusions of ketamine and/or lidocaine. METHODS: We retrospectively analyzed traumatic rib fracture admissions to a level 1 trauma center between January 2018 and December 2020. We evaluated two treatment groups: TEC only and continuous infusion only (drip only). A linear mixed-effects model evaluated the association of MME with treatment group. An interaction term of treatment group by time (days 1-7) was included to allow estimating potential time-dependent treatment effect on MME. A zero-inflated Poisson mixed-effects model evaluated the association of treatment with MPS. Both models adjusted for confounders. RESULTS: A total of 1,647 patients were included. After multivariable analysis, a significant, time-varying dose-response relationship between treatment group and MME was found, indicating an opioid-sparing effect favoring the TEC-only group. The opioid-sparing benefit for TEC-only therapy was most prominent at day 3 (27.4 vs 36.5 MME) and day 4 (27.3 vs 36.2 MME) (p < 0.01). The drip-only group had 1.21 times greater MPS than patients with TEC only (p < 0.001). CONCLUSION: Drip-only analgesia is associated with higher daily MME use and MPS, compared with TEC only. The maximal benefit of TEC therapy appears to be on days 3 and 4. Prospective, randomized comparison between groups is necessary to evaluate the magnitude of the treatment effect. LEVEL OF EVIDENCE: Therapeutic/Care Management; Level III.

Association of Opioid Disposal Practices with Parental Education and a Home Opioid Disposal Kit Following Pediatric Ambulatory Surgery.

BACKGROUND: The majority of opioid analgesics prescribed for pain after ambulatory pediatric surgery remain unused. Most parents do not dispose of these leftover opioids or dispose of them in an unsafe manner. We aimed to evaluate the association of optimal opioid disposal with a multidisciplinary quality improvement (QI) initiative that proactively educated parents about the importance of optimal opioid disposal practices and provided a home opioid disposal kit before discharge after pediatric ambulatory surgery. METHODS: Opioid disposal behaviors were assessed during a brief telephone interview pre- (Phase I) and post-implementation (Phase II) after surgery. For each phase, we aimed to contact the parents of 300 pediatric patients ages 0 to 17 years who were prescribed an opioid after an ambulatory surgery. The QI initiative included enhanced education and a home opioid disposal kit including DisposeRX®, a medication disposal packet that renders medications inert within a polymeric gel when mixed with water. Weighted segmented regression models evaluated the association between the QI initiative and outcomes. We considered the association between the QI initiative and outcome significant if the beta coefficient for the change in intercept between the end of Phase I and the beginning of Phase II was significant. Safe opioid disposal and any opioid disposal were evaluated as secondary outcomes. RESULTS: The analyzed sample contained 161 pediatric patients in Phase I and 190 pediatric patients in Phase II. Phase II (post-QI initiative) cohort compared to Phase I cohort reported higher rates of optimal (58%, n = 111/190 vs 11%, n = 18/161) and safe (66%, n = 125/190 vs 34%, n = 55/161) opioid disposal. Weighted segmented regression analyses demonstrated significant increases in the odds of optimal (odds ratio [OR], 26.5, 95% confidence interval [CI], 4.0-177.0) and safe (OR, 4.4, 95% CI, 1.1-18.4) opioid disposal at the beginning of Phase II compared to the end of Phase I. The trends over time (slopes) within phases were nonsignificant and close to 0. The numbers needed to be exposed to achieve one new disposal event were 2.2 (95% CI, 1.4-3.7]), 3.1 (95% CI, 1.6-7.4), and 4.3 (95% CI, 1.7-13.6) for optimal, safe, and any disposal, respectively. CONCLUSIONS: A multidisciplinary approach to educating parents on the importance of safe disposal of leftover opioids paired with dispensing a convenient opioid disposal kit was associated with increased odds of optimal opioid disposal.

Early Complications of Planned Resection Versus Unplanned Excision of Sarcomas in the Distal Upper Extremity.

Purpose: Unplanned excisions are defined as excisions of malignant tumors performed without preoperative cross-sectional imaging or diagnostic biopsy, frequently resulting in residual disease and re-excision secondary to positive surgical margins. The purpose of this study was to compare the relative morbidity of planned versus unplanned upper-extremity sarcoma excisions. Methods: A single tertiary referral hospital pathology database was queried from January 2015 through 2022 for primary upper-extremity sarcomas (forearm, wrist, hand, and finger). Demographics, tumor features, survival characteristics, and outcomes were retrospectively reviewed. Results: Forty-two upper-extremity sarcoma patients were identified, two-thirds of whom had unplanned excisions. Those with unplanned excisions were more likely to be female (relative risk [RR]: 1.9; P = .002), undergo initial excision at a nonsarcoma center (RR: 14.0; P < .001), have masses distal to the forearm (RR: 1.6; P = .02), and have smaller masses (4.8 vs 7.4 cm, P = .03). 71.4% of tumors were high grade, and 60.7% less than 5 cm in size.Unplanned excisions had positive margins in 96.4% of cases and were more likely to undergo re-excision (odds ratio [OR]: 20.0; P = .001), more total resections (2.7 vs 1.4, P = .009), sacrifice of neurovascular structures (OR: 6.1; P = .04), adjuvant radiation therapy (OR: 4.5; P = .05), adjuvant systemic therapy (OR: 10.9; P = .03), or experience a complication (OR: 17.6; P = .002) at an average of 38.0 months of follow-up.Nearly half of all unplanned excision patients developed a local recurrence or metastatic disease. Six patients required an amputation versus one in the planned cohort (P = .17), and 26.5% of patients died at an average of 32.5 months from presentation. Conclusions: Distal upper-extremity sarcoma excisions are frequently unplanned, with high rates of morbidity compared with planned excisions. Surgeons should have a low threshold for cross-sectional imaging and core needle biopsy of atypical lesions, irrespective of size, with referral to a sarcoma center. Type of study/level of evidence: Prognostic IV.

Treatment of a Carpal Giant Cell Tumor of Bone With Curettage and Cemented Capitohamate Fusion.

Carpal giant cell tumor of bone spanning multiple bones is a rare condition. We present a case of a man in his fifth decade with wrist pain who was found to have giant cell tumor of bone involving his capitate and hamate bones. This condition was successfully treated with intralesional curettage, argon beam coagulation, chemical cauterization and a cemented limited carpal fusion with satisfactory outcomes and no recurrence at 1-year postoperative follow-up.

Price transparency at the point of prescribing with real-time prescription benefits.

PURPOSE: Real-time prescription benefits (RTPB) shows prescribers patient-, medication-, and pharmacy-specific information on medication pricing, prior authorization requirements, and lower-cost alternatives. RTPB is intended to improve patient satisfaction and prescription fill rates by decreasing out-of-pocket costs for prescriptions. Therefore, we evaluated how RTPB affects prescribing patterns by examining acceptance and subsequent fill rates for RTPB alternative suggestions. METHODS: RTPB was implemented in February 2022 using external vendor interfaces. Prescribing data from March 2022 to March 2023 were analyzed. RTPB displayed alerts for medications requiring prior authorization or when alternative medications would result in cost savings. Patients were included if their prescription received an RTPB response and they had a subsequent encounter with pharmacy fill data. Primary outcomes were alert acceptance rates and prescription fill rates across RTPB alert groups, with a secondary outcome of monthly copay savings for accepted alerts. RESULTS: RTPB requests received a response for 88% of prescriptions, with price estimates provided for 77.9% of them. Lower-cost alternatives accounted for 67.2% of alerts, while prior authorization requirements represented 15% of alerts. Prescribers selected a lower-cost alternative 32% of the time. For those with an RTPB alert, patients filled prescriptions 68% of the time when an alternative was chosen, compared to 59% of the time when the original prescription was retained (odds ratio, 1.5; 95% confidence interval, 1.5-1.6; P < 0.001). Patients saved an average of $27.77 per month on copay costs when alternatives were selected. CONCLUSION: Implementation of RTPB was found to result in significant improvements in prescription fill rates and decrease patient copay costs, despite low alert acceptance rates.

Improving Medication Management for Inpatients with a Secondary Diagnosis of Parkinson Disease.

ABSTRACT: Patients who have Parkinson disease require individualized medication regimens to optimize care. A review of the medication management of patients admitted to a tertiary care hospital with a secondary diagnosis of Parkinson disease found significant departures from the patients' home regimen. Medication regimens are often altered by health care teams unfamiliar with Parkinson disease-specific care in order to conform to standard hospital medication orders and administration times, potentially resulting in increased patient falls, delirium, and mortality.A nurse-led multidisciplinary team consisting of pharmacy, nursing, informatics, neurology, and quality personnel implemented a quality improvement (QI) project between July 2020 and July 2022 to identify patients with Parkinson disease, including those with a secondary diagnosis and those undergoing deep brain stimulation, and customize medication management in order to reduce length of stay, mortality, falls, falls with harm, and 30-day readmissions. The QI project team also evaluated patient satisfaction with medication management.Among patients with a secondary diagnosis of Parkinson disease, the proportion who had medication histories conducted by a pharmacy staff member increased from a baseline of 53% to more than 75% per month. For all patients with Parkinson disease, those whose medication history was taken by a pharmacy staff member had orders matching their home regimen 89% of the time, whereas those who did not had orders matching the home regimen only 40% of the time. Among patients with a secondary diagnosis of Parkinson disease, the length-of-stay index decreased from a baseline of 1 to 0.94 and observed-to-expected mortality decreased from 1.03 to 0.78. The proportion of patients experiencing a fall decreased from an average of 5% to 4.08% per quarter, while the proportion of patients experiencing a fall with harm decreased from an average of 1% to 0.75% per quarter. The rate of 30-day readmissions decreased from 10.81% to 4.53% per quarter. Patient satisfaction scores were 1.95 points higher for patients who had medication histories taken by pharmacy than for those who did not (5 versus 3.05).

Why do users override alerts? Utilizing large language model to summarize comments and optimize clinical decision support.

OBJECTIVES: To evaluate the capability of using generative artificial intelligence (AI) in summarizing alert comments and to determine if the AI-generated summary could be used to improve clinical decision support (CDS) alerts. MATERIALS AND METHODS: We extracted user comments to alerts generated from September 1, 2022 to September 1, 2023 at Vanderbilt University Medical Center. For a subset of 8 alerts, comment summaries were generated independently by 2 physicians and then separately by GPT-4. We surveyed 5 CDS experts to rate the human-generated and AI-generated summaries on a scale from 1 (strongly disagree) to 5 (strongly agree) for the 4 metrics: clarity, completeness, accuracy, and usefulness. RESULTS: Five CDS experts participated in the survey. A total of 16 human-generated summaries and 8 AI-generated summaries were assessed. Among the top 8 rated summaries, five were generated by GPT-4. AI-generated summaries demonstrated high levels of clarity, accuracy, and usefulness, similar to the human-generated summaries. Moreover, AI-generated summaries exhibited significantly higher completeness and usefulness compared to the human-generated summaries (AI: 3.4 ± 1.2, human: 2.7 ± 1.2, P = .001). CONCLUSION: End-user comments provide clinicians' immediate feedback to CDS alerts and can serve as a direct and valuable data resource for improving CDS delivery. Traditionally, these comments may not be considered in the CDS review process due to their unstructured nature, large volume, and the presence of redundant or irrelevant content. Our study demonstrates that GPT-4 is capable of distilling these comments into summaries characterized by high clarity, accuracy, and completeness. AI-generated summaries are equivalent and potentially better than human-generated summaries. These AI-generated summaries could provide CDS experts with a novel means of reviewing user comments to rapidly optimize CDS alerts both online and offline.

Leveraging explainable artificial intelligence to optimize clinical decision support.

OBJECTIVE: To develop and evaluate a data-driven process to generate suggestions for improving alert criteria using explainable artificial intelligence (XAI) approaches. METHODS: We extracted data on alerts generated from January 1, 2019 to December 31, 2020, at Vanderbilt University Medical Center. We developed machine learning models to predict user responses to alerts. We applied XAI techniques to generate global explanations and local explanations. We evaluated the generated suggestions by comparing with alert's historical change logs and stakeholder interviews. Suggestions that either matched (or partially matched) changes already made to the alert or were considered clinically correct were classified as helpful. RESULTS: The final dataset included 2 991 823 firings with 2689 features. Among the 5 machine learning models, the LightGBM model achieved the highest Area under the ROC Curve: 0.919 [0.918, 0.920]. We identified 96 helpful suggestions. A total of 278 807 firings (9.3%) could have been eliminated. Some of the suggestions also revealed workflow and education issues. CONCLUSION: We developed a data-driven process to generate suggestions for improving alert criteria using XAI techniques. Our approach could identify improvements regarding clinical decision support (CDS) that might be overlooked or delayed in manual reviews. It also unveils a secondary purpose for the XAI: to improve quality by discovering scenarios where CDS alerts are not accepted due to workflow, education, or staffing issues.

Revealing the dynamic landscape of drug-drug interactions through network analysis.

Introduction: The landscape of drug-drug interactions (DDIs) has evolved significantly over the past 60 years, necessitating a retrospective analysis to identify research trends and under-explored areas. While methodologies like bibliometric analysis provide valuable quantitative perspectives on DDI research, they have not successfully delineated the complex interrelations between drugs. Understanding these intricate relationships is essential for deciphering the evolving architecture and progressive transformation of DDI research structures over time. We utilize network analysis to unearth the multifaceted relationships between drugs, offering a richer, more nuanced comprehension of shifts in research focus within the DDI landscape. Methods: This groundbreaking investigation employs natural language processing, techniques, specifically Named Entity Recognition (NER) via ScispaCy, and the information extraction model, SciFive, to extract pharmacokinetic (PK) and pharmacodynamic (PD) DDI evidence from PubMed articles spanning January 1962 to July 2023. It reveals key trends and patterns through an innovative network analysis approach. Static network analysis is deployed to discern structural patterns in DDI research, while evolving network analysis is employed to monitor changes in the DDI research trend structures over time. Results: Our compelling results shed light on the scale-free characteristics of pharmacokinetic, pharmacodynamic, and their combined networks, exhibiting power law exponent values of 2.5, 2.82, and 2.46, respectively. In these networks, a select few drugs serve as central hubs, engaging in extensive interactions with a multitude of other drugs. Interestingly, the networks conform to a densification power law, illustrating that the number of DDIs grows exponentially as new drugs are added to the DDI network. Notably, we discovered that drugs connected in PK and PD networks predominantly belong to the same categories defined by the Anatomical Therapeutic Chemical (ATC) classification system, with fewer interactions observed between drugs from different categories. Discussion: The finding suggests that PK and PD DDIs between drugs from different ATC categories have not been studied as extensively as those between drugs within the same categories. By unearthing these hidden patterns, our study paves the way for a deeper understanding of the DDI landscape, providing valuable information for future DDI research, clinical practice, and drug development focus areas.

Evaluation of inpatient medication guidance from an artificial intelligence chatbot.

PURPOSE: To analyze the clinical completeness, correctness, usefulness, and safety of chatbot and medication database responses to everyday inpatient medication-use questions. METHODS: We evaluated the responses from an artificial intelligence chatbot, a medication database, and clinical pharmacists to 200 real-world medication-use questions. Answer quality was rated by a blinded group of pharmacists, providers, and nurses. Chatbot and medication database responses were deemed "acceptable" if the mean reviewer rating was within 3 points of the mean rating for pharmacists' answers. We used descriptive statistics for reviewer ratings and Kendall's coefficient to evaluate interrater agreement. RESULTS: The medication database generated responses to 194 (97%) questions, with 88% considered acceptable for clinical correctness, 76% considered acceptable for completeness, 83% considered acceptable for safety, and 81% considered acceptable for usefulness compared to pharmacists' answers. The chatbot responded to only 160 (80%) questions, with 85% considered acceptable for clinical correctness, 65% considered acceptable for completeness, 71% considered acceptable for safety, and 68% considered acceptable for usefulness. CONCLUSION: Traditional search methods using a drug database provide more clinically correct, complete, safe, and useful answers than a chatbot. When the chatbot generated a response, the clinical correctness was similar to that of a drug database; however, it was not rated as favorably for clinical completeness, safety, or usefulness. Our results highlight the need for ongoing training and continued improvements to artificial intelligence chatbots for them to be incorporated reliably into the clinical workflow. With continued improvement in chatbot functionality, chatbots could be a useful pharmacist adjunct, providing healthcare providers with quick and reliable answers to medication-use questions.

The Impact of Clinical Decision Support on Health Disparities and the Digital Divide.

OBJECTIVES: This literature review summarizes relevant studies from the last three years (2020-2022) related to clinical decision support (CDS) and CDS impact on health disparities and the digital divide. This survey identifies current trends and synthesizes evidence-based recommendations and considerations for future development and implementation of CDS tools. METHODS: We conducted a search in PubMed for literature published between 2020 and 2022. Our search strategy was constructed as a combination of the MEDLINE®/PubMed® Health Disparities and Minority Health Search Strategy and relevant CDS MeSH terms and phrases. We then extracted relevant data from the studies, including priority population when applicable, domain of influence on the disparity being addressed, and the type of CDS being used. We also made note of when a study discussed the digital divide in some capacity and organized the comments into general themes through group discussion. RESULTS: Our search yielded 520 studies, with 45 included at the conclusion of screening. The most frequent CDS type in this review was point-of-care alerts/reminders (33.3%). Health Care System was the most frequent domain of influence (71.1%), and Blacks/African Americans were the most frequently included priority population (42.2%). Throughout the literature, we found four general themes related to the technology divide: inaccessibility of technology, access to care, trust of technology, and technology literacy.This survey revealed the diversity of CDS being used to address health disparities and several barriers which may make CDS less effective or potentially harmful to certain populations. Regular examinations of literature that feature CDS and address health disparities can help to reveal new strategies and patterns for improving healthcare.

The landscape of drug sensitivity and resistance in sarcoma.

Sarcomas are a family of rare malignancies composed of over 100 distinct histological subtypes. The rarity of sarcoma poses significant challenges in conducting clinical trials to identify effective therapies, to the point that many rarer subtypes of sarcoma do not have standard-of-care treatment. Even for established regimens, there can be substantial heterogeneity in responses. Overall, novel, personalized approaches for identifying effective treatments are needed to improve patient out-comes. Patient-derived tumor organoids (PDTOs) are clinically relevant models representative of the physiological behavior of tumors across an array of malignancies. Here, we use PDTOs as a tool to better understand the biology of individual tumors and characterize the landscape of drug resistance and sensitivity in sarcoma. We collected n=194 specimens from n=126 sarcoma patients, spanning 24 distinct subtypes. We characterized PDTOs established from over 120 biopsy, resection, and metastasectomy samples. We leveraged our organoid high-throughput drug screening pipeline to test the efficacy of chemotherapeutics, targeted agents, and combination therapies, with results available within a week from tissue collection. Sarcoma PDTOs showed patient-specific growth characteristics and subtype-specific histopathology. Organoid sensitivity correlated with diagnostic subtype, patient age at diagnosis, lesion type, prior treatment history, and disease trajectory for a subset of the compounds screened. We found 90 biological pathways that were implicated in response to treatment of bone and soft tissue sarcoma organoids. By comparing functional responses of organoids and genetic features of the tumors, we show how PDTO drug screening can provide an orthogonal set of information to facilitate optimal drug selection, avoid ineffective therapies, and mirror patient outcomes in sarcoma. In aggregate, we were able to identify at least one effective FDA-approved or NCCN-recommended regimen for 59% of the specimens tested, providing an estimate of the proportion of immediately actionable information identified through our pipeline. Highlights: Standardized organoid culture preserve unique sarcoma histopathological featuresDrug screening on patient-derived sarcoma organoids provides sensitivity information that correlates with clinical features and yields actionable information for treatment guidanceHigh-throughput screenings provide orthogonal information to genetic sequencingSarcoma organoid response to treatment correlates with patient response to therapyLarge scale, functional precision medicine programs for rare cancers are feasible within a single institution.

Fibroblast Activation Protein Expression in Sarcomas.

Objectives: Fibroblast activation protein alpha (FAP) is highly expressed by cancer-associated fibroblasts in multiple epithelial cancers. The aim of this study was to characterize FAP expression in sarcomas to explore its potential utility as a diagnostic and therapeutic target and prognostic biomarker in sarcomas. Methods: Available tissue samples from patients with bone or soft tissue tumors were identified at the University of California, Los Angeles. FAP expression was evaluated via immunohistochemistry (IHC) in tumor samples (n = 63), adjacent normal tissues (n = 30), and positive controls (n = 2) using semiquantitative systems for intensity (0 = negative; 1 = weak; 2 = moderate; and 3 = strong) and density (none, <25%, 25-75%; >75%) in stromal and tumor/nonstromal cells and using a qualitative overall score (not detected, low, medium, and high). Additionally, RNA sequencing data in publicly available databases were utilized to compare FAP expression in samples (n = 10,626) from various cancer types and evaluate the association between FAP expression and overall survival (OS) in sarcoma (n = 168). Results: The majority of tumor samples had FAP IHC intensity scores ≥2 and density scores ≥25% for stromal cells (77.7%) and tumor cells (50.7%). All desmoid fibromatosis, myxofibrosarcoma, solitary fibrous tumor, and undifferentiated pleomorphic sarcoma samples had medium or high FAP overall scores. Sarcomas were among cancer types with the highest mean FAP expression by RNA sequencing. There was no significant difference in OS in patients with sarcoma with low versus high FAP expression. Conclusion: The majority of the sarcoma samples showed FAP expression by both stromal and tumor/nonstromal cells. Further investigation of FAP as a potential diagnostic and therapeutic target in sarcomas is warranted.

FAPI PET Signal in Hibernoma Reflects FAP Expression in Tumor Vasculature Cells.

ABSTRACT: A 43-year-old man with a growing mass in the right groin, concerned for liposarcoma, underwent MRI and 68 Ga-fibroblast activation protein inhibitor (FAPI)-46 PET/CT before surgery. Fibroblast activation protein inhibitor PET/CT demonstrated increased uptake (SUV max , 3.2) predominantly in the solid portion, where MRI showed gadolinium enhancement. The patient subsequently underwent surgery and was diagnosed with hibernoma. The immunohistochemistry of the tumor revealed the fibroblast activation protein expression in the fibrovascular network and myofibroblastic cells of the tumor. This case suggests that the FAPI uptake can be affected by the vascular cells, and thus, a careful interpretation of the FAPI PET signal may be needed.

Using AI-generated suggestions from ChatGPT to optimize clinical decision support.

OBJECTIVE: To determine if ChatGPT can generate useful suggestions for improving clinical decision support (CDS) logic and to assess noninferiority compared to human-generated suggestions. METHODS: We supplied summaries of CDS logic to ChatGPT, an artificial intelligence (AI) tool for question answering that uses a large language model, and asked it to generate suggestions. We asked human clinician reviewers to review the AI-generated suggestions as well as human-generated suggestions for improving the same CDS alerts, and rate the suggestions for their usefulness, acceptance, relevance, understanding, workflow, bias, inversion, and redundancy. RESULTS: Five clinicians analyzed 36 AI-generated suggestions and 29 human-generated suggestions for 7 alerts. Of the 20 suggestions that scored highest in the survey, 9 were generated by ChatGPT. The suggestions generated by AI were found to offer unique perspectives and were evaluated as highly understandable and relevant, with moderate usefulness, low acceptance, bias, inversion, redundancy. CONCLUSION: AI-generated suggestions could be an important complementary part of optimizing CDS alerts, can identify potential improvements to alert logic and support their implementation, and may even be able to assist experts in formulating their own suggestions for CDS improvement. ChatGPT shows great potential for using large language models and reinforcement learning from human feedback to improve CDS alert logic and potentially other medical areas involving complex, clinical logic, a key step in the development of an advanced learning health system.

Assessing the Value of ChatGPT for Clinical Decision Support Optimization.

Objective: To determine if ChatGPT can generate useful suggestions for improving clinical decision support (CDS) logic and to assess noninferiority compared to human-generated suggestions. Methods: We supplied summaries of CDS logic to ChatGPT, an artificial intelligence (AI) tool for question answering that uses a large language model, and asked it to generate suggestions. We asked human clinician reviewers to review the AI-generated suggestions as well as human-generated suggestions for improving the same CDS alerts, and rate the suggestions for their usefulness, acceptance, relevance, understanding, workflow, bias, inversion, and redundancy. Results: Five clinicians analyzed 36 AI-generated suggestions and 29 human-generated suggestions for 7 alerts. Of the 20 suggestions that scored highest in the survey, 9 were generated by ChatGPT. The suggestions generated by AI were found to offer unique perspectives and were evaluated as highly understandable and relevant, with moderate usefulness, low acceptance, bias, inversion, redundancy. Conclusion: AI-generated suggestions could be an important complementary part of optimizing CDS alerts, can identify potential improvements to alert logic and support their implementation, and may even be able to assist experts in formulating their own suggestions for CDS improvement. ChatGPT shows great potential for using large language models and reinforcement learning from human feedback to improve CDS alert logic and potentially other medical areas involving complex, clinical logic, a key step in the development of an advanced learning health system.