Community-engaged artificial intelligence research: A scoping review.

The degree to which artificial intelligence healthcare research is informed by data and stakeholders from community settings has not been previously described. As communities are the principal location of healthcare delivery, engaging them could represent an important opportunity to improve scientific quality. This scoping review systematically maps what is known and unknown about community-engaged artificial intelligence research and identifies opportunities to optimize the generalizability of these applications through involvement of community stakeholders and data throughout model development, validation, and implementation. Embase, PubMed, and MEDLINE databases were searched for articles describing artificial intelligence or machine learning healthcare applications with community involvement in model development, validation, or implementation. Model architecture and performance, the nature of community engagement, and barriers or facilitators to community engagement were reported according to PRISMA extension for Scoping Reviews guidelines. Of approximately 10,880 articles describing artificial intelligence healthcare applications, 21 (0.2%) described community involvement. All articles derived data from community settings, most commonly by leveraging existing datasets and sources that included community subjects, and often bolstered by internet-based data acquisition and subject recruitment. Only one article described inclusion of community stakeholders in designing an application-a natural language processing model that detected cases of likely child abuse with 90% accuracy using harmonized electronic health record notes from both hospital and community practice settings. The primary barrier to including community-derived data was small sample sizes, which may have affected 11 of the 21 studies (53%), introducing substantial risk for overfitting that threatens generalizability. Community engagement in artificial intelligence healthcare application development, validation, or implementation is rare. As healthcare delivery occurs primarily in community settings, investigators should consider engaging community stakeholders in user-centered design, usability, and clinical implementation studies to optimize generalizability.

Incorporating Patient Values in Large Language Model Recommendations for Surrogate and Proxy Decisions.

BACKGROUND: Surrogates, proxies, and clinicians making shared treatment decisions for patients who have lost decision-making capacity often fail to honor patients' wishes, due to stress, time pressures, misunderstanding patient values, and projecting personal biases. Advance directives intend to align care with patient values but are limited by low completion rates and application to only a subset of medical decisions. Here, we investigate the potential of large language models (LLMs) to incorporate patient values in supporting critical care clinical decision-making for incapacitated patients in a proof-of-concept study. METHODS: We simulated text-based scenarios for 50 decisionally incapacitated patients for whom a medical condition required imminent clinical decisions regarding specific interventions. For each patient, we also simulated five unique value profiles captured using alternative formats: numeric ranking questionnaires, text-based questionnaires, and free-text narratives. We used pre-trained generative LLMs for two tasks: 1) text extraction of the treatments under consideration and 2) prompt-based question-answering to generate a recommendation in response to the scenario information, extracted treatment, and patient value profiles. Model outputs were compared with adjudications by three domain experts who independently evaluated each scenario and decision. RESULTS AND CONCLUSIONS: Automated extractions of the treatment in question were accurate for 88% (n = 44/50) of scenarios. LLM treatment recommendations received an average Likert score by the adjudicators of 3.92 of 5.00 (five being best) across all patients for being medically plausible and reasonable treatment recommendations, and 3.58 of 5.00 for reflecting the documented values of the patient. Scores were highest when patient values were captured as short, unstructured, and free-text narratives based on simulated patient profiles. This proof-of-concept study demonstrates the potential for LLMs to function as support tools for surrogates, proxies, and clinicians aiming to honor the wishes and values of decisionally incapacitated patients.

APRICOT-Mamba: Acuity Prediction in Intensive Care Unit (ICU): Development and Validation of a Stability, Transitions, and Life-Sustaining Therapies Prediction Model.

On average, more than 5 million patients are admitted to intensive care units (ICUs) in the US, with mortality rates ranging from 10 to 29%. The acuity state of patients in the ICU can quickly change from stable to unstable, sometimes leading to life-threatening conditions. Early detection of deteriorating conditions can assist in more timely interventions and improved survival rates. While Artificial Intelligence (AI)-based models show potential for assessing acuity in a more granular and automated manner, they typically use mortality as a proxy of acuity in the ICU. Furthermore, these methods do not determine the acuity state of a patient (i.e., stable or unstable), the transition between acuity states, or the need for life-sustaining therapies. In this study, we propose APRICOT-M (Acuity Prediction in Intensive Care Unit-Mamba), a 1M-parameter state space-based neural network to predict acuity state, transitions, and the need for life-sustaining therapies in real-time among ICU patients. The model integrates ICU data in the preceding four hours (including vital signs, laboratory results, assessment scores, and medications) and patient characteristics (age, sex, race, and comorbidities) to predict the acuity outcomes in the next four hours. Our state space-based model can process sparse and irregularly sampled data without manual imputation, thus reducing the noise in input data and increasing inference speed. The model was trained on data from 107,473 patients (142,062 ICU admissions) from 55 hospitals between 2014-2017 and validated externally on data from 74,901 patients (101,356 ICU admissions) from 143 hospitals. Additionally, it was validated temporally on data from 12,927 patients (15,940 ICU admissions) from one hospital in 2018-2019 and prospectively on data from 215 patients (369 ICU admissions) from one hospital in 2021-2023. Three datasets were used for training and evaluation: the University of Florida Health (UFH) dataset, the electronic ICU Collaborative Research Database (eICU), and the Medical Information Mart for Intensive Care (MIMIC)-IV dataset. APRICOT-M significantly outperforms the baseline acuity assessment, Sequential Organ Failure Assessment (SOFA), for mortality prediction in both external (AUROC 0.95 CI: 0.94-0.95 compared to 0.78 CI: 0.78-0.79) and prospective (AUROC 0.99 CI: 0.97-1.00 compared to 0.80 CI: 0.65-0.92) cohorts, as well as for instability prediction (external AUROC 0.75 CI: 0.74-0.75 compared to 0.51 CI: 0.51-0.51, and prospective AUROC 0.69 CI: 0.64-0.74 compared to 0.53 CI: 0.50-0.57). This tool has the potential to help clinicians make timely interventions by predicting the transition between acuity states and decision-making on life-sustaining within the next four hours in the ICU.

Association of Sociodemographic Factors With Overtriage, Undertriage, and Value of Care After Major Surgery.

Objective: To determine whether certain patients are vulnerable to errant triage decisions immediately after major surgery and whether there are unique sociodemographic phenotypes within overtriaged and undertriaged cohorts. Background: In a fair system, overtriage of low-acuity patients to intensive care units (ICUs) and undertriage of high-acuity patients to general wards would affect all sociodemographic subgroups equally. Methods: This multicenter, longitudinal cohort study of hospital admissions immediately after major surgery compared hospital mortality and value of care (risk-adjusted mortality/total costs) across 4 cohorts: overtriage (N = 660), risk-matched overtriage controls admitted to general wards (N = 3077), undertriage (N = 2335), and risk-matched undertriage controls admitted to ICUs (N = 4774). K-means clustering identified sociodemographic phenotypes within overtriage and undertriage cohorts. Results: Compared with controls, overtriaged admissions had a predominance of male patients (56.2% vs 43.1%, P < 0.001) and commercial insurance (6.4% vs 2.5%, P < 0.001); undertriaged admissions had a predominance of Black patients (28.4% vs 24.4%, P < 0.001) and greater socioeconomic deprivation. Overtriage was associated with increased total direct costs [$16.2K ($11.4K-$23.5K) vs $14.1K ($9.1K-$20.7K), P < 0.001] and low value of care; undertriage was associated with increased hospital mortality (1.5% vs 0.7%, P = 0.002) and hospice care (2.2% vs 0.6%, P < 0.001) and low value of care. Unique sociodemographic phenotypes within both overtriage and undertriage cohorts had similar outcomes and value of care, suggesting that triage decisions, rather than patient characteristics, drive outcomes and value of care. Conclusions: Postoperative triage decisions should ensure equality across sociodemographic groups by anchoring triage decisions to objective patient acuity assessments, circumventing cognitive shortcuts and mitigating bias.

Resident and Fellow Performance and Supervision in Surgical Oncology Procedures.

BACKGROUND: Previous research has highlighted concerns among trainees and attendings that general surgery training and fellowship are inadequately preparing trainees for practice. Providing trainees with supervision that matches their proficiency may help bridge this gap. We sought to benchmark operative performance and supervision levels among senior surgery residents (post-graduate year 4 or 5) and fellows performing general surgical oncology procedures. STUDY DESIGN: Observational data were obtained from the Society for Improving Medical Procedural Learning (SIMPL) OR application for core general surgical oncology procedures performed at 103 unique residency and fellowship programs. Procedures were divided into breast and soft tissue, endocrine, and hepatopancreatobiliary (HPB). Case evaluations completed by trainees and attendings were analyzed to benchmark trainee operative performance and level of supervision. RESULTS: There were 4,907 resident cases and 425 fellow cases. Practice-ready performance, as assessed by trainees and faculty, was achieved by relatively low proportions of residents and fellows for breast and soft tissue cases (residents: 38%, fellows: 48%), endocrine cases (residents: 22%, fellows: 41%), and HPB cases (residents: 10%, fellows: 40%). Among cases in which trainees did achieve practice-ready performance, supervision only was provided for low proportions of cases as rated by trainees (residents: 17%, fellows: 18%) and attendings (residents: 21%, fellows 25%). CONCLUSION: In a sample of 103 residency and fellowship programs, attending surgeons rarely provided senior residents and fellows with levels of supervision commensurate to performance for surgical oncology procedures, even for high performing trainees. These findings suggest a critical need for surgical training programs to prioritize providing greater levels of independence to trainees that have demonstrated excellent performance.

Longitudinal clinical decision support for assessing decisions over time: State-of-the-art and future directions.

Objective: Patients and clinicians rarely experience healthcare decisions as snapshots in time, but clinical decision support (CDS) systems often represent decisions as snapshots. This scoping review systematically maps challenges and facilitators to longitudinal CDS that are applied at two or more timepoints for the same decision made by the same patient or clinician. Methods: We searched Embase, PubMed, and Medline databases for articles describing development, validation, or implementation of patient- or clinician-facing longitudinal CDS. Validated quality assessment tools were used for article selection. Challenges and facilitators to longitudinal CDS are reported according to PRISMA-ScR guidelines. Results: Eight articles met inclusion criteria; each article described a unique CDS. None used entirely automated data entry, none used living guidelines for updating the evidence base or knowledge engine as new evidence emerged during the longitudinal study, and one included formal readiness for change assessments. Seven of eight CDS were implemented and evaluated prospectively. Challenges were primarily related to suboptimal study design (with unique challenges for each study) or user interface. Facilitators included use of randomized trial designs for prospective enrollment, increased CDS uptake during longitudinal exposure, and machine-learning applications that are tailored to the CDS use case. Conclusions: Despite the intuitive advantages of representing healthcare decisions longitudinally, peer-reviewed literature on longitudinal CDS is sparse. Existing reports suggest opportunities to incorporate longitudinal CDS frameworks, automated data entry, living guidelines, and user readiness assessments. Generating best practice guidelines for longitudinal CDS would require a greater depth and breadth of published work and expert opinion.

TRANSCRIPTOMIC DIFFERENCES IN PERIPHERAL MONOCYTE POPULATIONS IN SEPTIC PATIENTS BASED ON OUTCOME.

ABSTRACT: Postsepsis early mortality is being replaced by survivors who experience either a rapid recovery and favorable hospital discharge or the development of chronic critical illness with suboptimal outcomes. The underlying immunological response that determines these clinical trajectories remains poorly defined at the transcriptomic level. As classical and nonclassical monocytes are key leukocytes in both the innate and adaptive immune systems, we sought to delineate the transcriptomic response of these cell types. Using single-cell RNA sequencing and pathway analyses, we identified gene expression patterns between these two groups that are consistent with differences in TNF-α production based on clinical outcome. This may provide therapeutic targets for those at risk for chronic critical illness in order to improve their phenotype/endotype, morbidity, and long-term mortality.

Identifying acute illness phenotypes via deep temporal interpolation and clustering network on physiologic signatures.

Using clustering analysis for early vital signs, unique patient phenotypes with distinct pathophysiological signatures and clinical outcomes may be revealed and support early clinical decision-making. Phenotyping using early vital signs has proven challenging, as vital signs are typically sampled sporadically. We proposed a novel, deep temporal interpolation and clustering network to simultaneously extract latent representations from irregularly sampled vital signs and derive phenotypes. Four distinct clusters were identified. Phenotype A (18%) had the greatest prevalence of comorbid disease with increased prevalence of prolonged respiratory insufficiency, acute kidney injury, sepsis, and long-term (3-year) mortality. Phenotypes B (33%) and C (31%) had a diffuse pattern of mild organ dysfunction. Phenotype B's favorable short-term clinical outcomes were tempered by the second highest rate of long-term mortality. Phenotype C had favorable clinical outcomes. Phenotype D (17%) exhibited early and persistent hypotension, high incidence of early surgery, and substantial biomarker incidence of inflammation. Despite early and severe illness, phenotype D had the second lowest long-term mortality. After comparing the sequential organ failure assessment scores, the clustering results did not simply provide a recapitulation of previous acuity assessments. This tool may impact triage decisions and have significant implications for clinical decision-support under time constraints and uncertainty.