Hartford Hospital

BackgroundMachine learning (ML) was used to predict pain scores and opioid consumption after elective spine surgery in the presence and absence of erector spinae plane block (ESP).MethodsA single-center retrospective chart review of 2796 cases was conducted. These cases were divided into the control group (N=1255) consisting of patients who did not receive the ESP blocks and the treatment group consisting of patients who received the blocks (N=1541). The gradient boosting ensemble tree methodology was employed to develop the AI predictive models. Feature importance for each optimized gradient boosting model was quantified using impurity-based importance scores, as implemented in the scikit-learn library. Partial dependence analysis was conducted to characterize the direction, magnitude, and non-linear nature of predictor-outcome relationships across clinically relevant ranges.ResultsOn unadjusted univariate analysis, the ESP block was associated with a statistically significant (p=0.01) yet clinically irrelevant 1% increase in average postsurgical pain scores. Conversely, ESP block was associated with a statistically non-significant (p=0.13) but clinically relevant 6.7% reduction in opioid consumption (MME/kg/day). These associations are exploratory and should not be interpreted as causal. Three AI models were developed to predict postsurgical pain and opioid consumption. The best-performing model, which predicts average postsurgical pain, achieved a mean absolute error of 1.24 on a 10-point scale (approximately 12.4%). High-importance predictors across the models included preoperative pain scores, serum glucose, and white blood cell count, as well as age.ConclusionIt is feasible to use machine-learning approaches to identify risk factors for postoperative pain and predict population-level pain scores and opioid consumption in spine surgery using large datasets; these models are not intended for individual-level prediction. The role of ESP in spine surgery, however, remains uncertain, and ESP block findings should be interpreted as exploratory associations only.

BackgroundLookUpTrials is a clinician-facing digital platform designed to support point-of-care navigation of institution-specific oncology clinical trials, incorporating artificial intelligence–assisted summarization and search functionalities. While embedding curated trial knowledge tools in community oncology workflows is feasible, the human infrastructure required to sustain high-quality, up-to-date trial data over time remains underexplored.ObjectiveThis study aimed to evaluate the feasibility of a structured training program embedding trainees within a supervised human-in-the-loop workflow for oncology clinical trial data curation in support of the LookUpTrials digital platform.MethodsWe conducted a descriptive feasibility evaluation of a cohort-based training and workflow model. A total of 10 undergraduate trainees curated publicly available trial information from institutional portals and ClinicalTrials.gov across 5 academic medical centers. Trial throughput and participant experiences were summarized descriptively.ResultsOver 10 months, trainees curated 2503 oncology clinical trial entries across 5 institutions, with processing rates increasing over time. Participants reported that structured onboarding and peer support facilitated engagement with data curation workflows.ConclusionsStandardized human-in-the-loop workflows for clinical trial data stewardship can be implemented with individuals without prior clinical trials or informatics experience when supported by structured training and quality assurance. This study complements prior feasibility work on embedding trial knowledge tools in community oncology settings by focusing on the sustainability of the underlying content pipeline. Further evaluation is needed to assess scalability across institutions and durability over longer periods of platform maturation.

INTRODUCTION: The readability, accuracy, and completeness of urogynecology health information provided by large language models (LLMs) has been found to be comparable to currently available education materials. However, LLM responses to patient-initiated questions have not yet been examined. OBJECTIVE: To investigate the accuracy, completeness, and readability of information provided to patients by an LLM in response to patient questions at the time of the first urogynecology consultation. METHODS: This is a secondary analysis of a randomized trial examining use of an LLM, Chat Generative Pre-trained Transformer (ChatGPT 4o; OpenAI) by patients at their initial urogynecology visit. Participants were recruited if they presented with prolapse, incontinence, or lower urinary tract symptoms (LUTS) and were English- or Spanish-speaking. They were randomized into one of three arms: use of ChatGPT prior to their appointment (Arm 1), use of ChatGPT following their appointment (Arm 2), or no use of an LLM (Arm 3). Participants in Arms 1 and 2 were provided ChatGPT on a tablet, and instructed to ask (typing or speaking) the program anything they’d like about their most important urogynecology problem, with up to five follow-up questions. English and Spanish ChatGPT transcripts were analyzed by two independent physician reviewers for accuracy, completeness, actionability/understandability [Patient Education Materials Assessment Tool (PEMAT)], and readability [Simple Measure of Gobbledygook (SMOG), Spanish Orthographic Length (SOL), Fry Readability Graph, Flesch-Kincaid (FK) Grade Level, Fernández Huerta (FH) Readability Index]. Significant results were declared at P<0.05. RESULTS: 125 patients were randomized from July to December, 2024, and 79 conversation transcripts were collected (41 Arm 1, 38 Arm 2). Average participant age was 58.4±13.8 years, and the majority of participants identified as non-Hispanic White (77%). 94% of participants had at least a high school education, and median health literacy measured by Short Assessment of Health Literacy (SAHL) score was 17 (out of 18). 91% of participants communicated with the chatbot in English (Table 1). Reviewers found the majority of chatbot responses were equally as accurate or complete as information that would be provided by a urogynecology specialist, and PEMAT scores indicated moderate understandability and low actionability (Table 2). Averaged readability scores of the chatbot responses were 12.2±2.3 for Arm 1 and 11.4±2.6 for Arm 2 (Figure 1). Readability scores did not significantly correlate with participants’ health literacy (SAHL) (r=0.086, p=0.450). Moderate or good agreement (kappa, κ) was noted between urogynecologist-documented primary diagnosis and themes patient discussed with chatbot for Arm 1 (prolapse κ=0.739, P<0.001; incontinence κ=0.564, P<0.001; LUTS κ=0.663, P<0.001) and Arm 2 (prolapse κ=0.713, P<0.001; incontinence κ=0.653, P<0.001; LUTS κ=0.418, P=0.011). CONCLUSIONS: ChatGPT responses to patient questions were overall as accurate and complete as those provided by urogynecologists and were easy to understand, but showed poor actionability. Responses also had higher-than-recommended readability levels for patient education materials (above the 5th to 6th-grade level). Unlike printed materials, however, the AI chatbot can simplify responses on request. Future studies may focus on the variability in accuracy of such simplified information when using LLMs for patient education.Figure 1Table 1Table 2

The patient was in his late 70s with a prior history of aortic valve replacement, recent emergent percutaneous coronary intervention for the right coronary artery (RCA), and a chronic type A aortic dissection and a 6-cm ascending aortic aneurysm. He developed Staphylococcus epidermidis bacteraemia and was found to have an aortic root abscess. Echocardiography demonstrated dehiscence of the aortic bioprosthesis with an aortic root abscess and moderate-to-severe mitral regurgitation. Radical debridement was performed, including resection of the aorto-mitral curtain through an extended transseptal approach. The protruding RCA stent was partially resected to allow selective cardioplegia. Mitral valve repair was performed with implantation of a 30-mm partial annuloplasty ring. A rectangular double-patch technique was used. The first bovine pericardial patch was sewn to the dome of the left atrium, and the second patch reconstructed the aortic annulus, with reinforcement of the fibrous trigones using pledgeted sutures. A handmade composite Bentall graft was constructed using a 30-mm Dacron graft with a skirted 25-mm bioprosthesis, and implanted with non-everting sutures. Both coronary buttons were reimplanted, followed by hemiarch replacement. The right atrium was closed with a separate pericardial patch, and a saphenous vein graft was placed to the distal RCA.

There is a gap in objective methods to review applications for advanced medical training. The objective of this study was to evaluate the accuracy of a new machine learning-based residency and fellowship applicant review program, Halsted (Medicratic) in urogynecology fellowship applicant selection compared with program director (PD) review. This Institutional Review Board-approved study compared PD's standard assessment of fellowship applicants to the Halsted-based assessment at 3 programs in the 2023-2024 application cycle. Each program provided a score for each candidate on a 100-point scale in several domains. After the conclusion of the match, each PD completed a profile within Halsted that identified their preferred qualities in applicants. Halsted scores were obtained, which were compared with PD scores. A total of 126 applications were reviewed, with 59 applicants reviewed by more than 1 program. Program 1 ( r =0.60; P =0.0019) and Program 2 ( r =0.58; P <0.001) exhibited a significantly strong positive correlation between PD-assigned overall application scores and Halsted scores, while Program 3 exhibited a weak positive correlation between scores ( r =0.33; P =0.0225). There were significant differences in the scoring of the same applicant between programs for PD-assigned mean overall scores ( P <0.001) and Halsted scores ( P <0.001). A significant positive correlation was found between Halsted rankings of applicants and rankings assigned by PDs. In addition, significant differences in interprogram rankings of applicants suggest that there is a range of qualities that each program values and that application review is individualized. Machine learning assistance in application review is a developing tool with the potential to reduce bias and decrease program administrative burden.

Despite advances in guideline-directed medical therapy (GDMT) for heart failure (HF) with reduced ejection fraction, challenges with uptitration may delay optimal treatment. Cardiac resynchronization therapy (CRT) improves clinical outcomes, yet optimal sequencing of CRT and GDMT remains uncertain. We sought to characterize longitudinal GDMT titration and evaluate associations between CRT timing and outcomes among US patients. A retrospective cohort study of 6334 Medicare beneficiaries with an identifiable incident HF diagnosis who will receive CRT between 2018 and 2022. Early CRT was defined as implantation within 1 year of diagnosis. Medication use was assessed by medication possession ratios of ≥0.8 in quarterly intervals. Outcomes were compared using propensity score overlap-weighted models. GDMT uptake increased sharply after the incident HF diagnosis but plateaued quickly with consistent medication use remaining low. Early CRT (n = 3777) was associated with lower adjusted risk of composite all-cause death or HF exacerbation (hazard ratio 0.80, 95% confidence interval [CI] 0.72-0.89), all-cause death alone (hazard ratio 0.79, 95% CI 0.69-0.90), and fewer HF exacerbations (incident rate ratio 0.66, 95% CI 0.58-0.75) vs late CRT (n = 2557). Early CRT was also associated with lower inpatient and emergency department use, Medicare expenditures, and HF-related out-of-pocket costs after CRT implant. Early CRT was associated with lower mortality, fewer HF exacerbations, and lower health care use, supporting reconsideration of GDMT and CRT treatment sequencing.

To develop and evaluate an internally validated natural language processing (NLP) model to determine guideline adherence of antibiotic decisions in the emergency department (ED) for skin and soft tissue infections (SSTIs). This cross-sectional pilot study developed and applied a random forest (RF)/NLP model to classify clinical narratives of patients with skin infections as requiring either methicillin-resistant Staphylococcus aureus (MRSA) or non-MRSA antibiotic coverage. Conducted at a quaternary care children's hospital with an annual volume of 50,000 ED visits, our study included patients aged 1 to 18 years presenting for SSTIs to the ED from July 1, 2018, through June 30, 2023. Main outcomes included the NLP model's sensitivity, specificity, and receiver operator characteristic (ROC) curve compared to gold standard manual physician review of the medical records. A total of 1588 patients were part of the training data set, with an additional 423 patients utilized for validation. The RF model achieved an area under the curve (AUC) of 0.99, 97.0% sensitivity, and 94.9% specificity. In the validation data set, the model achieved 96.6% sensitivity and 90.1% specificity. Performance remained strong despite absent/missing patient history in certain narratives. The NLP model demonstrated that automated analysis of clinical narratives for determining guideline adherence is feasible. Despite missing data in narratives, the model's high performance suggests potential for broader application. Given the rise in antibiotic-resistant infections and the role of judicious antibiotic use, developing automated systems through NLP can significantly contribute to health care delivery and patient safety. The methodology of this study provides a feasible, sustainable path for similar applications in the emergency setting.

Suicidal thoughts among physicians are common and increased during the COVID-19 pandemic, yet many hesitate to seek mental health support. We sought to examine factors influencing seeking help for suicidality among physicians and trainees, their narrative descriptions of suicidality, and the impacts of losing a physician in their lives to suicide. We conducted a qualitative analysis of published narratives from physicians and trainees describing experiences in seeking help for suicidality. We systematically searched MEDLINE and PsycInfo (March 2000 to March 2023) and manually reviewed reference lists of included articles. We analyzed narratives and generated themes using reflexive thematic analysis. Across 52 narratives, authors commonly described preceding pressures, ways of coping, and intense emotional distress and altered self-perceptions. Barriers to seeking help were rooted in pressures that contributed to suicidality, including a medical culture that values achievement and stoicism, the growing severity of illness, stigma regarding mental health, fears about losing one's license, reluctance to burden others, and limited access to timely, confidential care. In contrast, facilitators to seeking help included access to confidential services, mental health literacy and, most notably, the care and role modelling of seeking help by others. Narratives advocated for cultural and systemic changes within the profession. Intense pressures to achieve in medicine discourage physicians from acknowledging distress and seeking help, but role modelling of seeking help can attenuate the stigma. Understanding these intersecting factors can inform changes needed to develop targeted suicide prevention strategies for physicians and promote physician well-being.

A 56‐year‐old woman with a history of C4–C5 myelomeningocele repair as a newborn and cervical syringomyelia presented with one week of rapidly worsening bilateral lower extremity weakness and numbness, saddle anesthesia, and bladder incontinence. MRI of the entire spine revealed a 1.5 × 0.5 cm homogenously enhancing intramedullary lesion at T7–T8 with associated cord edema and rostral syrinx formation. MRI of the brain and FDG PET–CT scan were unremarkable. She was taken to the operating room for a T6–T8 laminectomy and biopsy. H&E staining revealed a relatively dense mononuclear‐cell infiltrate, which comprised numerous medium‐sized cells with round, irregular, or reniform nuclei, slightly dispersed chromatin, and relatively abundant eosinophilic, and occasionally, somewhat vacuolated cytoplasm. Immunohistochemical staining was strongly positive for CD163, PU.1, CD68, and ALK. FISH studies demonstrated ALK rearrangement in 70% of nuclei. Next‐generation sequencing including both DNA and RNA testing on a formalin‐fixed, paraffin‐embedded tissue sample detected a KIF5B/ALK gene fusion. She received radiotherapy with 2000 cGY in 200 cGy fractions to T6–T9 with no change in lesional size or enhancement. She was started on the ALK inhibitor alectinib. Subsequent MRI showed a complete response. She has had no evidence of disease recurrence on alectinib for 18 months. ALK‐positive histiocytosis is a recently described distinct clinicopathologic entity. Our case is notable for older age at diagnosis, isolated intramedullary involvement, and radioresistance but later marked targeted‐therapy response, thus furthering the understanding of the spectrum of ALK‐positive histiocytosis biology.

The role of coronary computed tomography angiography (CCTA) in diagnostic cardiology is steadily increasing. Normal CCTAs successfully reclassify patients to a low-to-intermediate risk category with marked reduction in cardiovascular events. However, hepatic steatosis has been associated with adverse cardiovascular events and can be routinely evaluated on non-contrast CCTAs. The prevalence of hepatic steatosis in the setting of normal CCTAs is not known and may be of clinical significance. We conducted a retrospective, single-center analysis to evaluate for the presence and incidence of hepatic steatosis in the setting of normal CCTAs. Non-contrast images were used to identify/measure the presence or absence of hepatic steatosis. The presence of hepatic steatosis was defined based on attenuation measurements in Hounsfield units (HU), by comparing the liver and spleen. Of 297 patients screened, a cohort of 99 patients were identified with a coronary calcium score of zero. Ten patients (10%) were found to have non-calcified plaque on CCTAand were excluded from the study population. Of the remaining 89 patients, 13 patients (14%) were found to have evidence of hepatic steatosis. Patients with normal CCTAs are often reassured, supported by data predicting an improved prognosis with low rates of cardiovascular events. Wepropose that evaluating and identifying patients with underlying hepatic steatosis, a known associated marker of atherosclerosis, even in patients with "normal"CCTAs is important, as this correlates with higher cardiovascular risk.