Quality, Accuracy, and Bias in ChatGPT-Based Summarization of Medical Abstracts.

PURPOSE: Worldwide clinical knowledge is expanding rapidly, but physicians have sparse time to review scientific literature. Large language models (eg, Chat Generative Pretrained Transformer [ChatGPT]), might help summarize and prioritize research articles to review. However, large language models sometimes "hallucinate" incorrect information. METHODS: We evaluated ChatGPT's ability to summarize 140 peer-reviewed abstracts from 14 journals. Physicians rated the quality, accuracy, and bias of the ChatGPT summaries. We also compared human ratings of relevance to various areas of medicine to ChatGPT relevance ratings. RESULTS: ChatGPT produced summaries that were 70% shorter (mean abstract length of 2,438 characters decreased to 739 characters). Summaries were nevertheless rated as high quality (median score 90, interquartile range [IQR] 87.0-92.5; scale 0-100), high accuracy (median 92.5, IQR 89.0-95.0), and low bias (median 0, IQR 0-7.5). Serious inaccuracies and hallucinations were uncommon. Classification of the relevance of entire journals to various fields of medicine closely mirrored physician classifications (nonlinear standard error of the regression [SER] 8.6 on a scale of 0-100). However, relevance classification for individual articles was much more modest (SER 22.3). CONCLUSIONS: Summaries generated by ChatGPT were 70% shorter than mean abstract length and were characterized by high quality, high accuracy, and low bias. Conversely, ChatGPT had modest ability to classify the relevance of articles to medical specialties. We suggest that ChatGPT can help family physicians accelerate review of the scientific literature and have developed software (pyJournalWatch) to support this application. Life-critical medical decisions should remain based on full, critical, and thoughtful evaluation of the full text of research articles in context with clinical guidelines.

Design of a pragmatic randomized implementation effectiveness trial testing a health system wide hypertension program for older adults.

Hypertension control remains poor. Multiple barriers at the level of patients, providers, and health systems interfere with implementation of hypertension guidelines and effective lowering of BP. Some strategies such as self-measured blood pressure (SMBP) and remote management by pharmacists are safe and effectively lower BP but have not been effectively implemented. In this study, we combine such evidence-based strategies to build a remote hypertension program and test its effectiveness and implementation in large health systems. This randomized, controlled, pragmatic type I hybrid implementation effectiveness trial will examine the virtual collaborative care clinic (vCCC), a hypertension program that integrates automated patient identification, SMBP, remote BP monitoring by trained health system pharmacists, and frequent patient-provider communication. We will randomize 1000 patients with uncontrolled hypertension from two large health systems in a 1:1 ratio to either vCCC or control (usual care with education) groups for a 2-year intervention. Outcome measures including BP measurements, cognitive function, and a symptom checklist will be completed during study visits. Other outcome measures of cardiovascular events, mortality, and health care utilization will be assessed using Medicare data. For the primary outcome of proportion achieving BP control (defined as systolic BP < 130 mmHg) in the two groups, we will use a generalized linear mixed model analysis. Implementation outcomes include acceptability and feasibility of the program. This study will guide implementation of a hypertension program within large health systems to effectively lower BP.

Multiple Pharmacotherapy Adaptations for Smoking Cessation Based on Treatment Response in Black Adults Who Smoke: A Randomized Clinical Trial.

Importance: Adapting to different smoking cessation medications when an individual has not stopped smoking has shown promise, but efficacy has not been tested in racial and ethnic minority individuals who smoke and tend to have less success in quitting and bear a disproportionate share of tobacco-related morbidity and mortality. Objective: To evaluate efficacy of multiple smoking cessation pharmacotherapy adaptations based on treatment response in Black adults who smoke daily. Design, Setting, and Participants: This randomized clinical trial of adapted therapy (ADT) or enhanced usual care (UC) included non-Hispanic Black adults who smoke and was conducted from May 2019 to January 2022 at a federally qualified health center in Kansas City, Missouri. Data analysis took place from March 2022 to January 2023. Interventions: Both groups received 18 weeks of pharmacotherapy with long-term follow-up through week 26. The ADT group consisted of 196 individuals who received a nicotine patch (NP) and up to 2 pharmacotherapy adaptations, with a first switch to varenicline at week 2 and, if needed, a second switch to bupropion plus NP (bupropion + NP) based on carbon monoxide (CO)-verified smoking status (CO ≥6 ppm) at week 6. The UC group consisted of 196 individuals who received NP throughout the duration of treatment. Main Outcomes and Measures: Anabasine-verified and anatabine-verified point-prevalence abstinence at week 12 (primary end point) and weeks 18 and 26 (secondary end points). The χ2 test was used to compare verified abstinence at week 12 (primary end point) and weeks 18 and 26 (secondary end points) between ADT and UC. A post hoc sensitivity analysis of smoking abstinence at week 12 was performed with multiple imputation using a monotone logistic regression with treatment and gender as covariates to impute the missing data. Results: Among 392 participants who were enrolled (mean [SD] age, 53 [11.6] years; 224 [57%] female; 186 [47%] ≤ 100% federal poverty level; mean [SD] 13 [12.4] cigarettes per day), 324 (83%) completed the trial. Overall, 196 individuals were randomized to each study group. Using intent-to-treat and imputing missing data as participants who smoke, verified 7-day abstinence was not significantly different by treatment group at 12 weeks (ADT: 34 of 196 [17.4%]; UC: 23 of 196 [11.7%]; odds ratio [OR], 1.58; 95% CI, 0.89-2.80; P = .12), 18 weeks (ADT: 32 of 196 [16.3%]; UC: 31 of 196 [15.8%]; OR, 1.04; 95% CI, 0.61-1.78; P = .89), and 26 weeks (ADT: 24 of 196 [12.2%]; UC: 26 of 196 [13.3%]; OR, 0.91; 95% CI, 0.50-1.65; P = .76). Of the ADT participants who received pharmacotherapy adaptations (135/188 [71.8%]), 11 of 135 (8.1%) were abstinent at week 12. Controlling for treatment, individuals who responded to treatment and had CO-verified abstinence at week 2 had 4.6 times greater odds of being abstinent at week 12 (37 of 129 [28.7%] abstinence) than those who did not respond to treatment (19 of 245 [7.8%] abstinence; OR; 4.6; 95% CI, 2.5-8.6; P < .001). Conclusions and Relevance: In this randomized clinical trial of adapted vs standard of care pharmacotherapy, adaptation to varenicline and/or bupropion + NP after failure of NP monotherapy did not significantly improve abstinence rates for Black adults who smoke relative to those who continued treatment with NP. Those who achieved abstinence in the first 2 weeks of the study were significantly more likely to achieve later abstinence, highlighting early treatment response as an important area for preemptive intervention. Trial Registration: ClinicalTrials.gov Identifier: NCT03897439.

Machine Learning Prediction of Urine Cultures in Primary Care.

Context: Antibiotics for suspected urinary tract infection (UTI) is appropriate only when an infection is present. Urine culture is definitive but takes >1 day to result. A machine learning urine culture predictor was recently devised for Emergency Department (ED) patients but requires use of urine microscopy ("NeedMicro" predictor), which is not routinely available in primary care (PC). Objective: To adapt this predictor to use only features available in primary care and determine if predictive accuracy generalizes to the primary care setting. We call this the "NoMicro" predictor. Study Design and Analysis: Multicenter, retrospective, observational, cross-sectional analysis. Machine learning predictors were trained using extreme gradient boosting, artificial neural networks, and random forests. Models were trained on the ED dataset and were evaluated on both the ED dataset (internal validation) and the PC dataset (external validation). Setting: United States (US) academic medical centers emergency department and family medicine clinic. Population Studied: 80387 (ED, previously described) and 472 (PC, newly curated) US adults. Instrument: Physicians performed retrospective chart review. The primary outcome extracted was pathogenic urine culture growing ≥100,000 colony forming units. Predictor variables included age; gender; dipstick urinalysis nitrites, leukocytes, clarity, glucose, protein, and blood; dysuria; abdominal pain; and history of UTI. Outcome Measures: Predictor overall discriminative performance (receiver operating characteristic area under the curve, ROC-AUC), performance statistics (e.g., sensitivity, negative predictive value, etc.), and calibration. Results: The "NoMicro" model performs similarly to the "NeedMicro" model in internal validation on the ED dataset: NoMicro ROC-AUC 0.862 (95% CI: 0.856-0.869) vs. NeedMicro 0.877 (95% CI: 0.871-0.884). External validation on the primary care dataset also yielded high performance (NoMicro ROC-AUC 0.850 [95% CI: 0.808-0.889]), despite being trained on Emergency Department data. Simulation of a hypothetical, retrospective clinical trial suggests the NoMicro model could be used to avoid antibiotic overuse by safely withhold antibiotics in low-risk patients. Conclusions: The hypothesis that the NoMicro predictor generalizes to both PC and ED contexts is supported. Prospective trials to determine the real-world impact of using the NoMicro model to reduce antibiotic overuse are appropriate.

Adaptation and External Validation of Pathogenic Urine Culture Prediction in Primary Care Using Machine Learning.

BACKGROUND: Urinary tract infection (UTI) symptoms are common in primary care, but antibiotics are appropriate only when an infection is present. Urine culture is the reference standard test for infection, but results take >1 day. A machine learning predictor of urine cultures showed high accuracy for an emergency department (ED) population but required urine microscopy features that are not routinely available in primary care (the NeedMicro classifier). METHODS: We redesigned a classifier (NoMicro) that does not depend on urine microscopy and retrospectively validated it internally (ED data set) and externally (on a newly curated primary care [PC] data set) using a multicenter approach including 80,387 (ED) and 472 (PC) adults. We constructed machine learning models using extreme gradient boosting (XGBoost), artificial neural networks, and random forests (RFs). The primary outcome was pathogenic urine culture growing ≥100,000 colony forming units. Predictor variables included age; gender; dipstick urinalysis nitrites, leukocytes, clarity, glucose, protein, and blood; dysuria; abdominal pain; and history of UTI. RESULTS: Removal of microscopy features did not severely compromise performance under internal validation: NoMicro/XGBoost receiver operating characteristic area under the curve (ROC-AUC) 0.86 (95% CI, 0.86-0.87) vs NeedMicro 0.88 (95% CI, 0.87-0.88). Excellent performance in external (PC) validation was also observed: NoMicro/RF ROC-AUC 0.85 (95% CI, 0.81-0.89). Retrospective simulation suggested that NoMicro/RF can be used to safely withhold antibiotics for low-risk patients, thereby avoiding antibiotic overuse. CONCLUSIONS: The NoMicro classifier appears appropriate for PC. Prospective trials to adjudicate the balance of benefits and harms of using the NoMicro classifier are appropriate.

Implementing a home-based virtual hypertension programme-a pilot feasibility study.

INTRODUCTION: Implementing a health system-based hypertension programme may lower blood pressure (BP). METHODS: We performed a randomized, controlled pilot study to assess feasibility, acceptability, and safety of a home-based virtual hypertension programme integrating evidence-based strategies to overcome current barriers to BP control. Trained clinical pharmacists staffed the virtual collaborative care clinic (vCCC) to remotely manage hypertension using a BP dashboard and phone "visits" to monitor BP, adherence, side effects of medications, and prescribe anti-hypertensives. Patients with uncontrolled hypertension were identified via electronic health records. Enrolled patients were randomized to either vCCC or usual care for 3 months. We assessed patients' home BP monitoring behaviour, and patients', physicians', and pharmacists' perspectives on feasibility and acceptability of individual programme components. RESULTS: Thirty-one patients (vCCC = 17, usual care = 14) from six physician clinics completed the pilot study. After 3 months, average BP decreased in the vCCC arm (P = 0.01), but not in the control arm (P = 0.45). The vCCC participants measured BP more (9.9 vs. 1.2 per week, P < 0.001). There were no intervention-related adverse events. Participating physicians (n = 6), pharmacists (n = 5), and patients (n = 31) rated all programme components with average scores of >4.0, a pre-specified benchmark. Nine adaptations in vCCC design and delivery were made based on potential barriers to implementing the programme and suggestions. CONCLUSION: A home-based virtual hypertension programme using team-based care, technology, and a logical integration of evidence-based strategies is safe, acceptable, and feasible to intended users. These pilot data support studies to assess the effectiveness of this programme at a larger scale.

Protocol from a randomized clinical trial of multiple pharmacotherapy adaptations based on treatment response in African Americans who smoke.

Background: The standard of care in tobacco treatment is to continue individuals who smoke on the same cessation medication, even when they do not stop smoking. An alternative strategy is to adapt pharmacotherapy based on non-response. A handful of studies have examined this approach, but they have adapted pharmacotherapy only once and/or focused on adaptation distal rather than proximal to a failed quit attempt. Few studies have included racial/ethnic minorities who have less success in quitting and bear a disproportionate share of tobacco-related morbidity and mortality. Methods: The current study is comparing the efficacy of optimized (OPT) versus enhanced usual care (UC) for smoking cessation in African Americans (AA) who smoke cigarettes. AAs who smoke (n = 392) are randomized 1:1 to OPT or UC. Participants in both groups receive 7 sessions of smoking cessation counseling and18-weeks of pharmacotherapy with long-term follow-up through Week 26. OPT participants receive nicotine patch and up to two pharmacotherapy adaptations to varenicline and bupropion plus patch based on carbon monoxide verified smoking status (≥6 ppm) at Weeks 2 and 6. UC participants receive patch throughout the duration of treatment. We hypothesize that OPT will be more effective than UC on the primary outcome of biochemically verified abstinence at Week 12. Discussion: If effective, findings could broaden the scope of tobacco dependence treatment and move the field toward optimization strategies that impro ve abstinence for AA who smoke. Trial registration: NCT03897439.

NMR structure/function relationships of peptides corresponding to the C1B1 Region of PKC gamma.

The region (101-112) of C1B domain in PKC gamma plays a crucial role in the activation of the enzyme and subsequent gap junction inhibition. Substitution studies on peptides correlating to the C1B region show that a flexible structure and ability to be phosphorylated on serine 109 are critical for this purpose.

Protein kinase C gamma mutations in the C1B domain cause caspase-3-linked apoptosis in lens epithelial cells through gap junctions.

Failure to control oxidative stress is closely related to aging and to a diverse range of human diseases. We have reported that protein kinase C gamma (PKCgamma) acts as a primary oxidative stress sensor in the lens. PKCgamma has a Zn-finger C1B stress switch domain, residues 101-150. Mutation, H101Y, in the C1B domain of PKCgamma proteins causes a failure of the PKCgamma oxidative stress response [Lin, D., Takemoto, D.J., 2005. Oxidative activation of protein kinase Cgamma through the C1 domain. Effects on gap junctions. J. Biol. Chem. 280, 13682-13693]. Some human neurodegenerative spinocerebellar ataxia type 14 are caused by mutations in the PKCgamma C1B domain. In the current study we have investigated the effects of these mutations on lens epithelial cell responses to oxidative stress. The results demonstrate that PKCgamma C1B mutants had lower basal enzyme activities and were not activated by H(2)O(2). Furthermore, the PKCgamma mutations caused a failure of endogenous wild type PKCgamma to be activated by H(2)O(2). These PKCgamma mutations abolished the effect of H(2)O(2) on phosphorylation of Cx43 and Cx50 by H(2)O(2) activation of PKCgamma. The cells with PKCgamma C1B mutations had more Cx43 and/or Cx50 gap junction plaques which were not decreased by H(2)O(2). Since open gap junctions could have a bystander effect this could cause apoptosis to occur. H(2)O(2) (100 microM, 3 h) activated a caspase-3 apoptotic pathway in the lens epithelial cells but was more severe in cells expressing PKCgamma mutations. The presence of 18alpha-glycyrrhetinic acid (AGA), an inhibitor of gap junctions, decreased Cx43 and Cx50 protein levels and gap junction plaque number. This reduction in gap junctions by AGA resulted in inhibition of H(2)O(2)-induced apoptosis. Our results demonstrate that there is a dominant negative effect of PKCgamma C1B mutations on endogenous PKCgamma which results in loss of control of gap junctions. Modeled structures suggest that the severity of C1B mutation effects may be related to the extent of loss of C1B structure. Mutations in the C1B domain of PKCgamma result in increased apoptosis in lens epithelial cells. This can be prevented by a gap junction inhibitor. Thus, propagation of apoptosis from cell-to-cell in lens epithelial cells may be through open gap junctions. The control of gap junctions requires PKCgamma.

PKCgamma knockout mouse lenses are more susceptible to oxidative stress damage.

Cataracts, or lens opacities, are the leading cause of blindness worldwide. Cataracts increase with age and environmental insults, e.g. oxidative stress. Lens homeostasis depends on functional gap junctions. Knockout or missense mutations of lens gap junction proteins, Cx46 or Cx50, result in cataractogenesis in mice. We have previously demonstrated that protein kinase Cgamma (PKCgamma) regulates gap junctions in the lens epithelium and cortex. In the current study, we further determined whether PKCgamma control of gap junctions protects the lens from cataractogenesis induced by oxidative stress in vitro, using PKCgamma knockout and control mice as our models. The results demonstrate that PKCgamma knockout lenses are normal at 2 days post-natal when compared to control. However, cell damage, but not obvious cataract, was observed in the lenses of 6-week-old PKCgamma knockout mice, suggesting that the deletion of PKCgamma causes lenses to be more susceptible to damage. Furthermore, in vitro incubation or lens oxidative stress treatment by H(2)O(2) significantly induced lens opacification (cataract) in the PKCgamma knockout mice when compared to controls. Biochemical and structural results also demonstrated that H(2)O(2) activation of endogenous PKCgamma resulted in phosphorylation of Cx50 and subsequent inhibition of gap junctions in the lenses of control mice, but not in the knockout. Deletion of PKCgamma altered the arrangement of gap junctions on the cortical fiber cell surface, and completely abolished the inhibitory effect of H(2)O(2) on lens gap junctions. Data suggest that activation of PKCgamma is an important mechanism regulating the closure of the communicating pathway mediated by gap junction channels in lens fiber cells. The absence of this regulatory mechanism in the PKCgamma knockout mice may cause those lenses to have increased susceptibility to oxidative damage.