Poor performance of ChatGPT in clinical rule-guided dose interventions in hospitalized patients with renal dysfunction.

PURPOSE: Clinical decision support systems (CDSS) are used to identify drugs with potential need for dose modification in patients with renal impairment. ChatGPT holds the potential to be integrated in the electronic health record (EHR) system to give such dosing advices. In this study, we aim to evaluate the performance of ChatGPT in clinical rule-guided dose interventions in hospitalized patients with renal impairment. METHODS: This cross-sectional study was performed at Tergooi Medical Center, the Netherlands. CDSS alerts regarding renal dysfunction were collected from the electronic health record (EHR) during a 2-week period and were presented to ChatGPT and an expert panel. Alerts were presented with and without patient variables. To evaluate the performance, suggested medication interventions were compared. RESULTS: In total, 172 CDDS alerts were generated for 80 patients. Indecisive responses by ChatGPT to alerts were excluded. For alerts presented without patient variables, ChatGPT provided "correct and identical" responses to 19.9%, "correct and different" responses to 26.7%, and "incorrect responses to 53.4% of the alerts. For alerts including patient variables, ChatGPT provided "correct and identical" responses to 16.7%, "correct and different" responses to 16.0%, and "incorrect responses to 67.3% of the alerts. Accuracy was better for newer drugs such as direct oral anticoagulants. CONCLUSION: The performance of ChatGPT in clinical rule-guided dose interventions in hospitalized patients with renal dysfunction was poor. Based on these results, we conclude that ChatGPT, in its current state, is not appropriate for automatic integration into our EHR to handle CDSS alerts related to renal dysfunction.

Clinical rule-guided pharmacists' intervention in hospitalized patients with hypokalaemia: A time series analysis.

WHAT IS KNOWN AND OBJECTIVE: Physicians' response to moderate and severe hypokalaemia in hospitalized patients is frequently suboptimal, leading to increased risk of cardiac arrhythmias and sudden death. While actively alerting physicians on all critical care values using telephone or electronic pop-ups can improve response, it can also lead to alert fatigue and frustration due to non-specific and overdue alerts. Therefore, a new method was tested. A clinical rule built into a clinical decision support system (CDSS) generated alerts for patients with a serum potassium level (SPL) <2.9 mmol/L without a prescription for potassium supplementation. If the alert was deemed clinically relevant, a pharmacist contacted the physician. The aim of this study was to evaluate the impact of the clinical rule-guided pharmacists' intervention compared to showing passive alerts in the electronic health records on outcome in patients who developed hypokalaemia (<2.9 mmol/L) during hospitalization. METHODS: A before (2007-2009) and after (2010-2017) study with time series design was performed. Pre-intervention, physicians were shown passive alerts for hypokalaemia in the electronic health records. During the intervention period, in addition to these passive alerts, a pharmacist provided the physician with a specific advice on patients with untreated hypokalaemia, guided by the generated alerts. Unique patients >18 years with SPL <2.9 mmol/L measured at least 24 hours after hospitalization in whom no potassium supplementation was initiated within 4 hours after measurement and normalization of SPL was not achieved within these 4 hours were included. Haemodialysis patients were excluded. The percentage of hypokalaemic patients with a subsequent prescription for potassium supplementation, time to subsequent potassium supplementation prescription, the percentage of patients who achieved normokalaemia (SPL ≥ 3.0 mmol/L), time to achieve normokalaemia and total duration of hospitalization were compared. RESULTS AND DISCUSSION: A total of 693 patients were included, of whom 278 participated in the intervention phase. The percentage of patients prescribed supplementation as well as time to prescription improved from 76.0% in 31.1 hours to 92.0% in 11.3 hours (P < .01). Time to achieve SPL ≥3.0 mmol/L improved, P < .009. No changes, however, were observed in the percentage of patients who achieved normokalaemia or time to reach normokalaemia, 87.5% in 65.2 hours pre-intervention compared to 90.2% (P = .69) in 64.0 hours (P = .71) in the intervention group. A non-significant decrease of 8.2 days was observed in the duration of hospitalization: 25.4 compared to 17.2 days (P = .29). WHAT IS NEW AND CONCLUSION: Combining CDSS alerting with a pharmacist evaluation is an effective method to improve response rate, time to supplementation and time to initial improvement, defined as SPL ≥3.0 mmol/L. However, it showed no significant effect on the percentage of patients achieving normokalaemia, time to normokalaemia or hospitalization. The discrepancy between rapid supplementation and improvement on the one hand and failure to improve time to normokalaemia on the other warrants further study.

The use of an electronic clinical rule to discontinue chronically used benzodiazepines and related Z drugs.

PURPOSE: The chronic use of benzodiazepines and benzodiazepine-related drugs (BZ/Z) in older people is common and not without risks. The objective of this study was to evaluate whether the implementation of a clinical rule promotes the discontinuation of chronically used BZ/Z for insomnia. METHODS: A clinical rule, generating an alert in case of chronic BZ/Z use, was created and applied to the nursing home (NH) setting. The clinical rule was a one-off intervention, and alerts did not occur over time. Reports of the generated alerts were digitally sent to NH physicians with the advice to phase out and eventually stop the BZ/Z. In cases where the advice was adopted, a follow-up period of 4 months on the use of BZ/Z was taken into account in order to determine whether the clinical rule alert led to a successful discontinuation of BZ/Z. RESULTS: In all, 808 NH patients were screened. In 161 (19.1%) of the patients, BZ/Z use resulted in a clinical rule alert. From these, the advice to phase out and stop the BZ/Z was adopted for 27 patients (16.8%). Reasons for not following the advice consisted of an unsuccessful attempt in the past (38 patients), patients family and/or patient resistance (37 patients), the non-continuous use of BZ/Z (32 patients) and indication still present (27 patients). Of the 12 NH physicians, seven adopted the advice. CONCLUSIONS: The success rate of a clinical rule for discontinuation of chronically used BZ/Z for insomnia was low, as reported in the present study. Actions should be taken to help caregivers, patients and family members understand the importance of limiting BZ/Z use to achieve higher discontinuation rates.

A study on several critical problems on arrhythmia detection using varying-dimensional electrocardiography.

Objective. This work tries to provide answers to several critical questions on varying-dimensional electrocardiography (ECG) raised by the PhysioNet/Computing in Cardiology Challenge 2021 (CinC2021): can subsets of the standard 12 leads provide models with adequate information to give comparable performances for classifying ECG abnormalities? Can models be designed to be effective enough to classify a broad range of ECG abnormalities?Approach. To tackle these problems, we (challenge team name 'Revenger') propose several novel architectures within the framework of convolutional recurrent neural networks. These deep learning models are proven effective, and moreover, they provide comparable performances on reduced-lead ECGs, even in the extreme case of 2-lead ECGs. In addition, we propose a 'lead-wise' mechanism to facilitate parameter reuse of ECG neural network models. This mechanism largely reduces model sizes while keeping comparable performances. To further augment model performances on specific ECG abnormalities and to improve interpretability, we manually design auxiliary detectors based on clinical diagnostic rules.Main Results. In the post-challenge session, our approach achieved a challenge score of 0.38, 0.40, 0.41, 0.40, 0.35 on the 12, 6, 4, 3, 2-lead subsets respectively on the CinC2021 hidden test set.Significance. The proposed approach gives positive answers to the critical questions CinC2021 raises and lays a solid foundation for further research in the future on these topics.

Improved pre-test likelihood estimation of coronary artery disease using phonocardiography.

Aims: Current early risk stratification of coronary artery disease (CAD) consists of pre-test probability scoring such as the 2019 ESC guidelines on chronic coronary syndromes (ESC2019), which has low specificity and thus rule-out capacity. A newer clinical risk factor model (risk factor-weighted clinical likelihood, RF-CL) showed significantly improved rule-out capacity over the ESC2019 model. The aim of the current study was to investigate if the addition of acoustic features to the RF-CL model could improve the rule-out potential of the best performing clinical risk factor models. Methods and results: Four studies with heart sound recordings from 2222 patients were pooled and distributed into two data sets: training and test. From a feature bank of 40 acoustic features, a forward-selection technique was used to select three features that were added to the RF-CL model. Using a cutoff of 5% predicted risk of CAD, the developed acoustic-weighted clinical likelihood (A-CL) model showed significantly (P < 0.05) higher specificity of 48.6% than the RF-CL model (specificity of 41.5%) and ESC 2019 model (specificity of 6.9%) while having the same sensitivity of 84.9% as the RF-CL model. Area under the curve of the receiver operating characteristic for the three models was 72.5% for ESC2019, 76.7% for RF-CL, and 79.5% for A-CL. Conclusion: The proposed A-CL model offers significantly improved rule-out capacity over the ESC2019 model and showed better overall performance than the RF-CL model. The addition of acoustic features to the RF-CL model was shown to significantly improve early risk stratification of symptomatic patients suspected of having stable CAD.

A clinical rule for the prediction of meningitis in HIV patients in the era of combination antiretroviral therapy.

BACKGROUND: The diagnosis of meningitis in HIV patients is challenging due to altered immune responses. Diagnostic scoring systems were recently proposed for use in research settings to help prompt and easy differential diagnosis. The objective of this study was to create a clinical prediction rule (CPR) for meningitis in HIV-infected patients and to address the enigma of differentiating bacterial (BM), TB (TBM) and cryptococcal (CCM) meningitis based on clinical features alone, which may be enhanced by easy-to-obtain laboratory testing. METHODS: We retrospectively enrolled 352 HIV patients presenting with neurological manifestations suggesting meningitis over the last 18 y (2000-2018). Relevant clinical and laboratory information were retrieved from inpatient records. The features independently predicting meningitis or its different types in microbiologically proven meningitis cases were modelled by multivariate logistic regression to create a CPR in an exploratory data set. The performance of the meningitis diagnostic score was assessed and validated in a subset of retrospective data. RESULTS: AIDS clinical stage, injecting drug use, jaundice and cryptococcal antigen seropositivity were equally important as classic meningitic symptoms in predicting meningitis. Arthralgia and elevated cerebrospinal fluid Lactate dehydrogenase (LDH) were strong predictors of BM. Patients with cryptococcal antigenemia had 25 times the probability of having CCM, whereas neurological deficits were highly suggestive of TBM. CONCLUSION: The proposed CPRs have good diagnostic potential and would support decision-making in resource-poor settings.

Pre-test probability of urinary tract infection in dogs with clinical signs of lower urinary tract disease.

Clinical signs of lower urinary tract disease in dogs are characteristic but non-specific for infection. It has been hypothesized that age, sex and neuter status influences the prevalence of urinary tract infection (UTI), but the predictive value of the combined clinical presentation has not been explored in dogs. The aim of the study was to assess clinical predictors (sex/neuter status, age, dysuria/stranguria, pollakiuria, macroscopic hematuria, malodorous urine and history of recurrent UTI) for bacterial cystitis, and to develop a clinical decision rule. Data was retrieved from medical records (retrospective cases) or from standardized recording sheets (prospective cases). Bacterial cystitis was defined as significant bacteriuria on quantitative bacterial culture in dogs with compatible clinical signs of urinary tract disease. Dogs of any breed, sex and age were included. A total of 1727 microbiology records were screened and 424 samples were included in the analysis. Bacterial cystitis was confirmed in 46% of the cases. Four variables predicted bacterial cystitis: sex/neuter status, age, pollakiuria and hematuria. A score was designated to each variable and a clinical rule was constructed. This rule attained an AUC of 0.75 and had sensitivity of 83% and specificity of 55% at its optimal cut-off (score ≥2.0). A score cut-off of ≥3.0 had a positive predictive value of 70%. Several factors predicted bacterial cystitis, but the clinical rule had only modest predictive value. Other variables or point-of-care test results should be included in future research to optimize overall precision.

A Clinical Rule for Preoperative Prediction of BRAF Mutation Status in Craniopharyngiomas.

BACKGROUND: Papillary craniopharyngiomas are characterized by BRAFV600E mutations. Targeted therapy can elicit a dramatic radiographic regression of these tumors. Therefore, prediction of BRAF mutation status before definitive surgery could enable neoadjuvant treatment strategies. OBJECTIVE: To establish preoperative prediction criteria to identify patients with a BRAF mutant craniopharyngioma. METHODS: Sixty-four patients with craniopharyngioma were included in this study. We determined BRAF mutation status by targeted sequencing. After scoring interobserver variability between presurgical clinical data and radiographic features, we established a diagnostic rule for BRAF mutation in our discovery cohort. We then validated the rule in an independent cohort. RESULTS: The BRAFV600E mutation was detected in 12 of 42 patients in the discovery cohort. There were no patients under age 18 with BRAF mutation. Calcification was rare in tumors with BRAF mutation (P < .001), and 92% of them were supradiaphragmatic in location. Combining these 3 features-older than 18 years, absence of calcification, and supradiaphragmatic tumor location-we established a rule for predicting BRAF mutation. In cases where all 3 criteria were fulfilled, the sensitivity and specificity for the presence of BRAF mutation were 83% and 93%, respectively. In the validation cohort (n = 22), the sensitivity was 100% and specificity was 89%. CONCLUSION: We propose predictive criteria for a BRAF mutation in craniopharyngioma using preoperative clinical and radiographic data. This rule may be useful in identifying patients who could potentially benefit from neoadjuvant BRAFV600E-targeted systemic therapies.