Harnessing the Power of Generative AI for Clinical Summaries: Perspectives From Emergency Physicians.

STUDY OBJECTIVE: The workload of clinical documentation contributes to health care costs and professional burnout. The advent of generative artificial intelligence language models presents a promising solution. The perspective of clinicians may contribute to effective and responsible implementation of such tools. This study sought to evaluate 3 uses for generative artificial intelligence for clinical documentation in pediatric emergency medicine, measuring time savings, effort reduction, and physician attitudes and identifying potential risks and barriers. METHODS: This mixed-methods study was performed with 10 pediatric emergency medicine attending physicians from a single pediatric emergency department. Participants were asked to write a supervisory note for 4 clinical scenarios, with varying levels of complexity, twice without any assistance and twice with the assistance of ChatGPT Version 4.0. Participants evaluated 2 additional ChatGPT-generated clinical summaries: a structured handoff and a visit summary for a family written at an 8th grade reading level. Finally, a semistructured interview was performed to assess physicians' perspective on the use of ChatGPT in pediatric emergency medicine. Main outcomes and measures included between subjects' comparisons of the effort and time taken to complete the supervisory note with and without ChatGPT assistance. Effort was measured using a self-reported Likert scale of 0 to 10. Physicians' scoring of and attitude toward the ChatGPT-generated summaries were measured using a 0 to 10 Likert scale and open-ended questions. Summaries were scored for completeness, accuracy, efficiency, readability, and overall satisfaction. A thematic analysis was performed to analyze the content of the open-ended questions and to identify key themes. RESULTS: ChatGPT yielded a 40% reduction in time and a 33% decrease in effort for supervisory notes in intricate cases, with no discernible effect on simpler notes. ChatGPT-generated summaries for structured handoffs and family letters were highly rated, ranging from 7.0 to 9.0 out of 10, and most participants favored their inclusion in clinical practice. However, there were several critical reservations, out of which a set of general recommendations for applying ChatGPT to clinical summaries was formulated. CONCLUSION: Pediatric emergency medicine attendings in our study perceived that ChatGPT can deliver high-quality summaries while saving time and effort in many scenarios, but not all.

Jerusalem's CoVID-19 Experience-The Effect of Ethnicity on Disease Prevalence and Adherence to Testing.

BACKGROUND: The management of the SARS-CoV-2 pandemic depends amongst other factors on disease prevalence in the general population. The gap between the true rate of infection and the detected rate of infection may vary, especially between sub-groups of the population. Identifying subpopulations with high rates of undetected infection can guide authorities to direct resource distribution in order to improve health equity. METHODS: A cross-sectional epidemiological survey was conducted between April and July 2021 in the Pediatric Emergency Department of the Shaare Zedek Medical Center, Jerusalem, Israel. We compared three categories: unconfirmed disease (UD), positive serology test result with no history of positive PCR; confirmed disease (CD), history of a positive PCR test result, regardless of serology test result; and no disease (ND), negative serology and no history of PCR. These categories were applied to local prevailing subpopulations: ultra-orthodox Jews (UO), National Religious Jews (NRJ), secular Jews (SJ), and Muslim Arabs (MA). RESULTS: Comparing the different subpopulations groups, MAs and UOs had the greatest rate of confirmed or unconfirmed disease. MA had the highest rate of UD and UO had the highest rate of CD. UD significantly correlated with ethnicity, with a low prevalence in NRJ and SJ. UD was also associated with larger family size and housing density defined as family size per number of rooms. CONCLUSION: This study highlights the effect of ethnicity on disease burden. These findings should serve to heighten awareness to disease burden in weaker populations and direct a suitable prevention program to each subpopulation's needs. Early awareness and possible intervention may lower morbidity and mortality.

Calculating Relative Lung Perfusion Using Fluoroscopic Sequences and Image Analysis: The Fluoroscopic Flow Calculator.

BACKGROUND: Maldistribution of pulmonary blood flow in patients with congenital heart disease impacts exertional performance and pulmonary artery growth. Currently, measurement of relative pulmonary perfusion can only be performed outside the catheterization laboratory. We sought to develop a tool for measuring relative lung perfusion using readily available fluoroscopy sequences. METHODS: A retrospective cohort study was conducted on patients with conotruncal anomalies who underwent lung perfusion scans and subsequent cardiac catheterizations between 2011 and 2022. Inclusion criteria were nonselective angiogram of pulmonary vasculature, oblique angulation ≤20°, and an adequate view of both lung fields. A method was developed and implemented in 3D Slicer's SlicerHeart extension to calculate the amount of contrast that entered each lung field from the start of contrast injection and until the onset of levophase. The predicted perfusion distribution was compared with the measured distribution of pulmonary blood flow and evaluated for correlation, accuracy, and bias. RESULTS: In total, 32% (79/249) of screened studies met the inclusion criteria. A strong correlation between the predicted flow split and the measured flow split was found (R2=0.83; P<0.001). The median absolute error was 6%, and 72% of predictions were within 10% of the true value. Bias was not systematically worse at either extreme of the flow distribution. The prediction was found to be more accurate for either smaller and younger patients (age 0-2 years), for right ventricle injections, or when less cranial angulations were used (≤20°). In these cases (n=40), the prediction achieved R2=0.87, median absolute error of 5.5%, and 78% of predictions were within 10% of the true flow. CONCLUSIONS: The current study demonstrates the feasibility of a novel method for measuring relative lung perfusion using conventional angiograms. Real-time measurement of lung perfusion at the catheterization laboratory has the potential to reduce unnecessary testing, associated costs, and radiation exposure. Further optimization and validation is warranted.

The value of parental medical records for the prediction of diabetes and cardiovascular disease: a novel method for generating and incorporating family histories.

OBJECTIVE: To determine whether data-driven family histories (DDFH) derived from linked EHRs of patients and their parents can improve prediction of patients' 10-year risk of diabetes and atherosclerotic cardiovascular disease (ASCVD). MATERIALS AND METHODS: A retrospective cohort study using data from Israel's largest healthcare organization. A random sample of 200 000 subjects aged 40-60 years on the index date (January 1, 2010) was included. Subjects with insufficient history (<1 year) or insufficient follow-up (<10 years) were excluded. Two separate XGBoost models were developed-1 for diabetes and 1 for ASCVD-to predict the 10-year risk for each outcome based on data available prior to the index date of January 1, 2010. RESULTS: Overall, the study included 110 734 subject-father-mother triplets. There were 22 153 cases of diabetes (20%) and 11 715 cases of ASCVD (10.6%). The addition of parental information significantly improved prediction of diabetes risk (P < .001), but not ASCVD risk. For both outcomes, maternal medical history was more predictive than paternal medical history. A binary variable summarizing parental disease state delivered similar predictive results to the full parental EHR. DISCUSSION: The increasing availability of EHRs for multiple family generations makes DDFH possible and can assist in delivering more personalized and precise medicine to patients. Consent frameworks must be established to enable sharing of information across generations, and the results suggest that sharing the full records may not be necessary. CONCLUSION: DDFH can address limitations of patient self-reported family history, and it improves clinical predictions for some conditions, but not for all, and particularly among younger adults.

The Amplatzer duct occluder (ADOII) and Piccolo devices for patent ductus arteriosus closure: a large single institution series.

Purpose: Evaluate Piccolo and ADOII devices for transcatheter patent ductus arteriosus (PDA) closure. Piccolo has smaller retention discs reducing risk of flow disturbance but residual leak and embolization risk may increase. Methods: Retrospective review of all patients undergoing PDA closure with an Amplatzer device between January 2008 and April 2022 in our institution. Data from the procedure and 6 months follow-up were collected. Results: 762 patients, median age 2.6 years (range 0-46.7) years and median weight 13 kg (range 3.5-92) were referred for PDA closure. Overall, 758 (99.5%) had successful implantation: 296 (38.8%) with ADOII, 418 (54.8%) with Piccolo, and 44 (5.8%) with AVPII. The ADOII patients were smaller than the Piccolo patients (15.8 vs. 20.5 kg, p < 0.001) and with larger PDA diameters (2.3 vs. 1.9 mm, p < 0.001). Mean device diameter was similar for both groups. Closure rate at follow-up was similar for all devices ADOII 295/296 (99.6%), Piccolo 417/418 (99.7%), and AVPII 44/44 (100%). Four intraprocedural embolizations occurred during the study time period: two ADOII and two Piccolo. Following retrieval the PDA was closed with an AVPII in two cases, ADOI in one case and with surgery in the fourth case. Mild stenosis of the left pulmonary artery (LPA) occurred in three patients with ADOII devices (1%) and one patient with Piccolo device (0.2%). Severe LPA stenosis occurred in one patient with ADOII (0.3%) and one with AVPII device (2.2%). Conclusions: ADOII and Piccolo are safe and effective for PDA closure with a tendency to less LPA stenosis with Piccolo. There were no cases of aortic coarctation related to a PDA device in this study.

An efficient landmark model for prediction of suicide attempts in multiple clinical settings.

Growing evidence has shown that applying machine learning models to large clinical data sources may exceed clinician performance in suicide risk stratification. However, many existing prediction models either suffer from "temporal bias" (a bias that stems from using case-control sampling) or require training on all available patient visit data. Here, we adopt a "landmark model" framework that aligns with clinical practice for prediction of suicide-related behaviors (SRBs) using a large electronic health record database. Using the landmark approach, we developed models for SRB prediction (regularized Cox regression and random survival forest) that establish a time-point (e.g., clinical visit) from which predictions are made over user-specified prediction windows using historical information up to that point. We applied this approach to cohorts from three clinical settings: general outpatient, psychiatric emergency department, and psychiatric inpatients, for varying prediction windows and lengths of historical data. Models achieved high discriminative performance (area under the Receiver Operating Characteristic curve 0.74-0.93 for the Cox model) across different prediction windows and settings, even with relatively short periods of historical data. In short, we developed accurate, dynamic SRB risk prediction models with the landmark approach that reduce bias and enhance the reliability and portability of suicide risk prediction models.

Improving risk prediction for target subpopulations: Predicting suicidal behaviors among multiple sclerosis patients.

Several recent studies have applied machine learning techniques to develop risk algorithms that predict subsequent suicidal behavior based on electronic health record data. In this study we used a retrospective cohort study design to test whether developing more tailored predictive models-within specific subpopulations of patients-would improve predictive accuracy. A retrospective cohort of 15,117 patients diagnosed with multiple sclerosis (MS), a diagnosis associated with increased risk of suicidal behavior, was used. The cohort was randomly divided into equal sized training and validation sets. Overall, suicidal behavior was identified among 191 (1.3%) of the patients with MS. A Naïve Bayes Classifier model was trained on the training set to predict future suicidal behavior. With 90% specificity, the model detected 37% of subjects who later demonstrated suicidal behavior, on average 4.6 years before the first suicide attempt. The performance of a model trained only on MS patients was better at predicting suicide in MS patients than that a model trained on a general patient sample of a similar size (AUC of 0.77 vs. 0.66). Unique risk factors for suicidal behavior among patients with MS included pain-related codes, gastroenteritis and colitis, and history of smoking. Future studies are needed to further test the value of developing population-specific risk models.

The risk of serious bacterial infections among young ex-premature infants with fever.

Background and Objectives: To determine the rate of serious-bacterial-infections (SBI) in young ex-premature infants with fever, and to develop a risk-stratification algorithm for these patients. Methods: A retrospective cohort study including all infants who presented to the pediatric emergency department (ED) of a tertiary-care university-hospital between 2010 and 2020 with fever (≥38°C), were born prematurely (<37-weeks), had post-conception age of <52-weeks, and had available blood, urine, or CSF cultures. The rates of SBI by age-of-birth and age-at-visit were calculated and compared to a cohort of matched full-term controls. Results: The study included a total of 290 ex-premature cases and 290 full-term controls. There were 11 cases (3.8%) with an invasive bacterial infection (IBI) of either bacteremia, meningitis or both and only six controls (2.1%) with IBI (p = 0.32). Over 28-days chronologic-age, there were 10 (3.6%) IBIs among cases and no IBIs among the controls (p = 0.02). There were eight (3%) cases and three (1%) controls with IBI who were well-appearing on physical examination (p = 0.19). All eight well-appearing ex-premature infants were under 60-days adjusted-age, seven of whom (88%) were also under 28-days adjusted-age. There were 28 (10.6%) cases and 34 (12%) controls with urinary tract infection (UTI) (p = 0.5). Among cases under 60-days adjusted-age, urinalysis was not reliable to exclude UTI (50% negative). Conclusions: Well-appearing ex-preterm infants have a significant risk for IBI until the adjusted age of 28-days and for UTI until the adjusted age of 60-days. Further studies are needed to evaluate the approach to fever in this unique population.

Transthoracic intracardiac lines-A double edged sword.

OBJECTIVE: Routine use of central venous access is needed in children undergoing open heart surgery for pressure monitoring and inotrope infusion. We sought to evaluate the efficiency and safety of routine use of transthoracic intracardiac lines (ICLs) in patients undergoing cardiac surgery and to compare them to patients who have been previously treated with traditional central venous lines (non-ICLs). METHODS: Retrospective review of charts of all patients who underwent cardiac surgery and had an ICL inserted in the operating room. Case control matching was done with similar patient in which ICL was not inserted. Patients characteristics, diagnosis, operative, and intensive care data were collected for each patient and analyzed. RESULTS: A total number of 376 patient records were reviewed (198 ICL patients and 178 non-ICL patients). Umbilical line and non-ICL durations were longer in the non-ICL group. ICL duration was the longest of all lines, averaging 12.87 ± 10.82 days. The necessity for multiple line insertions (˃2 insertions) was significantly higher in the non-ICL group, with a relative risk ratio of 3.24 (95% confidence interval: 1.617-6.428). There was no statistical difference of infections rate and line complications between the two groups. CONCLUSION: ICLs are safe in infants undergoing cardiac surgery and can be kept in place for a long period of time with a low rate of line complications and infection. Routine use of ICLs reduces the number of central venous catheter placement in this complex patient population.

"Feed and Swaddle" method of Infants Undergoing Head CT for minor head injury in the pediatric emergency department - A comparative case review.

INTRODUCTION: Brain imaging for suspected significant head injuries in pediatric emergency departments is an important and time-sensitive procedure. The use of sedation to successfully complete imaging can be limited due to young age and other injury related factors. Using a non-pharmacological method using feeding and swaddling can be used. This may obviate the need for sedation but can be time consuming. METHODS: A retrospective study of all children undergoing brain imaging for head injury during the years 2016-2021. Use of sedation, time to completion and imaging findings were compared. RESULTS: Of 281 children requiring brain imaging, 268 (95.4%) were completed using the feed and swaddle method. Time to imaging completion was similar between sedation and feed and swaddle groups (85.5 min vs. 86 min). Abnormal findings on imaging were found in 186 (69.4%) in the feed and swaddle group and in 10 (77%) of the sedation group. No adverse events were seen in the sedation group. CONCLUSION: Using the feed and swaddle method can help lower the need for sedation in the under 1 year age group with a successful and timely completion of brain imaging.

CMR Imaging 6 Months After Myocarditis Associated with the BNT162b2 mRNA COVID-19 Vaccine.

Temporal association between BNT162b2 mRNA COVID-19 vaccine and myocarditis (PCVM) has been reported. We herein present early and 6-month clinical follow-up and cardiac magnetic resonance imaging (CMR) of patients with PVCM. A retrospective collection of data from 15 patients with PCVM and abnormal CMR was performed. Clinical manifestation, laboratory data, hospitalizations, treatment protocols, and imaging studies were collected early (up to 2 months) and later. In nine patients, an additional CMR evaluation was performed 6 months after diagnosis. PCVM was diagnosed in 15 patients, mean age 17 ± 1 (median 17.2, range 14.9-19 years) years, predominantly in males. Mean time from vaccination to onset of symptoms was 4.4 ± 6.7 (median 3, range 0-28) days. All patients had CMR post diagnosis at 4 ± 3 (median 3, range 1-9) weeks, 4/5 patients had hyper enhancement on the T2 sequences representing edemaQuery, and 12 pathological Late glandolinium enhancement. A repeat scan performed after 5-6 months was positive for scar formation in 7/9 patients. PCVM is a rare complication, affecting predominantly males and appearing usually within the first week after administration of the second dose of the vaccine. It usually is a mild disease, with clinical resolution with anti-inflammatory treatment. Late CMR follow up demonstrated resolution of the edema in all patients, while some had evidence of residual myocardial scarring.

Prediction of Suicide Attempts Using Clinician Assessment, Patient Self-report, and Electronic Health Records.

Importance: Half of the people who die by suicide make a health care visit within 1 month of their death. However, clinicians lack the tools to identify these patients. Objective: To predict suicide attempts within 1 and 6 months of presentation at an emergency department (ED) for psychiatric problems. Design, Setting, and Participants: This prognostic study assessed the 1-month and 6-month risk of suicide attempts among 1818 patients presenting to an ED between February 4, 2015, and March 13, 2017, with psychiatric problems. Data analysis was performed from May 1, 2020, to November 19, 2021. Main Outcomes and Measures: Suicide attempts 1 and 6 months after presentation to the ED were defined by combining data from electronic health records (EHRs) with patient 1-month (n = 1102) and 6-month (n = 1220) follow-up surveys. Ensemble machine learning was used to develop predictive models and a risk score for suicide. Results: A total of 1818 patients participated in this study (1016 men [55.9%]; median age, 33 years [IQR, 24-46 years]; 266 Hispanic patients [14.6%]; 1221 non-Hispanic White patients [67.2%], 142 non-Hispanic Black patients [7.8%], 64 non-Hispanic Asian patients [3.5%], and 125 non-Hispanic patients of other race and ethnicity [6.9%]). A total of 137 of 1102 patients (12.9%; weighted prevalence) attempted suicide within 1 month, and a total of 268 of 1220 patients (22.0%; weighted prevalence) attempted suicide within 6 months. Clinicians' assessment alone was little better than chance at predicting suicide attempts, with externally validated area under the receiver operating characteristic curve (AUC) of 0.67 for the 1-month model and 0.60 for the 6-month model. Prediction accuracy was slightly higher for models based on EHR data (1-month model: AUC, 0.71; 6 month model: AUC, 0.65) and was best using patient self-reports (1-month model: AUC, 0.76; 6-month model: AUC, 0.77), especially when patient self-reports were combined with EHR and/or clinician data (1-month model: AUC, 0.77; and 6 month model: AUC, 0.79). A model that used only 20 patient self-report questions and an EHR-based risk score performed similarly well (1-month model: AUC, 0.77; 6 month model: AUC, 0.78). In the best 1-month model, 30.7% (positive predicted value) of the patients classified as having highest risk (top 25% of the sample) made a suicide attempt within 1 month of their ED visit, accounting for 64.8% (sensitivity) of all 1-month attempts. In the best 6-month model, 46.0% (positive predicted value) of the patients classified at highest risk made a suicide attempt within 6 months of their ED visit, accounting for 50.2% (sensitivity) of all 6-month attempts. Conclusions and Relevance: This prognostic study suggests that the ability to identify patients at high risk of suicide attempt after an ED visit for psychiatric problems improved using a combination of patient self-reports and EHR data.

Predictive structured-unstructured interactions in EHR models: A case study of suicide prediction.

Clinical risk prediction models powered by electronic health records (EHRs) are becoming increasingly widespread in clinical practice. With suicide-related mortality rates rising in recent years, it is becoming increasingly urgent to understand, predict, and prevent suicidal behavior. Here, we compare the predictive value of structured and unstructured EHR data for predicting suicide risk. We find that Naive Bayes Classifier (NBC) and Random Forest (RF) models trained on structured EHR data perform better than those based on unstructured EHR data. An NBC model trained on both structured and unstructured data yields similar performance (AUC = 0.743) to an NBC model trained on structured data alone (0.742, p = 0.668), while an RF model trained on both data types yields significantly better results (AUC = 0.903) than an RF model trained on structured data alone (0.887, p < 0.001), likely due to the RF model's ability to capture interactions between the two data types. To investigate these interactions, we propose and implement a general framework for identifying specific structured-unstructured feature pairs whose interactions differ between case and non-case cohorts, and thus have the potential to improve predictive performance and increase understanding of clinical risk. We find that such feature pairs tend to capture heterogeneous pairs of general concepts, rather than homogeneous pairs of specific concepts. These findings and this framework can be used to improve current and future EHR-based clinical modeling efforts.

COVID-19 Vaccine Is Effective in Inflammatory Bowel Disease Patients and Is Not Associated With Disease Exacerbation.

BACKGROUND & AIMS: Studies have shown decreased response to coronavirus disease 2019 (COVID-19) vaccinations in some populations. In addition, it is possible that vaccine-triggered immune activation could trigger immune dysregulation and thus exacerbate inflammatory bowel diseases (IBD). In this population-based study we used the epi-Israeli IBD Research Nucleus validated cohort to explore the effectiveness of COVID-19 vaccination in IBD and to assess its effect on disease outcomes. METHODS: We included all IBD patients insured in 2 of the 4 Israeli health maintenance organizations, covering 35% of the population. Patients receiving 2 Pfizer-BioNTech BNT162b2 vaccine doses between December 2020 and June 2021 were individually matched to non-IBD controls. To assess IBD outcomes, we matched vaccinated to unvaccinated IBD patients, and response was analyzed per medical treatment. RESULTS: In total, 12,109 IBD patients received 2 vaccine doses, of whom 4946 were matched to non-IBD controls (mean age, 51 ± 16 years; median follow-up, 22 weeks; interquartile range, 4-24). Fifteen patients in each group (0.3%) developed COVID-19 after vaccination (odds ratio, 1; 95% confidence interval, 0.49-2.05; P = 1.0). Patients on tumor necrosis factor (TNF) inhibitors and/or corticosteroids did not have a higher incidence of infection. To explore IBD outcomes, 707 vaccinated IBD patients were compared with unvaccinated IBD patients by stringent matching (median follow-up, 14 weeks; interquartile range, 2.3-20.4). The risk of exacerbation was 29% in the vaccinated patients compared with 26% in unvaccinated patients (P = .3). CONCLUSIONS: COVID-19 vaccine effectiveness in IBD patients is comparable with that in non-IBD controls and is not influenced by treatment with TNF inhibitors or corticosteroids. The IBD exacerbation rate did not differ between vaccinated and unvaccinated patients.

Prediction across healthcare settings: a case study in predicting emergency department disposition.

Several approaches exist today for developing predictive models across multiple clinical sites, yet there is a lack of comparative data on their performance, especially within the context of EHR-based prediction models. We set out to provide a framework for prediction across healthcare settings. As a case study, we examined an ED disposition prediction model across three geographically and demographically diverse sites. We conducted a 1-year retrospective study, including all visits in which the outcome was either discharge-to-home or hospitalization. Four modeling approaches were compared: a ready-made model trained at one site and validated at other sites, a centralized uniform model incorporating data from all sites, multiple site-specific models, and a hybrid approach of a ready-made model re-calibrated using site-specific data. Predictions were performed using XGBoost. The study included 288,962 visits with an overall admission rate of 16.8% (7.9-26.9%). Some risk factors for admission were prominent across all sites (e.g., high-acuity triage emergency severity index score, high prior admissions rate), while others were prominent at only some sites (multiple lab tests ordered at the pediatric sites, early use of ECG at the adult site). The XGBoost model achieved its best performance using the uniform and site-specific approaches (AUC = 0.9-0.93), followed by the calibrated-model approach (AUC = 0.87-0.92), and the ready-made approach (AUC = 0.62-0.85). Our results show that site-specific customization is a key driver of predictive model performance.

Temporally informed random forests for suicide risk prediction.

OBJECTIVE: Suicide is one of the leading causes of death worldwide, yet clinicians find it difficult to reliably identify individuals at high risk for suicide. Algorithmic approaches for suicide risk detection have been developed in recent years, mostly based on data from electronic health records (EHRs). Significant room for improvement remains in the way these models take advantage of temporal information to improve predictions. MATERIALS AND METHODS: We propose a temporally enhanced variant of the random forest (RF) model-Omni-Temporal Balanced Random Forests (OT-BRFs)-that incorporates temporal information in every tree within the forest. We develop and validate this model using longitudinal EHRs and clinician notes from the Mass General Brigham Health System recorded between 1998 and 2018, and compare its performance to a baseline Naive Bayes Classifier and 2 standard versions of balanced RFs. RESULTS: Temporal variables were found to be associated with suicide risk: Elevated suicide risk was observed in individuals with a higher total number of visits as well as those with a low rate of visits over time, while lower suicide risk was observed in individuals with a longer period of EHR coverage. RF models were more accurate than Naive Bayesian classifiers at predicting suicide risk in advance (area under the receiver operating curve = 0.824 vs. 0.754, respectively). The proposed OT-BRF model performed best among all RF approaches, yielding a sensitivity of 0.339 at 95% specificity, compared to 0.290 and 0.286 for the other 2 RF models. Temporal variables were assigned high importance by the models that incorporated them. DISCUSSION: We demonstrate that temporal variables have an important role to play in suicide risk detection and that requiring their inclusion in all RF trees leads to increased predictive performance. Integrating temporal information into risk prediction models helps the models interpret patient data in temporal context, improving predictive performance.

SARS-CoV-2 antibodies started to decline just four months after COVID-19 infection in a paediatric population.

AIM: We evaluated the prevalence of paediatric severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections using antibody testing and characterised antibody titres by time from exposure. METHODS: This was a single-centre, prospective, cross-sectional cohort study. Patients under 18 years old were eligible to participate if they attended the paediatric emergency department at the tertiary Shaare Zedek Medical Center, Jerusalem, Israel, from 18 October 2020 to 12 January 2021 and required blood tests or intravenous access. SARS-CoV-2 seropositivity and antibody levels were tested by a dual-assay model. RESULTS: The study comprised 1138 patients (56% male) with a mean age of 4.4 years (interquartile range 1.3-11.3). Anti-SARS-CoV-2 antibodies were found in 10% of the patients. Seropositivity increased with age and 41% of seropositive patients had no known exposure. Children under 6 years of age had higher initial antibody levels than older children, followed by a steeper decline. The seropositivity rate did not vary during the study, despite schools re-opening. The findings suggest that children's immunity may start falling 4 months after the initial infection. CONCLUSION: Immunity started falling after just 4 months, and re-opening schools did not affect infection rates. These findings could aid decisions about vaccinating paediatric populations and school closures.

Prediction of patient disposition: comparison of computer and human approaches and a proposed synthesis.

OBJECTIVE: To compare the accuracy of computer versus physician predictions of hospitalization and to explore the potential synergies of hybrid physician-computer models. MATERIALS AND METHODS: A single-center prospective observational study in a tertiary pediatric hospital in Boston, Massachusetts, United States. Nine emergency department (ED) attending physicians participated in the study. Physicians predicted the likelihood of admission for patients in the ED whose hospitalization disposition had not yet been decided. In parallel, a random-forest computer model was developed to predict hospitalizations from the ED, based on data available within the first hour of the ED encounter. The model was tested on the same cohort of patients evaluated by the participating physicians. RESULTS: 198 pediatric patients were considered for inclusion. Six patients were excluded due to incomplete or erroneous physician forms. Of the 192 included patients, 54 (28%) were admitted and 138 (72%) were discharged. The positive predictive value for the prediction of admission was 66% for the clinicians, 73% for the computer model, and 86% for a hybrid model combining the two. To predict admission, physicians relied more heavily on the clinical appearance of the patient, while the computer model relied more heavily on technical data-driven features, such as the rate of prior admissions or distance traveled to hospital. DISCUSSION: Computer-generated predictions of patient disposition were more accurate than clinician-generated predictions. A hybrid prediction model improved accuracy over both individual predictions, highlighting the complementary and synergistic effects of both approaches. CONCLUSION: The integration of computer and clinician predictions can yield improved predictive performance.

The effect of C-reactive protein on chest X-ray interpretation: A decision-making experiment among pediatricians.

INTRODUCTION: Clinical decision-making is complex and requires the integration of multiple sources of information. Physicians tend to over-rely on objective measures, despite the lack of supportive evidence in many cases. We sought to test if pediatricians over-rely on C-reactive protein (CRP) results when managing a child with suspected pneumonia. METHODS: A nationwide decision-making experiment was conducted among 337 pediatricians in Israel. Each participant was presented with two detailed vignettes of a child with suspected pneumonia, each with a chest X-ray (CXR) taken from a real-life case of viral pneumonia. Participants were randomly assigned to one of three groups: Controls-where no lab tests were provided, and two intervention groups where the vignettes also noted a high or a low CRP value, in varying orders. Between-participant and within-participant analyses were conducted to study the effect of CRP on CXR interpretation. The three groups were presented with identical medical history, vital signs, findings on physical examination, blood count, and CXR. RESULTS: Three-hundred and one pediatricians (89.3% of those approached) completed the study. Pediatricians were 60%-90% more likely to diagnose viral pneumonia as bacterial when presented with high CRP levels versus low CRP levels, despite the identical clinical data and CXR (62% vs. 39% and 58% vs. 31% of physicians; p = .002). Accordingly, they were 60%-90% more likely to prescribe antibiotics in these cases (86% vs. 53% and 78% vs. 41% of physicians; p < .001). CONCLUSIONS: CRP by itself may modify the way in which pediatricians interpret a CXR, leading to the overprescription of antibiotics.