The Deterioration Risk Index: Developing and Piloting a Machine Learning Algorithm to Reduce Pediatric Inpatient Deterioration.

OBJECTIVES: Develop and deploy a disease cohort-based machine learning algorithm for timely identification of hospitalized pediatric patients at risk for clinical deterioration that outperforms our existing situational awareness program. DESIGN: Retrospective cohort study. SETTING: Nationwide Children's Hospital, a freestanding, quaternary-care, academic children's hospital in Columbus, OH. PATIENTS: All patients admitted to inpatient units participating in the preexisting situational awareness program from October 20, 2015, to December 31, 2019, excluding patients over 18 years old at admission and those with a neonatal ICU stay during their hospitalization. INTERVENTIONS: We developed separate algorithms for cardiac, malignancy, and general cohorts via lasso-regularized logistic regression. Candidate model predictors included vital signs, supplemental oxygen, nursing assessments, early warning scores, diagnoses, lab results, and situational awareness criteria. Model performance was characterized in clinical terms and compared with our previous situational awareness program based on a novel retrospective validation approach. Simulations with frontline staff, prior to clinical implementation, informed user experience and refined interdisciplinary workflows. Model implementation was piloted on cardiology and hospital medicine units in early 2021. MEASUREMENTS AND MAIN RESULTS: The Deterioration Risk Index (DRI) was 2.4 times as sensitive as our existing situational awareness program (sensitivities of 53% and 22%, respectively; p < 0.001) and required 2.3 times fewer alarms per detected event (121 DRI alarms per detected event vs 276 for existing program). Notable improvements were a four-fold sensitivity gain for the cardiac diagnostic cohort (73% vs 18%; p < 0.001) and a three-fold gain (81% vs 27%; p < 0.001) for the malignancy diagnostic cohort. Postimplementation pilot results over 18 months revealed a 77% reduction in deterioration events (three events observed vs 13.1 expected, p = 0.001). CONCLUSIONS: The etiology of pediatric inpatient deterioration requires acknowledgement of the unique pathophysiology among cardiology and oncology patients. Selection and weighting of diverse candidate risk factors via machine learning can produce a more sensitive early warning system for clinical deterioration. Leveraging preexisting situational awareness platforms and accounting for operational impacts of model implementation are key aspects to successful bedside translation.

Patient-centered design criteria for wearable seizure detection devices.

INTRODUCTION: Epilepsy is a common neurological condition. Seizure diary reports and patient- or caregiver-reported seizure counts are often inaccurate and underestimated. Many caregivers express stress and anxiety about the patient with epilepsy having seizures when they are not present. Therefore, a need exists for the ability to recognize and/or detect a seizure in the home setting. However, few studies have inquired on detection device features that are important to patients and their caregivers. METHODS: A survey instrument utilizing a population of patients and caregivers was created to obtain information on the design criteria most desired for patients with epilepsy in regard to wearable devices. RESULTS: One thousand one hundred sixty-eight responses were collected. Respondents thought that sensors for muscle signal (61.4%) and heart rate (58.0%) would be most helpful followed by the O2 sensor (41.4%). There was more interest in these three sensor types than for an accelerometer (25.5%). There was very little interest in a microphone (8.9%), galvanic skin response sensor (8.0%), or a barometer (4.9%). Based on a rating scale of 1-5 with 5 being the most important, respondents felt that "detecting all seizures" (4.73) is the most important device feature followed by "text/email alerts" (4.53), "comfort" (4.46), and "battery life" (4.43) as an equally important group of features. Respondents felt that "not knowing device is for seizures" (2.60) and "multiple uses" (2.57) were equally the least important device features. Average ratings differed significantly across age groups for the following features: button, multiuse, not knowing device is for seizures, alarm, style, and text ability. The p-values were all<0.002. Eighty-two point five percent of respondents [95% confidence interval: 80.0%, 84.7%] were willing to pay more than $100 for a wearable seizure detection device, and 42.8% of respondents [95% confidence interval: 39.8%, 45.9%] were willing to pay more than $200. CONCLUSIONS: Our survey results demonstrated that patients and caregivers have design features that are important to them in regard to a wearable seizure detection device. Overall, the ability to detect all seizures rated highest among respondents which continues to be an unmet need in the community with epilepsy in regard to seizure detection. Additional uses for a wearable were not as important. Based on our results, it is important that an alert (via test and/or email) for events be a portion of the system. A reasonable price point appears to be around $200 to $300. An accelerometer was less important to those surveyed when compared with the use of heart rate, oxygen saturation, or muscle twitches/signals. As further products become developed for use in other health arenas, it will be important to consider patient and caregiver desires in order to meet the need and address the gap in devices that currently exist.

Pilot study using machine learning to improve estimation of physical abuse prevalence.

BACKGROUND: International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes have been shown to underestimate physical abuse prevalence. Machine learning models are capable of efficiently processing a wide variety of data and may provide better estimates of abuse. OBJECTIVE: To achieve proof of concept applying machine learning to identify codes associated with abuse. PARTICIPANTS AND SETTING: Children <5 years, presenting to the emergency department with an injury or abuse-specific ICD-10-CM code and evaluated by the child protection team (CPT) from 2016 to 2020 at a large Midwestern children's hospital. METHODS: The Pediatric Health Information System (PHIS) and the CPT administrative databases were used to identify the study sample and injury and abuse-specific ICD-10-CM codes. Subjects were divided into abused and non-abused groups based on the CPT's evaluation. A LASSO logistic regression model was constructed using ICD-10-CM codes and patient age to identify children likely to be diagnosed by the CPT as abused. Performance was evaluated using repeated cross-validation (CV) and Reciever Operator Characteristic curve. RESULTS: We identified 2028 patients evaluated by the CPT with 512 diagnosed as abused. Using diagnosis codes and patient age, our model was able to accurately identify patients with confirmed PA (mean CV AUC = 0.87). Performance was still weaker for patients without existing ICD codes for abuse (mean CV AUC = 0.81). CONCLUSIONS: We built a model that employs injury ICD-10-CM codes and age to improve accuracy of distinguishing abusive from non-abusive injuries. This pilot modelling endeavor is a steppingstone towards improving population-level estimates of abuse.

Physical abuse re-reporting during the COVID-19 pandemic: The kids are not okay.

BACKGROUND: Child physical abuse (PA) is a significant societal concern with limited research into predictors of re-reports. OBJECTIVE: Our research explores correlations between sociodemographic variables and re-reported PA. Our aim was to characterize populations at higher risk and identify changes in presentation during the COVID-19 pandemic. PARTICIPANTS AND SETTING: This retrospective descriptive study focused on 238 patients with re-reports of PA made by a pediatric hospital from January 2019 through April 2021. METHODS: We analyzed sociodemographic information and details of reports made to child protective services (CPS) obtained from the electronic health record. RESULTS: Females were 2.5 years older than males (mean 11.0 and 8.5 years, respectively) (p < .001, 95%CI 1.21-3.76). Males were more likely to have observable injuries (OR 2.61, p < .001) and a CPS response (OR = 2.70, p = .003). Patients categorized as "Other" races were less likely to have observable injuries (OR = 0.32, p = .006). Presentation changed during the pandemic: a quadrupling of re-reports by behavioral health clinicians caused the percentage of reports made by them to increase significantly (OR = 3.46, p < .001) and the mean age increased by 2.0 years (8.2 years before, 10.2 years during) (p = .009, 95%CI 0.5-3.5), though females remained approximately 2.2 years older than males (p = .003, 95%CI 0.8-3.7). CONCLUSIONS: Males experienced higher rates of re-reported PA and were younger at the time of re-report. Changes to presentation during the pandemic suggest an increase in PA among older children. Future research should further explore differences in sex/race, while current prevention efforts should focus on children receiving behavioral health care.

Prioritizing Hormone Therapy Over Vigabatrin as the First Treatment for Infantile Spasms: A Quality Improvement Initiative.

BACKGROUND AND OBJECTIVES: Infantile spasms (IS) are early childhood seizures with potentially devastating consequences. Standard therapies (adrenocorticotropic hormone [ACTH], high-dose prednisolone, and vigabatrin) are strongly recommended as the first treatment for IS. Although this recommendation comes without preference for one standard therapy over another, early remission rates are higher with hormone therapy (ACTH and high-dose prednisolone) when compared with vigabatrin. Using quality improvement (QI) methodology that included hormone therapy as the first treatment, we sought to increase the percentage of children with new-onset nontuberous sclerosis complex (TSC)-associated IS achieving 3-month electroclinical remission from a mean of 53.8% to ≥70%. METHODS: This was an observational consecutive sample cohort study at a single academic tertiary care hospital that compared a prospective intervention cohort (May 2019-January 2022, N = 57) with a retrospective baseline cohort (November 2015-April 2019, N = 67). Our initiative addressed key drivers such as the routine use of vigabatrin over hormone therapy as first treatment and the common initiation of a second treatment after 14 days for initial nonresponders. We included consecutive children without TSC presenting with new-onset IS diagnosed and treated between ages 2 and 24 months. We displayed our primary outcome and process measures as control charts in which the centerline is the quarterly (previous 3 months) mean based on statistical process control methodology. RESULTS: QI interventions that included the standardization of hormone therapy as the first treatment resulted in higher rates of 3-month remission, rising from 53.8% (baseline cohort) to 75.9% (intervention cohort). Process measure results included an increased rate of children receiving hormone therapy as first treatment (mean, 44.6%-100%) and a decreased number of days to both clinical follow-up after first treatment (mean, of 16.3-12.6 days) and starting a second treatment within 14 days for initial nonresponders (mean, 36.3-17.2 days). DISCUSSION: For children with IS, improved rates of 3-month electroclinical remission can be achieved with QI methodology. Implementation of similar QI initiatives at other centers may likewise improve local remission rates.

Lung nodule CAD software as a second reader: a multicenter study.

RATIONALE AND OBJECTIVES: The purpose of this multicenter, multireader study was to evaluate the performance of computed tomography (CT) lung nodule computer-aided detection (CAD) software as a second reader. METHODS AND MATERIALS: The study involved 109 patients from four sites. The data were collected from a variety of multidetector CT scanners and had different scan parameters. Each chest CT scan was divided into four quadrants. A group of three expert thoracic radiologists identified nodules between 4 and 30 mm in maximum diameter within each quadrant. The standard of reference was established by a consensus read of these experienced radiologists. The cases were then interpreted by 10 other radiologist readers with varying degrees of experience, without and then with CAD software. These readers identified nodules and assigned an actionability rating to each quadrant before and after using CAD software. Receiver operating characteristic curves were used to measure the performance of the readers without and with CAD software. RESULTS: The average increase in area under the curve for the 10 readers with CAD software was 1.9% for a 95% confidence interval (0.8-8.0%). The area under the curve without CAD software was 86.7% and with CAD software was 88.7%. A nonsignificant correlation was observed between the improvement in sensitivity and experience of the radiologists. The readers also showed a greater improvement in patients with cancer as compared to those without cancer. CONCLUSIONS: In this multicenter trial, CAD software was shown to be effective as a second reader by improving the sensitivity of the radiologists in detecting pulmonary nodules.

Application of statistical modeling to optimize a coastal water quality monitoring program.

The long-term water quality monitoring program implemented by the Massachusetts Water Resources Authority in 1992 is extensive and has provide substantial understanding of the seasonality of the waters in both Boston Harbor and Massachusetts Bay and the response to improvements in effluent quality and offshore transfer of the effluent in September 2000. The monitoring program was designed with limited knowledge of spatial and temporal variability and long-term trends within the system. This led to an extensive spatial and temporal sampling program. The data through 2003 showed high correlation within physical parameters measured (e.g., salinity, dissolved oxygen) and in biological measures such as chlorophyll fluorescence. To address the potential sampling redundancies in the measurement program, an assessment of the impact of reduced levels of monitoring on the ability to make water quality decisions was completed. The optimization was conducted by applying statistical models that took into account whether there was evidence of a seasonal pattern in the data. The optimization used model survey average readings to identify temporal fixed effects, model survey-average-corrected individual station readings to identify spatial fixed effects, corrected the individual station readings for temporal and spatial fixed effects and derived a correlation model for the corrected data, and applied the correlation model to characterize the correlation of annual average readings from reduced monitoring programs with true parameter levels. Reductions in the number of sampling stations were found less detrimental to the quality of the data for annual decision-making than reductions in the number of surveys per year, although there is less of a difference in this regard for dissolved oxygen than there is for chlorophyll. The analysis led to recommendations for a substantially lower monitoring effort with minimal loss of information. The recommendation supported an annual budget savings of approximately $183,000. Most of the savings was from fewer surveys as approximately $21,000 came from the reduction in the number of stations monitored from 21 to 7 and associated laboratory analytical costs.

Transcriptional analysis of protective antigen-stimulated PBMC from non-human primates vaccinated with the anthrax vaccine absorbed.

The transcriptional responses in recombinant protective antigen (PA)-stimulated peripheral blood mononuclear cells (PBMCs) from Anthrax Vaccine Absorbed (AVA)-vaccinated rhesus macaques were evaluated using Affymetrix HGU133 Plus 2.0 GeneChips. PBMCs from animals vaccinated at 0, 4, and 26 weeks were harvested at week 30, stimulated with PA, and RNA isolated. The expression of 295 unigenes was significantly increased in PA-stimulated compared to non-stimulated PBMCs; no significant decrease in gene expression was observed. These upregulated transcripts encoded for proteins functioning in both innate and adaptive immunity. Results were corroborated for several genes by real-time RT-PCR. This study provides information on the potential underlying transcriptional mechanisms in the immune response to PA in AVA-vaccinated rhesus macaques.