Background:
Medication exposures and poisonings are a major cause of pediatric morbidity and mortality. A recent survey of pediatric primary care clinicians shows that most (82.5%) reported having at least one patient, and nearly half (45.4%) reported having more than 10 patients, who experienced an unintentional or intentional medication exposure or poisoning event. Although the American Academy of Pediatrics recommends pediatric primary care clinicians provide medication safety counseling, lack of time, training and resources are barriers. We aim to describe the incidence of, and characteristics of children with, medication exposures and poisoning events in a large pediatric primary care sample to guide medication safety counseling strategies.

Methods:
This is a retrospective observational study of pediatric primary care patients in two mid-Atlantic health care systems, an integrated system (IS) and a university-affiliated system (US). We queried the electronic medical record for patients ≤ 19 years with a pediatric primary care clinic visit between 1/1/2018-12/31/2019, and identified those with a clinical diagnosis and/or visit for a medication exposure or poisoning event (i.e., clinical encounter) in the two-year period using ICD-10 codes: T36-T50 “Poisoning by, adverse effect of and underdosing of drugs, medicaments and biological substances”. Data extracted included sex, age, race, ethnicity, insurance type, area deprivation index and active medical diagnoses on the problem list during the study period. Descriptive statistics were used to summarize the data, which were reported separately from the two systems.

Results:
There were 211,173 total patients included, with 156,165 IS and 55,008 US patients. Of these, 452 IS patients (0.29%) and 196 US patients (0.36%) had a clinical encounter for a medication exposure or poisoning event. In aggregate, this represents 307 cases per 100,000 patients over the 2-year period. The mean age at the documented time of the clinical encounter was 8.1 years (IS) and 7.4 years (US). A higher proportion of IS children with an encounter were in the 3-4 year (13.9% vs 10.8%) and 15-19 year age groups (23.7% vs 16.8%), whereas a higher proportion of US children were in the 0-2 year age group (42.9% vs 28.4%), when compared to those without an encounter. Within both systems, a higher proportion of children with an encounter were white (IS:30.3% vs 22.9%; US:49.5% vs 33.4%), or in the highest (least deprived) quartile of the area deprivation index (IS:38.2% vs 36.4%; US:43.4% vs 37.4%), when compared to those without an encounter. Regardless of encounter status, the most common other medical diagnoses for patients in both systems included diseases of the respiratory system and diseases of the skin and subcutaneous tissue.

Conclusions:
In this large pediatric primary care sample, approximately 1 in 345 IS patients and 1 in 278 US patients had a clinical encounter for a medication exposure or poisoning event. Limitations include potential variations in clinician documentation, coding practices, underreporting and exclusion of events that were not medically attended. While these findings support the need for primary care-based medication safety counseling, a better understanding of the context of the events may help design appropriate counseling efforts.

Background/Purpose

Timely and accurate surveillance of Emergency Department (ED) accesses is crucial to ensure patient safety and quality of care. However, such data are often in free-text format, and manual coding is time-consuming and prone to errors. This issue is even more significant in developing countries where epidemiological surveillance is already lacking, although it would be particularly relevant given that they present specific characteristics and needs different from Western countries. In this era of rapidly evolving artificial intelligence, the application of Generative Pre-trained Transformer (GPT) models offers a promising solution to simplify this process. OpenAI is an extensive language model based on the GPT architecture. The present study aims to implement an automatic coding system using OpenAI to extract and classify data on discharge diagnoses from a database of 1,789 pediatric ED visits, in Spanish language free-text format, collected in the EDs of nine hospitals in Nicaragua.

Methods

The GPT-3.5 variant was chosen for the present work, guided by awareness of the scarcity of economic resources. OpenAI was accessed through a public application programming interface (API) using R software, through which users can provide prompts to the trained model, hosted in the cloud, and receive the models’ response. The free-text discharge diagnoses were manually revised and classified by an independent peer-review group of expert pediatricians. The resulting classification, consisting of 10 disease classes, was considered as the gold standard to evaluate the performance of the GPT-based classification task.

Results/Outcomes

Preliminary results showed a classification accuracy of 92.4% percent. The sensitivity, i.e. the ability of the tool to correctly classify diagnoses, was 93%, and the specificity was 95.7%.

Conclusions

Monitoring ED visits serves as a powerful tool for public health surveillance, as it enables the analysis of the frequency and distribution of diseases, early outbreak detection, assessment of health service quality, and evaluation of intervention program effectiveness. Alongside the development of a standardized and computerized data collection system for ED visits, it is crucial to adopt classification approaches that facilitate the exploitation of unstructured text-based data for epidemiological surveillance purposes. In the present study, the performance of classification task was excellent, contributing to the awareness that OpenAI represents a promising opportunity to improve health surveillance and patient safety also in low- and middle-income countries.

Background

Children with Autism Spectrum Disorder (ASD) face elevated risks of injury compared to their peers. This study aggregated 22 years of Medical Expenditure Panel Survey (MEPS) data and applied machine learning algorithms to identify significant risk factors for injury in children with ASD.

Methods

We aggregated MEPS data from 2000 to 2021. We ascertained consistent predictors across these years. Using ICD10 code (F.84), we identified 1294 children with ASD. From 297 variables considered relevant, we only used responses from the first round of questions, as some were asked three times each year. This was to avoid potential bias, as injuries may have already happened before the later rounds of data collection. The final set for modeling included 130 variables. The outcome variable, injury occurrence across the study period (12.8% with injury and 87.2% without injury), imbalanced data. To improve model accuracy, we used RandomOverSampler in Python 3.10 to oversample the minority class (with injury), achieving a balanced sample size between those with and without injury. The new sample size was therefore increased change to 2,254 (50% with injury and 50% without injury). We then applied Boruta in R 3.14, a feature selection algorithm, to remove the variables with low importance scores. Subsequently, a classification decision tree (DT) model with 5-fold cross-validation was run to rank and visualize important predictors and their interactions and to construct the prediction model. Logistic regression was incorporated to compute the P-value and odds ratio of selected predictors on injury risk.

Results

A total of 93 variables were selected by the Boruta algorithm for the DT model. The overall accuracy was 93%, with a precision of 90%, recall of 100%, and an AUC of 94%, without setting up threshold numbers for predictors in the DT model. We presented the top 12 important predictors in the DT plot, based on the impurity measure method, which also maintained good model performance with an overall accuracy of 70%, precision of 76%, recall of 74%, and AUC of 74%. The high-risk factors of injury in autistic children included: (a) low education level, (b) male gender, (c) children with emotional, developmental, or behavioral problems requiring treatment or counseling, (d) children living in small family sizes, (e) children with no work limitations, (f) children using assistive devices, or (g) children living in the Midwest or West. Additionally, the results expressed that: (a) children with Health Maintenance Organization insurance, (b) those not receiving food stamps, or (c) children who can use more than two languages have a lower risk of injury.

Conclusion

In summary, this study contributes to (a) demonstrating how to preprocess complex medical data, (b) providing a high-performance prediction model. and (c) suggesting significant predictors that can support the design of injury prevention plans for children with ASD. We believe this study can mitigate the injury risk of children with ASD.

Keywords: machine learning, injury prevention, children with ASD, MEPS data

Vulnerable communities are disproportionally impacted during emergency events. A community's capacity to prepare for, manage, and recover from a disaster determines its level of resilience. a disaster determine a community’s level of resilience. The Federal Emergency Management Agency (FEMA) recommends engaging community members in resilience planning to ensure relevant and sustainable strategies are implemented. The Mobilizing communities to strengthen resilience using a Community Asset Inventory Dashboard (CAssI Dashboard), aims to build an interactive dashboard incorporating a comprehensive list of resources that contribute to the resilience of North Texas communities and its ability to address the short- and long-term impacts of emergency events.

This project uses community-based asset mapping to identify resources that would go unnoticed at a larger scale. An asset map documents a community’s existing resources, highlighting its strengths that aid in the positive influence of its social determinants of health. A community’s assets include emergency services, education access, healthcare, local and state government, social and cultural organizations, and neighborhood resources. Identifying the resources available in the community to mitigate the impacts of emergency events allows for an early assessment of a community’s resilience. Information used to develop the preliminary asset maps was gathered using secondary data, such as Google Maps, school district websites, and business/ establishment websites. Asset maps were created for urban and rural communities in North Texas and will be shared with community members to further explore assets not identified in secondary data sources.

Focus group discussions in both urban and rural communities will be conducted to obtain feedback and suggestions about existing assets and suggestions for additional assets to include on the maps. Participants will be asked to reflect on the community resources used during the most recent emergency events, specifically the non-traditional assets such as human, social, cultural, political, and natural assets. Findings from the focus group analysis will be used to update the asset map to include a comprehensive list of assets.

Using a systematic approach like asset mapping allows for an in-depth understanding and relevance to a community and its resources. A comprehensive map of community-based assets will more accurately reflect resources that are useful to communities in emergency events and identify areas of opportunity within the community. The asset map will be used to develop an easy-to-use online interactive dashboard to be used by communities and emergency managers in disaster planning. Local response organizations and government officials will be encouraged to use the new tool to identify how to best support and collaborate with their communities while engaging individuals in capacity-building efforts.

Engaging community members and leaders in the identification of community-based assets creates opportunities to learn about traditional and non-traditional resources, as well as identify challenges communities face before, during, and after an emergency event. A collaborative approach to community resilience building can foster relationships between individuals and emergency planners to create sustainable preparedness, response, and recovery strategies.

Background/Purpose

The frequency of dance injuries has increased over the last few decades. Previous research involving adolescent and adult dancers has shown overuse ankle injuries to be more common among female dancers, and acute shoulder, knee, and face injuries to be more common among male dancers, likely due to gender-based dance roles. Our study aimed to describe the type and rate of dance injuries seen in pediatric patients in the last two decades.

Methods

The Public Health Information System (PHIS) was queried for patients aged 0-18 years who suffered dance-related injuries and presented to participating children’s hospitals between 2004 and 2023. Patients with non-orthopedic diagnoses were excluded. Demographics, diagnosis, injury severity, visit type, physician specialty, and complications were collected from single encounters. Diagnoses were categorized into seven locations (ankle/lower leg, foot/toe, knee, neck/back, pelvis/upper leg, upper extremity, and other) and eight injury types (knee/lower leg fracture, ankle/foot fracture, elbow fracture, wrist fracture, dislocations/instability, ankle/foot sprain, knee sprain, and other). Descriptive statistics were calculated, and chi-square tests were performed.

Results/Outcomes

19,955 dance injuries (16,901 female) with mean patient age 11.6 ± 4.1 years, presented to attending physicians mostly in Emergency Medicine (63%), Pediatrics (18%), and Orthopedic Surgery (7%) settings. The number of injuries per year increased from 2009 (761) to 2023 (1,555) and peaked in 2018 (1,960). 146 encounters were flagged for infection at presentation and 7 were flagged for surgical complication. Most injuries were of minor severity (95%). Injuries of moderate and major severity were more common among males (7% and 1% of male injuries) than females (5% and <1% of female injuries). The relationship between gender and severity level was significant (p<0.001). The relationship between age group and severity level was not significant (p = 0.489). Lower-body injuries (66%) outnumbered upper-body injuries (32%). The most frequently injured body parts were the ankle (18%) and knee (16%). Lower-body injuries were more common among females (11,401, 68%) than among males (1,674, 55%). Upper-body injuries were more common among males (1,312, 43%) than females (5,156, 31%). The relationship between body part injured and gender was significantly different (p<0.001). Among females, sprains (5,182, 31%) outnumbered fractures (4,146, 24%). Among males, fractures (1,109, 37%) outnumbered sprains (656, 22%). Dislocations/instability were more common among males (313, 10.3%) than females (1,415, 8.4%). The relationship between injury type and gender was significant (p<0.001).

Conclusions

Most dance injuries involve the lower body, specifically the ankle and knee, and are minor in severity. Injury severity, location, and type vary by gender. Male dancers tend to experience more upper body injuries, fractures, dislocation events, and moderate-severity injuries. Female dancers tend to experience more sprains, lower body, and minor severity injuries, though these three categories of injuries are the most common amongst both male and female dancers. Age group does not appear to influence the severity of injury. These findings may help focus injury prevention programs for youth to reduce the risk of lower-body injury, particularly at the ankles and knees.

Introduction: Flag football has become a popular alternative to tackle football within the United States. Likely driven by concern for concussion, youth participation in tackle football has declined by 29% from 2016-2021, while flag football participation increased 15% within the same time period. Despite the notable increase in flag football popularity, limited research has been conducted regarding the epidemiology of flag football injuries.

Methods: The National Electronic Injury Surveillance System (NEISS) database was queried to identify all cases of flag football injuries in children aged 0-18 years, from 2008 through 2022. All injuries attributed to football, miscellaneous or “other” ball sports, and generalized sports or recreational activity were identified and filtered to include only those related to flag football. NEISS weights were applied to gather national estimates. Descriptive statistics summarized injury type and trends. The Mann-Kendall test was used to determine the trend of injury estimates over time (statistics from 2020 were excluded from this test) and Pearson’s correlation coefficient was used to understand the relationship of injury types over time.

Results: After applying NEISS weights, the national estimate for flag football injuries from 2008-2022 was 69,045 injuries, or 4603 injuries annually. Patients aged 12 years and younger, classified as children, suffered 42% of injuries while patients aged 13-18 years, labeled as adolescents, accounted for 58% of injuries. Injuries were most commonly sustained to the head, neck, and face (31% of cases), however concussions accounted for only 10% of overall injures. The majority of injuries, 98.5%, were labeled as not severe. Fractures and dislocations made up 31% of all injuries and sprains and strains accounted for 24% of cases. The most common fractures were those of the finger or hand while the most common strains were those of the ankle, foot, and lower leg. Mann-Kendall testing revealed there was not a significant upward trend in injuries during the 15-year period (p=0.07). Generated Pearson-values indicate an association between year and type of injury (p=0.0269) as well as year and severity of injury (p=0.0054).

Conclusion: The increase in flag football’s popularity over the last 15 years has not been accompanied by a significant increase in overall injury totals, but has been correlated with an increase in injury severity, although injuries remain overwhelmingly classified as not severe. While nearly one third of injuries are sustained to the head, neck, or face, concussions account for a low proportion of overall injuries sustained during flag football. Overall, flag football injuries are classified as low severity with most injuries over the last 15 years being fractures or dislocations to the upper extremity and foot/ankle sprains or strains. The increase in popularity and participation rates without a significant increase in overall injury totals provides evidence that flag football may be a safe alternative to tackle football. This information, as well as the trends in body part of injury and injury type can be used in decision making regarding sports participation, safety equipment, as well as sport rules and regulations.

Background/purpose: Ageism is a common form of discrimination that can negatively affect the health of older adults (ages 65+ years). Little is known about the relation of perceived ageism to falls, fall risk, and fall prevention engagement. Annually, 28% of older adults report ≥1 fall. The purpose of this study is to explore if perceived ageism is related to fall-related outcomes among older adults.

Methods: A total of 1,113 older adults across the United States participated in FallStyles, part of the 2023 Porter Novelli ConsumerStyles survey. The survey included the Everyday Ageism Scale (a validated ageism measure for adults aged 50-80 years), Three-Key Questions to determine fall risk from the Center for Disease Control and Prevention’s Stopping Elderly Accidents, Deaths and Injuries (STEADI) initiative, and other questions to gauge beliefs, knowledge, and intentions around falls and fall prevention. We conducted a preliminary analysis that included descriptive statistics and cross tabulations with 95% confidence intervals (CI). Total ageism scores were categorized as low, medium, and high for comparisons between groups. Analyses were conducted using SAS 9.4 to explore the relation of perceived ageism with fall-related variables.

Results/outcomes: Almost one-third (31.2% (95% CI, 28.2%-34.03%)) of older adult respondents reported ≥1 fall in the past year, and more than half (54.4% (95% CI, 51.3%-57.4%)) were at risk for future falls based on the Three-Key Questions. Half (49.6% (95% CI, 36.3%-62.9%)) of the older adults with high perceived ageism experienced one or more falls in the past year compared to 19.7% (95% CI, 15.4%-24.0%) with low perceived ageism. A higher proportion of older adults with high perceived ageism were at high risk for future falls (70.7% (95% CI, 58.5%-83.0%)) compared to those with low perceived ageism (38.3% (95% CI, 33.1%-43.6%)). A higher proportion of older adults with high perceived ageism also had an inaccurate perception of their fall risk (55.7% (95% CI, 42.5%-69.0%)) compared to those with low perceived ageism (37.2% (95% CI, 32.0%-42.3%)). Finally, a higher proportion of older adults with high perceived ageism reported being likely to talk to their healthcare provider about fall prevention (22.5% (95% CI, 11.6-33.4)) compared to those with low perceived ageism (9.4% (95% CI, 6.0%-12.8%)).

Conclusions: Higher perceived ageism was related to being at risk for falling, experiencing at least one fall in the past year, and having an inaccurate perception of fall risk. However, those with higher perceived ageism were more likely to talk to their health care provider about fall prevention which may increase their opportunity to receive prevention advice. These findings represent a preliminary analysis of a cross-sectional study. Additional longitudinal research is needed to understand the strength and direction of the relation between ageism and fall outcomes including how perceptions of ageism may inform acceptability and adoption of fall prevention strategies.

Background: Work-related injuries are likely underreported among older workers because they are less likely to receive workers' compensation for their injuries. A variety of methods have been used to identify work-related injuries from administrative databases, including workers' compensation payer data and ICD-10-CM external cause of injury codes. We show how Medicare administrative claims data can identify and describe work-related injuries among Medicare enrollees.

Methods: We identified injury claims from 2016-2019 Medicare Inpatient and Outpatient claims among fee-for-service Medicare enrollees aged 65 and older. We then identified work-related injuries using specific ICD-10-CM codes, along with employment-related and workers' compensation codes used in Medicare claims processing on the injury claims. We further describe enrollee demographics, injuries and their mechanisms, and healthcare encounter characteristics of Medicare enrollees with work-related injury claims.

Results: From 2016-2019, the average annual rate of work-related injuries from Medicare claims was 27.6 work-related injury claims per 100,000 Medicare fee-for-service enrollees aged 65 and older. Injury claims were filed most often for outpatient emergency department (ED) visits (58%), and less often for non-ED outpatient visits (20%), or hospitalizations (19%). Medicare was the primary payer for most claims (89%), followed by workers' compensation (7%), and other insurance (4%). Falls, transportation, and machinery-related mechanisms of injury each accounted for approximately 20% of injuries.

Conclusion: Using mechanism of injury, employment-related, and workers' compensation codes, Medicare claims can be used to identify work-related injuries. Injury claims submitted to Medicare are infrequently paid by workers’ compensation, even when a work-related cause is apparent. Work-related injuries in Medicare most often appear in outpatient settings, although hospitalizations involve the most extensive care. Future research should validate or expand use of these methods, and explore the costs and health outcomes of work-related injuries in older populations.