Presenter Profile

Aedan Villegas, BS

Medical Student
University of Houston and Baylor College of Medicine
aedanvillegas@gmail.com

Aedan Villegas is a graduate of the University of Houston with a B.S. in Biochemistry. Aedan was introduced to this project by Dr. Shenoi through a college program a year ago, and since then, has developed a deep interest in injury prevention and surveillance, with a focus on improving swimming protocols and practices.

Presentations

Studying the epidemiology of drowning from media reports using artificial intelligence

Aedan Villegas, BS
Shaila Zaman, PhD
Rohit P. Shenoi, MD

Part of session:
Platform Presentations
Drowning Prevention
Friday, December 6, 2024, 9:15 AM to 10:30 AM
Background:

Media reports are a source of information on fatal drowning. The narratives provide information about demographics and contextual information of victims. We sought to describe the epidemiology of fatal drowning in the metropolitan Houston region using artificial intelligence to analyze media report narratives.

Methods:

This was a cross-sectional study of victims of unintentional drownings of all ages who drowned in the 8-county metropolitan Houston region from 2016-2022. We queried the search platform – NexisUni using keywords “drown” and “submersion” to access online media archives of fatal drownings from the following formats: print; blog, video, and audio records. Temporal data, demographic information (age, sex, ethnicity, race), risk and protective factors were obtained from the narratives. We developed an automated method to extract the information using the natural language processing (NLP) capabilities of OpenAI's GPT-3.5 model. The OpenAI API significantly enhances the efficiency and accuracy of extracting structured information from unstructured text by implementing advanced NLP techniques, facilitating efficient handling of large datasets with precision that surpasses manual methods. Using OpenAI API, we focused on extracting key factors such as swimming ability, supervision status, alcohol use, and life vest use from narrative descriptions of drowning circumstances. Our approach involved prompt engineering that guided the model to output structured responses in a dictionary format, ensuring consistency in the extracted data. The method was implemented in Python, where we utilized the OpenAI API to process each narrative in our dataset. The responses were then parsed and organized into a structured format suitable for further analysis. We used descriptive statistics for reporting results.

Results:

There were 133 media reports of drowning between 2016-2022 in metropolitan Houston. We analyzed 110 cases of drowning after excluding non-fatal (n=16), homicides (n=4) and those external to the Houston region (n=3). The mean age of drowning victims was 25.5 years (Std. Dev. ±23). There were 49 (44.5%) children; males constituted 81% of reports. The majority of drownings occurred in natural water (73%). The rest occurred in confined water (swimming pools, bathtubs)(26%) and unknown (1%) Life jackets were used by 33% of drowning victims; their use was unknown in 67% of cases. Fourteen percent of drowning victims knew how to swim, whereas 19% did not know how to swim. Swimming ability was unknown in 67% of cases. A majority of drowning victims did not consume alcohol (65%). Only 1% consumed alcohol and its use was unknown in 34%. Among children, 65% were supervised, 10% were unsupervised. Supervision was unknown in 25%.

Conclusions:

Media reports are an alternate source of data on fatal drowning. Artificial intelligence is a novel method to extract drowning information from media report narratives for epidemiological purposes. However, the utility of media reports in drowning surveillance and prevention is limited, given biases in reporting and sub-optimal reporting of information on protective and risk factors. Engagement with media outlets to include drowning prevention information during reporting is encouraged.

Objectives:

1. Explain how artificial intelligence can be used in injury surveillance
2. Discuss the epidemiology of drowning obtained from media reports.
3. Recognize the limitations of media reports in describing the epidemiology of drowning