IRAIS 2026 Proceedings

Abstract

First responders, including police and fire, work under high-stress conditions with limited resources (Regehr et al., 2003). Behavioral health (BH)-related incidents place additional strain on the emergency response system. These responders often lack clear tools and methods to accurately identify BH-related calls from incident reports generated following 911 responses. Improving recognition and classification of the diverse types of BH-related cases (from acute mental health crisis to not needing follow-up) is essential for generating meaningful data, resource allocation, and timely, appropriate follow-up care (Kuehl et al., 2025; Satel, 2020).

From an information systems perspective, one core problem relates to the disparate mandates of responding agencies. Police are structured for law enforcement, while fire departments are designed for medical triage and hazard mitigation. Because these agencies operate on different reporting infrastructures, essential BH indicators such as suicidal ideation, substance use risk, and recurring crisis cycles may be obscured or lost in digital record-keeping. Data fragmentation can hinder users of existing datasets from forming a united view, limiting their ability to fully understand and respond effectively.

Existing research highlights significant challenges, particularly in identifying behavioral health (BH) crises within emergency dispatch systems. Studies show that standardized dispatch codes often fail to capture the nuance of behaviorally complex situations. While frameworks like those from the International Academies of Emergency Dispatch aim for uniformity, inconsistent implementation across jurisdictions, and even across veteran vs. newly trained first responders, leads to varied data quality and loss of BH indication data (Hashmi et al., 2023). Additionally, while co-responder models (police/fire and clinician responding to BH calls) advocate for inter-agency collaboration, these programs are often hindered by a lack of shared information-sharing agreements and integrated data practices (Horspool et al., 2016). The current literature underscores that without a unified data environment, the siloed nature of public safety records remains a primary barrier to proactive BH care management.

To better characterize this challenge, we conducted a study to match cross-agency incident reports with the aim of:

  1. Understanding the data matching problem and showing an example of how it occurs and how to solve it.
  2. Examining how cross-agency narratives provide a more complete representation of behavioral health incidents than either source alone.

This study used 200,000+ actual police and fire incident records spanning 2019–2022 through a cooperative MOU with a suburban county in the Southeastern US. ID matching was initially attempted using record IDs but was not feasible, as the systems did not share common identifiers despite operating within a unified IT infrastructure. Therefore, a multi-key, rule-based police-fire record linkage was applied using a truncated address key consisting of the first 10 standardized address characters, together with the incident date. This initial linkage produced 19,830 candidate police-fire record pairs. After further filtering to address duplicate addresses and many-to-many matching issues, where a record from one department could match multiple records in the other department or vice versa, only 1,252 record pairs with the closest occurrence times within a one-hour window were retained for analysis. As part of addressing the first aim, the matched narratives were manually reviewed for validation. The result was a final matched dataset of 1,180 valid matches and 72 mismatched observations, representing an estimated match accuracy rate for the address/datetime approach of about 94.3%.

To address the second aim, we used manual narrative review alongside a validated BH-detection algorithm (Brown et al., 2023) to closely examine cases with high BH likelihood to see what details were clear in one data source but not in another and then identify the value of connecting the two sources. Our incident report data reviews revealed varying narratives across agency records. In particular, police and fire narratives differed in terminology, classification, and level of detail. For instance, behavioral-health indicators were sometimes unclear, hidden, or inconsistently recorded across systems. Police records often focused on legal issues. Fire records focused on health and response speed issues. We can trace both motivations to the national reporting requirements these agencies use. Three primary insights emerged:

  1. Loss of Critical Clinical Nuance: High-risk markers were in one source but not another.
  2. Failure of Continuity and Cycle Detection: Recurring crisis cycles were lost.
  3. Algorithmic Implications: Any machine learning model trained on a single agency's data may suffer from high false-negative (or positive) rates regarding behavioral health risks.

This study offers an applied example of how fragmented data governance may affect police-fire operations. It also illustrates a practical record-linkage procedure for integrating institutional datasets in the absence of common identifiers. The findings suggest that responsible AI in public-sector crisis response depends not only on model design but also on the quality and governance of the underlying data infrastructure.

References

Brown, M., Khan, M. A. A. H., Thomas, D., Pei, Y., & Nandan, M. (2023). Detection of behavioral health cases from sensitive police officer narratives. 2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC), 1398–1403. https://doi.org/10.1109/COMPSAC57700.2023.00213

Hashmi, T., Thomas, D., & Nandan, M. (2023). First responders, mental health, dispatch coding, COVID-19: Crisis within a crisis. Journal of Emergency Management, 21(3), 233–240. https://doi.org/10.5055/jem.0664

Horspool, K., Drabble, S. J., & O’Cathain, A. (2016). Implementing street triage: A qualitative study of collaboration between police and mental health services. BMC Psychiatry, 16(1), 313. https://doi.org/10.1186/s12888-016-1026-z

Kuehl, S., Cooper, L., & Every-Palmer, S. (2025). “Able to stop things from escalating” – Stakeholders’ perspectives of police, ambulance and mental health co-response to 911-mental health calls. The Police Journal, 98(1), 167–184. https://doi.org/10.1177/0032258X241253965

Regehr, C., Hill, J., Knott, T., & Sault, B. (2003). Social support, self-efficacy and trauma in new recruits and experienced firefighters. Stress and Health, 19(4), 189–193. https://doi.org/10.1002/smi.974

Satel, S. (2020, October 1). Rethink crisis response: People who call 911 shouldn’t get an ill-trained police officer, especially when they’re dealing with a mental health emergency. Reason Magazine, 52(5), 28. Gale General OneFile.

Download

Abstract Only

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.