Funder: NIMH (MHRN III Feasibility Pilot Program)
Grant Number: U19MH121738
Project Period: 7/1/2022 – 6/30/2023
Background: Exposure to potentially traumatic events such as physical and sexual abuse/assault, serious accidental injury, mass shootings, and terrorism, and associated PTSD are major public health concerns (Magruder, McLaughlin & Elmore Borbon, 2017). It is estimated that over 20 million Americans develop PTSD at some point in their life (Kessler, Berglund et al., 2005). Inadequate treatment of PTSD may lead to chronic impairment and disability and have long-term and widespread familial and societal consequences (e.g., domestic violence, suicide, incarceration).
Incident rates of PTSD appear strikingly low in the health care system compared to estimates derived from representative epidemiological studies of the general public. Conservative estimates suggest that up to 80% of adults will experience a traumatic event during their lifetime. In a large nationally representative epidemiological study, it was estimated that PTSD impacts 3.6% of civilians each year, with a lifetime prevalence rate of 6.8% (Kessler, Berglund et al., 2005; Kessler, Chiu et al., 2005). However, in a recent examination of PTSD in six MHRN-affiliated health care systems we found less than 1% of the patient population had a diagnosis of PTSD when using ICD diagnosis codes only, suggesting patients may be underdiagnosed or inadequately captured using this method. Further, ICD diagnosis codes are limited in their ability to capture trauma exposure type (e.g., combat exposure, motor vehicle crash, sexual abuse, elder abuse, intimate partner violence, natural disaster) and may be underutilized by providers.
This project builds on previously MHRN-funded research conducted by Negriff and colleagues (Lynch, 2022) who examined incidence of child maltreatment comparing rates of those captured by ICD diagnosis codes versus natural language processing (NLP). In their investigation, NLP identified 10 times more children with child maltreatment than just using the diagnosis code. Building on this methodology the proposed pilot project will use NLP to identify patients within one health care system (Kaiser Permanente Hawaii) who experience PTSD compared to those identified using ICD diagnosis codes only. Further, we test the feasibility of using NLP to categorize patients based on exposure type (e.g., combat, motor vehicle crash, sexual abuse, etc.). NLP may help to identify additional trauma-exposed individuals with PTSD that are not documented/captured through ICD codes. This may lead to the identification of care gaps, novel treatment targets, and characteristics (e.g., age, sex, race/ethnicity, trauma exposure type) that may make it more/less likely to have ICD coded PTSD. To date, PTSD has been relatively underexamined within the Mental Health Research Network (MHRN) despite being identified as a priority area in this third funding cycle.
- Does NLP allow us to obtain estimates of the number of adults who experience PTSD that are more comparable to national epidemiologic data?
- Are there differences by group (e.g., age, sex, race/ethnicity, trauma type) of those captured through NLP versus ICD diagnosis code?
- Can we establish feasibility for systematically identifying trauma exposure using previously collected data within the health care system?
Methods: The project PI will convene a panel of interested MHRN investigators to discuss approach, assist in the identification of terms, interpretation and use of results, and future research. Drs. Frances Lynch, Jordan Braciszewski and Rob Penfold have expressed interest in serving on this panel and a larger invitation will be sent to all MHRN-affiliated investigators, if funded.
We propose to use simple NLP queries at 1 MHRN site to identify incidents of trauma exposure and PTSD and compare the number of cases identified through NLP compared to those identified using ICD codes only. We will Identify a cohort of adults (age 18 and over) at KPHI and develop a Bag of Words (concept unique identifiers), building off those developed by Negriff, Lynch and Penfold, to search chart notes. Following the initial search, the PI, a licensed clinical psychologist, will conduct chart review of up to 150 cases to manually review text for each concept unique identifier and flag confirmed cases (yes/no). This data will be used to retrain NLP and the process will be repeated a second time for quality assurance/validation. We will use standard methods for identifying patients based on ICD-codes only (comparison group). We will then conduct appropriate statistical analyses to examine differences in identification by groups.
Planned Product: The results of this pilot study will serve as the basis for an R01 application to the National Institute of Mental Health under the NOSI Secondary Analysis of Posttraumatic Psychopathology Data. In addition, results from this study will be presented via scientific conference presentation and/or peer-reviewed publication.
Lead Site: KPHI (PI Vanessa Simiola)
Participating Sites: N/A