Figure 1 illustrates the PRISMA article selection process. Initially, 527 unique citations were identified from searches of electronic databases and reviewing article references (snowballing). After the title and the abstract screening, 381 were excluded, with 146 full-text publications to be retrieved and assessed for eligibility. Of these, 86 did not meet the inclusion criteria. The remaining 60 citations were considered eligible for inclusion in this review.
Most citations focusing on stressful life events and EHR were coming from the US (n=47, 78.3%), Netherlands (n=5, 8.3%) and UK (n=3, 5%). Canada, China, Italy, Portugal, and South Korea were represented by one publication. Reference characteristics by country of origin and US states statistics are provided in Figure 2A.
Most citations can be categorized into five following areas by their focus:
- Finding statistical dependency, in which one of the variables is stressful life event(s) (n=24, 40%);
- Clinical utility of stressful life events in EHR (n=14, 23.3%);
- Extracting events from EHR (n=8, 13.3%);
- Discussing issues of storing life events in EHR (n=5, 8.3%), such as privacy;
- Proposing new EHR design, template, or feature to facilitate stressful life events capturing (n=4, 6.7%).
The remaining 5 publications were focused as follows. Three citations [25-27] spanned two areas in a single study: finding statistical dependency and extraction. One study [28] compared paper-based and EHR-based screening tools and found that capturing adverse life events improved significantly when using specialized tools in EHR: 77% of EHR had information on such events compared to 33% when using older paper-based tools. One more citation [9] is the already mentioned work of the Committee on the Recommended Social and Behavioral Domains and Measures for Electronic Health Records, which debated what measures should be considered for stressful life events and decided that a general measure of stress would be sufficient. Figure 3A depicts the focus areas of all included articles.
Most of the citations (n=43, 71.7%) discussed or utilized life events stored in various formats in EHR. However, a significant number of studies used events obtained via surveys conducted specifically for the study (n=10, 16.7%) – these were statistical studies, and EHR was most often used as a source of variables related to health outcomes. One more citation utilized the interview format for events identification and compared them to what is documented in EHR and found that health records “generally failed to include social contexts salient to patients” [29]. Another publication linked a credit history report to identify serious financial events in patients’ lives, such as bankruptcy [30]. One citation [31] discussed issues of using external data sources linked to EHR for capturing “financial, legal, life event and sociodemographic data” to improve suicide screening models. One more study [32] investigated the impact of the 2008 economic crisis on different health outcomes using the unemployment data from the census. Figure 3B demonstrates breakdown of the articles by source of stressful life events.
The publications that were focused on EHR as a source of stressful life events used structured data on events in more instances (n=22, 56.4%) than unstructured data (n=17, 43.6%). One additional paper [33] discussed a time chart feature inside EHR showing correlations between life events and body mass index. In citations using unstructured data natural language processing (NLP) was the most popular method for extraction (n=10, 58.8%), followed by manual abstraction (n=6, 35.2%), however, audio recording was used in one citation [34]. Figure 3C illustrates the data types for stressful life events in EHR.
In studies that were focused on finding statistical dependency the health outcome under question was most often in the domain of mental health (n=13, 39.4%), such as diagnosis of post-traumatic stress disorder, and anxiety; followed by behavioral outcome variables related to violent and self-harmful behavior (n=10, 30.3%), and physical health outcomes (n=6, 18.2%). Few outcome variables (n=4, 21.1%) were in the borderline area spanning across physical and mental health, such as in the case of chronic pain. Figure 3D summarizes observed outcome variables.
There were recurring stressful life events seen across multiple citations, such as child abuse and/or neglect, arrest/legal issue, divorce or relationship breakup, death of relative, loss of job, homelessness. Different types of stressful life events which were mentioned by two or more studies are presented in Figure 4. A sizable number of events were mentioned by one study only: abandonment by parent, captivity, child marriage, disappearance of family member, forced marriage, genital mutilation/cutting, life-threatening event, miscarriage, parental behavioral problems, unspecified psychological trauma, sex trafficking in childhood, sexual exploitation, something valuable lost or stolen, starvation, terrorism, taking drugs as a child, unwanted pregnancy, victimization experience, witnessing somebody's injury or death.
Child abuse and/or neglect was the most common type of stressful life event mentioned in citations (n=19, 31.7%). A sizable number of publications (n=14, 23.3%) considered a serious life event but did not specify what event it was. An example of this approach is a survey question “After the age of 17, did you experience any major upheaval that you think may have shaped your life or personality significantly?” [35]. Unspecified stress was used in a sizable portion of citations (n=11, 18.3%), a common question to capture it was “Stress means a situation in which a person feels tense, restless, nervous, or anxious, or is unable to sleep at night because his/her mind is troubled all the time. Do you feel this kind of stress these days?”. This question is used in various popular screening tools such as PRAPARE [36-38].
Among the articles that analyzed clinical utility, a half (n=7) were focused on child abuse and neglect. These publications discussed several clinical processes which are triggered when such events are detected: alert in EHR, starting an abuse-related order set, and reporting to child protection agencies. The other half (n=7) touched on the topic of possible clinical workflow alterations based on detected SDOH categories, among which some stressful life events were mentioned: homelessness, domestic violence. However, because these publications considered broad SDOH domains, we couldn’t extract specific workflows that would be related to stressful life events. Some of the general SDOH-related clinical interventions discussed in these papers included: referral to trauma recovery [39], sending educational materials to patient [39], alerts and triggers [36, 40], panel management [36], automatic and manual referrals to third party agencies [36, 38, 41], risk adjustments [42], modification of clinical practices [36, 41], adjusted provider recommendations [36, 41].
Individual citations characteristics in table format are provided in Supplementary Appendix S3. For some of the citations the table also includes additional extracted characteristics, such as dependent and independent variables in statistical studies, types of surveys used to capture events, NLP tools used, relevant information from results, discussion and conclusion sections of reviewed citations.