Correlates of Emergency Department Use by Individuals With Bipolar Disorder Enrolled in a Collaborative Care Implementation Study

This study involved a secondary outcome analysis of the Recovery-Oriented Collaborative Care (ROCC) study, a cluster-randomized controlled implementation trial from 2009–2014 involving five community-based clinics from Michigan or Colorado that was designed to improve the uptake of a CCM for bipolar disorder called Life Goals–Collaborative Care (LG-CC). The study received approval from local institutional review boards. Details regarding ROCC are available elsewhere (6,7). In brief, the study compared implementation strategies provided at sites that offered LG-CC to patients who were diagnosed as having bipolar disorder and who were able and willing to provide informed consent to complete assessments.

LG-CC is an evidence-based CCM (8) that includes four weekly self-management sessions during the first month that focus on managing depressive and bipolar symptoms and ongoing telephone-based care management contacts for up to six months after patient enrollment. Care management contacts are implemented by clinical social workers or master’s-level mental health clinicians at each site. Providers from participating clinics received an LG-CC toolkit and training, but they were randomly assigned at the clinic level to receive additional, regular coaching and active technical assistance on a proactive basis to address barriers to LG-CC uptake (9) or as-needed technical support.

Participants gave informed consent prior to attending LG-CC sessions and completed baseline, six-, and 12-month assessments. The primary outcome was self-reported emergency department visits occurring between six and 12 months after the end of the LG-CC intervention period. Because many of the clinics lacked comprehensive medical record data, self-reported data were used in order to capture consistent information across all sites.

Factors thought to influence emergency department visits were also ascertained from the baseline survey and were included in the analyses as covariates, including age, gender, race (white and nonwhite), education (high school or more and less than high school), employment (unemployed and employed or retired), recent (past month) history of homelessness, clinically significant depressive symptoms (score ≥10 on the Patient Health Questionnaire–9 [PHQ-9]) (10), presence of panic symptoms (yes-or-no response to the question “In the past year, did you ever have a spell or an attack when all of a sudden you felt frightened, anxious, or very uneasy?”) (11), alcohol use (Alcohol Use Disorders Identification Test [AUDIT-C] three-item survey score, with possible scores ranging from 0 to 12 and higher scores indicating greater alcohol use) (12), illicit substance use, current smoking status, prior hospitalization history, number of self-reported medical conditions, overall general medical health (physical component summary [PCS] score on the 12-item Short-Form Health Survey, with possible scores ranging from 0 to 100 and higher scores indicating greater physical health–related quality of life [PCS-12]) (13), and sites indicator (mental health vs. primary care). Finally, number of LG-CC care management contacts during the six-month intervention was ascertained from provider record reviews. Participants were categorized by number of care management contacts (≥5 or <5) on the basis of previous analyses describing variations in LG-CC completion during the first six months (6).

Descriptive statistics were calculated to summarize the baseline characteristics of the sample and their associations with the outcome variable. Multiple logistic regression analysis (SAS, version 9.3) was used to determine participant factors associated with probability of having an emergency department visit during the six- to 12-month period after the end of the LG-CC intervention period, adjusting for the aforementioned demographic and clinical variables from the survey. Sensitivity analyses were conducted to determine whether additional organizational factors, such as clinic type (primary or mental health care), influenced probability of emergency department use, resulting in virtually identical results.

Of the 385 study participants with complete baseline surveys, 139 did not complete the 12-month survey. Of those who completed both surveys, 27 had either missing outcome or covariate data. The final analytical sample included 219 participants (57% retention). The 166 participants excluded from the analysis were not significantly different from participants in the analytical sample in any of the characteristics surveyed at baseline, except homelessness history. Persons with a history of homelessness had higher odds of dropping out of the study compared with persons without such a history (odds ratio [OR]=2.16, 95% confidence interval [CI]=1.30–3.58, p=.003). Baseline homelessness history was included as a covariate in the analyses.

Overall 24% (N=53) of the sample reported at least one emergency department visit during the six to 12 months after the end of the LG-CC intervention period. At baseline, the mean±SD age of the analytical sample was 42.6±11.0, 32% (N=71) were male, 72% (N=158) were white, 88% (N=193) had a high school education or less, 73% (N=160) were unemployed, 47% (N=102) were current smokers, 15% (N=33) had a recent history of homelessness, 39% (N=85) had PHQ-9 scores ≥10, 85% (N=185) had moderate anxiety symptoms, 29% (N=64) reported substance abuse, and 24% (N=53) had a history of recent hospitalization. The mean PCS-12 score was 36.3±7.3, the mean AUDIT-C score was 1.2±2.1, and the mean number of comorbidities was 1.6±1.2.

After adjustment for participant demographic and clinical factors and site, the analyses found that participants with a recent history of homelessness at baseline (OR=3.76, p=.01) or five or more care management contacts (OR=2.62, p=.05) at six-month follow-up had a higher probability of an emergency department visit during the six-month period after the intervention ended (Table 1). In addition, patients with a higher PCS-12 score at baseline (indicating better physical health) had lower probability of having an emergency department visit (OR=.94, p=.05).

This study determined participant factors associated with probability of emergency department use after the implementation of a CCM (LG-CC) for bipolar disorder in community practices. We found that participants with a recent history of homelessness were more likely to have an emergency department visit compared with participants without such a history, after the analyses adjusted for other clinical and sociodemographic factors. Because homelessness history was associated with missing data or attrition, it is possible that participants with a history of homelessness who needed to be seen in the emergency department were more likely to participate in the 12-month survey compared with homeless participants who did not need emergency department visits. In prior research, bipolar disorder was associated with increased emergency department use and inpatient hospitalization, the latter representing 60% to 70% of the total direct health care costs for these individuals (14), and much of the increased critical service encounters was associated with instability from homelessness and poverty (1).

This study also found that participants who completed five or more care management contacts had a greater probability of an emergency department visit. Perhaps these individuals had more unmet health services needs that were not observed in the study assessments compared with participants with fewer care management contacts. Most of the individuals enrolled in the study, although primarily recruited from community mental health programs, were not eligible for more intensive case management services, such as assertive community treatment; hence, a subset of the sample, for example, persons with homelessness history, might have required more case management than provided in the relatively brief LG-CC intervention. Perhaps CCMs that emphasize psychoeducation and health care coordination may not fully address social determinants of health, notably homelessness or lack of employment, and for this study population, different types of services, such as social services, housing, or other support systems, might be needed.

There were limitations of this study that warrant discussion. The primary utilization measure was based on participant self-report and was not verifiable with actual utilization data from electronic medical records or claims. However, evidence suggests that patients are more accurate in reporting low-frequency events, such as emergency room visits, than high-frequency events, such as office visits (15). Because LG-CC was implemented for individuals between their baseline and six-month assessments, this study assessed medical care utilization between six and 12 months after receiving LGCC. Finally, there was limited information on the types of care management services provided.

The increased need among health care payers to promote more efficient care has led to a growing interest to implement more proactive models of care to better manage services for patients with chronic general medical and mental health conditions. At the same time, these models of care may not be adequate, especially given that they do not address the socioeconomic determinants of health that exist beyond the practice setting, such as poverty and homelessness. In this study we found a greater likelihood of emergency department use in a patient population with bipolar disorder among those a history of homelessness compared with those without a history of homelessness, despite enrollment in a CCM-based program. These individuals may have unmet service needs that require additional management beyond standard care management. Health care systems deploying CCM-based programs should invest in more resources that address these social determinants of health, which are important predictors of emergency department utilization. Future studies may also require the further intensification of evidence-based CCMs for individuals with serious mental illness in order to potentially mitigate critical service encounters and improve outcomes.