General Medical Outcomes From the Primary and Behavioral Health Care Integration Grant Program

We selected three PBHCI sites for the quasi-experiment from the subset of grantees that were expected to serve sufficient consumers during the study period for adequate study power (750 consumers per site). Anticipated volume was based on early utilization data and SAMHSA project officer assessments. We applied additional criteria to include a sample of PBHCI sites with diversity in geographical location (for example, different states and urban and rural areas), consumer demographic characteristics, and approach to implementing PBHCI (for example, sites with and without a formal health clinic partner). PBHCI sites were included only if we could also engage a behavioral health clinic to serve as a matched control—a clinic located in the same state that served a similar consumer population but provided no or very limited primary care services.

The final sample of PBHCI sites included one program funded in 2009 and two funded in 2010. Service utilization data from all 56 grantees showed that the PBHCI sites included in the quasi-experiment had higher-than-average rates of primary care provider contacts and general medical screenings but generally did not differ dramatically from the set of unselected sites. [Tables comparing characteristics of the sites that were or were not selected for the quasi-experiment are available as an online supplement to this article.]

Participants were at least 18 years of age; had a primary psychiatric diagnosis of schizophrenia or other psychotic disorder, bipolar disorder, or severe depression; and were enrolled in a PBHCI program or received behavioral health care at a matched control site. Individuals were excluded if they were unwilling or unable to provide written informed consent or complete study questionnaires in English or Spanish (with assistance, if requested).

Consumers in the PBHCI programs enrolled in the program during (approximately) the second year of grant funding (August 2010–July 2011 for the 2009-funded grantee and August 2011–July 2012 for the 2010-funded grantees). These enrollment windows allowed us to evaluate the effect of participating in a PBHCI program for a full year (±6 months) within the time frame of the evaluation contract, while allowing for program start-up time and keeping the number of years of PBHCI funding consistent across sites.

Grantees proposed various criteria for enrolling consumers in the PBHCI program, as described previously (15); however, because of challenges meeting enrollment targets, clinics typically relaxed criteria to include all adults with serious mental illness. Enrollment criteria originally proposed by the quasi-experimental sites were inadequate connection to primary care and psychotropic medication use (site 1), inadequate connection to primary care and high-risk clinical factors (site 2), and all adults with serious mental illness (site 3).

Baseline data for PBHCI clinics were collected by clinic staff at the time consumers enrolled in integrated care. We drew these data from SAMHSA’s online data repository (Transformation Accountability system). PBHCI follow-up data were collected by a biometric screening contractor during a weeklong data collection event held approximately one year (±6 months) after individual consumers enrolled. Baseline and follow-up data for the control sites were also collected by the contractor during weeklong data collection events held approximately one year apart. We used clinic-collected (as opposed to contractor-collected) data for baseline measurements for PBHCI participants to minimize participant burden, given considerable program- and consumer-level data requirements for all grantees (15). Clinic-collected data were expected to be more variable than contractor-collected data (thus potentially limiting power) but were not expected to be systematically biased, given that baseline data were collected prior to clinics’ being selected for the study.

Indicators of general medical health included body mass index (BMI), blood pressure (systolic and diastolic), and self-reported smoking status. Blood samples were collected for plasma glucose or hemoglobin A1c (HbA1c) and lipids, including cholesterol—total, high-density lipoprotein (HDL), and low-density lipoprotein (LDL)—and triglycerides. Blood samples were drawn from individuals who were fasting or nonfasting; fasting status was accounted for in analyses.

In lieu of using general medical diagnoses, we identified consumers as “at risk” of general medical illness if their baseline health indicators exceeded standard normal ranges (systolic blood pressure, ≥130 mmHg; diastolic blood pressure, ≥85 mmHg (16); BMI ≥25 kg/m² (17); total cholesterol, ≥240 mg/dL; HDL cholesterol, <40 mg/dL; LDL cholesterol, ≥130 mg/dL, triglycerides, ≥150 mg/dL (18); fasting plasma glucose, ≥100 mg/dL; nonfasting plasma glucose, ≥140 mg/dL, HbA1c, ≥5.7% (19), and self-reported current tobacco smoking).

Data about participants’ demographic and functioning variables were collected by using an abbreviated version of SAMHSA’s National Outcomes Measures (20), including measures of education, employment, social connectedness, housing stability, substance use, and arrests (21). Clinics provided participants’ primary DSM-IV psychiatric diagnosis.

We conducted difference-in-differences analyses to estimate the average treatment effect of PBHCI services on general medical outcomes (22–25). We used propensity-score weighting to account for the nonrandom assignment of consumers to treatment and to adjust for differences between consumers at the PBHCI and control clinics in the demographic and functioning variables described above. Each consumer was assigned a propensity weight according to the following model: the inverse of the estimated probability of receiving the treatment actually received (PBHCI or control), given individual demographic and functioning variable values. This approach is similar to survey weighting methods that adjust samples to be representative of target populations (25). Balance diagnostics were performed to assess the adequacy of the propensity-weight model described above (26).

For each general medical outcome of interest, analyses were limited to consumers whose baseline values for that outcome were in the “at risk” range. For example, the sample for estimating the effect of PBHCI on BMI included only consumers with a baseline BMI over 25 kg/m². These restrictions were used to identify individuals who were in need of intervention and for whom a change in clinical indicators was desired.

Difference-in-differences analyses were conducted for consumers with both baseline and follow-up general medical data. Sensitivity analyses were performed to compare these results to a cross-sectional approach that also included consumers with only baseline or only follow-up data (24). All analyses were conducted by using SAS, version 9.2, software.

PBHCI and control clinics each aimed to recruit 300 consumers for the study. Across the three control sites, 793 consumers participated at baseline and 492 completed the one-year follow-up, yielding a 62% follow-up rate. Among the control-clinic consumers retained at follow-up, 469 (95%) had complete data for propensity-weight calculation and were included in difference-in-differences analyses. No directly comparable retention metric was available for intervention sites because baseline data for PBHCI consumers were retrospectively drawn from consumers who agreed to participate in the study at the follow-up data collection event. Across the three intervention sites, 1,049 consumers enrolled in PBHCI during the study baseline period (one year [±6 months] prior to follow-up), and 343 (33%) agreed to participate in the research study (in addition to receiving clinical services) and attend the follow-up, contractor-administered data collection event. Among these PBHCI participants, 322 (94%) had complete data for propensity-weight calculation and were included in outcome analyses. Sensitivity analyses revealed similar results for cross-sectional and repeated-measures analyses, suggesting neither selection into the PBHCI group nor attrition in the control group significantly affected outcome results; we present results for the subset of consumers with both baseline and follow-up data.

Baseline characteristics of the sample are shown in Table 1. Among PBHCI consumers, the mean±SD age was 42±12, and 59% were female; 71% were white, 16% black, 5% Hispanic/Latino, and 8% other race-ethnicity. The most common primary psychiatric diagnoses among PBHCI consumers were schizophrenia (28%), major depressive disorder (26%), and bipolar disorder (25%); diagnoses were similarly distributed among consumers in the control group.

Compared with the PBHCI sites, the control sites had more white and fewer black participants (p<.001). Consumers at the control site were also older (p<.001) than PBHCI consumers, and they were more likely to have completed high school (p<.01), to be employed or enrolled in school or job training (p<.05), to report social connectedness (p<.05) and housing stability (p<.001), and to report recent binge drinking at baseline (p<.001) but less likely to report illegal substance use (p<.001).

To adjust for baseline case-mix differences, we computed propensity-score weights using all variables shown in Table 1 to estimate treatment probabilities. After adjusting comparisons for propensity weights, we found no significant differences between consumers at PBHCI and control sites for any variables listed in Table 1, indicating successful balance across groups.

Table 2 shows mean values and proportions of consumers exceeding standard thresholds for health indicators at baseline. Compared with consumers at the control sites, PBHCI consumers were more likely to have elevated rates of diastolic blood pressure (30% versus 22%, p<.05), LDL cholesterol (28% versus 21%, p<.05), and tobacco smoking (62% versus 54%, p<.05). Consumers at the control site were more likely to have a high BMI (81% versus 74%, p<.05) and elevated rates of triglycerides (52% versus 38%, p<.001). [These variables were not included in case-mix adjustments because the difference-in-differences analyses accounted for group differences in baseline outcomes.]

Change in general medical outcomes for each group and the difference in differences between consumers in each group are shown in Table 3. Over the study period, consumers served at PBHCI clinics had statistically significant improvements in all general medical outcomes for which they were at risk, except BMI and HbA1c. Among control clinics, consumers had statistically significant improvements in all outcomes for which they were at risk, except LDL cholesterol and HbA1c.

After the analyses adjusted for case mix, the mean reduction in total cholesterol among consumers with cholesterol-related risk factors was greater by 36 mg/dL among PBHCI consumers compared with consumers at the control clinics (p<.01). The adjusted mean reduction in LDL cholesterol was greater by 35 mg/dL among PBHCI consumers compared with consumers at the control clinics (p<.001), and the increase in HDL cholesterol was greater by 3 mg/dL among PBHCI consumers compared with consumers at the control clinics (p<.05). No other differences between PBHCI and consumers at the control sites were observed.