Glucose Control and Medication Adherence Among Veterans With Diabetes and Serious Mental Illness: Does Collocation of Primary Care and Mental Health Care Matter?

The prevalence of diabetes among individuals with serious mental illness (SMI) is estimated to be 16–25% or two to three times the rate of the average population (1,2). This increased prevalence has been noted for decades in people with SMI, notably in people with schizophrenia and other severe psychotic disorders (3). The exact cause of this relationship is unknown, but newer antipsychotic agents likely play some role (4). Regardless of the cause, diabetes and cardiovascular disease contribute significantly to the observed increased mortality and morbidity among persons with SMI (5). Within the Veterans Health Administration (VHA), the difference in the prevalence of diabetes and mortality rates between individuals with and without SMI are smaller and seem to be driven by, among other things, atherosclerosis and hypertension (6). Given that veterans with mental health needs are more likely to use the VHA for physical and mental health care than veterans without mental health needs and the high prevalence of diabetes in veteran populations (7), this creates an important subpopulation of patients with comorbid SMI and diabetes that rely heavily on the VHA for care.

Clinical care models that aim to deliver disease specific care (such as diabetes care) to veterans with SMI may improve both SMI and disease-specific healthcare outcomes. One approach is to collocate primary care healthcare professionals in mental health clinics (8). Collocation is thought to facilitate communication between healthcare professionals as well as coordination of care and has been shown to improve subjective general health for veterans with SMI (9,10). The VHA has invested heavily in collocated care, placing mental healthcare professionals in primary care clinics to help primary care healthcare professionals better manage patients with co-occurring mental illness (11); however, there has also been a move to increase the presence of primary care healthcare professionals within mental health clinics. Two recent evaluations of the VHA system have shown that when primary care healthcare professionals are collocated in the mental health clinics, veterans with SMI are more likely to receive cardiovascular risk factor evaluations and appropriate preventive care than veterans with SMI not seen in these clinics (12,13). Another evaluation from the same group indicates that veterans seen in collocated care are also less likely to be admitted for ambulatory-sensitive admissions such as for asthma, congestive heart failure, and diabetes (14).

By overcoming barriers to care as well as coordinating care, collocation inherently embodies many of the goals of the Patient Aligned Care Teams, the Veterans Affairs (VA) medical home initiative, which aims to improve the provision of accessible, coordinated, comprehensive, patient-centered care. Determining if they also improve clinical outcomes will make it clearer for the VHA the value added from collocated care. Toward this end, we sought to determine whether receiving collocated care is associated with better diabetic outcomes in veterans with comorbid diabetes and SMI. We compared glucose control and diabetes medication adherence in veterans from three different VA medical centers, comparing outcomes in individuals with diabetes and SMI receiving collocated care to those not receiving collocated care (which we call usual care). We hypothesized that veterans with SMI and diabetes receiving collocated care would have both better glucose control and medication adherence than those not receiving collocated care.

We performed a cross-sectional, observational cohort study of patients with type 2 diabetes and SMI seeking care from three VA medical facilities: two sites provide both collocated and usual care, and one site provides only usual care. Potential participants were identified through the VHA electronic medical record. To be eligible, the potential participant had to have: a diagnosis of diabetes (any 250 ICD-9 code); received a prescription through the VHA for an antihyperglycemic medication (oral or injectable) at some time between 12 and 24 months prior to enrollment; and a diagnosis of an SMI that we defined as schizophrenia, schizophreniform disorder, schizoaffective disorder, bipolar disorder, manic affective disorder, delusional disorder, or other nonorganic psychoses (ICD-9 code: 295.0–295.99; 296.0, .1, .4, .5, .6, .7, or .8; 297.0, .1, .2, .3, .8, or .9; and 298.0, .1, .2, .3, .4, .8, or .9). Due to potential miscoding of SMI (15), SMI was confirmed with the potential participant’s mental health healthcare professional. Healthcare professionals were given an opportunity to remove patients from contact lists if they did not think it was appropriate for us to contact the patient. We then mailed potential participants a letter describing the study including a number that they could call for more information. At two sites, this was followed by a phone call, and at the third site, patients were approached by clinic staff at scheduled appointments. Recruitment procedures at the different sites varied based on what each local Institutional Review Board considered acceptable.

Interested patients met individually with a research assistant who explained the study procedures, obtained informed consent, administered a quiz assessing comprehension of the consent (16), and administered the research questionnaires asking about demographics, diabetes history, general clinical characteristics, and psychiatric symptoms and functioning. Patients who did not comprehend the consent were reimbursed for their time but not included in the study. Questions were read to patients, and responses were recorded by the research assistant. Patients held large-font, laminated cards with answer options to help them remember and select appropriate responses. Following completion of the questionnaire, the research assistant escorted participants to the clinical laboratory where they had a glycosylated hemoglobin (HbA_1c) drawn. In addition, we obtained, from the medical record, the patients last documented HbA_1c and date, their BMI, and pharmacy records for all antihyperglycemic, mood-stabilizing, and antipsychotic medications for the 12 months preceding the survey administration date. The Institutional Review Board at each institution approved the study protocol.

Our outcomes of interest were glucose control as measured by the HbA_1c drawn on the day of the survey administration and antihyperglycemic medication adherence as measured by the medication possession ratio (MPR) (17) over the prior year for diabetes medications and the self-reported diabetes-specific Morisky Medication Adherence scale (18). We chose to include two measures of adherence since no single measure is considered a gold standard, and there are concerns that self-report might suffer from social desirability bias (19). We calculated a composite MPR for all antihyperglycemic medications including insulin from electronic pharmacy data for the 12 months prior to the day of the survey administration. The MPR calculates the percent of days in the past 365 in which the patient was in possession of the appropriate antihyperglycemic medications. To calculate the MPR for insulin, we used the methods developed by Kleinman et al. (20). The Morisky Medication Adherence scale includes four general questions about medication adherence without a specified timeframe. We evaluated HbA_1c and MPR as continuous variables. We dichotomized self-reported adherence as perfect versus nonperfect adherence.

Our main independent variable of interest was receipt of collocated care (yes/no). The three study sites were chosen because each provides a different model of primary care for patients with SMI. Table 1 summarizes the differences between each site. Site 1 integrates primary care healthcare professionals into the mental health clinics. At this site, the primary care healthcare professionals deliver care in the exact same space as the mental health healthcare professionals, participate in mental health clinic conferences, and discuss comanagement of care on a routine basis. We call this model integrated collocated care. Site 2 collocates primary care healthcare professionals in a specialized site where a majority of the mental health care is delivered such that the primary care healthcare professionals at this site predominantly care for veterans with mental illness. At this site, the primary care healthcare professionals are one floor above the mental health healthcare professionals, and although they only see patients who are seen in the mental health clinics, they do not formally comanage care. We call this model specialized collocated care. At Site 3, there is no special arrangement between primary care and the mental health services; however, as is the case in all VA medical centers, healthcare professionals use the same electronic medical record. We call this model usual care. While Site 1 and 2 both strive to accommodate most of their SMI patients in the collocated primary care clinics described, some SMI patients receive care from routine primary care clinics, and these patients were also designated as receiving usual care. When the study began, neither site had formal guidelines regarding who was referred to collocated care.

Covariates included demographic characteristics: age, sex, race, education completed, and marital status. Diabetes-specific covariates included: how long the person had diabetes; if they were prescribed oral antihyperglycemic medications, insulin, or both; the total number of diabetes medications prescribed in the previous 12 months; the number of diabetes-related comorbidities (including ophthalmologic disease, cardiovascular disease, peripheral vascular disease, cerebrovascular disease, renal disease, and complications from neuropathy); and whether they had seen an endocrinologist in the past year. Additional clinical covariates were smoking status, BMI, and a self-reported Charlson Comorbidity Index (a measure of overall comorbidity) (21,22). Mental health covariates included the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) memory index score (an indicator of the extent of cognitive impairment) (9,23) and two subscales from the revised Behavior and Symptom Identification Scale assessing psychotic symptoms and depression (24). The RBANS memory index score is expressed as an age-adjusted standard score with a mean of ∼100 and an SD of ∼15. Revised Behavior and Symptom Identification Scale subscales are weighted scores based on five-point rating scales ranging from 0 to 4. Finally, we also included whether the individual received all of their care from the VHA.

We evaluated the unadjusted association between collocation and all covariates as well as the outcomes of interest. We used t tests to evaluate differences in continuous variables and χ² tests to evaluate differences in categorical variables. Linear regressions were used for the adjusted analysis of HbA_1c and MPR, while logistic regression was used for the analysis of self-reported adherence. Because of the limited sample size and the large number of potential covariates, we performed stepped regressions, forcing collocation (yes/no) into the model to determine the association between collocation and each of the three outcomes. Covariates were retained in the final models if the association with the outcome of interest was P < 0.05. We used these same techniques to examine the associations between the type of collocation (integrated or specialized) and the outcomes of interest. All analyses used nonresponse weights that make survey responses representative of the eligible populations at each site based on available demographic and clinical characteristics. Our aim was to enroll 100 patients per group. Assuming a moderate effect in any outcome (Cohen d = 0.40), this would provide 80% power to detect statistically significant and clinically meaningful differences between groups with a two-tailed α = 0.05.

We enrolled 363 individuals (212 from usual care, 151 from collocated care [89 from integrated collocated and 62 from specialized collated care]) out of 681 eligible participants (response rate 53%). Compared with nonparticipants, participants were more likely to be black (43 vs. 38%; P < 0.001), have bipolar disorder (41 vs. 25%; P < 0.001), and be younger (mean age 59 vs. 62 years; P < 0.001); there was no difference by sex or electronic medical record–documented last HbA_1c.

Overall, on average participants were 59 years old and had diabetes for 10 years (Table 2). A total of 54% had completed high school or less education, 59% were on oral diabetes medications alone, 40% had seen an endocrinologist in the past year, the mean BMI was 33, the mean psychotic symptoms score was 0.99, and the mean depression score was 1.36. Patients seen in collocated care were more likely to be black (48 vs. 35%; P = 0.044), more likely to be a current smoker (49 vs. 44%; P < 0.001), less likely to have seen an endocrinologist in the last year (31 vs. 46%; P = 0.017), and had a lower RBANS score (70 vs. 76; P = 0.001). Eighty-eight percent of the sample was on a psychiatric medication, 70% were on an antipsychotic, 53% were on a mood-stabilizing medication, and 36% were on both. There were no differences in receipt of either class of psychiatric medication by care model.

In unadjusted analyses (Table 3), there were no glucose control or medication adherence outcome differences by collocation. The mean HbA_1c was 7.4% (57 mmol/mol), the mean MPR was 80%, and 51% of the sample self-reported perfect adherence to their diabetes medications. In adjusted analyses (Table 4), patients seen in collocated care tended to have better HbA_1c levels and MPR values, but this was not statistically significant at a P < 0.05 level. The specific type of collocation did not alter these findings.

Results did not substantially differ when the HbA_1c and MPR were dichotomized at standard VHA cut points (>8% or >64 mmol/mol for HbA_1c and ≥80% for MPR). Similarly, the analysis of MPR did not differ substantially when insulin was omitted from the MPR calculation.

We found that in a population of veterans with comorbid SMI and diabetes, patients on average had good medication adherence (≥80%) and adequate glucose control as defined by the VHA (HbA_1c ≤8% or ≤64 mmol/mol) regardless of where they received primary care (25). While we found a trend toward better outcomes in those receiving integrated collocated care, it was not statistically significant, possibly because outcomes were quite good even in usual care. In general, the VHA seems to be meeting VHA diabetic recommendations for their patients with both SMI and diabetes.

In this study, the mean HbA_1c for all groups was 7.5% (58 mmol/mol) or less, which is on par with or better than national VHA averages (26). This value is also consistent with current VHA recommended goals for patients with diabetes (25), indicating that even in usual care, patients with SMI are achieving diabetes goals. This may in part be because the VHA system as a whole already embodies many of the attributes specifically valued in collocated care. First, there is one electronic medical record. Thus, as long as the care is being provided within a VHA system, all records are available to every healthcare professional, and records can be flagged such that specific healthcare professionals are forced to acknowledge the content of specific notes or results. Second, even if not physically located in the same group of offices, most VA medical center clinics are located in the same building or set of buildings, and appointments, even when not collocated, can be planned for the same day without the need for additional transportation or parking. Third, while patients might not be discussed in team meetings with mental health and primary care healthcare professionals, given the proximity of providers, urgent mental health or primary care issues can be addressed by walking the patient over to the appropriate healthcare professional. Finally, with an internal encrypted e-mail system, healthcare professionals can discuss specific patient issues without security concerns.

Another explanation that might make our results specific to the VHA includes the patient population. For the most part, the patients are all veterans and as such come from a specific culture that could make patients, especially patients with SMI, more responsive to authority figures such as medical professionals. Thus, these results may not be generalizable to a nonveteran population.

It is also interesting to note that those in usual care were more likely to have seen an endocrinologist in the past year. This may reflect the better ability of collocated healthcare professionals to address barriers to adherence in patients with SMI while usual care healthcare professionals may more frequently turn to specialists when trying to help their patients gain better control of their diabetes. Thus, while they may achieve the same results, the means may be different.

However, good diabetes control among veterans with SMI may have nothing to do with the specifics of VHA care since even outside of the VHA, patients with and without SMI have similar control (27), and self-reported perfect adherence is almost identical to that observed in this study (18). Diabetes control in patients with SMI may have to do with the sheer numbers of appointments these patients have and the many opportunities there are to address different medical issues. Another possible explanation is patients’ established experience with managing a chronic medical condition (i.e., SMI). Both of these potential explanations are consistent with literature indicating that adherence to diabetes medications in veterans with SMI is better than those without SMI and improves with visit frequency (28,29).

In contrast, our results are inconsistent with a recent national study by Kilbourne et al. (13) in which patients seen in VA medical centers with integrated collocated care were significantly more likely to have good blood pressure control but worse diabetes control. Our findings might differ because the study by Kilbourne et al. (13) did not distinguish which patients seen at centers with integrated collocated care actually received integrated collocated care, while in our analyses, we verified that patients with SMI were receiving primary care from mental health clinics.

It is also important to note that some patients had low HbA_1c (<6.5%, and one had an HbA_1c of 4.6%). Regardless of how low the HbA_1c was on the day of the study, the mean reported years with diabetes was >5 years, and all participants had been prescribed a hypoglycemic medication (thus it is unlikely that we accidentally enrolled patients without diabetes). However, it does raise the possibility that primary care providers are not closely monitoring glucose control in their patients with SMI and that some patients may be overly aggressively taking their medication or should not be on medications at all. This in and of itself is a concern since hypoglycemia can also have serious consequences (30). However, there was no indication that low HbA_1c values differed whether or not the patient’s care was collocated.

While large, this study has limitations. As a cross-sectional study, we can only examine associations and cannot claim causality. We included only three VA medical centers; thus, we cannot comment on all collocated care within the VHA. However, given how we collected data, we are able to assess multiple individual-level potential confounders. In addition, our findings regarding HbA_1c levels are consistent with national data evaluating glucose control in patients with comorbid diabetes and SMI (31). Both sites with collocated care also had SMI patients in noncollocated usual care, making it likely that those with the most severe disease were seen in collocated care. Although we adjusted for several measures of psychiatric severity (the RBANS, psychotic symptoms, and depression symptoms), there may have been unmeasured confounding, perhaps leading us to underestimate the benefits of collocated care. While we had adequate power to detect differences between those in collocated care versus usual care, we were underpowered to detect equally large differences between the different types of collocated care and usual care. Given the direction and magnitude of the β estimate for integrated collocated care versus usual care, a larger sample may have found these differences statistically significant. Regardless, the entire sample exhibited adequate diabetes control and adherence. Finally, we did not measure resource use and thus cannot make any definitive statements about cost or cost-effectiveness of collocated care. Setting up collocated clinics within the VHA generally requires reallocation of providers and thus is cost-neutral to set up. Unless collocated care leads to greater resource use (such as laboratories, appointments, studies, etc.), these clinics likely place limited economic burden on the VHA but may decrease resources available to the general clinic population by diverting them to collocated clinics.

Collocated care is gaining wide favor both inside and outside of the VHA, and it achieves many of the goals set out by the VHA in its efforts to become more patient centric. Future studies should evaluate patients’ perceptions of and satisfaction with care across different models of collocation as well as other important intermediate diabetic outcomes such as blood pressure, lipid, and weight control. While we cannot say that collocated care improved diabetic outcomes, it certainly was as good as usual care for the outcomes we evaluated and was perhaps less resource intense. Examination of patients’ experiences would be valuable to include in future research efforts. If collocated care improves SMI patients’ experience of care, a strong argument can be made for the model.

Funding. This work was supported by VA Health Services Research & Development IIR-07-124-3.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.A.L. developed the study design, obtained funding, supervised the acquisition of data, analyzed and interpreted data, and drafted the manuscript. A.W. acquired data as well as supervised the acquisition of data, organized the data analysis and interpretation, and edited the manuscript. E.L.M. performed the data analysis and edited the manuscript. S.V.E. and A.J.G. assisted with study design, acquisition of data, data analysis and interpretation, and edited the manuscript. J.K. and S.C.M. assisted with study design, data analysis and interpretation, and edited the manuscript. J.A.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.