Use of Treatment Services in a Comparative Effectiveness Study of Bipolar Disorder

Bipolar disorder is a severe, chronic, and disabling mental illness characterized by episodes of mania, hypomania, and depression (1). Patients with bipolar disorder often have multiple psychiatric and general medical comorbidities (2–6). Their risk of having at least one general medical comorbidity is increased threefold compared with persons without bipolar disorder (7), and chronic medical disorders are correlated with a more severe and earlier course of illness (7,8). Thus bipolar disorder often requires extensive, lifelong care, making this illness one of the most difficult and expensive conditions to treat worldwide (9,10).

Persons with bipolar disorder who do not use counseling and medical services to help with the disorder’s impairment tend to experience a worse course of illness (4,8). Reports of how these patients use services are inconsistent. For example, in some studies, individuals with bipolar disorder used more general medical, psychiatric, and total services compared with the general population (3,11), whereas other studies concluded that these individuals sought fewer general medical services than sought by the general population, resulting in comorbid conditions that went undiagnosed and untreated (12–15). The Lithium Treatment–Moderate Dose Use Study (LiTMUS) found that only 61% of participants with hypertension and 48% of participants with hyperlipidemia had received diagnoses of these conditions (5). Furthermore, much of this limited work has been done with populations of military veterans; hence service utilization research with broader bipolar outpatient populations is much needed.

Studies have addressed specific factors that may affect service utilization. The following have been associated with increased service utilization: being middle-aged or Caucasian; having insurance; having a greater number of comorbid medical and psychiatric disorders; and having dyslipidemia, obesity, comorbid anxiety, more severe psychopathology, impaired functioning, and perceived need for help (6,7,15,16). People who are in a racial-ethnic minority group, single, and currently or recently homeless are less likely to seek services (15,17,18). Younger individuals and females are also more likely to report needing help but are less likely to seek services (15,17). In this study, we further investigated the prevalence and correlates of service utilization in bipolar disorder.

We expected that individuals who were older and Caucasian with greater severity of illness, symptoms, and side effects and lower functioning would use more services. We also expected that individuals with more psychiatric and general medical comorbidities would have a higher frequency of using all treatment services and that Caucasian, younger, college-educated, and single individuals would seek counseling services.

Methods

Procedure

LiTMUS was a six-month multisite, parallel-group, randomized comparative effectiveness trial. LiTMUS was designed to assess the efficacy of flexible, moderate doses of lithium in the context of optimized, personalized treatment (Li+OPT) compared with OPT without lithium. All participants were prescribed at least one mood stabilizer, as defined by the Texas Medication Algorithm Project (19). Details of the rationale, design, and methods of LiTMUS are reported elsewhere (20).

Participants

The study protocol was approved by the institutional review boards of the six study sites, and participants provided verbal and written informed consent prior to participation. A total of 283 individuals with bipolar disorder enrolled in LiTMUS between 2008 and 2009. The Cornell Service Index (CSI) was offered only at weeks 12 and 24; because of attrition before week 12, this report includes data from only 246 of the 283 participants. LiTMUS offered broad inclusion and limited exclusion criteria to achieve a diverse and relatively generalizable study population. Potential participants were excluded if they required acute or inpatient care or detoxification from alcohol, opiates, or barbiturates; were pregnant or breastfeeding; had renal impairment or high serum thyroid stimulating hormone concentration (either of which would put the patient at risk if assigned to the lithium condition); had previously experienced inadequate efficacy or tolerability in an adequate-lithium trial; required inpatient care; or could not comply with study procedures.

Measures

To assess current and lifetime DSM-IV diagnoses of bipolar and comorbid axis I disorders among participants, we used clinician-administered structured diagnostic interviews, the Extended Mini-International Neuropsychiatric Interview, the Structured Clinical Interview, and the Structured Clinical Interview for DSM-IV Substance Use Disorder Module (21,22).

Mood symptoms, side effects, and functioning were assessed at each study visit. Single-blind raters and unblinded clinicians measured severity of bipolar symptoms with the Clinical Global Impression Scale for Bipolar Disorder Severity (CGI-BP-S), a modified version of the Clinical Global Impression Scale tailored for bipolar disorder (23). Depressive symptoms were assessed with the Montgomery-Åsberg Depression Rating Scale (MADRS) (24), and manic symptoms were assessed with the Young Mania Rating Scale (YMRS) (25).

Medication side effects were monitored with the Frequency and Intensity of Side Effects Rating (FISER), a reliable and validated self-report measure of intensity, frequency, and burden of side effects (26). To evaluate overall functioning and life satisfaction, raters also completed the LIFE–Range of Impaired Functioning Tool (LIFE-RIFT) with participants at weeks 0, 12, and 24 (27).

Service utilization was assessed with the CSI (28), a standardized measure of health service use with good interrater and test-retest reliability among adults without cognitive impairment (28). This measure captures the previous three months of service use and records the type, provider, site, frequency, duration, and cost of visits. Every consultation, visit, or hospitalization, regardless of location or duration, was counted as one service used. Categories for all services include outpatient psychiatric or psychological, outpatient medical, professional support, and intensive services (including emergency department visit, inpatient hospitalization, and participation in a day treatment program). Providers listed on the CSI range from general physicians, psychiatrists, and psychologists to religious leaders, counselors, and peer supporters. Participants in LiTMUS completed the CSI at week 12 (services used in the first three months of the study) and week 24 (services used in the last three months of the study).

For the purpose of this article, general medical services are defined as medical visits and ambulatory visits, and counseling services encompass clinical case management, psychotherapy (individual, group, or couples-family therapy) and alcohol or drug treatment visits. Study visits were not included in the count of services utilized.

Statistical analysis

All statistical analyses were completed with version 9.2 of Statistical Analysis Software. Using marginal models, we examined the relationship of clinical and demographic features with participants’ use of services; model fit assumed a natural log link function, a Poisson variance, and a common within-patient association. Correlation matrices were estimated with generalized estimating equations. Assuming that each predictor variable had a similar effect in each 12-week period, we included in each model a fixed effect for visit in order to adjust for any variability found between the two 12-week periods. We assumed a Poisson variance because count variables often arise from a Poisson distribution, and the log link function allows for ease of interpretation. Using this model specification, one can interpret the coefficient as an estimated incidence rate ratio (IRR) between levels of the given predictor variable. Empirical standard error estimates were used to test whether the coefficients (IRRs) were equal to the null value of 1.

Simple linear regression was used to test whether the mean of each continuous variable at week 12 differed significantly between therapy groups. Logistic regression was used to test whether the odds of seeing a therapist at week 12 were significantly different between dichotomized groups of each unordered categorical variable. We also conducted multivariate analyses to examine the association of service use with clinical variables. We used forward selection to build these models and used an entry level of p<.10.

Of the 283 randomly assigned participants, 246 had outcome data at week 12, and 236 continued to week 24. To examine the possible effects of attrition, we conducted t tests for continuous variables and Fisher’s exact tests for categorical variables to compare the 37 participants who dropped out of the study before week 12 (not included in the above analyses for lack of CSI data) with the 246 participants who were included in our analyses.

A significance level of .05 was used to test each hypothesis, and because of the exploratory nature of this study, no adjustment for multiple hypothesis testing was made.

Results

Participants in the two treatment groups (Li+OPT versus OPT only) did not significantly differ in the total number of services used in either 12-week period. There were also no significant associations between treatment group and use of medical services or counseling services. At week 12, the mean±SD number of total services reported to be used in the prior three months was 5.40±10.00. The 246 participants used 1.07±2.17 medical services and 1.98±6.03 counseling services. At week 24 (N=236), they reported using 5.90±11.20 total services, 1.31±3.23 medical services, and 1.81±4.98 counseling services. The distribution of total services was positively skewed, with over 80% of participants (N=189) recording between zero and ten service visits (range=0–93). Number of medical services and number of counseling services were also positively skewed but had lower counts than total number of services. This was expected because we counted only a subset of total services.

We found that outpatients with higher depressive symptoms (on the MADRS) and manic symptoms (on the YMRS) had higher rates of using medical services (Table 1). Specifically, ten-unit increases in MADRS score and YMRS score resulted in a 20% and 48% increase, respectively, in rate of using medical services. MADRS, YMRS, and CGI-BP-S scores were marginally associated with total number of services (p<.06). Therefore, a ten-unit increase on the MADRS and YMRS resulted in 14% and 25% higher rates, respectively, of using all services. A one-unit increase on the CGI-BP-S resulted in a 13% higher rate of using medical services. Adjusting for visit, analyses indicated that both side-effect intensity and side-effect interference (FISER) were associated with the use of all services such that a one-unit increase in side-effect intensity and interference resulted in an 11% and 17% increased rate, respectively, of using all services (p<.01), but frequency of side effects was not associated with use of treatment services.

Participants with psychiatric conditions comorbid with bipolar disorder had higher rates of using all types of treatment services as well as medical and counseling services specifically (all p<.05) (Table 1). Each additional psychiatric comorbidity resulted in a 25% higher rate of using all services, a 35% higher rate of using counseling services, and an 11% higher rate of using medical services. Those who had an anxiety disorder (posttraumatic stress disorder, panic disorder, agoraphobia, obsessive-compulsive disorder, generalized anxiety disorder, social phobia, or a specific phobia) had a 25% higher rate of using all types of treatment services (Table 1). Given the association of anxiety disorders and use of services, we conducted follow-up analyses to examine specific anxiety disorders. We found a strong association between using more treatment services and having obsessive-compulsive disorder (IRR=2.21, p<.01) but not panic, phobia, generalized anxiety disorder, or posttraumatic stress disorder.

In regard to medical comorbidities, participants with high (exceeding 100 mg/dl) versus normal fasting plasma glucose used more than twice as many medical services but nearly half (46%) as many counseling services (Table 1).

Each multivariate model included an effect for time, forced into the model during variable selection, and considered the entry of all covariates in Table 1. The effects in the multivariate analysis were adjusted for all other covariates that entered the model. Analyses showed that number of psychiatric comorbidities (IRR=1.29, p=.003), obesity (IRR=.63, p=.03), and side-effect interference (IRR=1.19, p=.02) independently predicted use of all services. High glucose (IRR=2.25, p=.002) independently predicted use of medical services, and manic symptoms marginally predicted use of medical services (IRR=1.04, p=.07). High glucose (IRR=.53, p=.05) and side-effect interference (IRR=1.21, p=.02) predicted use of counseling services.

Participants who were older (middle age) (IRR=1.02, p=.01), had lived in the United States longer (IRR=1.02, p=.01), and were unemployed (IRR=1.84, p=.01) had higher rates of using all services, whereas use of psychotherapy was not significantly associated with any demographic or clinical variable (Tables 2 and 3).

In regard to the attrition analyses, we found that the dropout group had a higher proportion of men (p=.04), single or never-married individuals (p=.04), unemployed individuals (p=.04), and diagnosis of a substance use disorder (p=.02).

Discussion

Individuals with bipolar disorder experience a substantial burden of medical and psychiatric comorbidity, yet reports of how they use treatment services have been inconsistent (2–5,7,8). This study contributes to the limited but emerging literature on service utilization in a relatively representative population of outpatients with bipolar disorder. Neither randomization to low or moderate doses of lithium compared with no lithium nor progression from week 12 to week 24 of the study appeared to affect service utilization, but we found several predictors of differential service use among outpatients with bipolar disorder. This study furthers our understanding of health behavior (personal health practices and use of health services), a key component of understanding health service use and improving outcomes, including perceived and evaluated health status and consumer satisfaction (29).

Current psychological symptoms of bipolar disorder were associated with increased utilization of medical services, but not counseling services. Furthermore, compared with depressive symptoms, manic symptoms were more strongly associated with service utilization. Thus it seems that a more severe course of mental illness (particularly mania) may encourage seeking treatment for general physical but not psychological problems. Of note, only the association with mania persisted when the multivariate analyses were conducted, suggesting that several of the variables (depression and anxiety, for example) were collinear or did not independently have strong effects on service utilization (Table 1).

Participants who rated their side effects as more intense, frequent, and burdensome tended to seek not only medical services but also counseling and more overall services than sought by other participants—a finding that persisted in the multivariate analyses. These data suggest that physical symptoms, as opposed to psychiatric and mood symptoms, may prompt individuals with bipolar disorder to seek counseling services. Physical symptoms may have less mutable characteristics (degree to which they can change service utilization), but they still have some degree of mutability, and this finding can inform clinicians about whom to target for services (such as individuals with high psychological distress) (29). These discrepant findings highlight the need for future research to assess the demographic and psychiatric correlates of seeking counseling services, particularly because other studies have found that individuals with bipolar disorder do not tend to seek counseling services (30,31).

We also found that among participants with bipolar disorder, characteristics with low to no degree of mutability generally predicted their likelihood to seek services and specifically predicted utilization among those who were of middle age, who lived in the United States longer compared with others, and who were unemployed. These findings suggest that individuals who are younger, recent immigrants, or unemployed may underutilize services and could benefit from close monitoring by providers when they do seek services. We also found that men, single or never-married participants, unemployed participants, and substance users tended to drop out of the study at greater rates than others, which may suggest that these individuals are at higher risk for not adequately using medical services.

Psychiatric comorbidities, particularly anxiety disorders, predicted higher rates of using all treatment services, further suggesting that individuals with more comorbid diagnoses have more insight into their need for help and thus seek more treatment services than individuals with just one diagnosis (15). Anxiety has been shown to increase concern for one’s health and report of somatic symptoms (32). High glucose levels predicted the use of medical and counseling services (also shown in the multivariate models). In the multivariate model, obesity predicted the use of all services but not counseling or medical services specifically. These latter findings are particularly notable because obesity is associated with low utilization of counseling services (33), which in turn may contribute to how this condition affects individuals’ overall health (34). The known health risks of obesity alone (35), compounded by the worsened outcomes seen among obese patients with bipolar disorder (36), highlight the importance of understanding the complex relationships between obesity and use of services because this will further our knowledge of both predisposing (demographic characteristics and comorbid diagnoses) and need (perceived and evaluated illness) factors of health utilization (37).

These results are consistent with extant research showing that bipolar disorder renders high medical service utilization and associated costs for both patients and their insurance companies (10,38,39). Consequently, general medical providers may be treating symptoms or comorbid conditions of bipolar disorder, such as depressed mood, insomnia, fatigue, agitation, cardiovascular disease (40), or respiratory disorders (7), as well as observing patterns of increased treatment utilization without necessarily being aware of the bipolar diagnosis. This highlights the possibility of fragmented care for this group of patients, who often have multiple symptoms; coordinated prevention management of bipolar disorder would likely decrease its financial burden and improve the likelihood of sustained remission (10,41).

Interpretation of these results must be considered within the context of the methodological limitations. First, the CSI does not differentiate between using multiple services and using the same service multiple times, nor does it capture reasons for seeking a service or distinguish between general and specialized care. Second, the study relied on patient self-reports of service utilization and did not assess whether participants used services appropriate for their individual needs. Observations were not purely naturalistic, given that they were collected in the context of a clinical trial, and thus limit the generalizability of the findings. This context could influence service use, either increasing it as frequent study visits made participants more aware of health care needs or decreasing it as the clinical trial offered consistent clinical attention. Furthermore, although diverse, the study population comprised individuals willing to participate in research and thus cannot fully represent the entire population with bipolar disorder. The largely null results for the use of counseling services may be accounted for by a potential statistical power issue in that few participants sought such services and did so infrequently. Finally, the direction of causality of these relationships is speculative.

Conclusions

In short, our data show that outpatients with bipolar disorder sought medical and counseling services at specific times in their course of illness and overall health. Results suggest that treatment providers should pay special attention to characteristics with low mutability (middle age, recent immigration, or current unemployment) to ensure that persons have equitable access to health services (29). Data also highlight characteristics with higher mutability, such as worsening manic symptoms and high burden of side effects, to help providers determine areas of focus in their treatment planning and evaluations. Overall, these analyses helped to clarify the behavioral health service model for bipolar disorder by offering data on both population characteristics (predisposing and need factors) and health behavior (personal health practices and use of health services); this information will ultimately help to optimize overall outcomes (perceived and evaluated health outcomes and consumer satisfaction) (28,37).