Several aspects of these results are worth noting. Despite the observed improvements in quality of care, there is room for further improvement. In particular, the sharp decline in the likelihood of receiving psychotherapy is concerning. Although it is unclear what frequency or duration of therapy would be appropriate, it seems reasonable that receiving at least one session in a given year would have therapeutic value, given the chronic and disabling nature of bipolar I disorder.
The pharmacotherapy-specific measures indicated improvements in quality over time. In the case of anti-depressant prescribing, this improvement represented a reverse in the trend observed in 1994. These changes likely have at least one of two explanations. In 1996 valproate received approval from the Food and Drug Administration for acute mania. Thus the increase in lithium, valproate, or carbamazepine prescribing may be explained by the expansion of treatment options. Also, over the course of this time series, there was increasing consensus in the literature as to appropriate pharmacotherapy for bipolar disorder. This increased consensus could account for improved quality in lithium, valproate, or carbamazepine and antidepressant prescribing for bipolar disorder. This analysis did not include pharmacotherapy measures recommended by guidelines after the year 2002. The addition of newer pharmacotherapies with demonstrated bipolar disorder efficacy—such as lamotrigine and second-generation antipsychotics, including risperidone and olanzapine—likely would have magnified the improvements we observed over time in maintenance-phase pharmacotherapy for bipolar I disorder.
There was a decline in the likelihood of receiving any psychotherapy in 1994 and a further, sharper decline in 1999. It is possible that the decline in psychotherapy is due to differences in patient characteristics over time. In our data, compared with 1991, the bipolar I detection rate increased by 39 percent in 1994 (similar comparison is not possible for 1999 since it is a sample matched to 1994). However, with the exception of a decline in co-occurring psychiatric diagnoses across time, there is little difference in the observable patient characteristics in our cohort (which were also controlled for in the regression model). Still, patients may have differed in unobservable characteristics. However, these characteristics would not have affected this measure because at least one psychotherapy session in a given year would be reasonable. Additionally, these results are consistent with those of other studies that demonstrate a decline in psychotherapy for psychiatric illnesses, such as depression and schizophrenia (24
). Thus there is literature suggesting that our results are due to circumstances independent of the clinical characteristics of the patients in this study.
The decline in psychotherapy may represent a substitution of CPT-4 coding by psychiatrists without a real change in the visit content—that is, rather than coding for psychotherapy, psychiatrists are billing for medication management but are also providing psychotherapy. However, we believe that this is unlikely given the financial disincentive for psychiatrists to provide psychotherapy compared with medication management (26
). It would also be difficult to provide medication management and psychotherapy with demonstrated efficacy for bipolar disorder in the time constraints of a medication management visit. Although we cannot know that the psychotherapy coded as such in these claims also adheres to this evidence base, we know at least that psychotherapy is coded as being provided in that visit. This is not the case with psychotherapy delivered in a 90862 CPT code; in that circumstance the procedure is much less likely to conform to evidence-based practice. Finally, as previously mentioned, our results showing that psychotherapy visits declined over time are consistent with other literature (24
). Hence, the likelihood of changes in coding without changes in the actual services provided during the visits is unlikely.
Prior studies examining the effect of managed behavioral health care in treatment utilization specifically for major depression have observed a shift away from psychotherapy and toward increased antidepressant utilization (24
). Some investigators interpret these results as indicating there is a “substitutability” between the two treatment modalities for depression (27
)—that is, that therapy to at least some extent can be replaced by medication. However, our results suggest, at least for bipolar disorder, another possible explanation of the forces guiding treatments received. Our pharmacotherapy measures included one measure that recommends prescribing a medication (that is, lithium, valproate, or carbamazepine) and another that required refraining from a poor quality measure (that is, prescribing an anti-depressant without lithium, valproate, or carbamazepine). Both of these measures improved over time, whereas the therapy indicator worsened. On the basis of these results, an alternative hypothesis is that psychosocial treatments are constrained in ways that pharmacotherapies are not—rather than the hypothesis that pharmacotherapy acts as a substitute for psychotherapy.
It is also important to note that although one would expect to see similar results in managed behavioral health carve-out arrangements (because carve-outs typically are not financially at risk for pharmacotherapy but are for inpatient and outpatient visit utilization), these enrollees were served under a variety of insurance arrangements. The care of some, but not all, enrollees was managed by carve-outs. Regardless of the insurance arrangement for delivering the mental health care, this analysis indicates different psychotherapeutic modalities are under different constraints and respond to these constraints differently.
Our results differ from those of Blanco and colleagues (12
), who found no change in pharmacotherapy prescribing between 1992 and 1999 of non–dopamine-blocking antimanic agents (that is, lithium or anticonvulsants) and of antidepressants in the absence of lithium or anticonvulsant agents. Although both studies were similar in that bipolar diagnoses were determined by psychiatrists and occurred over similar study years, the two studies used different methodologies and are not directly comparable. Additionally, Blanco and colleagues employed an expanded definition of non–dopamine-blocking antimanic agents by including gabapentin, topiramate, and lamotrigine—three medications that as of 1999 had unclear therapeutic efficacy for bipolar disorder.
It is also notable that in this cohort of persons with bipolar I disorder, there was an improvement in guideline-concordant pharmacotherapy over time. Although one cannot assume that these improvements are directly attributable to the guidelines, guideline publication does signal an increasing consensus in the medical literature, as well as promotes awareness of that consensus. These observed improvements are in contrast to other medical conditions and procedures, such as diabetes or cesarean section rates, for which guideline publication has not been previously associated with improvements in clinician practice (29
). Similarly, in the treatment of other psychiatric illnesses, the literature suggests that guideline publication has rarely been associated with improvements in clinician practice (31
This study has several limitations. It used administrative claims to identify persons with bipolar I disorder. Although structured interviews with patients and chart reviews have greater accuracy in establishing a bipolar cohort, Unützer and colleagues (14
) demonstrated that claims data have demonstrated validity in analyses utilizing a bipolar diagnosis to establish the cohort and assess population-based quality of care. We used more stringent criteria in our cohort algorithm than Unützer and colleagues. Finally, although our algorithm was developed to balance maximizing the “true positive” rate while minimizing the “false negative” rate, it is likely that there are persons with bipolar I disorder who were not given the diagnosis in the claims, and therefore, we cannot comment on the quality of care for those enrollees.
Furthermore, the literature points to specific psychotherapies, such as psychoeducation and cognitive-behavioral therapy, as having demonstrated efficacy in treating bipolar disorder (32
). However, information about the content of psychotherapy is not obtainable in claims data, nor is it obtainable from chart review. Thus we cannot know if the psychotherapy services delivered to this cohort would have been these efficacious treatments. Still, it seems reasonable to expect, given the highly debilitating and chronic nature of bipolar disorder, that patients would receive at least one psychotherapeutic session that could provide psychoeducation to the patient or to his or her family.
Another limitation is that although these quality indicators were informed by published bipolar disorder guidelines, they are very loose indicators of quality, setting a minimum benchmark of whether patients received at least one prescription or therapy session. We did not measure the dosage or duration of these treatments.