Quality of Depression Care and Its Impact on Health Service Use and Mortality Among Veterans

Using a variety of data sources—including VHA outpatient and inpatient treatment files and pharmacy benefits manager, as well as Department of Veterans Affairs (VA) Beneficiary Identification Records Locator Subsystem and Social Security Administration death records—we constructed a database containing information from October 1999 through September 2005. Using the following methods, we extracted a final cohort of 205,165 patients with a new-onset depressive disorder who received at least one prescription for an antidepressant medication. This study was approved by the institutional review board at Baylor College of Medicine as well as the Michael E. DeBakey VA Medical Center Research and Development Committee in Houston.

Construction of our cohort began with examination of all VHA outpatient records from fiscal years 1999–2005. Patients with an ICD-9-CM code for depression (296.2x, major depressive disorder, single episode; 296.3x, major depressive disorder, recurrent episode; 300.4, neurotic depression and dysthymia; or 311.0, depression not otherwise specified) were extracted (N=764,510). Each patient was then assigned a depression index date to reflect the first date of a depression diagnosis. We then restricted our cohort to patients with new-onset depression by examining the 365 days before the identified index date. Depression was classified as new onset if patients had no ICD-9-CM diagnosis for depression in the prior 365 days and they had received no prescription for an antidepressant between 31 and 365 days before their initial diagnosis. We determined that patients who received an antidepressant 30 days or less before their depression diagnosis had received this treatment in relationship to the episode identified, and we therefore reclassified their depression index date as the date of the medication fill. Applying the new-onset criteria reduced our cohort to 363,347. Antidepressant medications included tricyclic antidepressants, monoamine oxidase inhibitor antidepressants, and other antidepressants (including selective serotonin reuptake inhibitors), as described in prior studies ( 23, 24, 25 ).

Because our primary aims focused on provision of and follow-up care related to antidepressant medications, we restricted our sample to patients who had filled a prescription for an antidepressant (N=234,370). Finally, we excluded 3,149 patients who died during the first 180 days after the depression index date, 7,908 patients who had a large number of inpatient bed days (36 or more days during the 180 days postindex date), and 18,157 patients who had comorbid diagnoses of schizophrenia or bipolar disorder.

Methods for measuring and interpreting application of depression care guidelines vary widely. However, current guidelines and the literature suggest that quality measures of depression care related to antidepressant medication supply ratios and adequate follow-up are potentially important predictors of patient and health service use outcomes ( 15, 21, 22, 26 ).

We dichotomized patients into categories representing adequate and inadequate supply of antidepressants. Supply was calculated for the number of days each patient was provided an antidepressant prescription (during the 84-day profiling period and based on pharmacy prescription fill data) divided by 84 days. Adequate supply of antidepressant medication, as in prior studies ( 10, 23 ), was established as a patient's possession of medication (medication possession ratio) for 80% or greater of the total number of days in a given period. Overlapping antidepressants were not counted in the days' supply.

The definition of the adequate number of follow-up visits for monitoring antidepressant use has fluctuated in the literature between three visits within the 84-day period after diagnosis ( 22 ) and three visits after the initial diagnosis ( 10 ). We used a criterion of three visits after the initial diagnostic visit as the standard for adequate depression care ( 10 ), consistent with current Health Plan Employer Data and Information Set indicators. Follow-ups were dichotomized as adequate and inadequate. A depression follow-up encounter was defined by Current Procedural Terminology codes on the basis of the following: any psychiatric or mental health visit regardless of accompanying depression diagnosis, any non-mental health visit when a depression ICD-9-CM code accompanied the visit, and any telephone-based service for which a depression ICD-9-CM code accompanied the encounter.

Several demographic and service use characteristics were used. Demographic factors included age, gender, race and ethnicity, marital status, and income (based on data from the Internal Revenue Service's average adjusted gross incomes by zip code). Race and ethnicity information was poorly populated in these national databases and therefore excluded from all predictive analyses.

Other information and health service characteristics from the medical record included patient service-connected disabilities; illness burden, represented by a relative risk (RR) score; and comorbid mental health conditions, which were restricted to anxiety, posttraumatic stress disorder (PTSD), and substance use disorders. Because of increased access to care provided to veterans with service-related disabilities, we dichotomized patients into no service-connected disability or any service connection. Illness burden was assessed with a diagnosis-based, risk-adjustment methodology (DxCG Company, Boston) ( 27 ), which has been validated in the VA population ( 28 ). An individual's RR score is the patient's total predicted health care cost divided by the average predicted health care cost for the population, with a score of 1.0 reflecting average risk ( 27, 28 ). RR score was broken down into tertiles with the following ranges: tertile 1, representing the lowest illness burden, .0–.6 (34.4%; N=68,423); tertile 2, .6–1.2 (33.3%; N=68,372); and tertile 3, representing the highest illness burden, 1.2–38.7 (33.3%; N=68,370).

To assess the impact of quality of depression care on outcomes 12 months after new onset, we extracted variables related to all-cause mortality, inpatient hospitalizations (medical and psychiatric), and outpatient service use (primary medical care, surgical care, mental health care, urgent care, and ancillary or other care).

Descriptive statistics were used to describe variables related to age, gender, ethnicity, marital status, income, illness burden, service-connected disability status, comorbid mental health diagnoses, adequate depression follow-up visits, adequate antidepressant medication supply, and 12-month outcome variables (relative to quality of depression care status, mortality, inpatient medical and psychiatric stays, and outpatient service use). To assess aggregate changes in quality of depression care over time, we examined the percentage of patients receiving adequate supply and adequate follow-up over time by using six-month intervals, and we subsequently tested for changes in study time periods by using Poisson regression analyses. Finally, logistic regression modeling procedures were used to predict inpatient medical stays, inpatient psychiatric stays, and 12-month all-cause mortality. To control for multiple logistic regression tests, a Bonferroni correction was used for data interpretation (p<.05 divided by three tests) and resulted in a significance threshold of .01.

Table 1 shows the demographic composition of the cohort and summary measures of service use during the 84-day period after the depression index date. The cohort was predominantly male (89.7%), and mean±SD age was 53.6±13.8. Less than half of the veterans were married (48.2%), and mean income was $40,368±$20,973. RR score for the cohort was 1.3±1.5, indicating a slightly higher illness burden than that of the average VHA service user. RR scores were also divided into tertiles for analysis. In terms of service-connected disability status, 68.2% of the cohort did not have this status, and 31.8% were documented to have a service-connected disability of 1% or greater. Comorbid mental disorders had been diagnosed for 34.1% of the patients, specifically anxiety, 13.3%; PTSD, 18.8%; and substance use disorder, 7.1%.

In regard to follow-up care for depression, 31.0% of the cohort had adequate follow-up with three or more visits, 16.9% had only two follow-up visits, 22.9% had only one visit, and 29.3% had no visits. Analysis of supply of antidepressant medication showed that 48.2% had a medication possession ratio of 80% or greater, classified as having adequate supply.

The 12-month mortality rate was 1.9% (however, we excluded 7,732 patients who died within six months of the depression index date). Service use characteristics at the 12-month follow-up revealed that 7.4% (N=30,586) of the cohort had one or more inpatient medical stays, and 3.9% (N=16,052) had one or more inpatient psychiatric stays. Outpatient service visits at the 12-month follow-up were as follows: primary medical care, 6.4±8.0; surgery, 5.1±6.7; mental health, 20.5±69.1; urgent care, 2.7±3.1; and ancillary and other services (such as for laboratory tests), 16.4±28.2.

Examining six-month intervals after the index diagnosis, we found that the percentage of patients with an adequate supply of medication ranged from 48% in October 1999 to 52% by April 2002 and remained close to this level for the remaining periods. Poisson regression confirmed that no time period was significantly different from the final time period (April 2005).

Thirty-six percent of patients had adequate follow-up at the first period, October 1999, but this declined steadily to 27% in April 2003. The percentages then increased over the remaining four periods, to 33% in April 2005. Poisson regression showed that follow-up care at the April 2005 time point was significantly better than care at the April 2003 time point ( χ ² =4.55, df=1, p=.03).

Table 2 shows summary measures for the cohort, with patients classified into adequate and inadequate follow-up visits and supply of antidepressant medication. Outpatient service use for the patients with adequate and inadequate follow-up was similar for primary medical care, surgery, urgent care, and ancillary and other care. Outpatient mental health service use was nearly three times higher among those who received adequate follow-up (33.8±92.4 visits 12 months after the index dates) compared with those who received inadequate follow-up (11.8±46.2 visits). Of those with adequate follow-up, 8.5% had one or more inpatient medical stays and 5.9% had an inpatient psychiatric stay, whereas for those with inadequate follow-up 7.0% had one or more inpatient medical stays and 3.0% had an inpatient psychiatric stay. Mortality was lower for those with adequate follow-up (1.4% versus 2.1%).

Comparisons of veterans with adequate and inadequate medication supply for depression suggest that both outpatient and inpatient service use in the two groups was similar for primary medical care, surgery, urgent care, mental health care, and ancillary and other care. Similarly, no differences were noted between these two subgroups in 12-month mortality.

Separate logistic regressions were conducted to determine which patient and clinical care factors (age, illness burden as measured by RR score, frequency of primary medical care visits, adequate follow-up care for depression, and adequate supply of antidepressant medication) predicted inpatient medical care, inpatient psychiatric care, and 12-month mortality.

For this model, significant predictors were RR score and adequate follow-up ( Table 3 ). Patients with the highest illness burden (RR score in the third tertile) were 2.1 times as likely to use inpatient services (odds ratio [OR]=2.10). The healthiest, with RR score in the first tertile, were 22% less likely to use services (OR=.79) relative to patients in the second tertile. Adequate follow-up and number of outpatient primary medical care visits had modest effects (OR=1.09 and OR=1.06, respectively), suggesting that increases in these variables equated to more inpatient service use.

The largest significant predictor of inpatient psychiatric care was presence of comorbid mental disorders (OR=3.47) ( Table 3 ). Other important predictors were RR score (highest tertile) (OR=1.67) and adequate follow-up (OR=1.38). Veterans in the lowest RR score tertile showed a modest increase in use of inpatient care (OR=1.08); primary medical care visits were also a modest predictor of inpatient care (OR=1.01). For each increasing year of age, odds of inpatient psychiatric use decreased by 4% (OR=.96).

Significant predictors of 12-month mortality were age, RR score, and adequate follow-up ( Table 3 ). The most ill patients (third tertile) were 2.5 times more likely to die, and the healthiest were 13% less likely to die (relative to patients in the second tertile). Notably, adequate follow-up had a protective effect on mortality (OR=.86). Each year of age increased risk of mortality by 6% (OR=1.06). Neither adequate supply nor number of outpatient primary medical care visits significantly affected mortality.