SEARCH

SEARCH BY CITATION

Keywords:

  • Cost of care;
  • Disease management system;
  • Healthcare utilization;
  • Program evaluation;
  • Severe mental illness

Abstract

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

Introduction: This article presents baseline findings that describe how nonclinical factors were associated with patient use of psychiatric and general medical care and how those relationships changed after patients enrolled in the 41-site Sequenced Treatment Alternatives to Relieve Depression study (STAR*D). Aims: STAR*D offered adult outpatients with major depression diligently delivered, measurement-based care. To achieve full remission within a tolerable medication dose, recommendations for treatment based on routine symptom and side-effect measurements were discussed with patients by clinical research coordinators and offered to clinicians who could flexibly tailor that guidance to accommodate individual patient needs. Medications were provided gratis. Pre- and post-enrollment data came from provider records and from patient face-to-face, telephone, and computer-assisted surveys. Two-part nested mixed models assessed patient likelihood and volume of mental and general medical care services. Results: Prior to enrollment, predisposing (gender, race, education, and care attitude), affordability (private insurance), and clinical factors (depressive symptoms and mental and physical functioning) were found to be important drivers of patient use of psychiatric and general medical care. After STAR*D enrollment, however, predisposing factors were less important drivers of psychiatric service use but remained important drivers of general medical care. Conclusions: Data suggest diligent, measurement-based mental health programs may reduce race, gender, and education disparities in the use of needed mental health care.


WHAT IS KNOWN

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

Prior studies often show disparity in patient use of mental health care on nonclinical factors, such as race, gender, education, insurance status, and attitudes about health care. Yet, little is known as to whether physicians in clinical practice can reduce this disparity by offering measurement-based care.

WHAT THIS STUDY FOUND

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

This study found that the association between nonclinical factors and mental health use rates changed after patients were enrolled in measurement-based care for depression in the large, multisite Sequenced Treatment Alternatives to Relieve Depression project, or STAR*D.

Introduction

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

Major depression disorder (MDD) poses a significant burden on the U.S. economy. [1,2] With a lifetime prevalence between 4.9% and 17.9%[3,4] and a 50% symptom reoccurrence rate following the first episode [5], the total cost for MDD in the United States [6,7] in 2005 dollars [8] was $24.6 billion in direct medical care, $35.5–48.8 billion in lost earnings from reduced worker productivity, and $11.1 billion from suicides. MDD is the fourth most disabling disease worldwide [9].

Despite its economic burden, there are patients with severe depression who are not receiving treatment [10], underscoring scientific findings that patient use of mental health care is associated not only with clinical symptoms [11] but also with “nonclinical” factors [12–17]. The health services literature [18,19] classifies these nonclinical predictors into enabling and predisposing factors. Enabling factors have been well studied, including financial burden or affordability[20,21] (health insurance and income) and time constraint or accommodation[22,23] (employment time burden, household responsibilities, and family advocates).

Predisposing factors focus on the perception of care and patient willingness to assume sick role behaviors. Predisposing characteristics are often traced to patient race, ethnicity, cultural background, education, and attitudes about care. Strategies designed to reduce use-of-care disparities across predisposing groups often focus on the patient–doctor relationship, including behavioral models [24], patient perception of clinical need [25], language proficiency [26], and psychological distress [27]. Other studies focus on the organized practice, including primary care settings [28–30] and managed care [31]. However, no study has focused on examining whether diligently delivered, measurement-based treatment for depression will diminish disparity in patient use of needed mental health care.

This study examines the use of mental and general medical care among patients participating in the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) trial (/http://www.star-d.org). Described elsewhere [32–36], STAR*D followed over 4000 outpatients during one or more acute treatments for nonpsychotic MDD, and for up to 1 year of continuing treatment following acute-phase treatment if successful (follow-up). The study involved over 400 clinicians at 41 sites in both specialty and primary care settings and both public and private sectors. STAR*D was based on an equipoise design [37]. Prior to randomization, the participants could opt out of selected treatment strategies in the second and third treatment steps that were found to be unacceptable [38]. For example, a participant could decline randomization to all three second-step augmentation treatments, while accepting all four treatment switch options. The subjects could not select a specific treatment. They all had to agree to randomization after the first step to at least two second- or third-step treatments to stay in the study.

The clinical aim of STAR*D was to achieve full symptom remission with a tolerable dose of medication. To accomplish this purpose, STAR*D offered diligently delivered, measurement-based care [34]. This included a clinically trained clinical research coordinator, or CRC. At each STAR*D psychiatric visit, the CRCs measured depressive symptoms and medication side effects and discussed with the patient the clinical meaning and treatment implications of those measurements. CRCs also discussed the STAR*D clinical protocol with the treating, study-participating psychiatrist. The clinical protocol included possible treatment options based on these measurements. These recommendations, however, were designed to be flexible so that clinicians could disregard or otherwise tailor the guidance to accommodate the needs of individual patients (e.g., medically fragile patients might receive a lesser dose adjustment than suggested). STAR*D also offered consultation advice directly to the treating psychiatrist. All psychiatric medication costs were gratis. Psychotherapy and all medication visits were paid for by public sector or private insurance, save for co-pays for the privately insured who continued to have coverage. If insurance coverage ended, all treatment costs were covered by STAR*D.

As envisioned by the developers, STAR*D offered diligently delivered, measurement-based care was expected to focus patients on the clinical purpose of seeking treatment to relieve their severe depressive symptoms. Thus, our primary hypothesis is: after adjusting for clinical, affordability, and accommodation factors, the strength of any association between predisposing factors and mental health use rates will diminish after patients become enrolled in STAR*D. For confirmatory purposes, since STAR*D focused on mental health care only, no change is expected following STAR*D enrollment in the association between predisposing factors and patient use rates of general medical care. Since STAR*D offered medication gratis and covered the costs for medication visits for uninsured patients, we also expected the strength of any association between insurance status and mental health use rates to diminish after patients became enrolled in STAR*D.

Methods

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

Study Data

STAR*D enrolled participants between the ages of 18 and 75 who satisfied the criteria for major depression based on the Diagnostic and Statistical Manual for Mental Disorders—4th Edition[39] (DSM-IV) criteria for nonpsychotic MDD and who scored 14 or higher on the 17-item Hamilton Depression Rating Scale [40]. Up to five treatment steps, with several treatment alternatives at each step (including cognitive therapy at step 2), were provided in sequenced, prospective, and randomized trials.

The patients initially received citalopram. Outcomes were assessed using the 16-item, clinician-rated Quick Inventory of Depressive Symptomatology (QIDS-C) [41] administered at each clinic visit. Patients who remitted (or who had responded sufficiently to justify follow-up) could consent to the 1-year post acute-treatment follow-up. Patients who did not remit after up to 14 weeks of treatment could enter the next and, if needed, subsequent treatment steps. At the beginning of each step, the patients were randomized to the acceptable treatment strategies. Those who did not achieve remission after the last step (level 4) exited the study.

Patient data were collected from written questionnaires, face-to-face interviews, telephone interviews, medical records abstracts, and a patient-initiated, telephone interactive voice response (IVR) system administered through a centralized server.

Factors Driving Use

Predisposing factors included self-reported ethnicity, race, gender, education, and attitudes about care. Care attitudes were based on whether patients agreed with the statement: “(If) … I can get the help I need from a doctor,…(then I would be) better able to enjoy things that interest me.” Care attitudes were classified as helpful (strongly agree or agree) or not helpful (neutral, disagree, or strongly disagree). Accommodation factors included paid employment status, marital status, and family support indicated by subjects reporting that, “… the current overall impact of your family and friends on your condition …” was helpful (very, moderately, or minimally helpful) or not helpful (neutral, minimally, moderately, or much more difficult). Affordability factors were based on whether patients had private insurance, on Medicaid, and family income. Clinical factors included the severity of baseline depressive symptoms measured using the 16-item Quick Inventory of Depressive Symptomatology or QIDS-SR16. Higher scores indicate a greater severity of symptoms [41–44]. Alcohol and drug abuse/dependence were assessed at baseline with the Psychiatric Diagnostic Screening Questionnaire [45] self-report, as described earlier [46]. Mental and physical functioning were assessed using the self-report 12-item Short Form (SF-12) [47] taken from face-to-face (or telephone) interviews at baseline. Higher scores indicate better functioning.

Healthcare Use

Healthcare use was measured as expenditures computed by multiplying service counts by a unit expenditure per count. The counts were recorded as the number of visits or days determined from patient survey responses. A unit expenditure per count was measured in dollars calculated from provider records and a schedule of market-based national prices. This strategy allowed the investigators to capture service use across all of the patients' providers, to compute expenditures in terms of the volume of care and service mix rather than the variation in local market prices across STAR*D's 41 sites, and to avoid selection biases by excluding patients who refused to sign medical records releases [48].

The counts (visits, days) were determined from the Utilization and Cost Questionnaire (UAC-Q) [49–53] scripted to an IVR format [54]. Beginning at baseline, the patients were asked to recall services used during the past 90 days for each of nine service classes defined by setting (outpatient visits, emergency room visits, or inpatient days) and clinical purpose (depression-related, other psychiatric, or general medical care).

Costs per count for each of the nine service classes were computed by estimating a mean unit cost per visit or day from provider records obtained from participants who signed a medical release. The estimates of unit costs were adjusted to reflect the characteristics of the final sample. Each patient encounter described in the medical record was classified into one of the nine service classes. For example, outpatient visits were classified “depression-related” if the record specified a depression-related diagnosis (DSM-4 nos. 296 and 311). Otherwise, the visit was classified as “other psychiatric” if the record specified at least one nondepression psychiatric condition (DSM-4 nos. 290–319 excluding nos. 296 and 311) or a psychiatric procedure listed under the American Medical Association's Current Procedural Terminology (CPT) [55] or level-1 Healthcare Common Procedure Coding System [56] (CPT nos. 90801–90899). Otherwise, the visit was considered general medical care. The visit was considered “emergency” if at least one emergency procedure was recorded (CPT nos. 99281–99285 and 99288); otherwise, the record encounter was considered a nonemergency outpatient visit. For inpatient records, “depression-related” stays must have depression-related diagnosis-related groupings [57] (DRG nos. 426 and 430). If a reliable code could not be ascertained from the record or chart, the stay was considered “depression-related” if 25% or more of reported diagnoses were depression-related (DSM-4 nos. 296 and 311). Otherwise, the stay is “other psychiatric” if: (1) the stay was nondepression psychiatric care (DRG nos. 424–438 except nos. 426 and 430), (2) no DRG was listed but 25% or more of the reported diagnoses were nondepression psychiatric conditions (DSM-4 nos. 290–319 excluding nos. 296 and 311), or (3) a psychiatric procedure was specified (CPT nos. 90801–90899). Otherwise, the stay was considered “general medical.” Physician visits to hospitalized patients were considered hospital expenditures and were included under the costs per diem.

A DRG-day and CPT procedure was priced based on national market-based rates obtained through the Department of Veterans Affairs Reasonable Charges program (version 2.4, April 11, 2005) [58]. These prices represented the 80th percentile of national paid transaction charges for 2005 and have been used in prior studies [49,53]. Since patients supplying provider records differed slightly from those who refused a medical release [48], the mean unit costs were adjusted to reflect the average STAR*D patient based on baseline demographic characteristics (age, ethnicity, race, gender, marital status, and high school graduate), resource (private insurance, paid employment, Medicaid status, student status, service volunteer status, and household size), and clinical factors (baseline QIDS-SR, physical functioning, mental functioning, family history of depression, presence of other Axis I conditions, and drug and alcohol abuse or dependence).

Analyses

Expenditure data are often bimodal, skewed, and heteroskedastic and require analytic designs that reflect study purposes [59–64]. Ninety-day use rates were measured in two parts [65,66]: dichotomous variables representing use versus no use (likelihood) and expenditures of use among users (expenditures). The estimates were adjusted for all factors as covariates and corrected for facility-level nesting/clustering, repeated measures, and heteroskedastic random variates [67].

The effects on the likelihood of use were computed as odds ratios (ORs) for dichotomous factors and as ORs per unit change for continuous factors. Pre-post differences were computed as a ratio of ORs estimated from time–factor interaction coefficients. The effects on expenditures were computed as elasticities for continuous variables and as expenditure ratios for dichotomous variables. Elasticities are defined as the percent change in expenditures per 1% change in factors. Expenditure ratios are defined as the ratio of adjusted mean care expenditures for patients with and without the given characteristic. Pre-post differences in expenditures were computed as differences in elasticities for continuous factors and as a pre-to-post ratio of expenditure ratios for discrete factors.

Results

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

Among 4041 patients enrolled in STAR*D at 41 sites, 3789 (93.8%) completed at least one of a total 9864 calls (mean = 2.6 [±1.7], range = 1–8) to report care use over a 90-day period. Sample characteristics are described in Table 1.

Table 1.  Characteristics of IVR responders (n = 3789)
 Number or mean valuePercent of total (%) or standard deviation (±)
Nonclinical factors
 Predisposing
   Hispanica (3784)46412.3%
   Raceb (3784)
     White288276.2%
     African American65517.3%
     Asian American651.7%
     Native American320.8%
     Pacific Islander210.6%
     Multiracial1293.4%
   Female (3789)237762.7%
   Education (3784)
     None46912.4%
     High School Graduate154640.9%
     High School Graduate2897.6%
      Equivalent Degree  
     Associate degree49713.1%
     College degree66117.5%
     Master degree2386.3%
     Doctorate/professional degree842.2%
   Attitude: Agree care makes things enjoyable (3775)347091.9%
 Accommodation/access
   Employment (3784)
     Retired2205.8%
     Unemployed
       Not looking79821.1%
       Looking60916.1%
     Employed
       Part-time47512.6%
       Full-time149539.5%
       Self1874.9%
       All employed215757.0%
   Student status (3785)
     None323585.5%
     Part-time2376.3%
     Full-time3138.3%
   Volunteer status (3782)
     None323285.5%
     Part-time52613.9%
     Full-time240.6%
   Household size (4032)2.8 persons±2.1
   Marital status (3785)
     Married127033.6%
     Separated/divided96525.5%
     Widowed1203.2%
     Other143037.8%
   Attitude: Agree family/friends helpful (3773)221758.8%
   Residence
     Years at current residence5.5 years±8.2
     Residence type (3785)
       Detached198852.5%
       Row/town1614.3%
       Mobile2155.7%
       Apartment131534.7%
       Room631.7%
       Retirement80.2%
       Nursing home20.1%
       Homeless330.9%
 Accommodation/access
   Insurance status
     Medicare enrollee (3711)2978.0%
     Medicaid enrollee (3691)45812.4%
     Private insurance enrollee (3715)191151.4%
   Household monthly income (3648)$2381±3146
Clinical factors
 Mental status
   Baseline QIDS-SRc (3789)15.4 patients±5.2
   Mental statusd (3760)26.9±8.9
   Alcohol abuse/dependencee (3789)2095.5%
   Drug abuse/dependencee (3789)1082.9%
   Family history of depressionf (3750)205554.8%
 Physical status
   Physical statusd (3760)49.3±12.0
 Other
   Medical/psychiatric leave (3780)2897.6%
   Age (3788)41.2 years±13.2
   Age (3788)
     <2546912.4%
     25–3492124.3%
     35–4487423.1%
     45–5489323.6%
     55–6447512.5%
     65–741483.9%
     75+80.2%
  Number or mean valuePercent of total (%) or standard deviation (±)
  1. aIncludes both Black and White.

  2. bNumber in parentheses following item is the number of patients reporting.

  3. cQuick Inventory of Depressive Symptomatology—patient self-report.

  4. DBased on SF-12.

  5. eBased on the Psychiatric Diagnostic Screening Questionnaire.

  6. fFamily history of major depression, including parent, sibling, or child.

Baseline 3-month utilization of health care
 Outpatient services (3789)
   Depression-relatedUser190450.3%
Visits per user3.05±4.09
   Other psychiatricUser46112.2%
Visits per user2.70±3.52
   General medicalUser130934.5%
Visits per user3.21±3.94
 Emergency services (3789)
   Depression-relatedUser3158.3%
Visits per user1.42±1.82
   Other psychiatricUser802.1%
Visits per user1.39±0.82
   General medicalUser46612.3%
Visits per user1.78±1.71
 Inpatient care (3789)
   Depression-relatedUser1183.1%
Days per user4.85±4.50
   Other psychiatricUser280.7%
Days per user4.61±4.15
   General medicalUser1303.4%
Days per user5.43±6.94

Unit costs per count were computed from provider records taken from the 2629 (65.1%) callers who provided medical records from 37 sites among 3116 (77.1%) callers who also signed record releases. Adjusted costs per count are displayed in Table 2 (inpatient days) and Table 3 (outpatient and emergency room visits). Standard deviations reveal variability across encounters, though low intraclass correlations suggest low facility clustering.

Table 2.  Unadjusted and adjusted inpatient cost per day
 Number of reporting unitsCPTDRGICCaPer Diem scalerb
FacilityPatientAdministrationDayPhysical hospital visitsAnc servRoom rateUnadjustedAdjustedc
  1. aIntraclass correlation coefficient, or the between-facility variance divided by total variance in encounter costs.

  2. bStandard deviation is represented by ±; standard error is in parentheses.

  3. cAdjusted based on a core variable set (age, Hispanic and African-American status, gender, marital status [married], education [high school graduate or higher], private insurance status, employment status, and baseline QIDS-SR), plus additional variables predicting per visit costs for patients with provider records using general medical care (physical functioning, mental functioning, Medicaid status, student status, household size, and family history of depression), plus additional variables predictive of having general medical care provider records among consenters (other Axis I conditions, drug abuse/dependence).

  4. dAdjusted for age and baseline QIDS-SR only.

  5. CPT, current procedural terminology; DRG, diagnosis-related grouping.

Depression-related1445502882076152311100.019$4710±5703$5563(778)
Other psychiatric367481334177711320.071$4244±2803$3923d(1157)
General medical1514723410074110215711740.028$7442±9728$7691(694)
Table 3.  Unadjusted and adjusted outpatient and emergency room cost per visit
 Number of reporting units$/procedureICCPer Visit scalera
FacilityPatientVisitProcedure  UnadjustedAdjustedb
  1. aStandard deviation is represented by ±; standard error is in parentheses.

  2. bAdjusted based on a core variable set (age, Hispanic and African-American status, gender, marital status [married], education [high school graduate or higher], private insurance status, employment status, and baseline QIDS-SR), plus additional variables predictive of per visit costs for record patients with one or more medical visits (physical functioning, mental functioning, Medicaid status, Student status, household size, and family history of depression), plus additional variables predictive of having provider records among consenters (other Axis I conditions, drug abuse/dependence).

Outpatient
 Depression-related37254815,97221,984$282±2190.015$346±640$345(5)
 Other psychiatric36797552511,467$289±2240.047$570±958$471(19)
 General medicine35163121,62248,333$533±9960.028$1022±2745$1092(21)
Emergency room
 Depression-related16106132475$704±5600.122$1948±3417$2295(381)
 Other psychiatric104655266$618±3610.139$2235±2696$2387(583)
 General medicine244559913883$618±4110.077$2075±2800$2184(128)

Use rates changed following STAR*D enrollment. Following STAR*D enrollment, the adjusted likelihood that patients would initiate mental health services increased by 5% (OR = 1.05, 95% CI = 1.02–1.09, t= 3.24, P= 0.002), while expenditures per user doubled (expenditure ratio = 2.16, 95% CI = 1.91–2.45, t= 12.06, P < 0.001). On the other hand, the participants were actually less likely to initiate general medical care (OR = 0.97, 95% CI = 0.94–0.99, t= 2.16, P= 0.031), though expenditures remained unchanged (expenditure ratio = 0.94, 95% CI = 0.84–1.06, t= 1.02, P= 0.307) after enrollment.

The adjusted estimates of the association between each factor on use rates and expenditures were computed at baseline and pre-post enrollment change and are presented in Tables 4 (psychiatric care) and 5 (general medical services). The estimates were adjusted to reflect influences from all other factors combined and corrected for facility nesting, repeated measures, and heteroskedastic random variates.

Table 4.  Adjusteda association between factors and mental health use (R) and user expenditure (E) rates for 90-day intervals at pre-baseline and pre-post changes
  Pre-STAR*D enrollmentPre-post enrollment change
Effect95% CI tPEffect95% CI tP
  1. aAdjusted for all other factors as covariates and corrected for facility nesting and repeated measures. Linear regressions are also corrected for heteroskedasticity.

  2. bFor discrete variables, factor effects on the likelihood of use are measured as odds ratios. Pre-post enrollment change is measured as a ratio of odds ratios. Factor effects on expenditures among users are measured as an expenditure ratio. Pre-post enrollment change is measured as a ratio of expenditure ratios or the post-enrollment expenditure ratio divided by the pre-enrollment expenditure ratio. Dichotomous factors include Hispanic versus not Hispanic; African American versus not African American, male vs. female, less than a high school graduate versus high school graduate or greater, care no help versus believed care was helpful, employed (full-, self-, and part-time) versus not employed, single (separated, divided, widowed, and other) versus not single (married), no family assistance versus claims family provided assistance, private insurance versus no private insurance coverage (though may have government-sponsored coverage such as Medicaid or Medicare), Medicaid versus no Medicaid coverage, alcohol versus no positive screen for alcohol dependence/abuse screen, and drugs versus no positive screen for drug dependence/abuse.

  3. cHispanic includes white and black racial groups.

  4. dFor continuous variables, factor effects on the likelihood of use are measured as odds ratios per $1000 income, per 10 points SF-12 mental status, per 10 points SF-12 physical status, and per 10 years of age. Pre-post enrollment change on the likelihood is measured by a ratio of odds ratio or the post-enrollment odds ratio divided by the pre-enrollment odds ratio. Thus, one indicates no change in effects on the likelihood. Factor effects on expenditures among users are measured as elasticities computed by ∂ln(y|y > 0)/∂ln(xc), where xo is a specific factor scaled accordingly and y is expenditure. Pre-post enrollment change is measured as the pre-enrollment elasticity minus the post-enrollment elasticity. Thus, zero indicates no change in effects on expenditures.

  5. eHigher QIDS-SR scores indicate worse depressive symptoms. Lower SF-12 scores indicate worse functioning.

Nonclinical factors
 Predisposing
   Hispanicb, cR0.820.661.011.870.0611.010.781.300.090.927
E1.090.931.271.050.2930.960.801.160.430.664
   African AmericanbR0.780.630.962.370.0181.371.061.782.400.017
E1.130.981.301.640.1020.960.801.160.420.678
   MalebR0.800.700.913.300.0011.411.171.713.570.001
E1.281.111.483.440.0010.880.751.021.730.084
   <High school graduatebR0.700.570.853.460.0011.030.771.370.180.855
E1.180.901.541.220.2221.331.081.672.600.010
   Care: no helpbR0.680.550.853.440.0011.250.961.631.690.090
E1.040.861.250.410.6790.660.550.84.200.000
 Accommodation/access
   EmployedbR0.870.770.992.120.0341.160.991.371.830.067
E0.890.800.982.350.0190.780.700.855.200.000
   SinglebR0.890.741.081.170.2410.950.781.160.500.617
E1.100.971.251.50.1331.110.991.271.750.079
   No family assistancebR0.970.821.150.370.7091.090.921.280.950.344
E0.900.830.992.10.0361.181.051.322.880.004
 Affordability
   Private insurancebR1.391.141.703.310.0010.780.630.972.250.024
E0.760.650.893.440.0010.950.821.100.730.468
   MedicaidbR1.030.811.300.230.8211.000.731.370.010.992
E1.210.971.521.680.0940.970.711.320.220.824
   IncomedR1.020.991.051.210.2260.990.961.030.360.720
E−0.01−0.040.010.960.336−0.01−0.040.020.480.633
Clinical factors
 Mental statuse
   QIDS-SRdR1.020.991.041.440.1491.000.981.020.120.906
E0.230.070.402.740.0070.07−0.100.240.830.407
   SF-12-menb, dR0.860.740.992.070.0380.960.831.120.490.626
E0.07−0.150.290.610.545−0.10−0.390.190.660.507
   AlcoholbR1.301.041.632.260.0240.790.601.041.670.095
E1.431.051.942.260.0240.760.511.111.430.152
   DrugsbR1.571.052.332.200.0280.640.371.111.600.110
E1.370.971.921.790.0740.660.470.932.390.017
 Physical status
   SF-12-physicaldR0.870.790.962.680.0080.920.851.011.800.072
E−0.18−0.440.081.370.171−0.01−0.240.220.060.949
   AgedR0.960.911.011.500.1321.141.071.223.840.000
E0.210.420.002.000.0460.190.000.381.920.054
Table 5.  Adjusteda association between factors and general medical use (R) and user expenditures (E) rates for 90-day intervals at pre-baseline and pre-post changes
  Pre-STARD enrollmentPre-post enrollment change
Effect95% CI tPEffect95% CI tP
  1. aAdjusted for all other factors as covariates and corrected for facility nesting and repeated measures. Linear regressions are also corrected for heteroskedasticity.

  2. bFor discrete variables, factor effects on the likelihood of use are measured as odds ratios. Pre-post enrollment change is measured as a ratio of odds ratios. Factor effects on expenditures among users are measured as an expenditure ratio. Pre-post enrollment change is measured as a ratio of expenditure ratios or the post-enrollment expenditure ratio divided by the pre-enrollment expenditure ratio. Dichotomous factors include Hispanic versus not Hispanic; African American versus not African American, male versus female, less than a high school graduate versus high school graduate or greater, care no help versus believed care was helpful, employed (full-, self-, and part-time) versus not employed, single (separated, divided, widowed, and other) versus not single (married), no family assistance versus claims family provided assistance, private insurance versus no private insurance coverage (though may have government sponsored coverage such as Medicaid or Medicare), Medicaid versus No Medicaid coverage, alcohol versus no positive screen for alcohol dependence/abuse screen, and drugs versus no positive screen for drug dependence/abuse.

  3. cHispanic includes white and black racial groups.

  4. dFor continuous variables, factor effects on the likelihood of use are measured as odds ratios per $1000 income, per 10 points SF-12 mental status, per 10 points SF-12 physical status, and per 10 years of age. Pre-post enrollment change on the likelihood is measured by a ratio of odds ratio or the post-enrollment odds ratio divided by the pre-enrollment odds ratio. Thus, one indicates no change in effects on the likelihood. Factor effects on expenditures among users are measured as elasticities computed by ∂ln(y|y > 0)/∂ln(xc), where xo is a specific factor scaled accordingly and y is expenditure. Pre-post enrollment change is measured as the pre-enrollment elasticity minus the post-enrollment elasticity. Thus, zero indicates no change in effects on expenditures.

  5. eHigher QIDS-SR scores indicate worse depressive symptoms. Lower SF-12 scores indicate worse functioning.

Nonclinical factors
 Predisposing
   Hispanicb, cR0.670.500.902.660.0081.160.911.471.180.238
E1.140.921.411.210.2270.890.651.200.770.443
   African AmericanbR0.730.600.902.950.0040.960.751.210.380.705
E1.261.131.404.280.0000.840.711.001.920.055
   MalebR0.810.700.932.940.0041.070.911.250.790.428
E1.261.141.384.730.0000.960.851.090.670.502
   <High school graduatebR0.650.530.804.070.0000.920.721.160.700.482
E1.140.931.391.180.2371.060.841.350.510.608
   Care: no helpbR1.130.901.421.050.2960.990.731.330.080.936
E0.890.701.130.980.3281.110.801.540.610.545
 Accommodation/access
   EmployedbR0.920.771.110.880.3801.050.871.280.530.594
E0.810.710.923.230.0010.850.721.011.850.065
   SinglebR0.970.841.120.420.6721.030.871.220.330.744
E1.161.031.302.410.0160.970.841.120.410.681
   No family assistancebR1.281.121.473.590.0011.020.861.200.180.855
E0.890.801.002.030.0430.970.871.090.460.648
 Affordability
   Private insurancebR1.441.181.763.610.0011.000.831.190.030.978
E0.880.780.982.300.0210.810.680.952.560.011
   MedicaidbR1.240.961.581.670.0940.990.781.260.080.937
E1.591.341.885.270.0000.970.761.230.250.804
   IncomedR1.010.991.040.840.4041.000.971.030.080.939
E−0.02−0.040.011.360.1750.01−0.010.040.970.330
Clinical factors
 Mental statuse
   QIDS-SRdR1.000.981.020.090.9281.010.991.040.890.373
E0.11−0.010.231.820.069−0.04−0.200.110.520.606
   SF-12-mendR0.870.810.953.300.0011.040.921.190.660.508
E−0.15−0.340.051.450.1490.10−0.100.300.960.336
   AlcoholbR0.960.681.370.210.8370.900.621.310.540.592
E1.271.001.611.970.0491.020.751.380.100.921
   DrugsbR1.401.001.961.960.0490.600.281.251.360.174
E1.411.021.952.080.0371.430.752.741.090.277
 Physical status
   SF-12-physicaldR0.610.570.6513.620.0001.030.951.120.660.511
E1.101.280.9211.870.0000.390.170.613.480.001
   AgedR0.990.931.060.170.8681.101.031.182.990.003
E−0.06−0.270.150.580.5590.12−0.110.341.040.300

Consistent with the primary hypothesis, predisposing factors (male, African American, lack of education, and attitudes toward care) were negatively associated with baseline use of mental health care. Following STAR*D enrollment, the strength of these associations significantly diminished for gender, race, and education and, contrary to expectation, increased for attitudes about care.

Consistent with the confirmatory hypotheses, predisposing factors were found at baseline to be associated with less likelihood to initiate general medical care (Hispanic, African American, male, and education); however, these associations did not change after patients enrolled in STAR*D. Patients with private health insurance (affordability) were more likely at baseline to initiate mental health and medical services, while incurring fewer expenditures once care was initiated, than their counterparts who did not report having private health insurance coverage. After STAR*D enrollment, however, the impact of private insurance on mental health use rates was substantially diminished.

Exploring these data further, employed patients were less likely to initiate mental care and to incur less mental and medical expenditures than their unemployed counterparts. These trends continued even after patients had enrolled in STAR*D. In fact, differences in psychiatric expenditures between employed and unemployed patients widened as the 89% pre-enrollment expenditures ratio dropped to a post-enrollment rate of 69%. This pattern is consistent with the theory that employed patients may be in better health with less need for care (selection) and face work schedules that impose time burdens whenever patients must leave work to access care (time cost). On further analyses, we found employed patients did have better baseline physical functioning (r = 0.34, P < 0.001). Patients who reported no satisfaction with the assistance they received for their health problems from family members tended to use fewer services once care had been initiated. After patients enrolled in STAR*D, however, the role of family assistance changed for mental health, but not for medical services.

Clinical factors uniformly drove use rates, though patterns varied by factor. More psychiatric symptoms were associated with greater mental health expenditures, while mental and physical functioning were associated with patients initiating care. Physical functioning had larger effects on general medical than on mental health care. The role of these clinical factors on use rates did not change after STAR*D enrollment.

Discussion

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

These data confirm previous findings that nonclinical factors are important drivers of patient use of mental health care. Efficacious psychiatric treatments will not work if patients in need do not show up for care. Under STAR*D diligently managed, measurement-based care, patient symptoms and side effects were scientifically assessed at each patient treatment visit by a CRC whose salary was covered by STAR*D. These assessments were used to inform both the patient and the provider about the patient's progress toward achieving clinical goals of full remission within tolerable medication doses. The CRC worked closely to help each patient understand his or her clinical status and available treatment options. STAR*D also contained a clinical protocol with training and consultation advice for the treating psychiatrist. STAR*D also covered patient out-of-pocket costs for MDD medications and for MDD medication visits to nonpublic facilities for patients who were not covered with private health insurance.

Following STAR*D enrollment, the role of predisposing factors such as race, gender, and education became less important in driving patient use of mental health care; thereby reducing the race, gender, and education disparity among patient groups in the use of needed mental health care. These findings build on earlier studies that suggest racial and cultural differences in the use of mental health care may arise from differences in how patients interpret symptoms [25] and underscore the public health importance to address issues of racial and cultural disparities in mental health care [68].

Following STAR*D enrollment, disparities in mental healthcare use rates between insured and uninsured groups were substantially reduced, suggesting that financial factors remain important considerations in patient decision to use mental health care. On the other hand, STAR*D did not change clinic hours or appointment scheduling. Thus, accommodation concerns, such as taking time off from work, continued to drive use rates, even after patients were enrolled in STAR*D.

The study incorporated several innovations. First, the sample was drawn from diverse sites, representing a wide range of treatment preferences [37–38] and willingness to release their medical records [48]. Second, use-of-care data were collected through an automated IVR system designed to ensure uniform responses across subjects and time periods [54]. Third, total expenditures were computed from structured survey responses and priced at standardized market-based rates computed from patient provider records. Thus, use rates reflected the volume and mix of services, rather than the differences in medical prices across local markets.

Finally, care use was divided into use rates (initiate use vs. no care use) and expenditures (the volume of care, given the patient initiates use). While common in health economics, separating use rates from volume expenditures allows researchers to detect important differences that may remain obscured when only total use is reported. For example, factors were often negatively associated with initiating use while being positively associated with volume after care was initiated (e.g., male and private insurance status on mental health use, African American, male, and no family assistance on general medical use). This mixed pattern was first noted by Donabedian [69], who offered an explanation that patients waited for care until their conditions worsened and thus required more care. This conclusion has some support with these baseline STAR*D data. For instance, private insurance (r = 0.20, P < 0.001) and African-American status (r =−0.17, P < 0.001) were associated with baseline physical functioning. Attitude about family assistance (r = 0.05, P= 0.002) and male gender (r =−0.06, P < 0.001) were associated with baseline mental functioning. Such mixed patterns contrast the role of employment status, which was associated with both lesser likelihood to initiate care and lower expenditure rates once care was initiated. This is consistent with the expectation that employed patients are healthier while facing job-related commitments that may interfere with accessing care and explains why in STAR*D employment status was positively associated with physical functioning (r = 0.34, P < 0.001) and negatively associated with mental functioning (r =−0.12, P < 0.001).

There were study limitations. The STAR*D protocol contained several elements, including symptom measurement, clinical coordinators, treatment protocol, staff consults, and reimbursement for some patient out-of-pocket costs. Offered as a single package, it is not empirically possible to link changes in patient use of care to any one program element. There were no direct measures of patient access to care in terms of travel time and distance. Patient characteristics were assessed at baseline and thus may introduce endogenity biases in pre-enrollment estimates, though none were detected statistically. The use of care was self-reported and subject to memory biases [54]. However, underreporting was not correlated with predisposing factors, suggesting that biases in the estimates of factor effect sizes are likely to be small. The patient sample is broad based, but is not a random sample of the U.S. population.

In conclusion, these data suggest that nonclinical factors continue to play important roles in affecting how patients access and use mental health care. However, it may be possible to reduce the disparity in use rates among predisposing patient groups with diligently managed, measurement-based care based on STAR*D clinical care model. Further research is recommended to replicate these findings in controlled, clinical trials and to better understand how STAR*D program elements may have each contributed to reducing use of care disparities. These findings have implication on national health policy in which the interest is to better match patient use of mental health treatment with patient health needs.

Acknowledgments

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

This project has been funded in whole or in part with federal funds from the National Institute of Mental Health, National Institutes of Health, under contract N01-MH-90003 (A.J. Rush MD, P.I.). The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services or the Department of Veterans Affairs, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

References

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix
  • 1
    Stoudemire A, Frank R, Hedemark N. The economic burden of depression. Gen Hosp Psychiatry 1986;8:387394.
  • 2
    Rice DP, Miller LS. The economic burden of affective disorders. Adv Health Econ Health Serv Res 1993;14:2137.
  • 3
    Weissman MM, Bruce ML, Leaf PJ, Florio LP, Holzer C III. Affective disorders. In: RobinsLN, RegierDA, editors. Psychiatric disorders in America: The Epidemiologic Catchment Area Study. New York : The Free Press, 1991;5380.
  • 4
    Kessler RC, McGonagle KA, Zhao S, et al. Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the United States. Results from the National Comorbidity Survey. Arch Gen Psychiatry 1994;51:819.
  • 5
    Keller MB, Boland RJ. Implications of failing to achieve successful long-term maintenance treatment of recurrent unipolar major depression. Biol Psychiatry 1998;44:348360.
  • 6
    Greenberg PE, Leong SA, Birnbaum HG. Cost of depression: Current assessment and future directions. Expert Rev Pharmacoecon Outcomes Res 2001;1:6976.
  • 7
    Greenberg PE, Stiglin LE, Finkelstein SN, Berndt ER. The economic burden of depression in 1990. J Clin Psychiatry 1993;54:405418.
  • 8
    U.S. Bureau of Labor Statistics, Consumer Price Index of urban wage earners and clerical workers for medical care and all items. Series CWUR0000SA0 and CWUR0000SAM. http://www.bls.gov, (accessed 03/04/2006).
  • 9
    Murray CJ, Lopez AD. Evidence-based health policy – lessons from the Global Burden of Disease Study. Science 1996;274:740743.
  • 10
    Gwynn RC, McQuistion HL, McVeigh KH, Garg RK, Frieden TR, Thorpe LE. Prevalence, diagnosis, and treatment of depression and generalized anxiety disorder in a diverse urban community. Psychiatr Serv 2008;59:641647.
  • 11
    Pearson SD, Katzelnick DJ, Simon GE, Manning WG, Helstad CP, Henk HJ. Depression among high utilizers of medical care. J Gen Intern Med 1999;14:461468.
  • 12
    Jones J, Amaddeo F, Barbui C, Tansella M. Predicting costs of mental health care: A critical literature review. Psychol Med 2007;37:467477.
  • 13
    Carr JV, Lewin TJ, Neil AL, Halpin SA, Holmes S. Premorbid, psychosocial and clinical predictors of the costs of schizophrenia and other psychoses. Br J Psychiatry 2004;184:517525.
  • 14
    Ginsberg G, Lerner Y, Mark M, Popper M. Prior hospitalization and age as predictors of mental health resource utilization in Israel. Soc Sci Med 1997;44:623633.
  • 15
    Knapp M, Beecham J, Anderson J et al. The TAPS project: Predicting the community costs of closing psychiatric hospitals. Br J Psychiatry 1990;157:661670.
  • 16
    Chisholm D, Knapp MR, Astin J, Lelliott P, Audini B. The mental health residential care study: Predicting costs from resident characteristics. Br J Psychiatry 1997;170:3742.
  • 17
    Simon GE, Katzelnick DJ. Depression, use of medical services and cost-offset effects. J Psychosom Res 1997;42:333344.
  • 18
    Andersen RM. Revisiting the behavioral model and access to medical care: Does it matter? J Health Soc Behav 1995;36:110.
  • 19
    Penchansky R, Thomas JW. The concept of access: Definition and relationship to consumer satisfaction. Med Care 1981;19:127140.
  • 20
    Pauly MV, Pagán JA. Spillovers and vulnerability: The case of community uninsurance. Health Aff 2007;26:13041314.
  • 21
    McWilliams JM, Meara E, Zaslavsky AM, Ayanian JZ. Use of health services by previously uninsured medicare beneficiaries. N Engl J Med 2007;357:143153.
  • 22
    Mechanic D, Cleary PD, Greenley JR. Distress syndromes, illness behavior, access to care and medical utilization in a defined population. Med Care 1982;20:361372.
  • 23
    Drapalski AL, Milford J, Goldberg RW, Brown CH, Dixon LB. Perceived barriers to medical care and mental health care among veterans with serious mental illness. Psychiatr Serv 2008;59:921924.
  • 24
    Pescosolido BA, Boyer CA. How do people come to use mental health services? Current knowledge and changing perspectives. In: HorwitzAV, ScheidTL, editors. A handbook for the study of mental health. Cambridge : Cambridge University Press, 1999;392411.
  • 25
    Zuvekas SH, Fleishman JA. Self-rated mental health and racial/ethnic disparities in mental health service use. Med Care 2008;46:915923.
  • 26
    Gilmer TP, Ojeda VD, Barrio C, Fuentes D, Garcia P, Lanouette NM, Lee KC. Adherence to antipsychotics among Latinos and Asians with schizophrenia and limited English proficiency. Psychiat Serv 2009;60:175182.
  • 27
    Manning W, Wells KB. The effects of psychological distress and psychological well-being on use of medical services. Med Care 1992;30:541553.
  • 28
    Simon G, Ormel J, Von Korff M, Barlow W. Health care costs associated with depressive and anxiety disorders in primary care. Am J Psychiatry 1995;152:352357.
  • 29
    Katzelnick DJ, Kobak KA, Greist JH, Jefferson JW, Henk HJ. Effect of primary care treatment of depression on service use by patients with high medical expenditures. Psychiatr Serv 1997;48:5964.
  • 30
    Simon G, Von Korff M, Barlow W. Health care costs of primary care patients with recognized depression. Arch Gen Psychiatry 1995;52:850856.
  • 31
    Henk HJ, Katzelnick DJ, Kobak KA, Greist JH, Jefferson JW. Medical costs attributed to depression among patients with a history of high medical expenses in a health maintenance organization. Arch Gen Psychiatry 1996;53:899904.
  • 32
    Rush AJ, Fava M, Wisniewski SR et al. Sequenced treatment alternatives to relieve depression (STAR*D): Rationale and design. Control Clin Trials 2004;25:119142.
  • 33
    Fava M, Rush AJ, Trivedi MH et al. Background and rationale for the sequenced treatment alternatives to relieve depression (STAR*D) study. Psychiatr Clin North Am 2003;26:457494.
  • 34
    Trivedi MH, Rush AJ, Wisniewski SR et al. Evaluation of outcomes with citalopram for depression using measurement-based care in STAR*D: Implications for clinical practice. Am J Psychiatry 2006;163:2840.
  • 35
    Rush AJ, Trivedi MH, Wisniewski SR et al. Bupropion-SR, sertraline, or venlafaxine-XR after failure of SSRIs for depression. N Engl J Med 2006;354:12311242.
  • 36
    Trivedi MH, Fava M, Wisniewski SR et al. Medication augmentation after the failure of SSRIs for depression. N Engl J Med 2006;354:12431252.
  • 37
    Lavori PW, Rush AJ, Wisniewski SR et al. Strengthening clinical effectiveness trials: Equipoise-stratified randomization. Biol Psychiatry 2001;50:792801.
  • 38
    Wisniewski SR, Fava M, Trivedi MH et al. Acceptability of second-step treatments to depressed outpatients: A STAR*D Report. Am J Psychiatry 2007;164:753760.
  • 39
    Diagnostic and statistical manual of mental disorders, 4th ed. Washington , DC : American Psychiatric Association, 1994.
  • 40
    Hamilton M. A rating scale for depression. J Neurol Neurosurg Psychiatry 1960;12:5262.
  • 41
    Rush AJ, Bernstein IH, Trivedi MH et al. An evaluation of the Quick Inventory of Depressive Symptomatology and the Hamilton Rating Scale for Depression: A Sequenced Treatment Alternatives to Relieve Depression trial report. Biol Psychiatry 2006;59:493501.
  • 42
    Rush AJ, Carmody TJ, Ibrahim HM et al. Comparison of self-report and clinician ratings on two inventories of depressive symptomatology. Psychiatr Serv 2006;57:829837.
  • 43
    Trivedi MH, Rush AJ, Ibrahim HM, Carmody TJ, Biggs MM, Suppes T, et al. The Inventory of Depressive Symptomatology, Clinician Rating (IDS-C) and SelfReport (IDS-SR), and the Quick Inventory of Depressive Symptomatology, Clinician Rating (QIDS-C) and Self-Report (QIDS-SR) in public sector patients with mood disorders: A psychometric evaluation. Psychol Med 2004;34:7382.
  • 44
    Rush AJ, Trivedi MH, Ibrahim HM et al. The 16-Item Quick Inventory of Depressive Symptomatology (QIDS), clinician rating (QIDS-C), and self-report (QIDS-SR): A psychometric evaluation in patients with chronic major depression. Biol Psychiatry 2003;54:573583; erratum p. 585.
  • 45
    Zimmerman M, Mattia JI. A self-report scale to help make psychiatric diagnoses: The Psychiatric Diagnostic Screening Questionnaire. Arch Gen Psychiatry 2001;58:787794.
  • 46
    Davis LL, Frazier E, Husain MM et al. Substance use disorder comorbidity in major depressive disorder: A confirmatory analyses of the STAR*D cohort. Am J Addict 2006;15:278285.
  • 47
    Ware JE, Kolinsky M, Keller SD. The 12-item ShortForm Health Survey: Construction of scales and preliminary tests of reliability and validity. Med Care 1996;34:220223.
  • 48
    Kashner TM, Trivedi MH, Wicker A, Fava M, ShoresWilson K, Wisniewski SR, Rush AJ. Release bias in accessing medical records in clinical trials: A STAR*D report. Int J Method Psychiatr Res, In press.
  • 49
    Kashner TM, Rush AJ, Crismon ML et al. Cost-outcome in the Texas Medication Algorithm Project: An empirical analysis. Psychiatr Serv 2006;57:648659.
  • 50
    Kashner TM, Suppes T, Rush AJ, Altshuler KZ. Measuring use of outpatient care among mentally ill individuals: A comparison of self- reports and provider records. Eval Program Plann 1999;22:3139.
  • 51
    Kashner TM, Rush AJ, Altshuler KZ. Measuring costs of guideline-driven mental health care: The Texas Medication Algorithm Project. J Mental Health Policy Econ 1999;2:111121.
  • 52
    Rush AJ, Crismon ML, Kashner TM et al. Texas Medication Algorithm Project – Phase 3 (TMAP-3): Rationale and study design. J Clin Psychiatry 2003;64:357369.
  • 53
    Kashner TM, Stensland MD, Lind L, Wicker A, Rush AJ, Golden RM, Henley SS. Measuring utilization and cost of care for patients with mood disorders: The utilization and cost inventory. Med Care 2009;47:184190.
  • 54
    Kashner TM, Trivedi MH, Wicker A et al. Voice response system to measure health care costs: A STAR*D report. Am J Manag Care 2009;15:153162.
  • 55
    American Medical Association. Current procedural terminology. Reston , VA : St. Anthony, 2004.
  • 56
    Centers for Medicare and Medicaid Services. Healthcare Common Procedure Coding System (HCPCS). http://www.cms.hhs.gov/medicare/hcpcs (accessed 11/01/2003).
  • 57
    St. Anthony's DRG guidebook (2004). Reston , VA : St. Anthony, 2004.
  • 58
    70 (68) F.R. 18465 (April 11, 2005);68(82) F.R. 22773 (4/29/2003), 68(82) F.R. 22966 (4/29/2003), 38 C.F.R. 17.101;66 (89) F.R. 23326 (5/8/2001), and 64 F.R. 22676 (10/1/1999), based on the Balanced Budget Act of 1997, PL 105–33, Sec. 8023(d), 38 USC 1729.
  • 59
    Mirandola M, Ammaddeo F, Dunn G, Tansella M. The effect of previous psychiatric history on the cost of care: A comparison of various regression models. Acta Psychiatr Scand 2004;109:132139.
  • 60
    Kilian R, Matschinger H, Loeffler W, Roick C, Angermeyer MC. A comparison of methods to handle skew distributed cost variables in the analysis of the resource consumption in schizophrenia treatment. J Mental Health Policy Econ 2002;5:2131.
  • 61
    Diehr P, Yanez D, Ash A, Hornbrook M, Lin DY. Methods for analyzing health care utilization and costs. Ann Rev Public Health 1999;20:125144.
  • 62
    Dunn G, Mirandola M, Amaddeo F, Tansella M. Describing, explaining or predicting mental health care costs: A guide to regression models. Br J Psychiatry 2003;183:398404.
  • 63
    McCrone P, Knapp M, Fombonne E. The Maudsley long-term follow-up of child and adolescent depression. Predicting costs in adulthood. Eur Child Adolesc Psychiatry 2005;14:407413.
  • 64
    Buntin MB, Zaslavsky AM. Too much ado about two-part models and transformation? Comparing methods of modeling Medicare expenditures. J Health Econ 2004;23:525542.
  • 65
    Duan N, Manning WG Jr., Morris CN, Newhouse JP. A comparison of alternative models for the demand for medical care. J Bus Econ Stat 1983;1:115126.
  • 66
    Mullahy J. Much ado about two: Reconsidering retransformation and the two-part model in health econometrics. J Health Econ 1998;17:247281.
  • 67
    Manning WG Jr. The logged dependent variable, heteroscedasticity, and the retransformation problem. J Health Econ 1998;17:283295.
  • 68
    Alegría M, Chatterji P, Wells K et al. Disparity in depression treatment among racial and ethnic minority populations in the United States. Psychiatr Serv 2008;59:12641272.
  • 69
    Donabedian A. Aspects of medical care administration. Cambridge , MA : Harvard University Press, 1973.

Appendix

  1. Top of page
  2. Abstract
  3. WHAT IS KNOWN
  4. WHAT THIS STUDY FOUND
  5. Introduction
  6. Methods
  7. Results
  8. Discussion
  9. Acknowledgments
  10. Conflict of Interest
  11. References
  12. Appendix

Mathematical Appendix

This section describes methods to compute predictor effects on care use rates when the log-transferred residual variances are heteroskedastic in the presence of facility-level nesting and patient-level repeated measures. In this two-part approach, we compute use versus no use separately from the volume of use among patients who use services:

  • image((A.1))
  • image((A.2))

where α's and β's are coefficients with dichotomous Xd and continuous Xc factors, y is care expenditures, and p is the probability that y > 0 versus y= 0 for patient i in facility f during quarter t following baseline (t= 0), with transformation T(t) = 0 whenever t= 0, and T(t) = 1 for all t > 0. Normal random variates s, z, u, v, and w were assumed to be independently distributed with zero mean. The conditional densities guarantee multiplicatively separable likelihood functions, allowing each model to be estimated separately.

The impact of each factor on the likelihood of use was computed as ORs (exp α11), with pre-post changes computed as ratios of ORs (exp α21). Factor effects on expenditures were computed as elasticities for continuous variables (Xc) and as expenditure ratios for dichotomous variables (Xd). Elasticities are defined as the percent change in expenditures per 1% change in factors. If the random variates in Eq. (A.1) have zero mean, an unbiased estimate of elasticity is

  • image((A.3))

Expenditure ratios are defined as the expenditures of care for group Xd= 1 divided by the expenditures of care for group Xd= 0. Taking into account heteroskedastic random variates as well as adjustments for facility- and patient-level nesting, an unbiased estimate of the expenditures ratio is

  • image((A.4))

where B represents the contribution of other terms, σ2g1 and σ2g0 are variances for the respective random effects, and g=u, v, or w, for groups defined by Xd= 1 and Xd= 0, respectively. Pre-post differences in expenditure were computed for continuous factors as differences between elasticities (β22) and for discrete factors as ratios of expenditure (β21) (adjusted for heteroskedasticity).