Treatment Selection for Depression: The Impact of Caseload Characteristics Marisa Elena Domino, Ph.D. The University of North Carolina David S. Salkever, Ph.D. The Johns Hopkins University We would like to thank the APA for providing data for this project and gratefully acknowledge support from NIMH. Context The tools of managed care are pervasive in general health as well as behavioral health These tools have been shown to have externalities at the macro level 2 Price sensitivity and managed care Pre-managed care: Physicians have little knowledge of and reaction to price Consumers are somewhat price sensitive RAND: drug prices matter, but only to quantity, not to switching 3 Price sensitivity and managed care managed care: May increase physicians’ response to price through information or incentives This response may spillover to non-managed care patients 4 Objectives: to determine the price sensitivity of psychiatrists when prescribing drugs for their depressed patients to determine the impact that various tools of managed care have on price sensitivity to investigate the existence of managed care “spillovers” where psychiatrists with a greater degree of managed care involvement demonstrate increased/decreased price sensitivity to their non-managed care patients 5 Setting: patients with depression Why? Highly prevalent Wide range in price of treatments Little difference in efficacy across all products 6 Dual Agency Model Insurers want physicians to chose treatments that minimize the expenses of treatment Consumers want physicians to chose treatments that improve their health and minimize their out-of-pocket expenses 7 Theoretical model Constrained utility maximization Physician maximizes own utility with arguments: patient utility, income (possibly a function of treatment), and effort (also a function of treatment) Patient utility a function of health, composite good, constrained by income 8 Theoretical model Drug price enters model through agency relationship with consumers and through own income 9 Empirical Methods Nested multinomial logit model on antidepressant drug choice Standard errors were adjusted for clustering on providers 10 Nesting Structure for Logit Models No Antidepressant Drug Prescribed TCAs SSRIs Others (MAOIs, Heterocyclics) Choice among 10 TCAs Choice among 4 SSRIs Choice among 7 Others 11 Methods the variable of interest is the price variable price alone determines price sensitivity price interacted with patient managed care variables determines additional influence on price sensitivity from managed care price interacted with managed care caseload variables determines managed care spillovers 12 Data: The American Psychiatric Association Practice Research Network The 1997 Study of Psychiatric Patients and Treatments linked by provider to: The 1996 National Survey of Psychiatric Practice 13 Sample Selection Persons with a clinical diagnosis of depression (DSM codes 296.2, 296.3, or 311) age 18 or over seen on an outpatient basis not receiving multiple concurrent antidepressant therapy 14 Physician and Patient Characteristics Variable Description (n) Patient Level Characteristics GAF Global Assessment of Functioning scale (n=301) Female Patient is Female(n=289) Educ Years of Education for Patient (n=300) Age Age of Patient(n=305) Inpatyr Patient has been admitted to a psychiatric facility for the treatment of a psychiatric disorder in the past 12 months (n=299) Subabu Patient has ever had substance abuse or dependence disorder, including alcohol or nicotine (n=281) Primdep Primary diagnosis of depression (n=305) Yearstx Number of years since first treatment (not necessarily for depression) (n=304) Provider Level Characteristics (weighted by number of patients in sample) Doc_age Age of provider (n=303) Doc_fem Provider is female (n=302) Ppatmon Number of patients in last month (asked in 1997 patient survey) (n=301) Mean/ Percent Standard Deviation 49.4 63.3% 14.4 44.9 14.7% 12.0 3.1 13.8 44.4% 77.7% 2.7 49.2 31.5% 137.6 4.0 9.9 93.2 15 Managed Care Variables Managed Care Variable Patient-level description and percent Managed Care Patient belongs to any type of managed care plan (43.1%; n=295) Carve-out Patient belongs to a behavioral health carve-out plan (12.9%; n=295) NA Managed Care Panels Discounted Fee-ForService (patients) Physician is compensated for treatment provided to this patient at this visit on a discounted fee-forservice basis (35.1%; n=305) Provider-level description, mean, and standard deviation (weighted by number of patients) Percentage of patients covered through a managed care plan (34.4 (27.8); n=294) Percentage of patients covered through a behavioral health carveout plan (19.8(23.1); n=294) Number of Managed Care Panels to which a Physician belongs (5.3(6.9); n=280) Percentage of patients for which fees are discounted. (40.8(34.4); n=291) 16 Distribution of Treatments 100% 90% 80% 70% 60% Other 50% SSRI 40% TCA 30% No Antidepressant 20% 10% 0% All Persons No Managed Care Managed Care 17 Explanatory Variable of Interest: Price Calculated as the median (of average) price per prescription by chemical entity (“drug”) for the first quarter of 1997 from CMS data on Medicaid Drug Utilization nationwide 18 Expectations of “Price” What we want to happen: » “price” picks up true price effects What could confound the results: » “price” picks up other effects which are correlated with the true price side effects episode costs 19 Model 1: Any type of Managed Care Lower Level Variable Price (*.01) Price*Patient MC Price*Physician MC (*.01) Price*Patient MC* Physician MC (*.01) Market Age (*.01) Market Age Squared (*.0001) Tricyclic Branch SSRI Branch Other Branch -0.684 (2.054) -1.088 (2.393) 0.778 (1.554) 2.411 (3.040) 0.495 (0.218) -0.097 (0.230) -0.735 (2.282) -0.014 (0.087) 0.290 (0.815) -0.498 (1.333) -0.179 (0.019) 3.753 (0.517) 0.599 (2.693) 0.021 (0.067) -0.415 (2.284) 0.494 (0.946) 2.505 (2.949) -0.705 (3.404) 20 Model 1: Any type of Managed Care Upper level coefficients Intercept GAF (*.01) Subabu Primdep Yearstx (*.1) Inpatyr Age (*.01) Age2 (*.0001) Female Patient Gender Missing Education Ppatmon (*.001) Doc_fem Doc_age (*.01) Doc_age2 (*.0001) Patient MC * Physician MC Patient MC Physician MC Inclusive Value log-Likelihood Tricyclic Branch SSRI Branch Other Branch -1.082 (0.130) 5.423 (2.277) 0.976 (1.509) 1.086 (1.642) 0.515 (0.925) 0.540 (0.556) -4.181 (0.339) 9.978 (0.780) 0.554 (0.901) 1.658 (1.295) -23.661 (2.219) 4.230 (0.710) 0.423 (0.457) -6.638 (0.235) -17.506 (0.616) 3.579 (0.570) -0.499 (0.218) -2.305 (1.247) -0.029 (0.040) -70.625 (0.254) -0.811 (0.402) 0.148 (0.301) 1.102 (2.258) -1.166 (1.933) 0.744 (1.052) -10.150 (1.161) 8.429 (0.873) 0.564 (1.135) -1.104 (1.090) -2.706 (0.348) 6.292 (1.148) 1.303 (2.368) 19.340 (1.089) -21.288 (1.287) -109.805 (0.425) -2.943 (0.076) 66.430 (0.539) -29.659 (0.360) - 515.49 -139.250 (1.420) 0.973 (0.450) 0.876 (1.590) 1.044 (2.027) -0.437 (0.738) 0.525 (0.646) -7.369 (0.709) 7.416 (0.668) 0.327 (0.575) -1.369 (0.981) 5.012 (0.656) 10.556 (2.064) 1.345 (2.189) 1.872 (0.103) -6.408 (0.387) -53.215 (0.909) -2.980 (0.072) 45.589 (1.425) 24.198 (1.311) 21 Figure 2: Estim ated Price Elasticity of TCAs With 90% Confidence Interval, based on Results from Model 1 6.000 4.000 2.000 Elasticity Patient not in Managed Care Patient in Managed Care 0.000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 CI-Upper around no MC CI-Low er around no MC -2.000 CI-Upper around MC CI-Low er around MC -4.000 -6.000 -8.000 Percent of Physician Caseload in Managed Care 22 Figure 3: Estim ated Price Elasticity of SSRIs With 90% Confidence Interval, Based on Results from Model 1 1.000 0.000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Estimated Elasticity -1.000 -2.000 Patient not in Managed Care Patient in Managed Care -3.000 CI-Upper no MC CI-Low er no MC -4.000 CI-Upper MC CI-Low er MC -5.000 -6.000 -7.000 -8.000 Percent of Physician Caseload in Managed Care 23 Figure 4: Estim ated Elasticity of SSRIs With 90% Confidence Interval, Based on Results from Model 2 2.000 0.000 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Estimated Elasticity -2.000 Patient not in Carve-out -4.000 Patient in Carve-out CI-Upper no CO -6.000 CI-Low er no CO CI-Upper CO -8.000 CI-Low er CO -10.000 -12.000 -14.000 Percent of Physician Caseload in Carve-outs 24 Conclusions: Price effects vary strongly among the three classes of antidepressants Psychiatrists do demonstrate price sensitivity when choosing among antidepressants Price elasticity estimates range from -0.4 to -4.2 25 Conclusions: Higher price elasticities for managed care patients Significant evidence of spillovers Not always in the expected direction Generally only to other managed care patients 26