 Pulmonary hypertension portends a poorer prognosis for blacks versus white populations, but the underlying reasons are poorly understood. We investigated associations of disease characteristics, insurance status, and race with clinical outcomes.

Retrospective cohort study of patients presenting for initial pulmonary hypertension evaluation at 2 academic referral centers.

 We recorded insurance status (Medicare, Medicaid, private, self-pay), echocardiographic, and hemodynamics data from 261 patients (79% whites, 17% blacks) with a new diagnosis of pulmonary hypertension. Subjects were followed for 2.3 years for survival. Adjustment for covariates was performed with Cox proportional hazards modeling.

Compared with white patients, blacks were younger (50 ± 15 vs 53 ± 12 years; 

= .04), with females representing a majority of patients in both groups (80% vs 66

 = .08) and similar functional class distribution (class 2/3/4: 30%/52%/16% blacks vs 33%/48%/14% whites; 

= .69). Blacks diagnosed with incident pulmonary hypertension were more frequently covered by Medicaid (12.5% vs 0.7%) and had less private insurance (50% vs 61%; 

= .007) than whites. At presentation, blacks had more right ventricular dysfunction (

= .04), but similar mean pulmonary arterial pressure (46 vs 45 mm Hg, respectively; 

= .66). After adjusting for age and functional class, blacks had greater mortality risk (hazard ratio [HR], 2.06; 95% confidence interval [CI], 1.18-3.44), which did not differ by race after additional adjustment for insurance status (HR, 1.74; 95% CI, 0.84-3.32; 

 In a large cohort of patients with incident pulmonary hypertension, black patients had poorer right-side heart function and survival rates than white patients. However, adjustment for insurance status in our cohort removed differences in survival by race.

Black race was associated with a 2-fold mortality risk in patients with incident pulmonary hypertension after adjustment for age and functional class. However, this relationship was no longer observed after additional adjustment for insurance. 

A total of 261 consecutive patients with pulmonary hypertension were included in our cohort at time of first evaluation at 2 large tertiary referral centers (Duke University Medical Center and Cleveland Clinic). 

African American patients had more severe pulmonary vascular disease at presentation and greater than a 2-fold risk of death over study follow-up. 

Pulmonary hypertension is a fatal disease and access to medications is a public health issue due to high costs and racial disparities.

Pulmonary hypertension (defined as mean pulmonary artery pressure ≥25 mm Hg) encompasses a broad spectrum of diseases and patient populations. Prior efforts to characterize patients with pulmonary hypertension focused on the subset of those with “pure” pulmonary vascular disease, defined as pulmonary arterial hypertension (pulmonary capillary wedge pressure ≤15 mm Hg and pulmonary vascular resistance ≥3 Woods units). Given the dismal 5-year survival rate of 34% in the National Institutes of Health pulmonary arterial hypertension cohort 3 decades ago,

 therapeutic advances targeting functional capacity, disease progression, and mortality have led to the availability of endothelin receptor antagonists (ERAs),

 and soluble guanylate cyclase stimulators.

 These advances in disease management have improved the care of patients with pulmonary arterial hypertension, but are expensive: the approximate annual costs are sildenafil, $12,761; bosentan, $55,890; ambrisentan, $56,736; iloprost, $92,146; epoprostenol (70 kg patient), $33,153; and treprostinil (70 kg patient), $97,615.

 Furthermore, 5-year survival, incorporating advances in modern therapies, remains low at 58% for patients with pulmonary arterial hypertension,

 and the development of pulmonary hypertension is among the most significant predictors of poor outcomes in patients with chronic heart failure and lung disease. 

Alhough there is a similar prevalence of pulmonary hypertension in black (6.6%) and white populations (6.8%),

 Black individuals have been shown to have a worse prognosis.

 Furthermore, age-standardized death rates for pulmonary arterial hypertension patients from 2 separate epidemiological studies, from 1980 to 1984 and 2000 to 2002, have diverged between whites (increasing from 5.0%-5.3%) and blacks (increasing from 4.8%-7.3%),

 with little insight into why these racial disparities currently exist. However, the Registry to Evaluate Early and Long-term Pulmonary Arterial Hypertension Disease Management (REVEAL) registry of patients with pulmonary arterial hypertension in the United States reported no difference in time-to-recognition of disease from symptom onset by race.

 We sought to investigate the association of race with survival in a population of patients referred to 2 large tertiary care centers for invasive evaluation of pulmonary hypertension. Additionally, given that cost may be a barrier to pulmonary arterial hypertension medications, associated equipment for intravenous/inhaled administration, and specialized care, we investigated the role of insurance status on mortality.

Consecutive patients who were evaluated at 2 pulmonary hypertension referral centers for initial hemodynamic assessment between 1998 and 2009, and found to have mean pulmonary artery pressure of at least 25 mm Hg, were prospectively entered in a database. Each patient underwent a diagnostic work-up, including echocardiography and cardiac catheterization with vasodilator testing, if indicated. Only patients with known insurance type were included in our study. The study was approved by the institutional review boards at Duke University Medical Center and the Cleveland Clinic Foundation. 

Right heart catheterization was performed by the same experienced cardiologist (RAK) in the cardiac catheterization laboratories at the 2 medical centers, using a single end-hole, balloon flotation catheter (either Bard Pulmonary Wedge Catheter [Medtronic; Minneapolis, Minnesota] or Balloon Wedge Pressure Catheter [Arrow International, Inc; Cleveland, Ohio]). Contrast injections during left heart catheterization, if performed, followed all hemodynamic measurements. Standard hemodynamic measurements from right heart catheterization were obtained, including mean right atrial pressure, right ventricular, systolic and diastolic pressures, pulmonary artery systolic, diastolic and mean pressures, and mean pulmonary capillary wedge pressure. Mean arterial blood pressure, cardiac index, pulmonary vascular resistance, systemic vascular resistance, and response to vasodilator challenge were also recorded in the database when indicated.

All echocardiograms were performed within 30 days of cardiac catheterization using a phase-arrayed scanner with a 4- or 8-MHz transducer, depending on body habitus and image quality. Standard 2-dimensional images were obtained in the parasternal long- and short-axis views, apical 2- and 4-chamber views, and subcostal view, and Doppler gradients were obtained to estimate valvular gradients. Assessment of right ventricular size was reported as a grade from 0 to 3, where 0 = normal, 1 = mildly enlarged, 2 = moderately enlarged, and 3 = severely enlarged. Right ventricular function was graded on a 4-point scale for systolic dysfunction, where 0 = no dysfunction, 1 = mild dysfunction, 2 = moderate dysfunction, and 3 = severe dysfunction. The degree of tricuspid regurgitation was determined using color-flow Doppler and assigned a grade from 0 to 4+, depending on the extent of color flow relative to the right atrial area. Estimated right ventricular pressures were obtained by applying continuous wave Doppler to the tricuspid regurgitation jet to the Bernoulli equation and added to the estimated right atrial pressure. If tricuspid regurgitation velocity could not be adequately visualized by color Doppler, saline microbubbles were injected to improve image quality. Saline microbubble injection was also used to assess for intracardiac shunting if flow across the interatrial septum was noted by color Doppler on subcostal imaging. Assessments were made by experienced imaging cardiologists at both academic medical centers.

Pulmonary hypertension work-up for all patients also included chest radiography, (chest computed tomography if chest radiography was abnormal), ventilation-perfusion scanning, full pulmonary function testing, electrocardiography, and echocardiography, in addition to catheterization. Test results, demographics, and medical histories were obtained from medical records and entered in the database. Insurance status at the time of the assessment was recorded as either Medicare, Medicaid, private, or self-pay, and confirmed by appointment schedulers at the time of the initial appointment. Race was self-reported at the time of catheterization. Long-term survival was assessed using electronic health records and confirmed using the Social Security Death Index. Variables of interest included the following: idiopathic pulmonary arterial hypertension (no attributable etiology of pulmonary hypertension), New York Heart Association (NYHA) class (functional class 1: no symptoms and class 4: severe symptoms), systemic hypertension and atrial fibrillation (by medical history), serum sodium and creatinine (venous laboratory values drawn at time of heart catheterization), and arterial oxygen saturation (obtained by femoral arterial blood sample at time of heart catheterization).

Only white and black patients were included in our study; patients who self-identified as other races were excluded. Descriptive statistics were presented as mean ± standard deviation (SD) for continuous variables and as a percentage for discrete variables. Comparison of categorical variables was performed using the χ

 test or Fisher’s exact test, where appropriate. Comparisons of continuous variables between groups were performed using 2-sided 

 tests and 1-way analysis of variance. Statistical significance was assumed with 

<.05. Survival analyses were performed using the Kaplan-Meier and Cox proportional hazards regression methods. Two proportional hazards models were utilized to investigate the effect of demographics, hemodynamics, echocardiographic measurements, and insurance status. Each model was constructed in a forward step-wise manner, investigating the effect of each covariate and potential interactions individually. Potential covariates included in the adjustment models were variables with baseline differences having 

<.20, including known predictors of death in the pulmonary hypertension population (including right atrial pressure, pulmonary artery pressure, pulmonary vascular resistance, right ventricular function, cardiac index, functional status), and race. Based on the size of the cohort, the maximum number of covariates included in each model was predetermined to be 5 to limit overfitting. Differences in the survival functions were assessed for significance using the Wilcoxon test. All analyses were performed using JMP version 12.0 software (SAS Institute; Cary, North Carolina).

A total of 250 patients (82% white, 18% black) were included in the study. An additional 11 patients self-identified as Asian/Pacific Islander, Hispanic, or Native American were excluded from the analysis. The median follow-up period was 2.3 years (whites: 2.0; interquartile range [IQR], 1.1-2.0 years; blacks: 1.7 years; IQR, 0.9-1.7; 

= .02). Overall, black patients were younger, more likely to have diabetes, and coverage with Medicaid versus private insurance compared with white patients (

). Etiology of pulmonary vascular disease was idiopathic pulmonary arterial hypertension in 42% of whites, but only 29% in blacks. Functional class distribution was similar for both races.

Comparison of pulmonary hypertension characteristics was also performed using invasive hemodynamics and echocardiography at presentation. Black individuals had higher right atrial pressure, decreased right ventricular function, and more severe tricuspid regurgitation than whites. Other markers of pulmonary hypertension severity, including pulmonary vascular resistance, cardiac index, and right ventricular size, did not differ by race (

No drug prescribing differences were seen between black and white patients. During follow-up, 55% of patients received at least 1 pulmonary arterial hypertension-specific therapy (ERA, prostanoid or phosphodiesterase 5-inhibitor) and 16% received combination pulmonary hypertension-specific therapy. The most commonly utilized pharmacologic agents in descending order of frequency included sildenafil (used in 29.6% of patients), bosentan (16.8%), epoprostenol (15.8%), treprostinil (6.8%), and iloprost (2.8%). None of these agents were associated with improved survival by univariate analysis.

Differences in baseline characteristics between patients who survived and those who died were examined (

). Older age, worse functional class, diabetes, serum sodium and creatinine values, and insurance status were associated with death by completion of the study follow-up period. Analysis by race revealed that 20 of 45 (44%) black patients and 67 of 205 (33%) white patients had died at the end of the 2.3-year follow-up period (

). With adjustment for baseline differences in age and NYHA functional class, blacks had increased risk for death (hazard ratio, 2.06; 95% confidence interval [CI], 1.18-3.44; 

). However, when insurance status was added to the model, blacks and whites had no statistically significant difference in survival rates (

Among 250 patients referred to 2 large pulmonary hypertension referral centers, blacks had worse survival from time of evaluation. However, additional adjustment for insurance status attenuated the association of race with outcomes, suggesting that insurance status plays an important role in pulmonary hypertension outcomes. blacks with pulmonary hypertension had more markers of advanced pulmonary vascular disease, decreased right ventricular function, and were more likely to have diabetes compared with whites. Patients who died were also more likely to have diabetes, which may be a marker for endothelial dysfunction and the resulting pulmonary venous hypertension.

 Despite increased disease severity and diabetes, pulmonary hypertension etiology (including pulmonary hypertension secondary to left-sided heart failure) did not vary between patients who survived and those who died. Finally, the percentage of our subjects lacking insurance was lower than that of the general US population.

The association of race and health outcomes in pulmonary hypertension, which carries a high mortality as an isolated disease and when associated with obstructive lung disease, pulmonary fibrosis, and left-sided heart failure,

 has not been well understood. This is the largest study to date to examine the association between insurance and racial disparities in outcomes for pulmonary hypertension, and the only study to examine this association among patients with pulmonary hypertension in WHO groups 2 to 5. Prior studies examining racial differences in pulmonary arterial hypertension have consistently shown that blacks have poorer outcomes, but the studies had significant limitations. Lilienfield et al found clear racial differences in pulmonary arterial hypertension survival using data from the CDC, but all diagnosis and disease characterization was based on 

International Classification of Diseases, Ninth Revision, Clinical Modification

 Furthermore, their analysis used data from 1979 to 1996 in the absence of many new pulmonary arterial hypertension therapies that have emerged in the last 2 decades, and are more difficult to interpret in the context of current management practices. A more recent study of pulmonary arterial hypertension prognosticators identified race as one of the most significant variables associated with death, with an adjusted survival hazard ratio of 4.3 for nonwhites compared with whites.

 However, further exploration of this strong association was limited by a small sample size of nonwhite patients (15 nonwhites, including 6 blacks), and socioeconomic and insurance status was not reported.

In the United States, blacks experience more life-years lost as a result of hypertension, HIV, diabetes, and trauma,

 and they are also less likely to undergo procedures for heart disease—including angiography and coronary artery bypass surgery&mdash;resulting in poorer outcomes following cardiothoracic surgery.

 These findings persist after controlling for socioeconomic factors. Thus, racial disparities in healthcare have been well established, and our study extends these findings to pulmonary hypertension.

 Hypotheses regarding discrepancies for other disease states have cited differences in genetics, income, education, access to medical insurance and providers, affordability of medical care, geography, and psychological biases among patients and providers.

In our population, blacks had poorer right ventricular function, but overall, no significant differences in clinicians’ assessment of functional class compared with whites. Right ventricular function is a known key determinant of outcome in pulmonary hypertension, and black patients without private insurance may have been referred later in their disease course, and presented with more difficult to treat disease. This discrepancy also suggests that clinical ascertainment of functional class, which is often used to base management decisions, may also be subject to important inaccuracies, as shown in heart failure populations.

Our results highlight the need for further public health study in understanding how to manage pulmonary hypertension most effectively. Although time-from-symptom onset to pulmonary arterial hypertension diagnosis likely does not differ by race,

 the poorer right ventricular function observed in blacks on presentation may be due to genetic factors, comorbid conditions, or fewer interactions with healthcare providers, leading to more aggressive disease at presentation to the pulmonary hypertension referral center. blacks are less likely to have private insurance, and we found that the type of insurance at the time of referral may interact with race for survival. Insurance may be a surrogate for overall socioeconomic status, and it also may have implications for ease of access to referral centers and potential pulmonary hypertension therapies. 

The granularity of our analysis is limited by lack of data on modes of death. However, distribution of pulmonary hypertension etiology did not significantly differ between survivor and nonsurvivor groups, and so it is less likely that our findings are driven by a single pulmonary hypertension subtype or associated comorbid conditions. Patient adherence rates and physician likelihood to prescribe pulmonary arterial hypertension therapies when indicated were not assessed, but, given their relationship with insurance status, would be difficult to interpret in the context of this study. Once the decision to prescribe pulmonary arterial hypertension therapy was made, we were able to assess that specific pulmonary arterial hypertension therapies did not differ by race alone. Finally, we noted a weak correlation between age and Medicare status in our population in our model: due to the discrete information provided by each variable and lack of significant multi—co-linearity, we kept both variables in the adjustment model.

Clinicians caring for patients with pulmonary hypertension should recognize that insurance status is an important prognosticator of increased mortality after adjustment for age and functional class, regardless of pulmonary hypertension WHO classification. Patients with pulmonary hypertension who are lacking private insurance represent a high-risk subpopulation that would benefit from further public health research and thorough evaluation of insurance options if necessary monitoring and treatments are not available.

Blacks referred for evaluation of pulmonary hypertension have worse right heart function and subsequent prognosis compared with whites after initial diagnosis. However, adjusting for insurance status weakens the racial outcomes disparities. Pulmonary hypertension workup and management often require multiple tests for diagnosis and monitoring, specialist care, and, if warranted, costly therapies. A better understanding of varying options available to patients with newly diagnosed pulmonary hypertension across the spectrum of insurance types may identify public health measures that would improve detection and morbidity and mortality associated with the disease.

 Duke Clinical Research Institute (KSP, RAK), Durham, NC; Division of Cardiology, Duke University Medical Center (KSP, TMB, RAK), Durham, NC; Cleveland Clinic Lerner College of Medicine, Case Western Reserve University (KAS), Cleveland, OH; University of Michigan (SAH), Ann Arbor, MI.

 Manuscript preparation supported by NIH grant 5T32GM086330-05 (KSP).

 Dr Krasuski has grants pending from Actelion and is a consultant for Actelion and Bayer. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article. 

 Concept and design (TMB, KSP, KAS, SAH, RAK); acquisition of data (SAH, RAK); analysis and interpretation of data (KSP, KAS, RAK); drafting of the manuscript (KSP, KAS, RAK); critical revision of the manuscript for important intellectual content (KSP, KAS, SAH, RAK); statistical analysis (KSP, KAS, RAK); obtaining funding (RAK); provision of patients or study materials (RAK); administrative, technical, or logistic support (TMB, SAH, RAK); and supervision (TMB). 

 Richard A. Krasuski, MD, Department of Cardiovascular Medicine, Duke University Medical Center, 2301 Erwin Rd, Durham, NC 27710. E-mail: Richard.krasuski@duke.edu. 

1. Rich S, Dantzker DR, Ayres SM, et al. Primary pulmonary hypertension. a national prospective study. 

2. Channick RN, Simonneau G, Sitbon O, et al. Effects of the dual endothelin-receptor antagonist bosentan in patients with pulmonary hypertension: a randomised placebo-controlled study. 

3. Liu C, Cheng J. Endothelin receptor antagonists for pulmonary arterial hypertension. 

4. McLaughlin VV, Gaine SP, Barst RJ, et al; Treprostinil Study Group. Efficacy and safety of treprostinil: an epoprostenol analog for primary pulmonary hypertension. 

5. McLaughlin VV, Sitbon O, Badesch DB, et al. Survival with first-line bosentan in patients with primary pulmonary hypertension. 

6. Rubin LJ, Badesch DB, Barst RJ, et al. Bosentan therapy for pulmonary arterial hypertension. 

7. Skoro-Sajer N, Lang I. The role of treprostinil in the management of pulmonary hypertension. 

8. Barst RJ, Rubin LJ, Long WA, et al; Primary Pulmonary Hypertension Study Group. A comparison of continuous intravenous epoprostenol (prostacyclin) with conventional therapy for primary pulmonary hypertension. 

9. McLaughlin VV, Genthner DE, Panella MM, Rich S. Reduction in pulmonary vascular resistance with long-term epoprostenol (prostacyclin) therapy in primary pulmonary hypertension. 

10. Olschewski H, Simonneau G, Galie N, et al; ; Aerosolized Iloprost Randomized Study Group. Inhaled iloprost for severe pulmonary hypertension. 

11. Benedict N, Seybert A, Mathier MA. Evidence-based pharmacologic management of pulmonary arterial hypertension. 

12. Sastry BK, Narasimhan C, Reddy NK, Raju BS. Clinical efficacy of sildenafil in primary pulmonary hypertension: a randomized, placebo-controlled, double-blind, crossover study. 

13. Singh TP, Rohit M, Grover A, Malhotra S, Vijayvergiya R. A randomized, placebo-controlled, double-blind, crossover study to evaluate the efficacy of oral sildenafil therapy in severe pulmonary artery hypertension. 

14. Ghofrani HA, D’Armini AM, Grimminger F, et al; CHEST-1 Study Group. Riociguat for the treatment of chronic thromboembolic pulmonary hypertension. 

. 2013;369(4):319-329. doi: 10.1056/NEJMoa1209657. 

15. Ghofrani HA, Galie N, Grimminger F, et al; PATENT-1 Study Group. Riociguat for the treatment of pulmonary arterial hypertension. 

2013;369(4):330-340. doi: 10.1056/NEJMoa1209655.

Articles

CLINICAL

In a multicenter cohort of patients with newly diagnosed pulmonary hypertension, increased risk for mortality among African Americans was no longer observed after adjusting for insurance status.

Health Insurance and Racial Disparities in Pulmonary Hypertension Outcomes

To examine the outcomes (ie, costs, hospitalizations, and mortality) associated with a Hospital-in-Home (HIH) program implemented in 2010 by the Veterans Affairs (VA) Pacific Islands Healthcare System in Honolulu, Hawaii.

 We obtained medical information for veterans who were enrolled in the HIH program in Honolulu, Hawaii, between 2010 and 2013. For purposes of comparison, we also gathered VA data to identify a cohort of hospitalized veterans in Honolulu who were eligible for, but not enrolled in, the HIH program. Using VA administrative data, we extracted a set of individual-level variables at baseline to account for the differences between program enrollees and comparators. In total, 99 HIH program enrollees and 322 unenrolled veterans were included. We identified 3 sets of outcome variables: total costs of care related to the index event (ie, HIH services for enrollees and hospitalizations for comparators), hospitalizations, and mortality after discharge from the index event. We used a propensity score-matching approach to examine the difference in related outcomes between enrollees and comparators.

The average medical cost was $5150 per person for veterans receiving HIH services, and $8339 per person for veterans receiving traditional inpatient services. The difference was statistically significant (

 <.01). There was no statistically significant difference in mortality or hospitalization rates after the index event.

 This study provides evidence of the potential benefits of a model that delivers acute care in patients’ homes. Considering the emergence of accountable healthcare organizations, interest in broader implementation of such programs may be worthy of investigation.

We evaluated a Hospital-in-Home (HIH) program implemented in Honolulu, Hawaii. Using Veterans Affairs administrative data and a propensity score-matching approach, this study provided further confirmation of the potential benefits of this new care delivery model.

		The costs of HIH services were 38% less than that of comparable inpatient hospitalizations. 

		There were no statistically significant differences in postdischarge hospitalization rates or mortality between HIH enrollees compared with veterans who were not enrolled in the program.

		With the reform of Medicare payment models and the emergence of accountable care organizations, there may be more interest in implementing HIH programs.

The hospital is the traditional place to provide acute care. However, hospitalizations can be costly and may lead to further deterioration in health status among the elderly.

The Hospital-in-Home (HIH) model offers an alternative approach to traditional hospital services.

 At the core of the HIH model is the goal of delivering hospital-level care to patients who have developed an acute episode that typically would require inpatient services, but who are medically stable enough to be treated at home.

 The common conditions that have been managed through HIH include congestive heart failure, chronic obstructive pulmonary disease, community-acquired pneumonia, and cellulitis.

 The HIH model has been associated with superior patient outcomes, such as reduced risks of delirium, improved functional status, improved satisfaction among patients and their family members, reduced mortality, reduced readmissions, and significantly lower costs.

 These studies vary in research design (eg, randomized trial or observational studies), study populations (eg, patients with different conditions in a variety of healthcare settings), and contrasting study methods and study outcomes. 

Despite promising findings, however, the HIH model has not been widely implemented in the United States. One of the barriers to the adoption of this model is the misalignment of financial incentives among patients, payers, and providers under the traditional fee-for-service payment system.

 Such disincentives may not be applicable to integrated healthcare delivery systems that are responsible for discrepancies in the costs and care of defined populations, such as managed care, accountable care, and the Veterans Health Administration (VHA) healthcare systems.

The Veterans Affairs (VA) health system is America’s largest integrated healthcare system. Its highly integrated medical delivery system aligns both financial incentives and quality care delivery, and thus is motivated to promote an HIH model. In 2010, the VA Office of Geriatrics and Extended Care (GEC) community-based transformational (T-21) programs piloted the HIH model as an alternative to inpatient services. One of the HIH programs was implemented at the VA Pacific Islands Health Care System (VAPIHCS) in Honolulu, Hawaii. This program is unique in that it does not have its own acute care hospital, and instead relies on the Tripler Army Medical Center (TAMC) in Honolulu, to provide inpatient care to its veterans. A steady flow of resources is transferred from the VA to the Department of Defense to reimburse the Army as veterans receive inpatient care from the TAMC, making the financial incentive for the implementation of the HIH program apparent at the VAPIHCS. The aim of our study was to evaluate the costs and related outcomes of this program. It was conducted as an operationally requested quality improvement project that was exempt from Institutional Resource Board review. Approval to publish these results was obtained from the GEC. 

When the VAPIHCS implemented the HIH program in September of 2010, the main targeted conditions included heart failure, pneumonia, chronic obstructive pulmonary disease, and cellulitis. To recruit patients, the HIH program staff made repeated and frequent visits to the TAMC and the VA Ambulatory Care Center in Honolulu. The HIH staff communicated with hospitalists, medical residents, discharge planners, and social workers at the TAMC, as well as VHA providers, nurses, and other staff at the facility, and maintained a close working relationship with a home-based primary care (HBPC) facility whose staff would refer eligible patients. A pamphlet was developed with basic information regarding the HIH program and outlining the patient referral process. 

One of the components of the Honolulu HIH program was the substitutive model (ie, it provided hospital-equivalent care in the veteran’s home for those who would have otherwise been hospitalized). The requirement to enroll in the program was that the veteran was living at home and had 1 of the required admitting diagnoses. Veterans were referred to the HIH program from outpatient clinics and the VA HBPC, and were screened by physicians or registered nurses for appropriate diagnoses to ensure that care could be safely provided in the home. Once admitted to the program, veterans received required intravenous (IV) infusions, respiratory treatments, laboratory tests, x-rays, and wound care in the home. They also received daily nursing and physician visits, as needed, and had 24-hour access to an on-call geriatrician. Although no specific age limit was imposed in this study, the HIH program is listed under geriatrics service, and all of the physicians covering HIH are geriatric-trained physicians. 

HIH staff ordered and delivered required supplies and medications, scheduled medical transportation, and coordinated care to meet the medical needs of veterans, their families, and caregivers. The HIH nursing staff took calls during duty hours on weekends and holidays, and worked with an outside pharmacy to provide all needed IV medications. Veterans were discharged from the program once symptoms improved; ultimately, they transitioned to their usual primary care providers (PCPs) or HBPC, if necessary. A discharge summary was placed in the Computerized Patient Record System, the patient’s Patient Aligned Care Team was notified, and follow-up care appointments were made with the PCP.

 To evaluate the program, we (the GEC Data Analyses Center) followed specific steps. First, basic information was requested from the HIH program director to identify veterans who received care through this program. Social security number (SSN), date of birth, and gender were used to obtain the veteran’s identifier number (ie, scrambled SSN) used in the VA data systems. Dates of HIH program enrollment and disenrollment were also used to define the “active” period within the program and to establish both the program pre-enrollment phase prior to the trial, and the follow-up period following the program’s completion. 

Data were then obtained from multiple VA administrative sources via the VA’s Corporate Data Warehouse. Specifically, data included vital status file, enrollment file, patient file, patient treatment file, outpatient file, Ward file, pharmacy file, inpatient treating specialty file, and fee-basis file from January 1, 2007, to September 30, 2013. These files contained demographic information (ie, age, gender, and race), VA enrollment status, and information about VA-provided services or VA-paid services, including the 

International Classification of Diseases, 9th Revision, Clinical Modification

) codes, dates of service, medication use, and the costs of VA-provided and VA-purchased services. These VA data were linked by individually scrambled SSNs and compiled chronologically into a single file to track veterans’ utilization of health services.

The study population included HIH enrollees who were admitted to the Honolulu HIH program from a noninstitutional location (eg, outpatient clinics or home) between September 2010 and June 2013, and they were compared with a group of veterans who were not enrolled in the program, but were comparable to the enrollees.  To identity potential comparators for the study, we used administrative data to include veterans who were not enrolled in the HIH program, but who were admitted to the TAMC for inpatient services from a noninstitutional location between January 2008 and September 2012 in Honolulu, Hawaii. We did not include nonenrollee veterans who had hospitalizations in fiscal year 2013 (FY13) as controls due to the concern of potential bias. Because of the expansion of the program, it is likely that the program enrolled most of the eligible veterans in FY13. Thus, those who were not enrolled during the same time period could be quite different from those who were enrolled. We then identified the primary diagnosis associated with hospital admissions and selected patients whose primary diagnosis was among the set of admission conditions stipulated for HIH enrollees. Furthermore, we identified the diagnosis-related group (DRG) and identified select hospital admissions with low-weight medical DRGs, with the assumption that the HIH enrollees did not have complications. This resulted in the 99 program enrollees and 322 nonenrolled veteran comparators who were identified for the study

Three sets of outcome variables were evaluated. The first outcome was the total cost of care, including inpatient, outpatient, or pharmacy costs, incurred during the index event. These services were either provided by the VA (ie, DSS data) or paid for by the VA (ie, TAMC-provided inpatient services were captured on fee-for-service files). The duration of the index event referred to the time between the admission and discharge dates of HIH services for enrollees and hospitalizations for nonenrollees. Costs were adjusted to 2013 dollars by the Consumer Price Index. The second set of outcomes included the incidence of 30- and 90-day hospitalizations after the discharge of the index event. The third set of outcomes entailed 30-, 90-, and 180-day mortality after the veteran had been discharged from the index event.

Based on VA administrative data, we extracted a set of individual-level variables at baseline (ie, the admission of the index event) to account for the potential differences between program enrollees and comparators. These variables included individual sociodemographic characteristics (eg, age, gender, race, and VA priority status that determined eligibility for Priority group 1 [veterans with 50% or more service-connected disabilities that precluded employment]), prior VA healthcare utilization (eg, number of inpatient events and total VA-paid costs in the 3, 6, and 12 months prior to the index event). In addition, we utilized VA claims data to obtain the 

 codes recorded within the 1-year period prior to the admission of the index event, and identified a set of chronic conditions for each veteran. We also calculated the number of different drug classes used by the veterans within 1 year prior to the index event. We categorized the number of drug classes into 3 groups, including fewer than 5 drug classes (lower 25th percentile), 5 to 10 drug classes, and more than 10 drug classes (upper 25th percentile), to allow for the potential nonlinear relationship between the number of drug classes and outcomes. These individual-level characteristics could be correlated with the outcomes and, thus, the differences in each factor between enrollees and other subjects were considered.

We used a propensity score-matching approach to examine the difference in related outcomes between enrollees and controls. Specifically, a logistic regression was first estimated to predict the probability of HIH enrollment, based on individual factors we identified at the baseline, as described previously. After obtaining the probabilities, nonenrollees were matched to program enrollees based on the enrollee’s probability of enrollment. (We used a radius-matching approach and required nonenrollees’ probabilities of enrollment to be within the radius of 0.01 of the enrollees’ probabilities). The outcomes were then compared between the 2 groups of veterans with matched probabilities. If there were multiple matched nonenrollees to 1 enrollee, the average difference in outcomes was calculated. The average treatment effect on the treated (ATT) population which represents the average difference in outcomes between the enrollees and matched nonenrollees) was then estimated by averaging over the unit-level treatment effects across matched pairs. The standard errors for the difference in outcomes (ATT) were obtained by a bootstrapping approach with 1000 iterations. The propensity score-matching process was performed by a STATA procedure, PSMATCH2. All the analyses were performed in STATA version 13 (StrataCorp; College Station, Texas) and SAS version 9.3 (SAS Institute Inc; Cary, North Carolina).

The average age of enrollees was 73 years. As expected, the majority of veterans were admitted for pneumonia, congestive heart failure, diabetes, unitary tract infection, or cellulitis. The remaining veterans were admitted for management of other heart conditions (eg, chronic ischemic heart disease), kidney conditions (eg, pyelonephritis), or infections (eg, osteomyelitis). Similarly, approximately 90% of nonenrollees were hospitalized with a primary diagnosis of cellulitis, pneumonia, unitary tract infection, and heart failure. The DRG weights of these hospitalizations varied between 0.6176 and 0.9938 for nonenrollees. On average, the LOS of HIH services was 9.24 days for enrollees and 6.21 days for inpatient events.

presents a comparison of individual characteristics between enrollees and nonenrollees before and after propensity score matching. Individual characteristics were reasonably balanced through the propensity score-matching approach, with the average age being 73 years for enrollees and 74 years for comparators. About 95% of enrollees and 97% of the comparators were male. The prevalence of chronic conditions for the 2 groups was similar, with approximately 24% of both enrollees and comparators experiencing an inpatient event within 90 days of the index event. Overall, the comparators were not sicker than enrollees, and the likelihood of overestimating the effect of the HIH program would be small.

Eight enrollees could not be matched with a nonenrollee veteran, and therefore, were not included in the propensity score-matched population. The unmatched enrollees seemed to be sicker than the other enrollees at baseline. For example, on average, they were 7 years older than the other enrollees and had more comorbidities. They also incurred higher healthcare costs prior to the index event than the other enrollees (data were not presented in the table).

The comparison of the outcomes between enrollees and comparable nonenrollees after the propensity score matching (ie, the ATT population) is shown in 

 The HIH program was associated with a 38% reduction in costs compared with an inpatient event. The average cost for HIH services per enrollee was $5150, and the average cost of an inpatient stay among the comparators was $8339 per person. The difference, $3189 (38.2%), was statistically significant (

<.01). The differences in the 30-day and 90-day rehospitalization rates after discharge were not statistically significant (ie, the 30-day rehospitalization rate was 14.3% for enrollees and 12.6% for the comparators; 

= .75), and the 90-day rehospitalization rate was 16.5% for program enrollees and 21.4% for comparators; 

 = .38). The 30-, 90- and 180-day mortality rates were 6.6%, 11,0%, and 13.2%, respectively, for enrollees, and 6.2 %, 14.1%, and 19.7%, respectively, for comparators. These differences were not statistically significant.

The growing population of aging veterans and the increasing demand for inpatient services present a challenge to the VA healthcare system. This study evaluated an HIH program implemented among veterans in Honolulu, Hawaii. It was an observational study, and the identification of the comparison group and evaluation of the outcomes were based on administrative data. A propensity matching approach was applied to account for the potential differences between the enrollees and the comparison group. Although the study population and the study approach were different from previous studies,

 the findings were consistent in that we found that the HIH program provided hospital-equivalent care to eligible patients at a lower cost. Unlike some of the similar programs,

 the VAPIHCS did not operate an inpatient hospital, and so, creating the program required diligence in planning the care and the provision of special services to be developed from the beginning, including a comprehensive set-up of medical equipment, supplies, and personnel. Many managed care organizations are in a similar position as the VAPIHCS in that they are based in physician groups rather than health systems that operate hospitals, and it is reassuring that a nonhospital can successfully implement an HIH program. 

We followed the HIH enrollees and the veterans chosen as comparators following their discharge from the HIH program or hospital, and did not find any statistically significant differences in mortality or rehospitalizations. On the other hand, it is likely that we underestimated the positive outcomes associated with the HIH program. For older adults, hospitalizations are associated with a high risk of adverse events (AEs), such as delirium, falls, and nosocomial infections.

 These hospital-acquired AEs not only lead to impairment in patients’ health status, but can also be costly.

 Providing acute services at a patient's home, when possible, is a logical way to limit these negative outcomes, and has been found to be associated with better functional status and better quality of life (QOL) compared with those who received hospital inpatient care.

The HIH model is well aligned with the goal to deliver patient-centered care, and the HIH team identifies each veteran’s specific needs and provides high-quality care to accommodate them in a uniquely personal setting: the veteran’s residence. Veterans can benefit from support of family members while receiving hospital-equivalent services, which could potentially improve QOL and general welfare. 

The HIH model also has the potential to reduce the extent of care discontinuity occurring during an acute episode. For example, instead of sending a veteran to TAMC, the Honolulu HIH staff collaborates with HBPC staff, geriatric healthcare providers, and primary and specialty care staff to provide necessary services in patients’ homes. Such collaboration and coordination facilitates the exchange of information among healthcare providers and ensures that appropriate medical measures are administered to veterans during and after participation in the program.

However, these findings should be interpreted with caution. Unfortunately, we were not able to assess change in veterans’ functional or cognitive status due to a lack of available data. In addition, we were not able to compare an individual’s satisfaction level with their experience with the HIH program versus hospitalizations. We are also likely to have underestimated the positive outcomes of HIH programs as we only selected hospitalizations with low-weight DRGs, and it is likely that the HIH program provided care to patients with more complicated diagnoses.

 Furthermore, the propensity score-matching approach only accounted for observed factors. This was not a randomized trial, and therefore, it is likely that there were some unobserved differences between enrollees and comparators that could have confounded the effect of the HIH program on outcomes. For example, the severity of individual conditions at admission—which can be related to costs, rehospitalizations, and mortality&mdash;may not be completely captured by administrative data. If this is the case, our findings may be biased. However, we have accounted for critical individual-level factors. Furthermore, based on the observed characteristics between enrollees and nonenrollees at baseline, nonenrollees did not seem to be sicker than enrollees, making the likelihood of overestimating the benefits of HIH services low. Last, we only investigated the costs incurred within the VA system. It is possible that veterans in both groups differentially used non—VA-paid services, such as Medicare or Medicaid, which were not captured by our study.

Earlier studies have demonstrated the success and promise of the HIH model. This program demonstrated that it was possible to build a program without a hospital, and provided a useful example to managed care systems wishing to develop such a cost-effective program. With Medicare payment reform (eg, accountable care organizations) and incentives for providers to administer more efficient care, there may be pronounced interest in implementing programs, such as HIH, more broadly.

The authors would like to thank the Veterans Health Administration’s Office of Geriatric and Extended Care (GEC) Transformational Initiatives for supporting the Hospital-in-Home program. The authors also thank Dr Kenneth Shay and Karen Massey in the VA Office of GEC for their support in the evaluation of the program and comments on the manuscript.

Geriatrics & Extended Care Data Analyses Center (SC, OI, RM, CSP, BK), Canadaigua, NY; Department of Public Health Sciences, University of Rochester (SC,OI), Rochester, NY; Canandaigua VA Medical Center (SC,OI), Canandaigua, NY; VA Pacific Island Health Care Systems (MLM, PL), Honolulu, HI; Center for Gerontology & Healthcare Research, Brown University School of Public Health (RM), Providence, RI; Corporal Michael Crescenz VA Medical Center (BK), Philadelphia, PA; Division of Geriatrics, University of Pennsylvania School of Medicine (BK), Philadelphia, PA; Health Economics Resource Center and Center for Innovation to Implementation, VA Palo Alto Health Care System (CSP), Menlo Park, CA; Department of Pediatrics, Division of Neonatal & Developmental Medicine, Stanford University School of Medicine (CSP), Stanford, CA. 

 The Hospital-In-Home program and the evaluation of the program conducted by the National Geriatrics & Extended Care Data & Analyses Center were supported by the Veterans Health Administration’s Office of Geriatric and Extended Care Transformational Initiatives.

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article. 

 Concept and design (SC, PL, MLM, BK, CSP, OI); acquisition of data (PL, RM, MLM, CSP, OI); analysis and interpretation of data (SC, RM, BK, CSP, OI); drafting of the manuscript (SC, BK); critical revision of the manuscript for important intellectual content (PL, MLM, BK, CSP, OI, RM); statistical analysis (SC, RM, CSP); obtaining funding (OI); administrative, technical, or logistic support (OI); and supervision (OI).

 Shubing Cai, PhD, Department of Public Health Sciences, University of Rochester, 265 Crittenden Blvd, Rochester NY 14642. E-mail: shubing_cai@urmc.rochester.edu. 

1. Creditor MC. Hazards of hospitalization of the elderly. 

2. Brennan TA, Leape LL, Laird NM, et al. Incidence of adverse events and negligence in hospitalized patients. results of the Harvard Medical Practice Study I. 

3. Leape LL, Brennan TA, Laird N, et al. The nature of adverse events in hospitalized patients. results of the Harvard Medical Practice Study II. 

4. Pfuntner A, Wier L, Steiner C. Costs for hospital stays in the United States, 2010. Healthcare Cost and Utilization Project website. http://www.hcup-us.ahrq.gov/reports/statbriefs/sb146.pdf. Published January, 2013. Accessed March 8, 2015.

The Hospital-in-Home program implemented at the Veterans Affairs Pacific Islands Health Care System in Honolulu, Hawaii, is associated with reduced costs with no compromise in quality.

Evaluation of a Hospital-in-Home Program Implemented Among Veterans

To estimate the long-run average cost (LAC) for a typical drug, accounting for the effects of generic competition and medical cost offsets. 

Descriptive analysis of retrospective cross-sectional survey data.

 We estimated the LAC for a drug as the average price per unit paid over the lifecycle of the drug, discounted across all time periods using Medical Expenditure Panel Survey data, and accounted for the effects of generic competition and medical cost offsets attributable to the use of pharmaceuticals.

 The average market-weighted price fell rapidly after generic entry. As a result, the brand price in the year prior to generic market entry was 39% (95% confidence interval [CI], 37%-43%) higher than the LAC per 30-day supply or package. When accounting for medical cost offsets, the brand price in the year prior to generic market entry was 75% (95% CI, 69%-79%) greater than the LAC per 30-day supply or package. The brand price at launch was 11% more than the LAC, and 40% more than the LAC net after adjusting for medical cost offsets. 

 Branded drug prices might overstate the true long-run cost of pharmaceuticals by 40% to 75%, accounting for generic price reductions and medical cost offsets. To ensure that all drugs providing long-run value end up entering the marketplace, market access and other policy decisions should consider the full range of long-term costs—and not just prices&mdash;at a particular point in time.

Using Medicare Expenditure Panel Survey data (1996-2013), this study estimated the long-run average cost (LAC) for an average pharmaceutical, accounting for the effects of generic competition and medical cost offsets. 

Focusing on branded drug prices may significantly overstate the LAC. 

Accounting for patent expiration, the loss of exclusivity price and the launch price overstate the LAC by 39% and 11%, respectively, and the LAC net of medical cost offsets by 75% and 40%, respectively. 

To ensure that all drugs providing long-run value end up entering the marketplace, value assessments or related market access decisions should consider these long-term costs.

The policy debate over the price of prescription drugs continues to intensify. For example, the recent introduction of very expensive and highly effective treatments for hepatitis C, along with innovative therapies for many cancers, has deepened concern about patients’ ability to afford innovative medicines.

 However, the debate does not consider the effect of patent expiration and the loss of exclusivity on drug prices. The price of prescription drugs depends on the nature and consequences of generic competition. Criticism has swirled around a cadre of generic drug manufacturers inflating prices for long-established compounds that, for various reasons, lack generic competitors. For instance, the price of digoxin, an “ancient” treatment for heart failure and atrial fibrillation, deemed an essential medicine by the World Health Organization, doubled over 6 months.

 However, for many widely used drugs, generic competition reduced prices significantly—and swiftly. Patent expiration lowered the price of simvastatin by 89% in 5 years and the price of clopidogrel by 46% in 1 month.

 These conflicting anecdotes provide little guidance to policy makers about the extent to which patent expiration lowers the long-term cost of drugs. Much has been written about the ability of drugs for specific diseases to reduce the need for other medical costs, yet little is known about the extent these “cost offsets” lower long-term costs associated with the average drug.

To address these knowledge gaps, we estimated the long-term cost of prescription drugs to patients, accounting for patent expiration and medical cost offsets. Focusing on the price of a drug at a given point in time can help inform discussions on which patients should use a drug and what the aggregate costs will be at that point

However, the value of innovation depends on the long-term cost of a new drug, and this has not been quantified previously. 

Drugs add long-term value to society when each unit sold produces a long-term benefit that exceeds long-term cost. To facilitate the assessment of long-term value, we estimated the long-run average cost (LAC), defined as the long-run cost per unit of the drug. We derived a formula for the LAC that formalizes this intuition (

The formula expresses the LAC as an average cost per unit of a drug. To illustrate, imagine that a branded drug costs $5 per fill and its generic equivalent costs $1 per fill. Imagine further that, over the lifecycle of the drug, 1000 branded prescriptions are filled and 5000 generic prescriptions are filled. The LAC is given by the weighted average cost per fill, which equals: ($5 × 1000 + $1 × 5000) / (1000 + 5000) = $1.40. 

Apart from generic price reductions, there are 2 additional factors we incorporated into this simple framework. First, a given drug may be produced as multiple dosages or by multiple manufacturers. Thus, the LAC calculates prices and quantities across a number of different dosages and product types. Continuing with the example above, suppose the drug comes in 5-mg and 10-mg doses, and each dosage accounts for half of branded fills; the 5-mg dose costs $2.50 and the 10-mg dose costs $7.50. In this more complex setting, we continued with our principle that the LAC was given by the weighted average cost per fill, which equals: ($2.50 × 500 + $7.50 × 500 + $1 × 5000) / (500 + 500 + 5000) = $1.40.

Second, timing matters. Using a prescription today is worth more than delaying until tomorrow. In economic analysis, this is known as the social discount factor, which measures the rate at which consumer well-being decreases from delaying consumption. Following the economics literature, we employed a 3% annual discount rate.

 Continuing with the earlier example, suppose that our hypothetical drug was on patent for 1 year only, generics entered after that year, and all drugs sold in year 2 were generic. The discount factor would enter our LAC calculation (in italics) as follows: ($2.50 × 500 + $7.50 × 500 + $1 × 5000 

) = $1.68. Discounting lowers the contribution of future prescriptions and thus shifts up the LAC in the direction of current period prices. In the eAppendix, we demonstrated proof that the LAC, as formulated here, possessed the following desirable and intuitive property: a drug adds incremental value to society if, and only if, the value per prescription exceeds the LAC.

We next incorporated medical cost offsets in the LAC. We took the average reduction in cost per prescription occurring as a result of avoided nondrug medical costs from the literature. We applied this cost reduction to the LAC to derive the LAC net of medical cost offsets (LAC net). A formal expression for the LAC net can be found in the eAppendix. As an example, we supposed that our hypothetical drug saved 25 cents in medical costs per prescription sold. This cost offset would enter our calculation (in italics) as follows: ([$2.50 — 

] × 5000 × [1 — 0.03]) / (500 + 500 + 5000 × [1 – 0.03]) = $1.43. Notice that adding the cost offset brought down the LAC by exactly 25 cents. In our empirical analysis, we were particularly interested in how the LAC and the LAC net compared with the price of the drug at the time of generic entry and at launch. 

The first step was to estimate prices for a wide sample of drugs. The host database for the price analysis was the Medical Expenditure Panel Survey (MEPS) Prescribed Medicines File from 1996 to 2013.

 The MEPS data are nationally representative, with detailed information on healthcare use, expenditures, sources of payment, and health insurance coverage. It has a household component that provides respondents’ self-reported information, including utilization of prescribed medicines. 

We defined the year of loss of exclusivity (LOE) as the first year we observed utilization of a generic National Drug Code (NDC) for a drug. The sample of medicines included those drugs whose LOE occurred in 1997 or later. Within this sample, we observed drug prices as early as 17 years before LOE and up to 16 years after LOE. For example, if a drug lost exclusivity in 2000, we observed prices for 3 years pre-LOE and 14 years post LOE. There were no biologics present in the sample because we did not observe LOE before 2013. We also excluded vaccines, medical devices, prescribed vitamins, and drug classes not meant for immediate patient use. 

The MEPS Household Component and pharmacy follow-back surveys measured pharmacy transaction prices. Rebates paid by the manufacturer to the insurer (eg, through the Medicaid Drug Rebate Program or contractual agreements negotiated between manufacturers and private insurers) were excluded. Prior literature has estimated that MEPS aggregate drug spending figures are modestly higher (3%) than national benchmarks,

 possibly due to the exclusion of rebates and other potential sources of measurement error. We conducted a sensitivity analysis to assess the potential effect of passed-through rebates on our estimates by lowering MEPS branded drug prices by 10% during the exclusivity period.

For each instance of drug utilization in the MEPS, the database records the associated NDC. As a single drug may have multiple NDCs, we used the Medi-Span Electronic Drug File to create a crosswalk from the NDC number to a drug—dose level identifier and linked this crosswalk to the MEPS data by NDC. Medi-Span is a database of drug and clinical information including active ingredients, generic equivalents, and therapeutic class. 

Prices were aggregated from the NDC level up to the drug—dose level in 3 stages. First, as each MEPS respondent could have had multiple purchases of the same NDC within a year, we calculated the average price paid per individual by year and NDC. Then, we constructed the national average by computing the survey weighted average price for each NDC by year across users of that NDC. Lastly, we calculated the NDC utilization weighted average of each drug by dose (or package size), type (brand or generic), and year.

Our sample included 132 therapeutic classes, 259 drugs, and 1229 different drug—dose combinations. The sample contained all drugs that experienced LOE during the sample window, with the exception of the exclusions noted above. This covers a broad set of drugs for a wide variety of diseases. We observed prices and utilization of these drugs from 1997 to 2013. 

We measured quantity as the number of prescription fills of a drug, dose (or package size), and type (brand or generic) per year. Quantities were aggregated up to the national level by summing the product of the MEPS survey weight and number of fills for each unique individual using a particular drug, dose, and type by year. 

Lastly, we estimated the direct medical cost offset per prescription fill. We conducted a targeted literature search for reviews and meta-analyses on the effects of pharmaceutical use on direct costs or medical/nonpharmaceutical spending. The goal was 2-fold: to estimate the direct cost offsets associated with pharmaceutical use and determine a plausible range of effects for medical cost offsets. We searched for estimates of the reduction in direct costs/medical expenditures associated with the use of a wide group of drugs, rather than only a few specific classes, for 1 month (30-day supply). 

Pharmaceutical Utilization and Prices Over Time

 illustrates that the average number of brand fills declined significantly after LOE (year post LOE of 0) and was more than offset by a concomitant increase in generic users in the following years. In the year generics entered the market, they captured 33% market share on average. This rapid capture of market share is consistent with previous research.

 After 5 and 10 years, generics possessed 74% and 77% of the market, respectively. 

 depicts the year-to-year time series for average brand or generic price relative to the price in the year before generic entry (LOE price). The average brand price increased throughout the drug’s lifecycle, but the average generic price tended to decrease for at least 10 years. The average generic price was 66% below the LOE price 5 years after generic entry and 80% below 10 years after generic entry. 

 presents the implications for the price of the average prescription (generic or branded) sold in the marketplace. This Figure demonstrates that the average market-weighted price fell rapidly after generic entry. In the year generics entered the market, the market price was 6% lower than the LOE price. After 5 years, it was 55% lower than the LOE price, and after 10 years, it was 71% lower.

Next, we explored the extent to which medical cost offsets reduced the long-term cost of pharmaceuticals. Our targeted literature search generated 13 articles on direct cost offsets associated with pharmaceutical use and/or pharmaceutical spending.

 Four of these studies contained information and estimates that allowed for the calculation of the effect of pharmaceutical use on medical (nonpharmacy) spending. Only 1 study, from the Congressional Budget Office (CBO), was representative of a large population (Medicare beneficiaries) and a broad set of therapeutic classes.

In 2012, the CBO estimated that a 1% increase in the number of prescriptions filled per year resulted in a 0.2% (95% confidence interval [CI], 0.1%&shy;-0.4%) decrease in medical expenditure per year. Based on the 2012 CBO study

 and information on the aggregate number of prescription fills and total healthcare expenditures in 2014, we estimated that each additional prescription filled (30-day supply) reduced medical expenditures by $94 per year ($8/month) (

 The cost savings in Medicare for specific chronic conditions, including diabetes and hypertension, were even higher.

We first calculated the LAC, which reflects the effects of generic entry and price declines. We reported this as a fraction of the LOE price and estimated that the LOE price was 39% (95% CI, 37%-43%) higher than the LAC per 30-day supply or package (

We calculated direct medical cost offsets of $94 per year ($8/month) then estimated the LAC net after applying this cost reduction to the drug prices in the data. On average, the LOE price was 75% (95% CI, 69%-79%) greater than the LAC net per 30-day supply or package. 

Pharmaceutical Costs Relative to Launch Prices

The analysis above compared the LOE price with LAC and LAC net. Another relevant comparison was to the brand price at launch. This calculation was more difficult, because the length of our sample window did not allow us to follow many drugs from launch well into the post—patent expiration period. As a rough approximation, we calculated the average percentage difference in price between the year prior to generic entry and the average year of launch. Using the FDA Orange Book Data, we found that for the average small molecule, 14 years elapse between the drug launch and the LOE. Moreover, according to Figure 3, the (real) brand price 14 years prior to generic entry was 80% of the price in the year just prior to generic entry. The LOE price was 39% more than the LAC and, thus, it follows that the launch price was 11% (= 1 – [0.8 × 139%]) more than the LAC. Similarly, the launch price is 40% (= 1 – [0.8 × 175%]) more than LAC net. 

As it is well understood that generic entry lowers prices, policy makers should consider the long-term cost of pharmaceuticals, rather than the cost at a particular point in time. We showed that these price reductions meaningfully lowered the average cost per prescription that patients pay over a drug’s lifetime. Branded drug prices were 11% above the LAC at launch and 39% above the LAC at patent expiration. Incorporating cost offsets brought these differentials to 40% and 75%, respectively, above the LAC net. Noting that long-run generic prices lie 80% below the branded price at patent expiration, the LAC net was roughly halfway between the pre-expiration branded price and the long-run generic price. 

Branded prices, generic prices, and the LAC all play important roles in economic decisions, which are made on the margin. Prices at a point in time matter to payers, who must decide if the benefit of treating 1 more patient outweighs the cost. The LAC and the LAC net, however, should matter to regulators, policy makers, and payers assessing whether a new drug can be marketed or reimbursed. In this context, overstating the eventual cost of a drug may lead to fewer drugs being made available, weaker incentives to innovate, and ultimately, fewer new drugs discovered. The use of the LAC and the LAC net could mitigate these effects by informing decision makers about the true value drugs add to society along all their lifecycle. For example, health technology assessments (HTAs) could include an assessment of the cost-effectiveness ratio that discounts the launch price—the most frequently used measure of drug cost&mdash;by 11% (excluding cost offsets) or 40% (including cost offsets). Not only should HTAs incorporate medical cost offsets, but they should also consider the long-run unit cost of a drug. 

As research has demonstrated, robust generic drug markets play a key role in holding down the LAC of drugs. Thus, policy makers should aim to mitigate barriers to entry for new generic drugs and ensure the continued safety of the generic drug supply. This is particularly relevant for specialized products that do not attract significant generic entry or competition. Recent events (eg, the 5000% overnight price increase of daraprim) demonstrate the importance of preserving competition within those market segments.

Medical cost offsets play a material role in holding down the LAC. This suggests the importance of taking an integrated view across pharmacy and medical benefits. Some have argued that standalone prescription drug insurers offer less-efficient benefit designs because they fail to internalize spillovers between prescription drugs and medical care.

Our estimate of the LAC may have overstated the actual LAC because we lacked data on rebates paid from manufacturers to payers. These may have been significant during the period of patent protection. Our analysis was also limited by the timeframe of the MEPS data. We were only able to characterize trends in price and utilization for drugs that lost exclusivity in 1997 or later; thus, we could only observe the first 15 years after LOE, even though many drugs were used for more than 15 years following LOE. On the other hand, we observed up to 16 years prior to LOE, likely covering the vast majority of the patent-protected period when market prices are highest. Therefore, our estimate might overstate the LAC. 

Our estimate covered only the population of drugs for which generics end up being used. Although this is not fully representative, this represented the vast majority of drug utilization. In 2013, for instance, drugs with generic alternatives represented at least 86.4% of prescription drug utilization in MEPS and these generic alternatives presented 81.0% of prescription drug utilization in MEPS. As some of the remaining 13.6% will inevitably include some patent-protected drugs that will eventually go generic, this is a conservative estimate of the fraction of use due to drugs with generic utilization. 

This study found that brand price at launch and generic entry overstates long-run average pharmaceutical costs, with and without accounting for medical cost offsets.

August 2017

Health Insurance and Racial Disparities in Pulmonary Hypertension Outcomes

Evaluation of a Hospital-in-Home Program Implemented Among Veterans

What Do Pharmaceuticals Really Cost in the Long Run?

The Effect of Implementing a Care Coordination Program on Team Dynamics and the Patient Experience

The Hospital Tech Laboratory: Quality Innovation in a New Era of Value-Conscious Care

Association Between Length of Stay and Readmission for COPD

Risk Stratification for Return Emergency Department Visits Among High-Risk Patients

Cost-Effectiveness Analysis of Vagal Nerve Blocking for Morbid Obesity

Impact of Formulary Restrictions on Medication Use and Costs

Geographic Variation in Medicare and the Military Healthcare System