Compliance with drug therapies for the treatment and prevention of osteoporosis

Transcription

1 Maturitas 48 (2004) Compliance with drug therapies for the treatment and prevention of osteoporosis Jeffrey S. McCombs a,, Patrick Thiebaud b, Connie McLaughlin-Miley c, Jinhai Shi c a Department of Pharmaceutical Economics and Policy, School of Pharmacy, University of Southern California, Los Angeles, CA, USA b Department of Economics, University of Southern California, Los Angeles, CA, USA c Amgen Inc., Thousand Oaks, CA, USA Received 6 June 2003; received in revised form 14 January 2004; accepted 23 February 2004 Abstract Objectives: This study used paid claims data from real-world treatment settings to investigate the impact of hormone replacement therapy (HRT), bisphosphonate and raloxifene on patients with a recorded diagnosis of osteoporosis. Methods: Data from a large health insurer were used to identify 58,109 osteoporosis patients who initiated drug therapy for osteoporosis. Multivariate statistical models were developed for duration of therapy, compliance at 1 year, time to discontinuation or a change in therapy, health care costs and risk of fracture over 1 year. Results: One-year compliance rates were below 25% for all osteoporosis therapies. The mean unadjusted duration of continuous therapy was 221 days for raloxifene, 245 days for bisphosphonate, 262 for estrogen-only and 292 days for estrogen plus progestin. Raloxifene patients were consistently less compliant than estrogen-only patients after adjusting for differences in patient characteristics. Estrogen plus progestin patients were generally more compliant while bisphosphonate did not differentiate from estrogen-only. Compliance reduced the risk of hip fracture (o.r. = 0.382, P<0.01) and vertebral fracture (o.r. = 0.601, P<0.05). Compliant patients used fewer physicians services ( US$ 56, P<0.0001), hospital outpatient services ( US$ 38, P<0.05) and hospital care ( US$ 155, P<0.01). Bisphosphonate patients were twice as likely as estrogen-only patients to experience vertebral, Colles and other fractures and experienced higher health care costs (+US$ 420, P<0.01). The effectiveness of both raloxifene and bisphosphonate medications relative to estrogen-only improved significantly with the age of the patient. Conclusions: Compliance with drug therapies for osteoporosis over 1 year is poor leaving patients at risk for fractures and higher health care costs Elsevier Ireland Ltd. All rights reserved. Keywords: Estrogen; Health care costs; Osteoporosis 1. Background Corresponding author. Tel.: ; fax: address: jmccombs@usc.edu (J.S. McCombs). Osteoporosis is a disease characterized by low bone mass density and micro-architectural deterioration of bone that increases the risk of bone fracture resulting in pain and deformity. Osteoporosis is considered a /$ see front matter 2004 Elsevier Ireland Ltd. All rights reserved. doi: /j.maturitas

2 272 J.S. McCombs et al. / Maturitas 48 (2004) significant public health concern that will be magnified in the future as the population ages in the developed world. The National Osteoporosis Foundation (NOF) [1] estimates that low bone density affects about 44 million men and women in the US based on 2000 census data. This represents 55% of people 50 or older in the US, of which 10 million (8 million women and 2 million men) suffer from osteoporosis. The condition disproportionately affects postmenopausal women, especially those who are Caucasian or Asian. Melton et al. [2] found that 45% of white women over age 50 in Minnesota had a bone density greater than 2 standard deviations below the mean of healthy young women. The major consequence of osteoporosis is bone fracture due to bone fragility [3]. The NOF report [1] estimates that 1.5 million fractures are caused annually by osteoporosis, including 300,000 hip fractures, 700,000 vertebral fractures, and 250,000 wrist (Colles) fractures. Lifetime risk of fracture has been estimated at: 6% men, 18% women (hip); 5% men, 16% women (vertebral); 3% men, 18% women (wrist) [6]. In any year, fracture occurs in 1 2% of women around age 65 and 6 10% of women age 75 [7]. Several studies have tackled the issues of excess mortality and morbidity caused by osteoporosis. The consequences of fracture depend on the type of fracture considered. Mortality increases substantially immediately after hip fracture and returns to normal after 2 years [6,8]. The pattern is reversed for vertebral fractures: mortality diverges from normal rates at an increasing pace from the fracture date. No significant difference was observed for forearm fractures. Hip fracture is associated with greater morbidity than vertebral or wrist fractures; it is particularly disabling and almost always requires hospital stays. Cooper [6] reports that of those who are independent before the fracture, 74% remain independent, 18% become dependent, and 8% enter a nursing home. Of those already dependent before the fracture, 50% enter a nursing home. A limited number of recent studies evaluate the cost of osteoporosis. In the US, the direct annual costs were estimated at US$ 14 billion in 1995 [4] of which 62% was for hospital services [5]. However, these estimates ignore the effect of bone fragility and fractures on quality of life. Martin et al. [3] describe the cost on Medicaid and Medicare of osteoporosis-prone Medicaid recipients. They evaluate the impact of fractures in women over 50 on hospital, physician, long-term care, drug use, and miscellaneous expenditures, controlling for patients prior health care utilization. They find an average cost increase of US$ 8986 per person-year for femur fracture and US$ 3326 per person-year for non-femur fracture. Johnell [9], reviewing studies conducted in different countries, observed that the direct cost of hip fracture is US$ 7000 in direct hospital cost and US$ 21,000 in total cost over 1 year after the occurrence of the fracture. Brainsky et al. [10] estimate the marginal cost of a hip fracture at between US$ 16,322 and US$ 18,727 while Cummings et al. [11] estimate that the number of hip fractures in the United States would increase from 261,000 in year 2000 to 512,000 in Hip fractures are by far the most expensive to treat as they almost always involve hospitalization and long rehabilitation periods. In contrast, forearm fractures do not usually require inpatient stays. Long-term compliance with osteoporosis treatment can be problematic. Kotzan et al. [12], in a study of postmenopausal Medicaid women, found that 54% of women remained on treatment for more than 29 months and 17% for more than 35 months. Cano [13] assessed compliance in 331 menopausal women treated with hormone replacement therapy (HRT) over a period of 5 years and reports that 9% of women never filled their prescription, 15% interrupted their treatment, and 14% followed their treatment intermittently. Faulkner et al. [14] conducted a retrospective study among 29,000 women aged who were new HRT users. Compliance was defined there as a medication possession ratio (MPR) greater than After 1 year, 54.4% of women were non-compliant with HRT. Cole et al. [15] assessed compliance in a sample of 222 women between the ages of 23 and 83 who had received a bone mineral density measurement (BMD). Compliance over a 9-month follow-up period ranged between 60% for HRT to 95.5% for Vitamin D, with the rates for alendronate sodium, calcitonin, and calcium being 82, 87, and 85%, respectively. Kayser et al. [16] evaluated the differences in continuation among 1394 women age 60 or above initiating treatment with either raloxifene or estrogen. At 24 months, 72% of the women who started on estrogen had discontinued their treatment versus

3 J.S. McCombs et al. / Maturitas 48 (2004) % of raloxifene patients. Marwick [17] reported that 27% of 1586 women enrolled in the Harvard Community Health Plan who received a new prescription for HRT terminated therapy within 100 days and 40% stopped taking HRT within 1 year. Bjorn and Backsrom [18] report that negative side-effects were a major cause of non-compliance with HRT. Ettigenger et al. [19] found that nearly 35% of 812 women using alendronate to treat osteoporosis discontinued therapy at 6 months based on prescription refill data. 2. Objectives This study is designed to investigate patient compliance with drug therapies use to prevent and treat osteoporosis in real-world treatment settings, and to estimate the cost-consequences associated with non-compliance. Particular attention will be paid to comparing the patient outcomes achieved using estrogen-only, estrogen plus progestin, raloxifene or bisphosphonate. This study is particularly important on three fronts. First, the cost-effectiveness of alternative drug therapies to treat osteoporosis in real-world clinical practice settings requires patients to maintain continuous drug therapy in excess of 10 years [20]. Second, more information is needed on the relative performance of newer alternatives to HRT for treating osteoporosis and preventing adverse patient outcomes, such as fractures. Finally, a recent randomized clinical trial of the long-term use of estrogen and progestin found increased risks for coronary heart disease, stroke, breast cancer, and pulmonary embolism compared to placebo. These risks exceeded the benefits from reduced risks for colorectal and endometrial cancer, and hip fractures after an average exposure of 5.2 years [21]. 3. Methods 3.1. Data Data for this analysis were derived from the historical paid claims files for a large health insurance company located in California. Paid claims from the period January 1, 1998 to August 30, 2001 were available for inclusion in the analysis. These data included claims for prescription drugs, hospitalizations, physicians services, home health care, laboratory tests, emergency room visits, physical therapy and durable medical equipment. Patients were covered under a variety of insurance plans including capitated health maintenance organizations (HMO), point-of-service and preferred provider options. The claims data included information on diagnosis, the amount paid by the plan, and patient co-payment obligations. However, data on patient height, weight and bone mineral density measurements were not available for study Inclusion and exclusion criteria The unit of observation was the initial observed treatment episode for any of the osteoporosis drug therapies listed in Table 1. Given the common use of HRT therapies to treat the symptoms of menopause, female patients had to be over the age of 55 to be included. Furthermore, all patients had to have a diagnosis of osteoporosis recorded on at least one paid Table 1 Study medications Medications Sample size Single HRT 46,109 Estradiol 7,332 Estrogens, conjugated 37,720 Estrogens, conjugated, synthetic 21 Estrogens, esterified 1,886 Estropipate 1,629 Medroxyprogesterone acetate a 2,349 Norethindrone a 25 Norethindrone acetate a 147 Two HRTs (estrogen plus...) 6,766 Medroxyprogesterone acetate a 6,491 Norethindrone a 27 Norethindrone acetate a 124 Estrogen 124 Bisphosphonate 3,720 Alendronate sodium 3,584 Etidronate disodium 131 Risedronate sodium 5 Raloxifene 1,514 Total sample 58,109 a Progesterone.

4 274 J.S. McCombs et al. / Maturitas 48 (2004) claim during the study period to be included. To insure that the treatment episode was new, drug treatment episodes were included in the study only if the patient was covered by insurance for a minimum of 6 months prior to the initiation of drug therapy. Therefore, it is possible that some patient episodes selected for study are not the first attempt by the patient to seek treatment of osteoporosis. Similarly, patients were included in the study only if their initiation of drug therapy began prior to August 30, 2000 and they maintained their health insurance coverage for the full year following treatment. The number of treatment episodes by the class of medication initially used by the patient are listed in Table 1. For purposes of the analysis, all patients using hormone replacement therapies were classified as filled either one or two HRT prescriptions on the first day of therapy. Over 98% of patients who filled two HRT-related prescriptions on the first day of therapy used a combination of estrogen and progestin. However, it is unclear to what extent patients filling a single HRT prescription on the first day were also estrogen plus progestin patients. For example, patients filling a prescription for progestin only could have filled a prescription for estrogen on a later day. Similarly, patients filling an initial prescription for estrogen may have filled a second prescription for progestin at a later date. Patients filling a single HRT prescription were designated as the comparison group in the multivariate analyses Outcome measures In this analysis, prescription drug data were used calculate the number of days of uninterrupted therapy achieve by the patient on the initial medication used and across all osteoporosis-related medications. Duration of therapy was measured by the count of days of therapy without an interruption of drug purchases greater than 2 weeks. Specifically, a refill prescription was considered to have been purchased without a break in therapy if the cumulative days supply for all previous prescriptions plus 14 days was greater than or equal to the number of days between the refill prescription s purchase date and the index date for the treatment episode. If the cumulative days supply plus 14 was less than the total days between the purchase date of the refill prescription and the index date, the count of continuous days of therapy was terminated. Separate counts of continuous days of therapy were calculated for the initial study drug used by the patient and for each additional study drug used thereafter. The total days of uninterrupted days of therapy achieved by the patient was then calculated across all drugs so long as the there was no gap in excess of 14 days between the termination of continuous use for one medication and the start of the following medication. The time to a change in therapy by the patient was also recorded. Finally, the total days of medication available during the first post-treatment year was calculated to capture returns to treatment or changes in therapy that occurred after a break in therapy in excess of 14 days. This latter outcome measure can be transformed into the medication possession ratio used in many studies as a measure of compliance by simply dividing days of therapy by 360 days. The costs associated with treating osteoporosis patients were investigated using the total cost of treating patients over the first year. Total costs were also broken down into the individual components of cost: drug costs, ambulatory care (physician services), hospital outpatient care, laboratory tests and hospital services. The drug therapies listed in Table 1 are typically used to prevent future bone fractures caused by osteoporosis. Three specific fractures are of particular interest in this study: fractures of the vertebrae, lower arm and wrist (Colles fractures), and the neck of the femur (hip fractures). Patient data for the 6 months prior to the initiation of drug therapy were screened for an ICD-9 diagnostic code indicating that the patient experienced a fracture of the vertebra (805.x, 806.x), lower arm (813.x, 814.x), hip (820.x), or any other fracture. Next, paid claims during the first post-treatment year were then screened for similar indicators of a fracture. A dichotomous variable indicating a new fracture in the post-treatment period was created only if the patient was not identified as having a similar type of fracture in the 6 month pre-treatment period. However, it is possible fractures first reported in the post-treatment period existed prior to treatment but were either undetected or their diagnosis was not yet definitive. This problem of pre-existing but undiagnosed fractures may be particularly troublesome for vertebral fractures that are often initially diagnosed as back pain.

5 J.S. McCombs et al. / Maturitas 48 (2004) Statistical methods Any comparison of compliant and non-compliant patients, or patients treated with different drug therapies using data from real-world clinical settings must use multivariate statistical models designed to take into account baseline differences in the populations being compared. Several different multivariate modeling techniques were required in this analysis given the broad range of patient outcome variables studied. Simple dichotomous patient outcome measures, such as whether or not a patient completed a year of uninterrupted therapy, switched drugs, or experienced a fracture, were analyzed using logistic regression models. Outcome measures based on the time to these events were analyzed using Cox proportional hazards models. Continuous outcome measures, such as health care costs over the first year, were analyzed using ordinary least squares (OLS) regression models. For hospital costs that may have a large number of patients with zero values, the two step approach developed by Duan et al. [22] was used. This approach first investigates the likelihood of being hospitalized using logistic regression models, then proceeds with an analysis of the hospital cost per hospitalized patient. The measures of the number of continuous days of therapy achieved were capped at 360 days, necessitating the use of the maximum likelihood estimation techniques using the Newton Raphson algorithm [23]. The adjusted R 2 goodness-of-fit values were reported for all OLS regression models and can be interpreted as the percentage of the dependent variable s variance explained by the model. All multivariate statistical modeling approaches are only as good as the data available to control for differences between the patient groups of interest. This analysis used the following variables to characterize the baseline characteristics of the study population: 1. Patient age and gender; 2. Type of health insurance coverage (FFS, HMO, point-of-service, PPO); 3. Use of health care services (cost) by type of service in the 6 months prior to initiation of the drug therapy episode; 4. Diagnostic profile as determined by the ICD-9 codes recorded on paid claims during the prior 6 months and the month in which treatment was initiated; 5. Prior fractures (vertebral, hip, Colles, all others); 6. Prior use of the following classes of prescription drugs: (a) NSAIDS; (b) Arthritis medications; (c) Drugs to treat diabetes; (d) Vitamin D therapy; (e) Anti-coagulation therapy; (f) Antidepressants; (g) Prescribed steroids. The availability of these data elements should significantly reduce the likelihood of bias in the multivariate statistical results due to unobserved factors that differ across patient groups and are correlated with patient outcomes. In particular, the diagnostic profile data were used to control for bias created by the possible inclusion of patients using HRT for its presumed cardiovascular benefits. Unfortunately, data on height, weight and bone density were not available for use in the analysis, thus creating the possibility of missing variable bias if physicians selected their initial therapy based on these factors Sensitivity analyses It is possible that female patients under the age of 55 may have used HRT primarily to relieve the acute symptoms of menopause, even after screening patients for a recorded diagnosis for osteoporosis. This is not likely to be the case for patients treated with bisphosphonates and raloxifene, thus creating a potential bias in estimates of the relative impact of alternative drugs. Moreover, raloxifene has been associated with a return of menopausal symptoms in symptom-free, younger female patients, thus reducing its effectiveness in these patients [24,25]. In an effort to document the extent to which bias may have been introduced by having included patients using HRT to treat menopause, sensitivity analyses were conducted to investigate if the impact of raloxifene and bisphosphonates relative to HRT vary with age. Cost data are often highly skewed, thus violating the normality assumptions of OLS regressions. Sensitivity analyses were also conducted using log-transform cost models to investigate the robustness of the OLS cost results.

6 276 J.S. McCombs et al. / Maturitas 48 (2004) Results 4.1. Descriptive statistics Baseline characteristics Table 2 presents data for the baseline demographic characteristics of patients by initial drug therapy. As expected, there are statistically significant and important differences across the four treatment groups under study. Based on this dataset, HRT is the dominant therapy among osteoporosis patients (91.0%). Only 6.4% of the patients used bisphosphonate; and 2.6% used raloxifene. Patients initiating hormone replacement therapy are younger than patients bisphosphonate or raloxifene and more likely to be members of an HMO. Patients treated initially with an HRT are also significantly less likely to have a reported co-morbidity and less likely to have experienced a fracture in the 6 months prior to starting therapy. Table 3 presents baseline data on prior use of health care services. Not surprisingly, HRT patients are less costly in terms of the health care use relative to Table 2 Baseline demographic characteristic Baseline characteristic Single HRT (N = 46,109) Two HRTs (N = 6,766) Bisphosphonate (N = 3,720) Raloxifene (N = 1,514) Demographics Age (mean, in years) Age categories (%) Gender (% male) Type of health insurance (%) Fee-for-service HMO Point-of-service Preferred provider Diagnostic profile at baseline (%) Endocrine disorders Blood disorders Nervous disorders Circulatory disorders Respiratory disorders Congenital disorders Digestive disorders Gentio-urinary disorders Infections Trauma/injury Mental disorders Muscle disorders Neoplasms Skin disorders Fractures prior to treatment (%) Vertebral Hip Colles Other

7 J.S. McCombs et al. / Maturitas 48 (2004) Table 3 Prior use of health care services Baseline characteristic Single HRT (N = 46,109) Two HRTs (N = 6,766) Bisphosphonate (N = 3,720) Raloxifene (N = 1,514) Prior use of health care (US$ per6 months) Physicians services Outpatient department Used DME (%) Durable medical equipment Use emergency room (%) Emergency room Laboratory tests Used physical therapy (%) Physical therapy Other services Used hospital services (%) Hospital care Nursing home care Co-payments for medical care Prescription drugs Plan payments Patient co-payments Total plan costs Total co-payments Total costs Drug profile (%) NSAIDS Arthritis medications Drugs for diabetes Anti-coagulation therapy Antidepressants Prescribed steroids P<0.01. patients initiating therapy on either bisphosphonate or raloxifene. Patients treated initially with an HRT have the lowest rates of prior use of NSAIDS, prescribed steroids, antidepressants, anti-coagulation therapy and prescription drugs used to treat diabetes Patient outcomes Table 4 presents descriptive data on patient outcomes across the four drug therapies under study. As before, patients treated with HRT continue to experience the lowest levels of health care use in the first post-treatment year compared to either raloxifene and bisphosphonates. Patients using two HRTs as initial therapy have the longest duration of therapy (191 days) while bisphosphonate patients have the highest rates of switching of therapy to a second medication (37.3%). Twenty-four percent of patients initiating therapy with bisphosphonates continue to use this therapy uninterrupted for 1 year as compared to only 17.9% of raloxifene patients. Patients using two HRTs are the most likely to complete 1 year of uninterrupted therapy (30.7%). The rate of fractures reported in the 1-year post-period are consistent with observed differences in pre-treatment rates across the four treatment groups. As a whole, HRT patients have the lowest rates of all types of fractures in the first post-treatment year, ranging from 0.13% for hip fractures patients using two initial HRTs to 0.57% for other fractures for single HRT patients. Raloxifene patients experience the next highest rates of fracture, ranging from 0.20% for Colles fractures to 0.46% for other fractures. Patients

8 278 J.S. McCombs et al. / Maturitas 48 (2004) Table 4 Patient outcomes over the first treatment year: health care costs, duration of therapy and number of fractures Baseline characteristic Single HRT (N = 46,109) Two HRTs (N = 6,766) Bisphosphonate (N = 3,720) Raloxifene (N = 1,514) Use of health care (US$ per 12 months post) Physicians services Outpatient department Durable medical equipment Emergency room Laboratory tests Physical therapy Other services Hospital care Nursing home care Co-payments for medical care Prescription drugs (all Rx) Plan payments Patient co-payments Total Plan Costs Total Co-Payments Total costs Use of osteoporosis drugs Days of continuous therapy on first Rx:1 year Days of continuous therapy, all Rx: year (this row contributes little but may cause some confusions) Total days of therapy, all Rx: 1 year Medication possession ratio Second therapy used-overall(%) 34.5% 26.5% 37.3% 27.9% Duration in days (initial therapy, %) > Fracture in first post-treatment year Vertebral 81 (0.18%) 9 (0.13%) 48 (1.29%) 4 (0.26%) Hip 62 (0.13%) %) 34 (0.91%) 6 (0.40%) Colles 141 (0.31%) 14 (0.21%) 40 (1.08%) 3 (0.20%) Other 265 (0.57%) 27 (0.40%) 80 (2.15%) 7 (0.46%) P< P<0.01.

9 J.S. McCombs et al. / Maturitas 48 (2004) treated initially with a bisphosphonate drug exhibit the highest rates of fractures in the post-treatment period, ranging from 0.91% for hip fractures to 2.15% for other fractures Multivariate statistical results Duration of therapy Three different types of analyses were used to investigate the impact of the initial medication class used by the patient on drug therapy outcomes. First, ordinary least squares (OLS) regression models were estimated for the number of days of continuous therapy achieved during the first year without a break in therapy either on the initial medication used or over all medications used. A third OLS model was estimated for the total days of therapy used during the first year regardless of breaks between prescriptions. Second, the likelihood that a patient would achieve 360 days of uninterrupted therapy was investigated using logistic regression models. Third, Cox proportional hazards models were estimated for time to break in therapy and time to the use of a second medication. This approach allows for censoring of the data at the end of the data period and was not limited to events occurring in the first year. The results of all four analyses are presented in Table 5. Bisphosphonate patients are 6.9% more likely to break therapy (P <0.05) and used 17.7 fewer days of total drug therapy over 1 year (P <0.0001) than patients treated initially with a single HRT. No other statistically significant differences in duration of uninterrupted therapy were found. Bisphosphonate patients were four times more likely to change therapies than single HRT patients. In comparison, raloxifene patients are uniformly less compliant and more likely to change therapies than single HRT patients. Patients using two HRTs as initial therapy were found to be more compliant than single HRT patients in four of five analyses and were no more likely to change therapies. Comments on several other results are warranted. Older patients were more compliant and more likely to change therapies than patients under 55 years of age. Prior hospital costs and having used durable medical equipment were associated with increased compliance while out of pocket drug costs in the prior period reduced compliance. The prior use of an emergency room visit was negatively associated with compliance. A history of a prior fracture was generally not found to affect compliance with the exception of vertebral fractures. Patient covered under preferred provider, point-of-service and fee-for-service insurance plans appear to achieve significantly better compliance than HMO patients, but were also more likely to change therapies. The presence of co-morbid conditions had a mixed pattern of effects. The use of NSAIDS, steroids and drugs for diabetes was correlated with reduced compliance while the use of antidepressants significantly increased compliance Fractures The results of the logistic analysis of the likelihood of a patient experiencing a hip, vertebral, Colles or other fracture in the first post-treatment year as a function of initial therapy are presented in Table 6. These results indicate that patients treated with bisphosphonates are twice as likely as single HRT patients to experience a vertebral, Colles or other fracture in the first post-treatment year. This result adjusts for the patient s fracture history and the other factors included in the analyses as independent variables. The risk of fracture was not found to differ between HRT and raloxifene patients. Several additional results in the models summarized in Table 6 are of interest. The risk of fracture increases with age, especially for hip fractures. For example, the likelihood of a hip fracture increased 22-fold for patients 65+ years of age relative to patients younger than age 55. HMO patients are significantly less likely to experience a fracture in the post-treatment period despite HMO patients being found to be less compliant. A history of fractures is not generally predictive of future fractures with the exception of the relationship between Colles fractures and future other fractures, and between other fractures and future hip fractures. The impacts of drug use patterns on the likelihood of fracture were also estimated using logistic regression models and these results are summarized in Table 7. While compliance with therapy for 1 year was found to reduce the likelihood of all types of fractures (o.r. < 1.00), the estimated impact is only statistically significant for hip fractures (o.r. = 0.382, P<0.01) and vertebral fractures (o.r. = 0.601, P<0.05). Switching therapies during the year was associated with an increased risk of all types of fractures except hip

10 280 J.S. McCombs et al. / Maturitas 48 (2004) Table 5 Impact of intial drud therapy on drug use patterns Baseline characteristic OLS: duration: initial Rx (R 2 = ) OLS: duration: all drugs (R 2 = ) OLS: total days in first year R 2 = Logistic: duratio(n > 360) Time to break in therapy: initial Rx Time to change in therapy Days of therapy Days of therapy Days of therapy Odds ratio Hazard ratio Hazard ratio Initial medication (wrt HRT) Bisphosphonates Raloxifene Two HRTs Days of post data available Gender (1: male) Age Prior use of health care Ambulatory care (US$ per 6 months) Drugs (insurance, US$ per 6 months) Drugs (copay, US$ per months) Hospital costs (US$ per 6 months) Medical equipment (yes: 1) Emergency room (yes: 1) Home health care (yes: 1) Physical therapy Prior fracture (1: yes) Hip Vertebral Colles Other Health insurance (wrt HMO) Fee-for-service Point-of-Service Preferred provider Diagnosis at baseline Digestive disorders Blood disorders Endocrine disorders Circulatory disorders Muscle disorders Genitourinary disorders Skin disorders Trauma/injury Drug profile NSAIDS Arthritis medications Drugs for diabetes Anti-coagulation therapy Antidepressants Prescribed steroids P<0.05. P<0.01.

11 J.S. McCombs et al. / Maturitas 48 (2004) Table 6 Impact of initial drug therapy on the likelihood of fracutres in the post-treatment period Baseline characteristic Hip fracture Vertebral fracture Colles fracture Other fractures Bisphosphonates Raloxifene Two HRT Days of post data available Gender (1: male) Age Prior use of health care Ambulatory care (US$ per 6 months) Drugs (insurance, US$ per 6 months) Drugs (copay, US$ per 6 months) Hospital costs (US$ per 6 months) Medical equipment (yes: 1) Emergency room (yes: 1) Physical therapy Prior fracture (1: yes) Hip n.a Vertebral n.a Colles n.a Other n.a. Health Insurance (wrt HMO) Fee-for-service Point-of-Service Preferred provider Diagnosis at baseline Digestive disorders Blood disorders Mental disorders Circulatory disorders Congenital disorders Genitourinary disorders Nervous disorders Trauma/injury Drug profile NSAIDS Arthritis medications Drugs for diabetes Anti-coagulation therapy Antidepressants Prescribed steroids P<0.05. P<0.01. fractures. However, the temporal relationship between the fracture event and the date of switching cannot be clearly determined using these data, thus calling into question the validity of this result Health care costs Three alternative models of health care cost were investigated (Table 8). The first drug use patterns model used dichotomous variables indicating whether or not

12 282 J.S. McCombs et al. / Maturitas 48 (2004) Table 7 Effects of drug use patterns and initial drug therapy on the likelihood of fracture Odds ratios Hip fractures Vertebral fractures Colles fractures Other fractures Drug use patterns: Model 1 One year of uninterrupted therapy Switched/added second drug Drug use patterns: Model 2 Total days of therapy in first year Switched/added second drug Alternative drugs Bisphosphonate Raloxifene Two HRTs P<0.05. P<0.01. the patient achieved 360 days of continuous therapy or switched therapies during the first treatment year. The second drug use patterns model used the count of days of therapy as its measure of compliance. The third cost model used independent variables for the initial therapy used by the patient with single HRT patients as the comparison group. Simple OLS models of all cost except hospital costs for which a two part model was estimated [22]. The results by type of service are reported in Table 9. The uninterrupted use of one or more medications for osteoporosis was not associated with lower total health care costs, inclusive of patient co-payments, in the year following the initiation of drug therapy. This result for total cost reflects the offsetting effects of a significant increase in drug costs of +US$ 266 (P < ) associated with continuous therapy against significant reductions in physician services ( US$ 56, P<0.0001), hospital outpatient services ( US$ 38, P<0.05), laboratory tests ( US$ 9, P<0.01) and hospital costs ( US$ 155, P<0.01) (see Table 9). The impact of compliance on total costs is significant and positive in Model 2 in which the total days of therapy used within 1 year is substituted for the dichotomous 1-year compliance variable. In this latter case, the savings in physicians services, outpatient care, laboratory services and hospital care per day of therapy used (Table 9) were not sufficient to fully offset the increased drug costs of US$ 1.58 per day of drug therapy. In both models, the switching of osteoporosis medications within 1 year was associated with significantly higher total costs due to higher drug costs and increased use if laboratory tests. Patients treated with raloxifene experienced a reduction in total costs in the year following treatment ( US$ 280) but this estimated effect was not statistically significant. However, raloxifene use was correlated with significantly lower costs for physicians services ( US$ 92, P<0.01) and laboratory tests ( US$ 21, P<0.05), and a reduction of 38% in the likelihood of a hospital admission (o.r. = 0.622, P< ) (Table 9). Conversely, patients treated with a bisphosphonate medication experienced significantly higher total costs than single HRT patients (+US$ 420, P<0.01), due to higher drug costs (+US$ 188, P<0.0001) and hospital costs per patient (+US$ 266, P<0.01), the latter due to a significant increase in the cost per hospitalized patient (US$ 2917, P< 0.05) (Table 9). The statistical results for other explanatory variables in Table 9 warrant discussion. Surprisingly, total cost in the first post-treatment year appears to decrease with age over 65. This un-intuitive result may reflect the fact that the model controls for each patient s profile of co-morbid disease states and baseline prescription drug use. Second, total cost is negatively correlated with hospital spending in the 6 months prior to the initiation of drug therapy. Third, a history of vertebral and Colles fractures are negatively correlated with future costs of between US$ 1200 and US$ 1500

13 J.S. McCombs et al. / Maturitas 48 (2004) Table 8 Factors affecting total costs of treatment over 1 year Baseline characteristic Effects of drug use patterns: Model 1 Effects of drug use patterns: Model 2 R 2 = R 2 = R 2 = Effect of alternative drugs Drug use profile One year of uninterrupted therapy 20 Added/switched to second drug 279 Total days of therapy in first year 0.99 Added/switched to second drug 293 Alternative drugs Bisphosphonate therapy (wrt HRT) 420 Raloxifene (wrt HRT) 280 Two HRTs 38 Gender(male: 1) Age Prior use of health care Ambulatory care (US$ per 6 months) Drugs (insurance, US$ per 6 months) Drugs (copay, US$ per 6 months) Hospital costs (US$ per 6 months) Medical equipment (yes: 1) Emergency room (yes: 1) Physical therapy (yes: 1) Home health (1: yes) Prior fracture Hip fracture (yes: 1) Vertebral (yes: 1) Colles (yes: 1) Other (yes: 1) Health insurance (wrt HMO) Fee-for-service Point-of-Service Preferred provider Diagnosis at baseline Digestive disorders Blood disorders Mental disorders Circulatory disorders Endocrine disorders Muscle disorders Nervous disorders Infections Congenital disorders Respiratory disorders Trauma/injury Drug profile NSAIDS Arthritis medications Drugs for diabetes Anti-coagulation therapy Antidepressants Prescribed steroids Days of data in post period P<0.05. P<0.01.

14 284 J.S. McCombs et al. / Maturitas 48 (2004) Table 9 Effects of drug use patterns and initial drug therapy on treatment cost by type of service Drugs Physicians Hospital outpatient Laboratory Likelihood of admission Hospital costs per admitted patient (N = 3502) Hospital costs (all patients) Drug use: Model 1 One year compliance , Switched Adjusted R Drug use: Model 2 Total days of therapy in first year Switched Adjusted R Alternative drug ITT model Bisphospahte Raloxifene Two HRTs Adjusted R P<0.05. P<0.01. (P <0.05 for five of six estimates). Finally, patients covered by the point-of-service or preferred provider insurance options exhibited higher total costs relative to HMO patients, even after controlling for the drug use patterns achieved or the initial therapy used by the patient Sensitivity analyses The results from the sensitivity analysis for age-dependent treatment effects are presented in Table 10. The impact of alternative therapies to HRT on the patient outcomes displays significant age dependency. Consider first the results for raloxifene. The difference in duration of initial therapy improved from a significant deficit of 47 days relative to single HRT in the under 55 age group to a deficit of only 6 days for patients over 65, an improvement of 41 days (P <0.01). Similar improvements are evident in duration on all therapies ( 46 days to 5 days), likelihood of completing 1 year of continuous therapy, the likelihood of discontinuation of therapy and the likelihood of changing therapies within 1 year. Not surprisingly, the trend in health care costs show a similar pattern with total costs relative to single HRT patients being +US$ 295 for patient under 55 (not significant) to a savings of US$ 599 (P <0.05) for patients over 65. The impacts of bisphosphonate medications relative to single HRT patients were also found to be age-dependent. Duration of therapy improves from a differential of 20 days for patients under 55 to a difference of +6 days for patients over 65, an improvement of 26 days (P <0.01). Cost differentials also improve in older bisphosphonate patients, decreasing from +US$ 1263 in higher total costs for bisphosphonate patients under 55 to a difference of +US$ 349 for patients over 65 (P <0.01 for the change). The sensitivity models for age-dependent treatment effects were also estimated using the log-transformation of total costs as the dependent variable. These results also display the pattern of improved performance of both raloxifene and bisphosphonate medications with age and were consistent with OLS results. 5. Discussion Retrospective database analyses using paid claims data present an array of advantages and limitations relative to randomized clinical trials. First, clinicians

15 J.S. McCombs et al. / Maturitas 48 (2004) Table 10 Effects of initial drug therapy on patient outcomes by age group [HRT as comparison medication within each age group] Patient outcome Raloxifene Bisphosphonates < a 65 a < a 65 a Drug use patterns Continuous days of Rx: initial drug Continuous days of Rx: all drugs Total days of Rx in first year Likelihood of 1 year of therapy Likelihood of discontinuation Likelihood of switching drugs One year health care costs Prescription drugs Physician services Hospital outpatient services Laboratory tests Hospital services (unconditional) Likelihood of admission Hospital cost per admitted patient Total costs Log-transform of total cost +10.1% +5.0% 5.0% +40.5% +20.1% +6.8% a Difference with respect to estrogen-only within each age group. Significance levels for the hypothesis that the impact of raloxifene or bisphosphonates relative to estrogen-only changes with age with the under 55 age group as the comparison group. P<0.05. P<0.01. require better data on how well alternative treatment options perform in unrestricted, real-world clinical settings. Compliance data from well controlled clinical trials does not correspond well to real-world practice as clinical trials are typically designed to minimize subject drop-out. It is also difficult to adequately measure health care utilization patterns and the incidence of rare events over long periods of time in the clinical trial environment. Finally, the limited sample size available for study and the selection criteria applied in clinical trials make it difficult to study the impact of alternative therapies across a wide range of patient characteristics. Conversely, retrospective database research cannot establish causality primarily due to the fact that patients are not randomly assigned to the treatment alternatives under study. At best, the multivariate statistical techniques used to estimate treatment effects adjusted for differences in patient characteristics only establish whether or not patient outcomes are correlated with the treatment option selected by the physician. The extent to which the user of retrospective database research is convinced that reported correlations are valid can be enhanced by several factors. First, multivariate models are only as good as the independent variables available to control for differences between treatment groups. While this analysis was careful to include as many clinically-relevant independent variables as were available from a patient s paid claim history, several important factors related to osteoporosis were not available, among them height, weight and BMD measurements. Second, patient outcomes should be measured using several alternative approaches and the estimated treatment effects should be robust across the multitude of analyses presented. This was the approach taken here as we analyzed six different dependent variables related to treatment pattern outcomes (Table 4), four outcomes related to fractures (Tables 5 and 6), and both total costs and the major components of total costs (Tables 7 and 8). Third, the multivariate models used to estimate treatment effects must display plausible results across

16 286 J.S. McCombs et al. / Maturitas 48 (2004) the independent variables included in the model. We have displayed these results for selected models for the reader s analysis (Tables 3 5 and 7). Finally, sensitivity analyses can be conducted that test the extent to which the results are sensitive to important issues not fully addressed by either the criteria used to select patients or the multivariate statistical models used to estimate treatment effects. In this case, since HRT therapies may be used to treat the symptoms of menopause in younger female patients, additional analyses were conducted to test whether or not the impact of alternatives to HRT therapy improved with the age of the patient (Table 9). In this case, the performance of raloxifene and bisphosphonates did improve with the age of the patient, as expected. Clinicians unfamiliar with retrospective database analyses may find it difficult to identify those results that are most relevant to clinical practice from the array of analyses presented here. We believe the critical take-home results for clinicians are: 1. Compliance with all of these therapies is intermittent, at best, and likely does not approach the 10 years of therapy estimated for these therapies to achieve cost-effectiveness [20]. 2. Bisphosphonates and raloxifene do not offer any compliance advantage relative to HRT. 3. While duration of uninterrupted therapy does improve with age, age-related incremental increases in duration were limited to a few days rather than additional months or years of therapy. 4. Continuous therapy over 1 year significantly reduced the risk of hip and vertebral fractures over 1 year and did not significantly increase total direct medical cost for compliant patients. 6. Limitaitons While great care has been taken to adjust estimated results for the baseline clinical characteristics of the patient population, other unobserved factors may exist that are correlated with the patient outcome measures studied here. Of particular concern is the lack of data for height, weight and bone density that are likely to be correlated with both patient outcomes and the selection of an initial therapy. Moreover, this analysis considers only the initial osteoporosis drug therapy used by the patient during the data period available for analysis. It is likely that many of these patients experienced prior treatment attempts using these medications and that these prior unobserved episodes of care would influence duration and switching patterns. However, one would think that duration of therapy would increase with the second or third treatment attempt by the patient. If so, the poor compliance reported here is even more discouraging. Hormone replacement therapy may have been intended for short-term use to treat the adverse effects of menopause, even in this population of patients with a diagnosis of osteoporosis. Conversely, the use of a bisphosphonate medication or raloxifene may have been intended for longer-term use to treat osteoporosis. If this is true, then one would expect to see longer duration of therapy for the alternative therapies relative to HRT. This was not found to be the case. Patients using HRT may have also been at significantly lower risk of fractures even after controlling for patient characteristics such as age, gender, co-morbid conditions and baseline drug profile. If this is the case, the estimated effects of raloxifene and bisphosphonates on fracture rates and costs may be significantly under-estimated. 7. Conclusions Taken as a whole, patients who achieve 1 year of uninterrupted drug therapy for osteoporosis achieve better patient outcomes than patients who terminate or interrupt therapy during the first year. Unfortunately, less than 25% of patients achieve in excess of 1 year without breaking therapy and approximately 30% of patients switch therapies. This is a mixed blessing. The Women s Health Initiative study [21] found that the long-term use of estrogen and progestin combination therapy (average 5.2 years) significantly increases a woman s risk of breast cancer, coronary heart disease, stroke and pulmonary embolisms to the point that these risks outweigh the benefits achieved. In this case, non-compliance with HRT in less than 1 year may be beneficial in the long term. However, therapies to replace HRT for the long-term treatment of osteoporosis are needed. Much remains to be done to improve long-term patient compliance with bisphosphonate medications and raloxifene before the full benefit of these therapies are achieved.

17 J.S. McCombs et al. / Maturitas 48 (2004) References [1] America s Bone Health: The State of Osteoporosis and Low Bone Mass. NOF: National Osteoporosis Foundation Report. [2] Melton LJ, Chrischilles EF, Cooper C, Lane AW, Riggs BL. How many women have osteoporosis? J Bone Miner Res 1992;7(9): [3] Martin BC, Chisholm MA, Kotzan JA. Isolating the cost of osteoporosis-related fractures for postmenoposal women. Gerontology 2001;47:21 9. [4] Lindsay R. The burden of osteoporosis: cost. Am J Med 1995;98(2A):9S 11S. [5] Ray NF, Chan JK, Thamer M, Melton LJ. Medical expenditures for the treatment of osteoporotic fractures in the United States in 1995: Report from the National Osteoporosis Foundation. J Bone Miner Res 1997;12(1): [6] Cooper C. The crippling consequence of fractures and their impact on quality of life. Am J Med 1997;103(2A):13S 5S. [7] Seeman E. Osteoporosis: trials and tribulations. Am J Med 1997;103(2A) Supplement 18:74S 87S. [8] Barrett-Connor E. The economic and human costs of osteoporotic fracture. Am J Med 1995;98(2A):3S 7S. [9] Johnell O. The socioeconomic burden of fractures: today and in the 21st century. Am J Med 1997;103(2A):20S 5S. [10] Brainsky A, Glick H, Lydick E, Epstein R, Fox KM, Hawkes W, et al. The economic cost of hip fractures in communitydwelling older adults: a prospective study. J Am Geriatr Soc 1997;45(3): [11] Cummins SR, Rubin SM, Black D. The future of hip fractures in the United States. Clin Orthop 1990;252: [12] Kotzan JA, Martin BC, Wade WE. Persistence with estrogen therapy in a postemenoposal Medicaid Population. Pharmacotherapy 1999;19(3): [13] Cano A. Compliance to hormone replacement therapy in menopausal women controlled in a third level academic center. Maturitas 1994;20(2-3):91 9. [14] Faulkner DL, Young C, Hutchins D, McCollam JS. Patient noncompliance with hormone replacement therapy: a nationwide estimate using a large prescription claims database. Menopause 1998;5(4): [15] Cole RP, Palushock S, Haboubi A. Osteoporosis management: physician s recommendations and women s compliance following osteoporosis testing. Women Health 1999;29(1): [16] Kayser J, Ettinger B, Pressman A. Postmenopausal hormonal support: discontinuation of raloxifene versus estrogen. Menopause 2001;8(5): [17] Marwick C. Hormone combination treats women s bone loss. J Am Med Assoc 1994;272(19):1487. [18] Bjorn I, Backsrom T. Drug related negative side-effects is a common reason for poor compliance with hormone replacement therapy. Maturitas 1999;32(2): [19] Ettinger B, Li DK, Klein R. Alendronate use among 812 women: prevalence of gastrointestinal complaints, noncompliance with patient instructions, and discontinuation. J Managed Care Pharmacy 1998;4(5): [20] Office of Technology Assessment, United States Congress. Effectiveness and costs of osteoporosis screening and hormone replacement therapy, volumes I and II. OTA-BP-H- 160, Washington, DC: US Government Printing Office; August [21] Women s health initiative. Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results from the Women s Health Initiative Randomized Controlled Trial. JAMA 2002;288(3): [22] Duan N, Manning WG, Morris CN, Newhouse JP. A comparison of alternative models for the demand of medical care. J Bus Econ Stat 1983;1(2): [23] SAS/STAT User s Guide, Version 8, volume 2. Cary NC: SAS Institute Inc.; 1999 [Chapter 36]. [24] Cohen FJ, Lu Y. Characterization of hot flashes reported by healthy postmenopausal women receiving raloxifene or placebo during osteoporosis prevention trials. Maturitas 1999;34: [25] O Regan RM, Gradishar WJ. Selective estrogen-receptor modulators in Oncology 2001;15(9):