Long-term health care costs for prostate cancer patients on androgen deprivation therapy.

BACKGROUND
Comparing relative costs for androgen deprivation therapy (adt) protocols in prostate cancer (pca) requires an examination of all health care resources, not only those specific to pca. The objective of the present study was to use administrative data to estimate total health care costs in a population-based cohort of pca patients.


METHODS
Patients in Ontario with pca who started 90 days or more of adt at age 66 years or older during 1995-2005 were selected from cancer registry and health care administrative databases. We classified patients (n = 21,818) by regimen (medical castration, orchiectomy, anti-androgen monotherapy, medical castration with anti-androgen, orchiectomy with anti-androgen) and indication (neoadjuvant, adjuvant, metastatic disease, biochemical recurrence, primary nonmetastatic). Using nonparametric regression methods, with inverse probability weighting to adjust for censoring, and bootstrapping, we computed mean 1-year, 5-year, and 10-year longitudinal total direct medical costs (2009 Canadian dollars).


RESULTS
Mean first-year costs were highest for metastatic disease, ranging from $24,400 for orchiectomy to $32,120 for anti-androgen monotherapy. Mean first-year costs for all other indications were less than $20,000. Mean 5-year and 10-year costs were lowest for neoadjuvant treatment: approximately $43,000 and $81,000 respectively, with differences of less than $4,000 between regimens. Annual costs were highest in the first year of adt. Orchiectomy was the least costly regimen for most time periods, but was limited to primary and metastatic indications. Outpatient drugs, including pharmacologic adt, accounted for 17%-65% of total first-year costs.


CONCLUSIONS
Compared with combined therapies, the adt monotherapies, particularly orchiectomy when clinically feasible, are more economical. Our methods exemplified the use of algorithms to elucidate clinical information from administrative data. Our approach can be adapted for other cancers to expand the range of studies using Canadian administrative data.

preparation of lhrh agonist cost approximately CA$1000 to CA$1500 11 .
Studies in the United States that estimated cumulative longitudinal costs for adt have included only pca-related costs 9 or included data only for the years that patients survived, thus excluding the high costs immediately before death 10 . Other studies compared the costs of various types of adt and included only the costs of the adt, assuming that the other health care costs would be similar [12][13][14] . A full understanding of the relative costs of various clinicallyrelevant adt regimens and indications requires consideration of other pca-related care, comorbid illness, treatment side effects 15 , and end-of-life care.
Administrative health care data are ideal for estimating long-term costs because they encompass a large number of patients-often an entire population-that can be followed for many years. The linked Surveillance, Epidemiology, and End Results-Medicare database 16,17 has been used for numerous cancer costing studies in the United States. It includes all claims, with their dates, payments, patient diagnosis codes, and procedure codes for all Medicare-covered patients and services. However, the Surveillance, Epidemiology, and End Results regions represent only approximately 14% of the U.S. population, and Medicare covers mainly patients 65 years of age and older 18 .
In Canada, universal health care insurance plans are managed by provincial and territorial governments and provide coverage for all permanent residents of all ages. Administrative records associated with those plans are an excellent source of utilization and costing data. Their lack of clinical detail can be overcome by developing algorithms that use information about diagnosis, drug prescriptions, surgical procedures, and other treatments.
The purpose of the present study was to estimate the total longitudinal health care cost for clinically relevant regimens related to common indications for adt in the population of pca patients in the province of Ontario who initiated adt during [1995][1996][1997][1998][1999][2000][2001][2002][2003][2004][2005]. No other study has considered both regimen and indication when examining total costs for pca patients treated with adt. The present study exemplifies the use of administrative data and provides a methodology for future research.

METHODS Ethics
Approval for the study was given by the Research Ethics Board of the University Health Network and the institutional review board at Sunnybrook Health Sciences Centre, Toronto, Ontario. Individual patient consent was not possible and was waived because the study used administrative databases and because patient identifiers were removed for the analyses.

Patients and Data
Using Canadian health care administrative databases, we selected patients in Ontario who began at least 90 consecutive days of lhrh agonist, estrogen, or anti-androgen, or who had a bilateral orchiectomy from 1 January 1995 to 31 December 2005 (Table i). We specified a 90-day period because that is the recommended minimum therapeutic duration of adt for any indication 19 . Applying standard methods, we allowed up to 50% of the number of days of drug supplied for the patients to refill their prescriptions 20,21 . Thus, a patient had up to 45 days to refill a prescription for a 30-day supply or to obtain a new adt prescription. We defined each patient's index date as the start of adt.
We used the Ontario Health Insurance Program unique patient number to link the selected patients to the Ontario Cancer Registry 22 , retaining those with an initial diagnosis of pca, but no other primary cancer, before or within 6 months after their adt index date (thus allowing for delays in reporting to the cancer registry). Outpatient prescription drug coverage is available under the Ontario Drug Benefit program, but only for Ontario residents 65 years of age and older and those on social assistance. To assess prior adt, we excluded patients younger than 66 years at index date.
Using unique encoded identifiers, the resulting cohort of patients was linked to other health care administrative databases; to data about radiation therapy obtained from the Radiation Oncology Research Unit at Queen's University in Kingston, Ontario; and to chemotherapy data obtained from Cancer Care Ontario. The linked data were analyzed at the Institute for Clinical Evaluative Sciences in Toronto, Ontario. Patients were followed from their index date until death or 31 December 2007.

ADT Regimens and Indications
We classified each patient according to the adt regimen received in the first 90 days after his index date: medical castration, orchiectomy, anti-androgen monotherapy, combined androgen blockade (cab)-medical, and cabsurgical. Based on the literature, clinical practice guidelines 8,9,23,24 , consultation with clinical experts, and the data described in Table i, we initia lly identif ied 6 mutually-exclusive indications for adt (Table ii): neoadjuvant therapy, adjuvant therapy, metastatic disease, biochemical recurrence, localized disease, and other. For the years covered by the study, the Ontario Cancer Registry contains limited staging information. We applied a previously used algorithm to identify patients with probable metastatic disease 25 .
For the analyses, we excluded patients who received adt for "other" indications, and we selected patients treated with clinically relevant protocols only. Treatment with orchiectomy has declined in recent years 3,6 , and the indications for which it is relevant are few. We therefore included only patients who received orchiectomy for metastatic disease or as primary therapy, and those who received cab-surgical for metastatic disease. We excluded patients who received adt outside of the relevant indications.

Data and Costing
Almost all medically necessary health care for permanent residents of Ontario is covered under the Ontario Health Insurance Plan, managed by the Ontario Ministry of Health and Long-Term Care (http://www.health.gov.on.ca/en/ public/programs/ohip). The exceptions are outpatient prescription drugs, covered only for residents 65 years of age and older and for those on social assistance, and most allied health services in community clinics. Patients have no co-payments or deductibles for most services that are covered by the Ontario Health Insurance Plan. Therefore, almost all the costs of health-related resources used by patients in the present study were covered by the public payer.
We used standard costing methods for administrative data in Ontario 26 . Table iii describes the health care resources, data, and costing methods. Hospital admissions were obtained using the Discharge Abstract Database, and emergency room visits, ambulatory procedures, and chemotherapy were obtained from the National Ambulatory Care Reporting System-both maintained by the Canadian Institute for Health Information. All records in both databases are assigned a resource intensity weight 28 , which we multiplied by the cost per weighted case in Ontario for the year of use to estimate costs. We included cancer drugs recorded in the New Drug Funding Program data obtained from Cancer Care Ontario. That program (https:// www.cancercare.on.ca/toolbox/drugs/ndfp) reimburses cancer centres and hospitals for expensive chemotherapy approved for use in specific cancers. Costs for pharmacologic adt and other outpatient prescription medications were identified from the database of the Ontario Drug Benefit program.
We used the Claims History Database of the Ontario Health Insurance Plan to identify costs for inpatient and outpatient physician services and outpatient diagnostic tests. Data from the Radiation Oncology Research Unit database, housed at Queen's University, Kingston, Ontario, were used to identify radiation therapy and its intent (curative or palliative). In Ontario, many radiation oncologists receive some of their income through alternative funding plans 29 . We estimated the income provided to radiation oncologists through alternative funding plans from responses to the 2004 and 2007 National Physician Surveys (http:// nationalphysiciansurvey.ca/).
We identified home care services using the Home Care Database and the Ontario Home Care Administrative System database. The average per-service cost was obtained from Community Care Access Centres in Toronto.
We included stays at complex continuing care facilities in Ontario. Each stay is assigned a case-mix index, indicating intensity of the required care. We multiplied that index value by the length of stay and by the provincial average cost per weighted chronic day for the year of the stay.
All costs were adjusted to 2009 Canadian dollars using the Consumer Price Index for health 26,30,31 . We summed all resource costs for each patient to estimate total health care costs.

Clinical and Demographic Characteristics
We described comorbid illness by categorizing all International Classification of Diseases diagnostic codes (revisions 9, 9 Clinical Modification, and 10) found for each patient in hospital records and physician billing data from 1 year before his index date into one of nearly 90 adjusted clinical groups, a population-patient case-mix adjustment system 32 . We were interested in the burden of comorbidity, and so we grouped the total number of adjusted clinical groups for each patient (comorbidity 0-3: low; 4-6: moderate; 7-9: high; and ≥10 very high) 32 .
We used the Statistics Canada Postal Code Conversion file and data from the 2001 Canadian Census to assign each patient's postal code to a standard geographic area, for which data on neighbourhood-level median household income (measured in quintiles) and rurality were obtained. Communities with a population less than 10,000 were defined as rural. Anti-androgen monotherapy ≥90 Days steroidal anti-androgen (cyproterone) or nonsteroidal anti-androgen (bicalutamide, flutamide, nilutamide), with no castration. A patient who uses both ≥90 days steroidal and nonsteroidal anti-androgen is classified as steroidal.
Combined androgen blockade Surgical Both orchiectomy and ≥90 days any anti-androgen. Date of orchiectomy can be within the first 90 days of anti-androgen, or the start of anti-androgen can be within 90 days after the date of orchiectomy. In the case of concurrent use of castration types, orchiectomy overrides medical castration.
Medical LHRH agonist or estrogen, but no orchiectomy, and any anti-androgen at the same time for ≥90 days. Start of LHRH agonist or estrogen can be within the first 90 days of anti-androgen, or start of anti-androgen can be within 90 days after start of LHRH agonist or estrogen.

Indications
Neoadjuvant therapy ADT beginning ≤8 months before primary curative treatment (date of radical prostatectomy, or start of curative radiation therapy), and ending within 6 months after the start of curative treatment.
Adjuvant therapy ADT beginning ≤8 months before primary curative treatment, or during or ≤6 months after curative treatment, and continuing for >6 months after the start of curative treatment.
Symptomatic metastatic (stage M1) disease Any one of the events that follow after a prostate cancer diagnosis and before or within 6 months after the index date:

Statistical Analysis
We would ideally have liked to follow all patients from start of adt until death. However, as with most analyses of longitudinal cost data, many patients are still alive at the end of the observation period, and other patients have short periods of observation because they enter the study too late to be followed for the maximum time. Such patients are censored, and without adjustment, their inclusion will bias the estimates of mean total cost 33 . Methods to account for censoring have been proposed. In the Kaplan-Meier sample average estimate from Lin et al. 33 , uncensored costs are weighted by survival probability. The entire study duration is partitioned, and for each time interval j, the mean cost for all patients Ê j observed at the start of the interval is multiplied by their probability of survival Ŝ j. The mean total cost is then estimated by summing costs for all time intervals: . When censoring occurs at interval boundaries, this approach provides consistent, asymptotically normal estimators of the mean total cost 34 .
Bang and Tsiatis account for censoring by weighting uncensored costs by the inverse probability of inclusion 35,36 . The study period is partitioned, and the total interval costs M ij are divided by the probability of not being censored K j at the beginning of the interval. The results are summed across all intervals and divided by the sample Case-mix index multiplied by length of stay and provincial average cost per weighted chronic day. TO BE RE-CREATEDnew one below As requested for paper 2953-29811-1-CE2 Krahn size n to yield the mean cost estimate. When partition boundaries occur at the censoring times, this estimator is equivalent to the Kaplan-Meier sample average estimator 37 . The Bang and Tsiatis partitioned estimator is defined by , where Δ j i = 1 if patient i is not censored at the beginning of interval j.
We applied the foregoing methods to adjust for censoring in our data. For each indication and regimen combination, we divided the observation time for the patients into censored time intervals defined as the time between one loss of 1 or more patients because of censoring and the next loss of 1 or more patients. Total health care costs were computed for each censored time interval until no patients remained. The number of patients who died in each interval was also computed. We produced 1000 bootstrapped samples for each indication-regimen. We ran the Bang-Tsiatis and the Lin mean cost estimation procedures in each of the 1000 bootstrapped samples to calculate mean 1-year, 5-year, and 10-year costs, with their 95% confidence intervals (cis) for each indication-regimen combination.
Analyses were performed using the SAS software application (version 9.1 for Unix and version 9.2 for Windows: SAS Institute, Cary, NC, U.S.A.) and the R software application (The R Foundation, Vienna, Austria). Figure 1 shows the cohort selection process. After selecting 26,809 pca patients who initiated adt during 1995-2005, we excluded 4991 patients whose regimens or indications were not clinically relevant, leaving 21,818 patients for the analyses. Mean age in this cohort was 75 years, and year of pca diagnosis ranged from 1965 to 2005. Only 0.2% of all patients were diagnosed before 1980; 55% were diagnosed during 1980-1999; and 45%, during 2000-2005. However, 87% of patients with recurrence and 61% of patients with metastatic disease were diagnosed during 1980-1999. Approximately 82% resided in urban areas. Patients were fairly evenly distributed among the middle-range neighbourhood income quintiles; only 17.6% were in the lowest quintile, and 22% in the highest. Patients receiving adt for metastatic disease or as primary therapy for nonmetastatic disease were the oldest, and those with metastatic disease had the most comorbidity. Differences in age, comorbidity, and status on 31 December 2007 were larger between the adt indications than between the regimens within indication groups (Table iv).

Patients
Likewise, length of survival after adt initiation differed between the indications. Patients who received adt as adjuvant or neoadjuvant treatment experienced similar survival (approximately 94% and 90% at 5 years, and 81% at 10 years). Patients who received adt for recurrence had 5-year and 10-year survival rates of 71% and 58% respectively, and rates for patients who received primary adt were 59% and 43% at 5 and 10 years respectively. The patients with metastatic disease had the shortest survival, with only 30% surviving to 5 years, and 23% surviving to 10 years.

Costs
Overall, variations in costs for each time period for patients who received different regimens within an indication were small compared with variations in costs between indications ( Figure 2). Mean first-year costs were highest for the patients with metastases (Figure 2), ranging from $24,400 for orchiectomy (95% ci: $20,790 to $27,330) to $32,120 for anti-androgen monotherapy (95% ci: $27,750 to $35,660). Mean first-year costs for all other indications were less than $20,000. The lowest first-year cost, $14,240 (95% ci: $11,990 to $ 16,190), was for patients who underwent orchiectomy as primary treatment.
Mean 5-year costs were also highest for metastatic and primary indications. For both indications, the highest costs were for the cab-medical and medical castration regimens, which ranged from $68,160 (95% ci: $63,230 to $73,420) to $70,290 (95% ci: $65,910 to $74,030). The cab-medical regimen was, for most indications, associated with the highest cost at 5 years ( Figure 2). The lowest mean 5-year costs were for the neoadjuvant regimens and ranged from $41,650 for anti-androgens (95% ci: $39,340 to $43,830) to $44,150 for cab-medical (95% ci: $41,850 to $46,260).
Most of the mean 10-year costs fell into the range of approximately $65,000-$109,000 ( Figure 2). Many of the trends already observed persisted. For example, mean costs for cab-medical were as high as $108,360 (95% ci: $94,180 to $120,410), among the highest of all the regimens. � ed across all intervals and divided by the sample size n to yield the When partition boundaries occur at the censoring times, this nt to the Kaplan-Meier sample average estimator 37 . The Bang and timator is defined by nt i is not censored at the beginning of interval j.   Orchiectomy and cab-surgical were associated with the lowest costs for all time periods, but were used only for primary and metastatic treatment. Medical castration and cab-medical, both of which include lhrh agonists, were usually the most costly regimens.
We estimated the percentage of total costs spent on outpatient drug prescriptions, including pharmacologic adt, for each regimen. Those percentages were highest in the first year of adt, when most patients would be receiving the most intense adt. Because of high costs for other resources, outpatient prescriptions represented only 23%-24% of the total costs for patients with metastatic disease. Outpatient prescription drugs represented a higher percentage of the total cost for patients with recurrence: for example, 38%-65% in the first year. Neoadjuvant adt is generally given for a short period before and after another primary treatment, and thus drug costs for patients on neoadjuvant regimens represented 27%-35% of the total cost in the first year and slightly less thereafter.

DISCUSSION AND CONCLUSIONS
Our study used administrative data to estimate total longitudinal health care costs for a population-based sample of patients who initiated clinically relevant types of adt for 5 common indications over an 11-year period in a large province in Canada. One consistent finding was lower overall health care costs for the surgical regimens (orchiectomy and cab-surgical).
The average difference in the mean first-year costs for those regimens compared with their medical equivalents was $3,277. From the point of view of overall cost, and assuming similar patients, if the 2771 patients in our cohort who received medical castration as primary treatment (n = 1951) or for metastases (n = 820) had received an orchiectomy, the first-year cost savings would be $8,160,580. Another 717 patients received cab-medical treatment for metastatic disease. Substituting a cab-surgical regimen would potentially save $3,255,180 in the first year. Overall, the exclusive use of orchiectomy would represent a savings of $11,415,760 for 3488 patients in the first year after initiation of adt. Those cost differences extend into later years, and the differences in cost between surgical and medical castration regimens at 10 years would total more than $75 million.
Our cost estimates include costs for all health care and might be conservative in terms of savings based on adt only. Survival rates were actually lower for patients who received surgical castration than for those who received medical castration; thus, high end-of-life costs are included in their total costs 25 . Previous studies that examined cost differences between medical and surgical castration estimated the annual savings to be approximately $5,000 12,14,38 (converted to 2009 Canadian dollars).
Medical castration, either alone or with anti-androgen (cab-medical), was the most costly regimen for most indications and time periods. However, anti-androgen monotherapy was associated with surprisingly high costs given the relatively low cost for those drugs in Ontario. We classified patients by their initial type of adt, and it is possible that patients remained on anti-androgen monotherapy for a longer period than they remained on lhrh agonists. Also, some patients using medical castration might have been on intermittent lhrh therapy, although most studies showing noninferiority of intermittent compared with continuous adt were published after 2005 39 .
Patients who received adt as neoadjuvant or adjuvant treatment were younger than the patients who received it for other indications. That difference probably accounts for their lower overall health costs. They received adt for a short time in conjunction with either radiation therapy or radical prostatectomy, and hence their first-year costs were high compared with the first-year costs for primary adt and recurrence, but their 5-year and 10-year costs were lower. The patients with metastasis were older and had a greater comorbidity score than most of the patients who received adt for other indications; many of their high health costs were therefore undoubtedly not directly associated with adt. However, our study represents the true clinical picture, in which patients who receive adt for localized disease differ from those who receive it for advanced and metastatic disease. Patients were similar within indications, and we were most interested in comparing costs for regimens within indications.
Our 5-year costs for the various regimens and indications ranged from $41,650 to $70,260 (2009 Canadian dollars). A study that used the capsure (Cancer of the Prostate Strategic Urologic Research Endeavor) database 40 and included only pca-related costs estimated that mean cumulative 5-year costs for patients initially treated with adt were $69,244 (2004 U.S. dollars) 9 , higher than our estimates for most regimens. Another study using Surveillance, Epidemiology, and End Results-Medicare data and a matched control group to calculate net pca-related costs estimated 5-year costs of hormonal therapy to be only $26,896 (2007 U.S. dollars), but it considered data from only the years during which the patients survived, thus excluding the high costs immediately before death 10 . Neither study described the adt regimens or their indications.
The methods used in the present study provide a model for similar economic evaluations for other pharmacologic or surgical therapies for cancers and other diseases. We used a rich selection of health care administrative data to estimate costs, and we applied algorithms and decision rules to classify patients by adt regimen and indication. We present estimates of real-world longitudinal costs for a large population-based cohort of pca patients who received adt, for use in economic analyses and models. Total health care costs reflect not only variations in adt indication and regimen, but also realistic differences in tumour factors, patient comorbidity, age, general health, and patient preference. Ideally, those factors-and not only cost-should be considered when determining the optimal type of adt for a patient or group of patients.