Yamini Kalidindi, MHA

 To compare cancer care spending and utilization by site of provider-administered chemotherapy in Medicare.

 A retrospective analysis using 2010-2013 Medicare claims.

The study population was a random sample of Medicare fee-for-service beneficiaries with cancer who initiated provider-administered chemotherapy in a hospital outpatient department (HOPD) or physician office (PO). We assessed the following outcomes during the 6-month follow-up period: (1) spending on cancer-related outpatient services excluding chemotherapy, (2) spending on cancer-related inpatient services, (3) utilization of select cancer-related outpatient services (evaluation and management, commonly used expensive billing codes, and radiation therapy sessions), and (4) the number of cancer-related hospitalizations. We used regression analyses to adjust for patient health risk factors and market characteristics.

 During the 6-month follow-up period, risk-adjusted spending on nonchemotherapy outpatient services was slightly lower among patients receiving chemotherapy in HOPDs than in POs ($12,183 [95% CI, $12,008-$12,358] vs $12,444 [95% CI, $12,313-$12,575]; 

 <.05). Risk-adjusted cancer-related inpatient spending was higher in the HOPD group than in the PO group ($3996 [95% CI, $3837-$4156] vs $3168 [95% CI, $3067-$3268]; 

 <.01). The HOPD group had fewer visits in all select outpatient services but had a higher number of hospitalizations than the PO group.

 Differences in cancer care spending by site of chemotherapy (HOPDs vs POs) vary by service type. Those differences are partially driven by utilization differences. As the site of chemotherapy shifts from POs to HOPDs, spending and utilization patterns in both settings need to be monitored.

Our study examined risk-adjusted differences in cancer care spending and utilization for the 6-month follow-up period after initiation of provider-administered chemotherapy by site of chemotherapy in Medicare. We found that:

Risk-adjusted spending on outpatient cancer services other than chemotherapy was slightly lower for patients who received chemotherapy in hospital outpatient departments (HOPDs) versus physician offices (POs).

Risk-adjusted inpatient cancer care spending was higher for patients who received chemo&shy;therapy in HOPDs versus POs.

Although utilization of select outpatient services was higher among patients who received chemotherapy in POs, the number of hospitalizations was higher in the HOPD group.

Medicare shoulders a large burden of the costs for cancer care delivered in the United States. In 2014, cancer care costs exceeded $87 billion,

 with costs projected to increase to more than $173 billion in 2020.

 Medicare’s share is roughly one-third.

 A main treatment modality for cancer is chemotherapy,

 which is mostly administered by providers.

Provider-administered chemotherapy is usually covered by medical benefits. In Medicare, Part B (coverage for outpatient medical services) pays for provider-administered chemotherapy. In recent years, there has been a shift in the site of provider-administered chemotherapy from physician offices (POs) to hospital outpatient departments (HOPDs).

 In 2016, nearly 50% of Part B chemotherapy administration claims occurred in HOPDs, a rise from less than 25% in 2008.

 A concern has been raised that this trend may lead to increased cancer care spending because of potential differences in spending patterns between HOPDs and POs.

Spending differences by site of chemotherapy have been extensively studied among commercially insured patients.

 Prior studies using commercial claims consistently found cancer-related spending on outpatient services among patients receiving chemotherapy in HOPDs to be substantially higher than that in POs.

 However, patients receiving chemotherapy in HOPDs were found to have only a slightly smaller number of office visits and outpatient services than those in POs.

 Thus, prior discussions of spending differentials in outpatient cancer spending by site of chemotherapy focused on differences in payments for outpatient cancer services between HOPDs and POs, especially in the commercially insured population.

 However, in Medicare, differences in payments for outpatient chemotherapy services by site of care are not profound.

 For example, Medicare pays the same for chemotherapy drugs between HOPDs and POs,

 whereas among commercial insurers, the average cost per chemotherapy drug claim was more than $2000 higher in HOPDs than in POs ($3799 vs $1466) after controlling for different distributions of chemotherapy drugs between HOPDs and POs.

 Thus, in Medicare, part of the difference in outpatient cancer care spending between HOPDs and POs may come from differences in utilization.

Two prior studies compared Medicare spending on provider-administered chemotherapy in HOPDs and POs and reached conflicting conclusions.

 A report by The Moran Company found that average per-patient spending on chemotherapy drugs was higher in HOPDs than in POs.

 However, Kalidindi et al found that spending per beneficiary was higher among patients receiving chemotherapy in POs, driven by higher chemotherapy utilization.

 It is important to note that Kalidindi et al used more rigorous risk adjustments than The Moran Company. For example, Kalidindi et al accounted for differences in the distribution of cancer type between the 2 care settings, whereas The Moran Company did not. However, both studies were limited to analyzing chemotherapy claims only. Neither study considered other services used by patients receiving chemotherapy. Patients who visit physicians to receive provider-administered chemotherapy often use additional services (eg, radiation therapy, computed tomography scans). Thus, analyzing only chemotherapy costs paints an incomplete picture of cost differences by site of care.

One report attempted to examine total care spending by site of chemotherapy in Medicare. The Milliman group analyzed 2006-2009 claims and found that patients receiving chemotherapy in HOPDs had approximately $6500 higher total annual costs than those receiving chemotherapy in POs.

 However, the Milliman study had the following limitations and did not allow us to identify sources of its main finding. First, the Milliman report calculated total costs by cancer type but did not adjust for other risk factors, such as metastasis or other chronic conditions. Second, it analyzed all care spending and not just the costs associated with chemotherapy and cancer care. Finally, it did not look at utilization differences between care settings. Thus, its estimates of cost differentials could result from (1) differences in Medicare payments for outpatient services between care settings, (2) differences in utilization of outpatient services, (3) differences in utilization of inpatient services, or (4) a combination of any of the aforementioned factors. Thus, drivers of cancer care cost differentials by site of chemotherapy in Medicare remain to be examined.

Our study fills this gap by expanding the study by Kalidindi et al of chemotherapy services to include other cancer services.

 For outpatient cancer care, we complement their study by focusing on spending on services other than chemotherapy. We also compare utilization of select outpatient cancer services by care site to understand whether site-specific cancer care spending in Medicare is entirely driven by payment differences or partially due to utilization differences. Further, because differences in utilization could affect patient outcomes, which can lead to hospitalizations, we examine cancer-related inpatient care spending and use. These analyses are important because they help us identify drivers of differential cancer care spending by site of chemotherapy in Medicare and thereby offer implications for patient care.

The study was conducted from the Medicare payer perspective. We used Medicare data including 2010-2013 MedPAR files (inpatient and skilled nursing facility claims), Carrier files (PO claims), Hospital Outpatient files (HOPD claims), and Medicare Master Beneficiary Summary files. We acquired these Medicare data from CMS. We also used publicly available American Community Survey data for zip code—level income, education, and unemployment rates.

The study population was a random sample of the national cohort of Medicare beneficiaries with cancer between 2010 and 2013, which was created as follows. CMS first identified all patients with cancer using 100% Medicare claims based on the standard algorithm used to create cancer indicators in the Medicare Chronic Condition Warehouse file: (1) having at least 1 inpatient or skilled nursing facility cancer claim or (2) having at least 2 Carrier or Outpatient cancer claims in a given year.

 Then, CMS selected a 10% random sample and provided us with the data for that sample.

We selected beneficiaries who initiated provider-administered chemotherapy between January 1, 2010, and June 30, 2013, using the Healthcare Common Procedure Coding System Level II (J-code) in the Carrier or Outpatient file (

]). We used a 1-year washout to identify new chemotherapy users and designated the date of the first claim for chemotherapy after the washout as the index date.

Patients were required to have continuous Part A and Part B coverage for 12 months. We excluded those who died within 3 months of the index date and Medicare Advantage enrollees. Patients were categorized into 2 groups by the location of chemotherapy (HOPD vs PO). We excluded patients who used both settings. The eAppendix presents sample sizes associated with each exclusion.

We measured outcomes for a 6-month period following the index date. Spending measures were (1) spending on nonchemotherapy cancer-related outpatient services (these include all cancer-related outpatient services, such as supportive medications/premedications for chemotherapy, but they exclude chemotherapy services themselves) and (2) spending on cancer-related inpatient services (including chemotherapy services). We measured spending by Medicare allowed payments, which include both patient out-of-pocket spending and Medicare reimbursements. Spending was adjusted to 2013 dollars.

We measured 3 utilization outcomes for outpatient cancer-related services: (1) number of evaluation and management (E&M) visits, (2) number of most commonly used expensive billing codes, and (3) number of radiation therapy sessions. Additionally, we assessed the number of cancer-related hospitalizations. Further details of these measures are included in the eAppendix.

The key explanatory variable was a binary indicator of chemotherapy receipt in HOPDs. We adjusted for patient risks, including (1) 22 comorbidities that are common among Medicare beneficiaries (eAppendix)

; (2) cancer-specific risk factors, including cancer type, metastasis, and cancer-related service use and spending during the 6 months before chemotherapy initiation

; (3) patient demographics and market characteristics; and (4) year dummies. The eAppendix reports a list of adjusters and describes the methods in detail.

We used generalized linear models with log-link and gamma distribution to estimate spending measures,

 and negative binomial models for utilization measures. We clustered standard errors by zip code.

We performed the analysis separately by year to identify whether differences existed across years. We also conducted the analysis separately by cancer type (breast, colon, leukemia, lung, lymphoma, and prostate) to check whether the results were consistent across cancer types.

A total of 61,993 patients were included in the analysis; 20,670 (33%) belonged to the HOPD group and 41,323 (67%) belonged to the PO group. The distribution of cancer types was different between the 2 groups: The proportion of patients with prostate cancer was lower in the HOPD group compared with the PO group (18% vs 45%), and the shares of patients with breast and lung cancer were higher in the HOPD group than the PO group (breast, 22% vs 15%; lung, 20% vs 14%, respectively) (

 illustrate the risk-adjusted results for spending and utilization measures, respectively. Risk-adjusted spending on outpatient cancer services other than chemotherapy was slightly lower among the HOPD group relative to the PO group ($12,183 vs $12,444; 

 <.05). Risk-adjusted spending on cancer-related inpatient services was higher among the HOPD group than the PO group ($3996 vs $3168; 

For risk-adjusted outpatient utilization, the HOPD group had lower numbers of cancer-related E&M visits (6.9 vs 7.7; 

 <.01), commonly used expensive billing codes (1.8 vs 2.1; 

 <.01), and radiation therapy sessions (2.8 vs 3.1; 

 <.01). However, the HOPD group had a higher number of cancer-related hospitalizations than the PO group (0.34 vs 0.31; 

All sensitivity analyses produced similar results to the primary findings except the analysis for breast and prostate cancer (

Analyzing 2010-2013 Medicare claims, we examined cancer-related spending and utilization by site of provider-administered chemotherapy in Medicare. We found that patients receiving chemotherapy in HOPDs had slightly lower risk-adjusted spending on nonchemotherapy outpatient services than those in POs during the 6-month follow-up period. Although the difference in the spending was small, relatively low spending in the HOPD group is unexpected because payment rates for nonchemotherapy outpatient services are generally higher in HOPDs than in POs.

 However, a lower level of utilization (ie, fewer nonchemotherapy outpatient visits) in HOPDs compared with POs indicated that the difference in outpatient spending was driven by the difference in utilization in Medicare.

Our finding of low outpatient care use and spending in HOPDs is consistent with the study by Kalidindi et al, which reported spending on outpatient provider-administered chemotherapy services in Medicare was lower in the HOPD group than in the PO group due to lower utilization of chemotherapy services in HOPDs.

 However, our finding differs from the results of previous studies of commercially insured patients.

 This is probably because the differences in payments for outpatient cancer services between HOPDs and POs are much larger in the commercial setting than in Medicare.

For cancer-related inpatient care, the HOPD group had higher spending than the PO group. This spending difference was also driven by utilization differences: The HOPD group had a higher number of hospitalizations than the PO group. Analyzing whether/how this finding is related to the difference in outpatient care utilization is important, but it was beyond the scope of our study. We leave it to future research.

We note limitations of our study. First, we did not have information on cancer stage—a factor that may influence cancer-related use and spending. However, we controlled for a comprehensive set of health-risk factors, including comorbidities, cancer type, metastasis, and prior cancer-related utilization and spending. Second, we did not assess implications of different utilization by care setting for patient outcomes, except the analysis of hospitalizations. Finally, our results may not be generalizable to patients with other types of insurance.

Our analysis is the first to report differences in utilization and spending on cancer-related services by site of chemotherapy in Medicare. Our study results suggest that differences in cancer care spending by site of chemotherapy in Medicare vary by service type. Those differences are not entirely driven by payment differences between HOPDs and POs but partially reflect differences in care utilization. As the number of patients receiving chemotherapy in HOPDs increases in Medicare, it will be important to continue examining differences in care patterns by care site and their implications for patient care and outcomes.

Pennsylvania State University (AS, YK, JJ), University Park, PA.

 National Institutes of Health (NIH)/National Institute on Aging grant number 1R01AG047934-01 and NIH grant number R24 HD041025.

 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 (AS, YK, JJ); acquisition of data (JJ); analysis and interpretation of data (AS); drafting of the manuscript (AS, YK); critical revision of the manuscript for important intellectual content (AS, YK, JJ); statistical analysis (AS, YK); obtaining funding (JJ); administrative, technical, or logistic support (YK); and supervision (JJ).

 Yamini Kalidindi, MHA, Pennsylvania State University, 501-G Ford Bldg, University Park, PA 16802. Email: ypk100@psu.edu.

1. Table 4: total expenses and percent distribution for selected conditions by source of payment: United States, 2014. Agency for Healthcare Research and Quality website. meps.ahrq.gov/data_stats/quick_tables_results.jsp?component=1&subcomponent=0&tableSeries=2&year=-1&SearchMethod=1&Action=Search. Accessed April 23, 2018.

2. Mariotto AB, Yabroff KR, Shao Y, Feuer EJ, Brown ML. Projections of the cost of cancer care in the United States: 2010-2020 [erratum in 

. 2011;103(2):117-128. doi: 10.1093/jnci/djr009.

Articles

Differences in cancer care spending and utilization between Medicare beneficiaries receiving chemotherapy in hospital outpatient departments versus physician offices vary by service type.

Cancer Care Spending and Use by Site of Provider-Administered Chemotherapy in Medicare

It is unfortunate that the key approach and message of our article were not received by the authors of the letter. We use this opportunity to clarify our risk-adjustment methods and to discuss the importance of continuing work on the topic.

The primary aim of our research was to provide the first information on risk-adjusted spending and utilization of chemotherapy by care site in Medicare because most prior work on this topic examined the commercial sector. We focused on chemotherapy drugs and administration, so researchers will be aware of differences in site-specific chemotherapy use and costs between Medicare and commercial settings. Future studies can expand our analysis to include cancer services other than chemotherapy.

Regarding the risk-adjustment methods, we lacked information on cancer stage. This is a well-known limitation of claims data that also affected the Winn et al

 study of the commercial setting. However, we adjusted for a comprehensive set of risk factors, including indicators of comorbid illnesses that are common among chemotherapy users (eg, heart disease, diabetes, hypertension, depression, chronic pulmonary disease). Moreover, we used cancer-specific risk factors, including cancer metastasis, cancer type, and prior-year cancer-related utilization variables, which are strong predictors of chemotherapy service use and spending. The authors of the letter mentioned the Deyo-Charlson Index, which predicts general health outcomes, including mortality, as a possible adjuster. However, we believe that the cancer-specific risk factors we used are more appropriate and better adjusters for the analysis of chemotherapy services.

It is also surprising that the authors of the letter viewed our analysis of and adjustment for differences in the distribution of cancer types between hospital outpatient departments (HOPDs) and physician offices as a limitation of the study. Most prior comparisons of cancer costs between HOPDs and physician offices did not adjust for differences in cancer types by setting. We clearly pointed out those differences and discussed the importance of adjusting for them by showing how the results changed after adjustment. This is a main contribution and takeaway of our study.

For the facility fees in HOPDs, the Medicare outpatient prospective payment system rates for spending on chemotherapy administration in HOPDs include hospital facility fees.

 Thus, using the Medicare payment rate to calculate chemotherapy administration spending in HOPDs is appropriate and incorporates the facility fees.

Finally, differences in payments by site of care are relatively small in Medicare compared with the commercial sector. However, Medicare is an important payer for cancer-related services, being responsible for one-third of all cancer costs.

 Given Medicare’s role in cancer care, Medicare cancer costs are an important issue to analyze. We offered the first evidence on site-specific Medicare spending for chemotherapy agents and administration. As the site of care continues to shift from physician offices to HOPDs, chemotherapy costs and total cancer costs by site will continue to be important in both Medicare and commercial settings. We hope to see future studies that address this issue.

 Pennsylvania State University (YK, JJ), University Park, PA; University of Minnesota (RF), Minneapolis, MN.

 None; the original research was supported by NIH/NIA grant number 1R01AG047934-01 and NIH grant number R24 HD041025.

 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 (YK, RF); analysis and interpretation of data (RF); drafting of the manuscript (YK, JJ); critical revision of the manuscript for important intellectual content (JJ, RF); administrative, technical, or logistic support (YK, JJ); and supervision (JJ). 

 Yamini Kalidindi, MHA, Department of Health Policy and Administration, Pennsylvania State University, 604 Ford Bldg, University Park, PA 16802. Email: ypk100@psu.edu.

1. Winn AN, Keating NL, Trogdon JG, Basch EM, Dusetzina SB. Spending by commercial insurers on chemotherapy based on site of care, 2004-2014. 

. 2018;4(4):580-581. doi: 10.1001/jamaoncol.2017.5544.

2. Guidi TU. Medicare’s Hospital Outpatient Prospective Payment System: OPPS 101 (part 2 of 2). 

. 2011;7(1):57-60. doi: 10.1200/JOP.2010.000191.

3. Chapter 3: hospital inpatient and outpatient services. In: Medicare Payment Advisory Commission. 

Report to the Congress: Medicare Payment Policy

4. Part B drugs payment systems. Medicare Payment Advisory Commission website. medpac.gov/docs/default-source/payment-basics/medpac_payment_basics_17_partb_final.pdf?sfvrsn=0. Published October 2017. Updated October 2017. Accessed August 22, 2018.

5. Total expenses and percent distribution for selected conditions by source of payment: United States, 2014. Agency for Healthcare Research and Quality website. meps.ahrq.gov/data_stats/tables_compendia_hh_interactive.jsp?_SERVICE=MEPSSocket0&_PROGRAM=MEPSPGM.TC.SAS&File=HCFY2014&Table=HCFY2014_CNDXP_D&_Debug=. Accessed August 22, 2018.

Our study provides the first evidence on site-specific Medicare spending on chemotherapy adjusting for patient comorbid illnesses, cancer type, and other cancer-related risks.

The Authors Respond and Stand Behind Their Findings

 To compare Medicare spending on provider-administered chemotherapy in hospital outpatient departments (HOPDs) and physician offices after controlling for cancer type.

 We used 2010-2013 claims data for a random sample of Medicare fee-for-service beneficiaries who had cancer and received chemotherapy services either in physician offices or in HOPDs. We constructed 2 spending measures: (1) spending on chemotherapy drugs and (2) spending on chemotherapy administration. Each spending measure was the allowed payment, which includes both Medicare reimbursement and patient out-of-pocket spending. We compared the spending measures in the 2 care settings using regression analysis to control for certain patient risk factors, including cancer type. We also compared the number of chemotherapy and administration claims per beneficiary and spending per claim by cancer type to understand differences in utilization patterns in the 2 care settings.

 Risk-adjusted chemotherapy drug spending per beneficiary was $2451 lower in HOPDs compared with physician offices. Risk-adjusted chemotherapy administration spending was $322 higher in HOPDs than in physician offices. Patients in physician offices received chemotherapy drugs more frequently than those in HOPDs. However, the chemotherapy spending per claim line was higher in HOPDs than physician offices.

 Chemotherapy drug spending per Medicare beneficiary was lower in HOPDs than in physician offices, driven by less frequent use of chemotherapy in HOPDs. As the site of provider-administered chemotherapy shifts from physician offices to HOPDs, continuing assessment of cancer care spending by site of care is necessary.

Using 2010-2013 Medicare claims data, this study's results demonstrate that:

Spending on chemotherapy drugs was $2451 lower for Medicare beneficiaries receiving chemotherapy in hospital outpatient departments (HOPDs) than in physician offices.

The spending on chemotherapy administration was $322 higher for Medicare beneficiaries receiving chemotherapy in HOPDs than in physician offices.

As chemotherapy infusions are increasingly provided in the hospital outpatient setting, policy makers and payers should be aware that this shift in the site of chemotherapy may influence cancer care patterns and spending.

Chemotherapy is a common cancer treatment modality and a significant contributor to the cost of cancer treatment.

 Many chemotherapy drugs are available in injectable forms, which are administered by providers in clinical settings.

 Provider-administered drugs are usually reimbursed under the medical benefit of an insurance policy instead of the pharmacy benefit.

 In Medicare, they are reimbursed by Part B coverage for outpatient medical services. Providers purchase drugs and then submit claims to Medicare for reimbursement of the drugs and associated administration costs.

 Most Part B—covered drugs are administered in physician offices or hospital outpatient departments (HOPDs).

Over the past decade, the site of provider-administered cancer drugs has shifted from physician offices to HOPDs.

 This trend has led to a concern that cancer care costs may increase because of differences in care costs between HOPDs and physician offices. Spending on cancer care in commercial settings is considerably higher in HOPDs than in physician offices, mainly due to higher payment rates for chemotherapy drugs and other services in HOPDs.

However, these findings may not apply to Medicare. Medicare typically reimburses hospitals and physicians the same fee for Part B—covered drugs: 106% of the manufacturer’s average sales price (ASP; the budget sequestration of 2013 reduced payments received by providers to 104.3% of ASP).

 No consistent pattern exists in Medicare’s reimbursement for drug administration. Some administration codes are paid more in HOPDs, whereas others are paid more in physician offices. In general, payments are higher in HOPDs. For example, in 2011, 14 of 20 administration codes payable in both settings were paid more in HOPDs.

 However, administration fees are much smaller than chemotherapy drug costs. Thus, differences in chemotherapy-related costs in Medicare Part B mainly come from differences in chemotherapy drug utilization, such as the quantity of chemotherapy or use of more expensive chemotherapeutic agents.

A report by The Moran Company compared spending on chemotherapy between HOPDs and physician offices in Medicare using 2009-2011 claims data.

 The report documented that the average number of chemotherapy claims per patient was slightly higher in HOPDs than in physician offices and average spending per patient on chemotherapy agents was substantially higher in HOPDs. Based on these findings, the Moran report concluded that more and costlier chemotherapy treatments are used in HOPDs than in physician offices, given the same Medicare fees for chemotherapy drugs in both settings. However, the Moran analysis did not adjust for differences in patient risk factors between the 2 settings. An important risk factor is cancer type. The distribution of cancer types differs by setting,

 and cancer drug utilization patterns differ by cancer type.

 If patients with specific cancer types requiring expensive chemotherapy are more likely to be treated in HOPDs, the Moran report’s conclusion is not valid.

To our knowledge, no study has examined chemotherapy-related spending in Medicare Part B after controlling for patient characteristics, such as cancer type. Our study fills this gap. We compared chemotherapy drug and administration spending in HOPDs and physician offices after controlling for cancer type. In addition, we explored differences in chemotherapy utilization patterns between the 2 settings.

The primary data sources were the 2010-2013 Medicare Outpatient file, which contains records for services in HOPDs, and the 2010-2013 Medicare Carrier file, which has claims for services by noninstitutional providers. Both files contain information on diagnosis, Healthcare Common Procedure Coding System (HCPCS) code, service date, and payments. Medicare Master Beneficiary Summary files provided beneficiaries’ demographic characteristics and disease indicators, including cancer type, and the American Community Survey supplied zip code—level income, education, and unemployment rates.

The study population is a random sample of Medicare fee-for-service beneficiaries with cancer between 2010 and 2013. To select the sample, CMS first identified all patients with cancer from 100% of Medicare claims based on the standard algorithm used to create cancer indicators in the Medicare Chronic Condition Warehouse: having at least 1 inpatient or skilled nursing facility claim with a cancer diagnosis or at least 2 Carrier or Outpatient claims with a cancer diagnosis in a given year. Next, CMS provided us with the data for a random sample of those patients.

We restricted the sample to patients who had at least 1 chemotherapy claim identified by HCPCS Level II (J-codes) in the Outpatient or Carrier data. Chemotherapy includes all antineoplastic drugs (immune, hormonal, and targeted therapy). We selected claims with both cancer diagnosis and chemotherapy J-codes to exclude cases using cancer drugs for other conditions. All cancer diagnosis codes and chemotherapy J-codes used are reported in 

). Claims for chemotherapy reported in both Carrier and Outpatient files using the same J-code on the same day were considered duplicates, and duplicate claims in the Carrier file were excluded to avoid double counting. We considered Carrier claims with the service place code of HOPDs as HOPD claims.

We further restricted the sample to patients with cancer who had both Medicare Part A and Part B coverage for the full year, and we excluded those who died within 3 months of diagnosis. We excluded enrollees in Medicare Advantage plans because their claims data are not available to researchers.

The study sample was categorized into 2 groups depending on the site of chemotherapy administration: HOPD-only if they received chemotherapy in HOPDs only and office-only if they received chemotherapy in physician offices only. Patients receiving chemotherapy in both settings, who accounted for 4.4% of the sample, were excluded to make a clean comparison of costs between HOPDs and physician offices.

We constructed 2 outcome measures: (1) chemotherapy drug spending per beneficiary and (2) chemotherapy administration spending per beneficiary. These spending variables are allowed payments, including both Medicare reimbursements and patient out-of-pocket spending. We constructed chemotherapy drug spending by summing the allowed payments across each patient’s chemotherapy claims with a cancer diagnosis code. Chemotherapy administration spending was created as the sum of the allowed payments across each patient’s claims with chemotherapy administration codes and a cancer diagnosis.

We began with a descriptive analysis of 6 cancer types (prostate, breast, lymphoma, colon, lung, and leukemia) for which Part B chemotherapy is frequently used. First, we compared the distribution of cancer types between HOPDs and physician offices. Second, we compared chemotherapy drug and administration spending per beneficiary between the 2 settings for the entire sample and for each cancer type.

We used a linear regression model with clustered standard errors within a zip code. Our unit of analysis was a patient-year. The dependent variables were chemotherapy drug spending and chemotherapy drug administration spending. The key explanatory variable was a binary indicator equal to 1 if the patient received provider-administered chemotherapy only in HOPDs and 0 if she/he received chemotherapy only in physician offices. Key control variables were cancer type indicators, an indicator of cancer metastasis, and the number of cancer-related hospitalizations and outpatient visits in the prior year. To identify metastasis, we used the criterion of at least 2 diagnosis codes of metastatic disease (

International Classification of Diseases, Ninth Revision, Clinical Modification 

] codes 196-199) separated by 30 days or more.

The regression also controlled for patient age, gender, race, state buy-in status (an indicator of whether Medicaid pays the patient’s Part B premium), indicators of chronic conditions (ischemic heart disease, diabetes, hypertension, hyperlipidemia, depression, heart failure, chronic pulmonary disease, and cataract), and number of chronic conditions. Area-level variables were average income, percent college educated, and unemployment rates at the zip code level. Finally, we used year dummies to control for year-specific effects that are common to all patients.

Using the regression results, we obtained risk-adjusted spending in each setting. To calculate risk-adjusted chemotherapy drug and administration spending in physician offices, we computed predicted spending by setting the HOPD indicator to 0 and all other covariates to their mean values. Similarly, we obtained risk-adjusted spending in HOPDs by computing predicted spending with the HOPD indicator equal to 1 and the means of all other covariates.

To explore whether chemotherapy utilization patterns differed between HOPDs and physician offices, we compared the number of chemotherapy and administration claims per beneficiary by cancer type. We also assessed spending per claim for chemotherapy drug and administration by cancer type.

We performed the following sensitivity checks. First, we performed the regression analysis by year to check if differences in a particular year were driving the overall regression results. We used the same variables as in the primary analysis (except year-specific dummies) and calculated risk-adjusted spending in HOPDs versus physician offices for each year.

Second, we performed the regression analysis separately for each of the 6 cancers (prostate, breast, lymphoma, colon, lung, and leukemia) to check if the results were consistent across major cancer types.

Third, we limited the analysis to separately reimbursable chemotherapy drugs (drugs that are not bundled into a payment group under the Medicare Hospital Outpatient prospective payment system). Medicare determines separately reimbursable drugs based on a threshold daily cost (>$80 in 2013). Chemotherapy drugs whose daily costs are below the threshold are considered a dependent or ancillary service to the drug administration. Their cost is “bundled” into an Ambulatory Payment Classification

 and arbitrarily allocated by hospitals. Including them in the analysis may lower the estimates of chemotherapy drug spending per beneficiary in HOPDs. We thus excluded nonseparately reimbursable drugs from both the Carrier and Hospital Outpatient files and checked the sensitivity of the results.

codes of surgeries for certain cancers for which there is evidence of better outcomes.

 Individuals who underwent these surgeries are likely to use chemotherapy drugs as adjuvant therapy. We conducted the regression analysis on this subpopulation, who were relatively homogenous in terms of cancer severity, and checked the sensitivity of the results.

 shows the distribution of cancer types among Medicare chemotherapy users in HOPDs and physician offices. Six cancers accounted for more than 90% of all chemotherapy users in both HOPDs and physician offices, but the distribution of cancer types differed by setting. Prostate cancer accounted for 25% of chemotherapy users in HOPDs but more than 55% of chemotherapy users in physician offices. The bottom panel of Figure 1 indicates that physician offices were the dominant place of services for all cancer types. Eighty-four percent of patients with prostate cancer received chemotherapy in physician offices, and about 60% of patients with other cancers used physician offices.

 shows that unadjusted average chemotherapy drug spending per beneficiary in the entire sample was about 34% higher in HOPDs than in physician offices ($15,058 vs $11,219). However, chemotherapy drug spending for patients with the same cancer type was higher in physician offices than in HOPDs for most cancer types except prostate cancer (bottom panel of Figure 2).

 depicts descriptive data on chemotherapy administration spending per beneficiary. Unadjusted average chemotherapy administration spending per beneficiary was higher in HOPDs compared with physician offices, both in the full sample and among patients with the same cancer type.

 presents risk-adjusted chemotherapy spending in each setting based on the regression results. Risk-adjusted chemotherapy drug spending showed very different patterns than unadjusted spending. Chemotherapy drug spending per beneficiary after risk adjustment was $2451 lower in HOPDs than in physician offices ($10,658 vs $13,109). Risk-adjusted chemotherapy administration spending per beneficiary was $322 higher in HOPDs compared with physician offices ($1543 vs $1221).

 reports the number of claims per beneficiary and spending per claim on chemotherapy drugs and administration by cancer type. These data help explain why chemotherapy drug spending was lower in HOPDs compared with physician offices after controlling for cancer type. The frequency of chemotherapy among chemotherapy users with the same cancer type was higher in physician offices than HOPDs for most cancer types except prostate cancer. For example, patients with colon cancer had 19 chemotherapy drug claims per beneficiary in physician offices versus 13 in HOPDs, on average. On the other hand, spending per claim among patients with colon cancer was $367 higher in HOPDs compared with physician offices. For other cancers, spending was between $257 and $737 higher in HOPDs than in physician offices. Thus, the difference in spending per claim between HOPDs and physician offices is much smaller compared with the difference in spending from adding 1 more claim, which exceeds $1000 across all cancer types in both settings. These data imply that additional drug claims are an important driver of total chemotherapy drug spending per patient, and more frequent use of chemotherapy led to higher spending in physician offices than HOPDs, after controlling for cancer type.

Similarly, the number of chemotherapy administration claims per beneficiary was higher in physician offices than in HOPDs for most cancer types, except prostate cancer. On the other hand, average spending per administration claim was almost twice as high in HOPDs compared with physician offices for most cancer types. For example, spending per chemotherapy administration claim for colon cancer was $182 in HOPDs compared with $100 in physician offices. This difference in spending per claim is large, considering that most administration claims were less than $200. Thus, higher chemotherapy administration costs per beneficiary in HOPDs compared with physician offices are largely driven by more costly administration claims in HOPDs.

The results of the sensitivity analysis supported the findings described above. Across all years, risk-adjusted chemotherapy drug spending per beneficiary was lower in HOPDs than in physician offices, and risk-adjusted chemotherapy administration spending per beneficiary was consistently higher in HOPDs compared with physician offices (

). Results from the analysis of each cancer (

) were also consistent with the main analysis. The analysis using only separately reimbursable chemotherapy produced very similar results to the primary analysis (

). Risk-adjusted chemotherapy drug spending per beneficiary was $2245 lower in HOPDs than in physician offices, driven by the smaller number of separately reimbursable chemotherapy claims in HOPDs. The analysis using patients undergoing a cancer-related surgery also produced results consistent with the main analysis.

Analyzing 2010-2013 Medicare claims data, we found that risk-adjusted chemotherapy drug spending was lower for patients receiving chemotherapy in HOPDs than for patients using physician offices. We explored 2 contributors to these differences: differences in the number of chemotherapy drug claims and spending per chemotherapy claim. We found that chemotherapy users in physician offices received more chemotherapy than those in HOPDs for most cancer types and that average spending per chemotherapy drug claim was slightly higher in HOPDs than physician offices for all cancer types. However, the differences in average spending per claim were not large enough to make substantial differences in total chemotherapy drug spending per beneficiary. These findings indicate that lower utilization per beneficiary was an important driver of lower risk-adjusted chemotherapy drug spending in HOPDs than in physician offices.

Our findings differ from those of the Moran report, which concluded that more and costlier chemotherapy treatments are used in HOPDs than in physician offices.

 It is important to note that the Moran report did not adjust for patient risk factors, including cancer type. As our results and prior literature indicate, the distribution of cancer types differs by setting

 Further, our data indicated that the frequency of chemotherapy among chemotherapy users with the same cancer was higher in physician offices than HOPDs for most cancer types except prostate cancer. Thus, adjusting for cancer type is of utmost importance in explaining the cost and utilization differences between the 2 settings.

Our findings also differ from prior research in commercial settings, which consistently found that chemotherapy costs were higher in HOPDs than physician offices.

 However, as mentioned earlier, spending differences in commercial settings are driven by price differences between HOPDs and physician offices rather than differences in the quantity of services.

 Medicare uses the same reimbursement rates for chemotherapy drugs in both settings. It is thus not surprising that chemotherapy spending per Medicare beneficiary is lower in HOPDs than physician offices.

Our analysis also showed that higher chemotherapy spending in physician offices was due to higher utilization. This result is consistent with research in commercial settings. Hayes et al found that the mean number of chemotherapy sessions in employer-sponsored plans was higher in community oncology clinics than in HOPDs.

 To our knowledge, our analysis is the first to explore differences in chemotherapy utilization by care setting and cancer type in Medicare.

We note several limitations of our study. First, we did not consider costs for other services that patients may have used when receiving chemotherapy. Prior research suggests that patients visiting HOPDs are likely to receive additional services (eg, laboratory tests) that might not be offered in physician offices.

 We did not analyze spending on those services. Second, our findings are not generalizable to the commercial sector, where payment rates for chemotherapy drugs differ substantially by care site.

 Third, we could not completely adjust for cancer severity, such as cancer stage, because detailed clinical information is not available in Medicare data. We partially addressed this issue by using a metastasis indicator, but our approach of identifying metastasis from diagnosis codes may have limited validity.

 Third, the choice of chemotherapy site could depend on patients’ preferences. Patients may prefer to use HOPDs because of the availability of other services or a short travel distance. Such patient characteristics might be related to chemotherapy use and spending to some extent. However, our study did not control for those factors. Finally, there was a shift in the site of cancer care from office-based to HOPD-based due to hospitals’ acquisition of physician practices during the study period.

 Although examining chemotherapy use and spending in those practices acquired by hospitals would be informative, it is beyond the scope of our analysis, and we leave it to future research.

Chemotherapy drug spending per Medicare beneficiary was lower in HOPDs than in physician offices, driven by less frequent use of chemotherapy. As the site of provider-administered chemotherapy shifts from physician offices to HOPDs, continuing assessment of cancer care spending by care site is necessary.

 Department of Health Policy and Administration, Pennsylvania State University (YK, JJ), University Park, PA; Division of Health Policy and Management, University of Minnesota (RF), Minneapolis,

Spending on chemotherapy drugs was lower among Medicare beneficiaries who received chemotherapy in hospital outpatient departments than among comparable beneficiaries receiving chemotherapy in physician offices.

Yamini Kalidindi, MHA

Articles

Cancer Care Spending and Use by Site of Provider-Administered Chemotherapy in Medicare

The Authors Respond and Stand Behind Their Findings

Differences in Spending on Provider-Administered Chemotherapy by Site of Care in Medicare