Vertical integration of oncologists and cancer outcomes and costs in metastatic castration-resistant prostate cancer

Abstract

Background

The share of oncology practices owned by hospitals (ie, vertically integrated) nearly doubled from 2007 to 2017. We examined how integration between hospitals and oncologists affected care quality, outcomes, and spending among metastatic castration-resistant prostate cancer (mCRPC) patients.

Methods

Using Surveillance, Epidemiology, and End Results–Medicare linked data and the Medicare Data on Provider Practice and Specialty, we identified Medicare beneficiaries who initiated systemic therapy for mCRPC between 2008 and 2017 (n = 9172). Primary outcomes included 1) bone-modifying agents (BMA) use, 2) time on systemic therapy, 3) survival, and 4) Medicare spending for the first 3 months following therapy initiation. We used a differences-in-differences approach to estimate the impact of vertical integration on outcomes, adjusting for patient and provider characteristics.

Results

The proportion of patients treated by integrated oncologists increased from 28% to 55% from 2008 to 2017. Vertical integration was associated with an 11.7 percentage point (95% confidence interval [CI] = 4.2 to 19.1) increased likelihood of BMA use. There were no satistically significant changes in time on systemic therapy, survival, or total per-patient Medicare spending. Further decomposition showed an increase in outpatient payment ($5190, 95% CI = $1451 to $8930) and decrease in professional service payment (−$4757, 95% CI = −$7644 to −$1870) but no statistically significant changes for other service types (eg, inpatient and prescription drugs).

Conclusions

Vertical integration was associated with statistically significant increased BMA use but not with other cancer outcomes among mCRPC patients. For oncologists who switched service billing from physician offices to outpatient departments, there was no statistically significant change in overall Medicare spending in the first 3 months of therapy initiation. Future studies should extend the investigation to other cancer types and patient outcomes.

The share of private medical practices owned by hospitals doubled in the past decade from 30% to 60% (1-3). Oncology is among the specialties that experienced the fastest growth in integration (1). Integration between physicians and hospitals, which is an example of vertical integration, may increase hospitals’ market power, allowing them to negotiate higher prices with private insurers (4). Although Medicare payment rates, which are set administratively, would be unaffected, integration could still lead to higher Medicare spending through Medicare’s site-based reimbursement policy (5,6). For example, an evaluation and management office visit and chemotherapy administration would be reimbursed 19%-70% more if delivered in the hospital outpatient department (HOPD) setting than in a freestanding oncology practice (5,6). Conversely, integration may reduce administrative costs, improve electronic health records interoperability, facilitate dissemination of practice standards, and allow seamless access to other specialties and ancillary services (7-9). These potential benefits could generate cost savings and improve patient outcomes (10).

Earlier studies found that physician–hospital integration led to increased health-care prices and spending (11,12). Jung et al. (13) showed that integration between hospitals and oncologists was associated with a 20% increase in Medicare spending for chemotherapy. A 6.8% decrease in the quantity of drugs administered was offset by a 28.4% increase in spending on intravenous chemotherapy per claim. Their study was limited to chemotherapy costs and did not consider spending on oral drugs or other services (eg, inpatient care), which comprise a large proportion of spending for some patients undergoing cancer treatment. If the incentives from the site-based reimbursement policy are substantial, oncologists may switch from oral to intravenous chemotherapy regimens after vertical integration, leading to increases in physician-administered chemotherapy spending and decreases in oral chemotherapy spending.

Studies examining the impact of physician–hospital integration found little effect on care quality and patient outcomes, including cancer mortality (13-16). Other cross-sectional studies comparing cancer care across different care settings showed that patients with prostate cancer treated in integrated systems received better quality of care (17); patients with metastatic cancers treated at hospital-based facilities remained on first chemotherapy episode longer, received fewer low-value supportive care drugs, and had similar end-of-life (EOL) utilization compared with those treated at private practices (18,19). However, cross-sectional studies may be biased by unobserved differences in patient characteristics.

To fill this gap, we used longitudinal population-based data to investigate the impact of oncologist-hospital integration on care quality, survival outcomes, and Medicare spending among a cohort of patients with metastatic castration-resistant prostate cancer (mCRPC) initiating systemic therapy in 2008-2017. Since 2010, several new drugs have been approved to treat mCRPC (20), resulting in more options with unclear guidance about the optimal course of treatment. The increasing complexity of treatment decisions for patients with mCRPC makes it an ideal condition to understand the impact of vertical integration.

Methods

Data and analytical sample

We used Surveillance, Epidemiology, and End Results (SEER)–Medicare linkage for patients with mCRPC. SEER is a population-based cancer registry system covering 21 regions and 34.6% of the US population (21). It provides information on patients’ cancer history, demographic information, and neighborhood-level characteristics from the American Community Survey. Medicare claims include patients’ health-care utilization, encrypted provider identifiers, and health-care costs. We obtained providers’ information from the Medicare Data on Provider Practice and Specialty file, including demographic characteristics, annual summary of Medicare service volume, up to 2 encrypted tax identification numbers that they billed most services to, and practice location (ie, core-based statistical area [CBSA]).

We extracted a cohort of Medicare beneficiaries aged 65 years and older who initiated systemic therapy for mCRPC in 2008-2017. We defined an index date for each patient as the first claim for systemic therapy including docetaxel, abiraterone, cabazitaxel, enzalutamide, radium RA 223, or sipuleucel-T (22,23). Procedure codes and drug names were used to identify these drugs (Supplementary Table 1, available online). We also required continuous Medicare fee-for-service (FFS) coverage and Part D enrollment from 1 year before to 1 year after the index date or death. Patients had to have valid histology, diagnosis, and death date if applicable. The treating oncologist was assigned based on the encrypted National Provider Identifier (NPI) or Unique Physician Identification Number (UPIN) that appeared most frequently in the patient’s outpatient or carrier claims within 30 days surrounding the index date. We excluded patients not treated by a medical or hematologic oncologist or by oncologists who de-integrated from hospitals (see below). Final sample consisted of 9172 patients. Supplementary Figure 1 (available online) provides the detailed sample derivation process. The Emory institutional review board approved the study and waived the requirement for written informed consent for participants.

Vertical integration

We used billing data to assess whether oncologists were vertically integrated with hospitals each year (24-27). Briefly, information from the Medicare Data on Provider Practice and Specialty file on the oncologist’s total services billed under different settings was used to calculate the percent of line items with an HOPD as the site of service (28). Vertical integration allows oncologists to bill from an HOPD while still providing services at the office under certain restrictions (5,29). Our primary analysis followed the approach by Saghafian et al. (26) to use a 10% cutoff, such that oncologists who switched from billing less than 10% to 10% or more services to HOPD were considered vertically integrated. Compared with higher cutoffs used in other studies (24-27), this lower cutoff identifies physicians experiencing a practice acquisition (ie, partial integration) and physicians becoming employed by a hospital (ie, full integration). Using a higher cutoff may misclassify oncologists who experienced partial integration as always independent.

Dependent variables

Following the Donabedian framework, we assessed the impact of vertical integration, ie “structure” on the “process” and “outcome” aspects of care delivery (30). “Process” measures included bone-modifying agents (BMA) use and patients’ time on systemic therapy. Overall survival and health-care costs were evaluated as “outcome” measures. BMA, including denosumab and zoledronic acid, are the gold standard for medical management of mCRPC to prevent cancer treatment–induced bone loss (31) and have been recommended by the National Comprehensive Cancer Network guidelines since 2010 (32). We examined any BMA use and the total number of administrations of BMA after the index date (33) (Supplementary Table 1, available online). Time on systemic therapy and survival were measured in months from the index date to the last systemic therapy claim and to death or end of follow-up (ie, December 2018), respectively. We also examined 2 dichotomous outcomes: alive at 1 year and time on systemic therapy exceeding 1 year.

As the time on treatment and survival time varied considerably for patients with mCRPC, Medicare payments were calculated for the first 3 months after the index date among patients who survived at least 3 months (n = 8233). We adjusted Medicare payments to 2019 US dollars using the Personal Health Care Index (34). We assessed EOL utilization as a secondary outcome among patients deceased by 2018 (n = 5641). Detailed measures are in Supplementary Table 1 (available online).

Covariates

Models were adjusted for characteristics that may affect care process and outcomes. Patients’ sociodemographic characteristics included age at index date, race and ethnicity (abstracted from medical records), metro residency, dual eligibility for Medicaid, and census tract–level socioeconomic status measured by the Yost index (35). The National Cancer Institute Comorbidity Index was calculated using claims within 1 year before the index date (36). We also measured time-varying provider characteristics, such as whether the oncologist’s total service volume was below the average in the CBSA in the index year, and organization size. They are included to proxy provider volume and experience, which are associated with cancer care quality (37-39).

Statistical analysis

Descriptive statistics for covariates and study outcomes were stratified by whether patients’ treating oncologists remained nonintegrated, were always integrated, or switched from nonintegrated to integrated status (hereafter, change-group).

A differences-in-differences (DID) approach was used to examine the impact of vertical integration on care process and outcomes (40). Each dependent variable was regressed on the treating oncologist’s integration status in the patient’s index year, controlling for oncologist and index year fixed effects, and covariates. A Weibull survival model was used for time-to-event outcomes, and a generalized linear model was used for continuous and dichotomized outcomes. This approach measures the adjusted changes in outcomes before and after oncologists became integrated relative to trends in outcomes among patients of oncologists whose integrated status did not change. We assessed the parallel trend assumption—that baseline trends were similar between groups prior to integration—by plotting yearly averages for each outcome for change-group patients treated in the preintegration period vs patients treated by oncologists who remained nonintegrated or always integrated.

Given the staggered timing of physicians attaining integration status, we conducted “event study” analyses to evaluate potential heterogeneous effects of vertical integration over time from 1 year up to 6 years after integration (41-43). Adjusted differences in outcomes were estimated for each year after integration relative to the second to last year before integration (as care may be interrupted in the last year before integration) compared with patients treated by oncologists who remained nonintegrated. Patients treated by always-integrated oncologists were not included in the event study analyses. Additional details are provided in the Supplementary Methods (available online).

Sensitivity analyses were conducted using alternative cutoffs to classify vertical integration. However, results should be interpreted with caution given the limited sample size of oncologists who experienced integration and the important distinctions between partial and full integration.

Results

Sample characteristics

Among 9172 eligible patients with mCRPC, 859 patients were treated by 168 unique oncologists who switched from nonintegrated to integrated status. Among the remaining patients, 4862 were treated by 1728 unique oncologists who remained nonintegrated, and 3451 were treated by 983 oncologists who were always integrated. The share of patients treated by integrated oncologists nearly doubled during the study period, from less than 28% in 2008 to 55% in 2017 (Figure 1). Noticeable differences in sociodemographic characteristics between the change-group and the other two groups were higher proportions of White patients, residency in nonmetro areas and areas with lower socioeconomic status, and a lower proportion with dual eligibility. Change-group patients were also more likely to be cared for by oncologists who had lower-than-average service volume in their CBSA (Table 1).

Figure 1.

Trend of patients treated by integrated oncologists (n = 9172). Integrated oncologists defined as billing more than 10% annual total services to hospital outpatient department.

Table 1.

Descriptive statistics of sample characteristics and dependent variables by treating oncologists’ integration status (n = 9172)

Sample characteristics and study outcomes	Treated by oncologists remained nonintegrated	Treated by oncologists who became integrated	Treated by oncologists always integrated	Pa
Total No. of patients, No. (%)b	4862 (53.0)	859 (9.4)	3451 (37.6)
Total No. of oncologists, No. (%)b	1728 (60.0)	168 (5.8)	983 (34.1)
Patient characteristics
Age at index date in years, mean (SD)c	76.2 (7.5)	75.9 (7.6)	75.1 (7.4)	<.001
Race and ethnicity, No. (%)				<.001
Hispanic	400 (8.2)	55 (6.4)	235 (6.8)
Non-Hispanic Black	457 (9.4)	58 (6.8)	380 (11.0)
Non-Hispanic other/unknownd	247 (5.1)	25 (2.9)	159 (4.6)
Non-Hispanic White	3758 (77.3)	721 (83.9)	2677 (77.6)
Marital status, No. (%)				<.001
Married	2614 (53.8)	483 (56.2)	1507 (43.7)
Single, divorced, widowed	936 (19.3)	162 (18.9)	552 (16.0)
Unknown	1312 (27.0)	214 (24.9)	1392 (40.3)
Metro residency, No. (%)e				.004
Metro	4176 (85.9)	701 (81.6)	2958 (85.7)
Nonmetro/unknownd	686 (14.1)	158 (18.4)	493 (14.3)
Dual eligibility, No. (%)f				<.001
No	3974 (81.7)	731 (85.1)	2821 (81.7)
Yes	888 (18.3)	128 (14.9)	630 (18.3)
NCI Comorbidity Index, mean (SD)	0.5 (0.6)	0.4 (0.5)	0.5 (0.6)	.004
Yost quintiles, No. (%)g				<.001
Q1	548 (11.3)	106 (12.3)	403 (11.7)
Q2	546 (11.2)	103 (12.0)	400 (11.6)
Q3	597 (12.3)	123 (14.3)	507 (14.7)
Q4	793 (16.3)	158 (18.4)	549 (15.9)
Q5	1239 (25.5)	179 (20.8)	906 (26.3)
Unknown	1139 (23.4)	190 (22.1)	686 (19.9)
Provider characteristics
Total service below average in the CBSA, No. (%)h	1053 (21.7)	303 (35.3)	2413 (69.9)	<.001
Total beneficiaries below average in the CBSA, No. (%)h	974 (20.0)	179 (20.8)	1576 (45.7)	<.001
No. of providers reported under the same TIN, mean (SD) h	195 (468)	331 (579)	538 (587)	<.001
Process
Any BMA, No. (%)i	3132 (64.4)	524 (61.0)	2009 (58.2)	<.001
No. of administrations of BMA, mean (SD)i	7.8 (10.8)	7.0 (10.3)	5.8 (9.1)	<.001
Months on systemic therapy, median (IQR)	11 [3, 27]	10 [10, 26]	14 [5, 31]	<.001
Outcomes
Survival in months, median (IQR)	18 [7-37]	17 [7-37]	22 [9-48]	<.001
Medicare payment in first 3 months, mean (SD)
Overall	$34 929 ($27 651)	$33 067 ($27 181)	$33 701 ($29 910)	.09
Inpatient payment	$4175 ($12 401)	$4177 ($12 529)	$3776 ($13 243)	.38
Outpatient payment	$4007 ($12 797)	$10 096 ($22 413)	$14 936 ($27 323)	<.001
Professional service payment	$15 748 ($24 225)	$7742 ($15 396)	$2626 ($8540)	<.001
Part D	$10 297 ($12 448)	$10 387 ($12 670)	$11 697 ($12 774)	<.001

a

χ² or Fisher exact tests were used to compare percentages, and 1-way analysis of variance was used to compare means across 3 groups of oncologists’ integration status. BMA = bone-modifying agents; CBSA = core-based statistical area; IQR = interquartile range; NCI = National Cancer Institute; Q = quintiles; SES = socioeconomic status; TIN = tax identification number.

b

Row percentage reported.

c

Index date and index year represents the date and year of the patient’s systemic therapy initiation.

d

Unknown group combined with other groups because of restrictions on small cell reporting. Sample sizes for unknown groups are less than 0.5% of the study sample.

e

Based on 2013 Rural Urban Continuum Code. Codes 1-3 were classified as metro, and Codes 4-9 were classified as nonmetro.

f

With state buy-in (Medicaid) from 12 months before to 12 months after systemic therapy initiation.

g

Yost index is a composite SES score for census tracts based on median household income, median house value, median rent, percent less than 150% of poverty line, education index, percent working class, and percent unemployed. Lower quintiles represent the lower SES groups. Details can be found at https://seer.cancer.gov/seerstat/databases/census-tract/index.html.

h

Statistics for the time-varying characteristics were reported corresponding to the index year, for example, oncologist’s age in patients’ index year.

i

BMA include denosumab and zoledronic acid.

BMA use

On average, approximately 61.8% of patients received denosumab and zoledronic acid. Patients received an average of 7 administrations during systemic therapy. Trends of any BMA and average administrations of BMA were mostly similar among change-group patients who initiated systemic therapy in preintegration periods and those treated by oncologists who remained independent (Figure 2, A and B). The DID estimates showed an adjusted increase of 11.7 (95% confidence interval [CI] = 4.2 to 19.1; P = .002) percentage points in the likelihood of BMA use associated with vertical integration, an approximately 20% relative increase in BMA use. The average number of BMA administrations after integration increased by 1.8 (95% CI = 0.4 to 3.2; P = .01) (Table 2). These increases in the probability of BMA use and the average number of administrations of BMA were consistent from 1 year up to 6 years postintegration (Figure 3, A and B).

Figure 2.

Yearly trends of study outcomes for change-group patients treated in preintegration periods and patients treated by oncologists remained nonintegrated and always integrated. A) Any BMA use. B) Number of administrations of BMA. C) Median months on systemic therapy. D) Median survival in months. E) Total Medicare payment. F) Total Medicare inpatient payment. G) Total Medicare outpatient payment. H) Total Medicare professional service payment. I) Total Medicare Part D payment. ^a Yearly trend of study outcomes reported up to 2015. All patients’ treating oncologists became integrated in 2017 in the change-group. Statistics for study outcomes in 2016 were suppressed because of small cell reporting restrictions (n < 11). ^b BMA include denosumab and zoledronic acid. BMA = bone-modifying agents.

Figure 2.

Continued.

Figure 3.

Adjusted differences in study outcomes by year relative to vertical integration. A) Adjusted changes in the probability of any BMA use. B) Adjusted changes in the number of administrations of BMA. C) Adjusted changes in the probability of time on systemic therapy at least 1 year. D) Adjusted changes in the probability of 1-year survival. E) Adjusted changes in total Medicare payment. F) Adjusted changes in the total Medicare inpatient payment. G) Adjusted changes in the total Medicare outpatient payment. H) Adjusted changes in the total Medicare professional service payment. I) Adjusted changes in the total Medicare Part D payment. Note: Analysis included patients treated by oncologists who became vertically integrated or remained nonintegrated. Patients treated by always-integrated oncologists were not included in the analysis. Event study models controlled for oncologist and year fixed effects. Time 0 represents the last year before vertical integration, and time 1 represents the first year of vertical integration. Time −1 represents the second last year before vertical integration, which was omitted as the reference year. Hence, all point estimates should be interpreted relative to the second last year before vertical integration. A negative point estimate means that vertical integration is associated with a decrease in the outcome (eg, a decrease in the probability of receiving BMA, a decrease in overall spending). Detailed model specification is provided in Supplementary Methods (available online). ^b BMA include denosumab and zoledronic acid. BMA = bone-modifying agents.

Figure 3.

Continued.

Table 2.

Adjusted changes in bone-modifying agent use and Medicare payments associated with vertical integration

Study outcomes	Adjusted difference (95% CI)a	P
Process measure
Any BMA, in percentage pointsb	11.7 (4.2 to 19.1)	.002
No. of administrations of BMAb	1.8 (0.4 to 3.2)	.01
Outcome measure
Medicare payment, first 3 moc
Overall	−$248 (−$4824 to $4329)	.92
Inpatient payment	$56 (−$1867 to $1979)	.96
Outpatient payment	$5190 ($1451 to $8930)	.007
Professional service payment	−$4757 (−$7644 to –$1870)	.001
Part D	−$579 (−$2153 to $996)	.47

a

Adjusted difference estimated using a difference-in-difference approach detailed in Supplementary Methods (available online). BMA = bone-modifying agents; CI = confidence interval.

b

BMA include denosumab and zoledronic acid.

c

Sample restricted to those who survived at least 3 months (n = 8233).

Time on systemic therapy and survival

The median time on systemic therapy and median survival of our study sample was 13  (interquartile range [IQR] = 4-28) months and 19 (IQR = 8-40) months. Patients in the change-group who initiated systemic therapy before integration had similar time on systemic therapy and cancer survival as patients treated by oncologists who remained nonintegrated (Figure 2, C and D). DID results showed small and statistically nonsignificant changes in time on systemic therapy (hazard ratio [HR] = 0.99, 95% CI = 0.81 to 1.21; P = .90) and survival (HR = 0.98, 95% CI = 0.81 to 1.19; P = .83) after integration (Table 3). The event study analyses assessing changes in the probability of time on systemic therapy at least 1 year and 1-year survival similarly showed statistically nonsignificant changes after vertical integration (Figure 3, C and D).

Table 3.

Adjusted changes in patients’ months on systemic therapy and survival associated with vertical integration^a

Study outcomes	Adjusted HR (95% CI)	P
Process measure
Months on systemic therapy	0.99 (0.81 to 1.21)	.90
Outcome measure
Survival in months	0.98 (0.81 to 1.19)	.83

a

Hazard ratios estimated using a difference-in-difference approach detailed in the Supplementary Methods. Hazard ratios should be interpreted as risks of discontinuing systemic therapy, and risks of death respectively. CI = confidence interval; HR = hazard ratio.

Health-care spending in the first 3 months

Average Medicare spending for the first 3 months of systemic therapy initiation was $34 929, $33 701, and $33 067 for patients treated by oncologists who remained nonintegrated, always integrated, and became vertically integrated, respectively (Table 1). Preintegration trends were similar across groups (Figure 2, E-I). Integration was associated with a statistically nonsignificant change in overall spending (−$248, 95% CI = −$4824 to $4329; P = .92) (Table 2). A similar pattern was observed in the event study analysis (Figure 3, E). A decline in professional service payments (−$4757, 95% CI = −$7644 to −$1870; P = .001) was offset by an increase of a similar magnitude in outpatient payments ($5190, 95% CI = $1451 to $8930; P = .007) in the first 3 months after systemic therapy initiation. The event study showed consistent changes in professional service and outpatient department payments in 1-3 years after vertical integration, but effects subsequently became noisier, likely because of small sample sizes (Figure 3, G and H). There were no statistically significant changes in spending for other services such as inpatient and prescription drugs (Table 2; Figure 3, F and I).

Secondary outcomes and sensitivity analyses

Adjusted differences in EOL utilization showed statistically nonsignificant changes associated with vertical integration (Supplementary Table 2, available online). Sensitivity analyses using alternative cutoffs to classify vertical integration showed consistent patterns for BMA use; time on systemic therapy; and Medicare outpatient, inpatient, and Part D payment. We observed modest reductions in risk of death and modest increases in overall Medicare spending using higher cutoffs. However, these estimates were not statistically different from estimates using our primary cutoff (Supplementary Figure 2, available online).

Discussion

Our study provides evidence about the impact of oncologist-hospital integration on care quality, survival outcomes, and Medicare spending in a cohort of Medicare beneficiaries with mCRPC. Consistent with previous literature, the percent of patients treated by integrated oncologists nearly doubled during 2008-2017 (1). Integration was associated with a clinically and statistically significant increase in BMA use by 11.7 percentage points but no statistically significant changes in survival outcomes and total Medicare spending.

The real-world use of BMA among patients with mCRPC that spanned similar study periods was less than 50%, likely because of physicians’ lack of awareness or concerns about severe adverse events (eg, jaw osteonecrosis) (44,45). A recent study found no statistically significant differences in BMA use across HOPD and office settings (46). Our study showed that more than 60% of patients received BMA, and vertical integration was associated with an 11.7 percentage point increase in BMA use. Although the data do not allow us to identify the mechanism for this increase, it is possible that vertical integration simplifies the process for patients obtaining dental clearance for BMA use (an essential step to mitigating jaw osteonecrosis).

Our finding of statistically nonsignificant changes in Medicare spending in our primary analyses is contrary to what has been reported in prior studies (5,12,13,47), though our analyses differ in multiple ways from those investigations. First, we measured total Medicare spending as opposed to spending on specific services such as physician-administered chemotherapy (13). Our goal was to capture the full impact of integration on cancer care. Even though Medicare pays more for services delivered in HOPD, recent analyses suggested that integration of specialist care reduced total spending, possibly by improving quality and reducing unnecessary services and care fragmentation (48,49). We included spending on Part D drugs, which could be important if integration affects incentives to switch from oral to physician-administered drugs or vice versa. Reassuringly, we did not find evidence of switching, as indicated by the minimal change in Part D spending. Our exploratory analyses also found no changes in the quantity of injectable vs oral chemotherapy claims. Second, previous studies focused on patient-year level spending or patients with heterogenous disease conditions. By focusing on a specific patient population over a narrow time frame (first 3 months of systemic therapy initiation), we reduce the potential bias from different patient types and disease trajectories. Moreover, Medicare site-based reimbursement differentials apply only to physician services and not to physician-administered or oral drugs. The magnitude of increases in Medicare spending on drug administration and routine office visits may be small relative to overall drug costs. Finally, we did not detect changes in service quantity (data not shown), another factor that could increase overall spending. The decrease in professional service payment and the corresponding increase in outpatient payment is likely because of shifts in the billing for injectable chemotherapy drugs to the hospital outpatient setting.

We did not detect statistically significant changes in time on systemic therapy or survival or resource utilization at EOL. Previous studies evaluating the impact of vertical integration and payment models (such as accountable care organizations and the oncology care model) also found minimal impacts on patient outcomes (14,50-52).

Our sensitivity analyses using higher cutoffs to define integration status show overall consistent results. Fluctuations could be because of limited sample sizes of patients with oncologists who became vertically integrated. Future studies that include more cancer sites and larger samples may generate more statistically reliable estimates, even with higher cutoffs to define integration.

Our study has some limitations. First, we focused on a cohort of Medicare FFS and Part D beneficiaries with mCRPC from 21 SEER regions. Hence, our results may not generalize to patients with Medicare Advantage plans or private insurance coverage, with other cancer types, and from other regions. Notably, Part D spending substantially increased as more oral cancer drugs became available to treat mCRPC. It is possible that the impact of vertical integration on physician-administered chemotherapy spending was small relative to overall health-care spending for mCRPC. Second, we could not attribute oncologists to a specific practice or hospital because provider identifiers were encrypted. Acquiring hospital characteristics such as teaching status, size, and cancer center designation may be important facilitators of quality and outcome improvements; this warrants future research. Encrypted identifiers also do not allow us to distinguish between oncologists becoming entirely employed by a hospital and oncologists whose practice got acquired by a hospital (26). Lastly, our analysis might have limited statistical power to detect changes in outcomes because of the modest sample size (859 patients treated by oncologists who experienced integration). Future investigations should extend to other cancer types and geographic areas and examine the impact on other patient outcomes.

In this population-based claims data analysis of older patients with mCRPC, oncologist-hospital integration was associated with increased use of BMA but did not have statistically significant impact on time on systemic therapy or survival. Although oncologists switched service billing from physician offices to hospital outpatient departments, we did not detect statistically significant changes in total Medicare spending within the first 3 months of therapy initiation. Amid the ongoing trend toward consolidation in US health-care markets, future studies should continue monitoring the structural changes in oncology markets and evaluating the effects on patients’ care quality, outcomes, and health-care spending.

Funding

This work was supported by a predoctoral fellowship grant from PhRMA Foundation.

Notes

Role of the funder: The PhRMA Foundation had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication.

Disclosures: Ms Hu received a dissertation grant from PhRMA Foundation. Dr Graetz received research support from Pfizer. No other disclosures were reported.

Author contributions: Conceptualization, funding acquisition, formal analysis, writing—original draft: XH; Project administration, methodology, resources: XH, JL, CJ, IG; Writing—review & editing: XH, JL, CJ, IG.

Acknowledgements: We are grateful to David Howard and Ian McCarthy for their expertise and guidance on the study design. This study used the linked Surveillance, Epidemiology, and End Results (SEER)–Medicare database. The interpretation and reporting of these data are the sole responsibility of the authors. The authors acknowledge the efforts of the National Cancer Institute; the Office of Research, Development and Information, CMS; Information Management Services (IMS), Inc; and the SEER Program tumor registries in the creation of the SEER-Medicare database. The collection of cancer incidence data used in this study was supported by the California Department of Public Health as part of the statewide cancer reporting program mandated by California Health and Safety Code Section 103885; the National Cancer Institute’s SEER Program under contract HHSN261201000140C awarded to the Cancer Prevention Institute of California, contract HHSN261201000035C awarded to the University of Southern California, and contract HHSN261201000034C awarded to the Public Health Institute; and the Centers for Disease Control and Prevention’s National Program of Cancer Registries, under agreement # U58DP003862-01 awarded to the California Department of Public Health. The ideas and opinions expressed herein are those of the author(s) and endorsement by the State of California Department of Public Health, the National Cancer Institute, and the Centers for Disease Control and Prevention or their Contractors and Subcontractors are not intended nor should be inferred. The authors acknowledge the efforts of the National Cancer Institute; the Office of Research, Development and Information, CMS; Information Management Services (IMS), Inc; and the SEER Program tumor registries in the creation of the SEER-Medicare database.

Prior presentations: Preliminary findings of this study was presented at the 2022 AcademyHealth Annual Research Meeting, the 2022 Georgia Clinical & Translational Science Alliance Health Services Research Day, and the 2022 Organization Theory in Health Care Association annual conference.

Data availability

Data underlying this article were provided by the National Cancer Institute by permission, which requires an Institutional Review Board approval and signed data use agreements with the National Cancer Institute. Data will be shared on request to the corresponding author with permission of the National Cancer Institute.