INTRODUCTION

Prostate cancer is the most common solid malignancy and the second leading cause of cancer-related deaths among men in the USA.1 Management options for prostate cancer can vary from watchful waiting to definitive treatment including radical prostatectomy or radiation therapy, but there is limited consensus regarding the superiority of alternative treatment options.2 There is wide variation in treatment patterns by geographic region and demographic and socioeconomic factors.3

Diffusion of new medical technologies has been recognized as a main driver of improvements in health outcomes as well as of increases in healthcare spending.4, 5 One example of a costly new technology is robotic surgery, which has been widely adopted and continues to replace the traditional open operative approach.6, 7 Though a recent Cochrane review of randomized controlled trials reported limited evidence of relative clinical benefits of robotic surgery for prostate cancer,8 surgical treatment for prostate cancer has largely transitioned from open to minimally invasive approaches in the last decade.9 Importantly, evidence also suggests that the introduction of robotic surgery has changed the choice of treatment for prostate cancer.10, 11

The cost implications of a new technology depend both on the volume of its use and on relative reimbursement rates. New York State (NYS) Medicaid reimbursement for prostate cancer care varies according to the type of treatment received. NYS Medicaid uses All Patient Refined Diagnosis Related Groups (APR-DRGs)–based payment for inpatient care both for those enrolled in its fee-for-service program and for health plans participating in Medicaid Managed Care (though managed care plans may vary the rate paid per APR-DRG).12 Surgeons performing prostatectomies are paid under Current Procedural Terminology (CPT) code 55810 (open retropubic) or 55840 (laparoscopic or robotic). Importantly, NYS Medicaid does not differentiate payment amounts based on whether prostate surgery is performed robotically, laparoscopically, or using an open procedure.13

In contrast, there is wide variation in Medicaid payment rates for radiation therapy within NYS. The majority of radiation therapy is furnished in freestanding radiation centers or hospital outpatient departments, and both bill through an outpatient patient classification system (Ambulatory Patient Groups (APGs)). When providers report all services provided during a visit, individual procedures are assigned to payment groups for reimbursement based on clinical similarity and resource cost. However, radiation treatment encounters may involve multiple services,14 which can be assigned to multiple APGs, and radiation therapy may be billed as a stand-alone procedure.

We hypothesized that increased availability of robotic procedures, measured as distance from zip code centroids of patient’s residence to the nearest hospital with robots and annual number of robotic prostatectomies conducted in the Health Referral Regions (HRRs), changed the treatment cancer patients received and their subsequent treatment costs. The share of acute general hospitals in NYS with robotic surgical systems increased from 22% in 2008 to 73% in 2017. As a result, the average distance from zip code centroids of patient’s residence to the nearest hospital with robots decreased from 10 mi in 2008 to 7 mi in 2017.

METHODS

Data and Study Sample

This retrospective cohort study used NYS Medicaid claims data for 2008–2018 which contain diagnostic and procedure codes for all inpatient stays, outpatient/emergency visits, office visits, and prescribed medications among Medicaid patients during the study period. Our data also contain all cost information including Medicaid Managed Care plans’ negotiated payments. We restricted the sample to adult men aged 40–64 years who underwent a prostate biopsy between 2008 and 2017 and were diagnosed with prostate cancer (as the primary diagnosis) within a year after biopsy. Patients younger than 40 years old were excluded due to low incidence rates of prostate cancer in this population. We also excluded those 65 years or older who were also enrolled in Medicare because these claims maybe incomplete in our Medicaid-based dataset. In sensitivity analyses, we exclude all dually eligible beneficiaries under 65 as well. Exclusion criteria also included incomplete demographic information and lack of 12 months of continuous enrollment in Medicaid before the biopsy and after the diagnosis date. Our final sample included 9564 men diagnosed with prostate cancer after biopsy. The American Community Survey provided 5-year estimates of median household income. The institutional review board of New York University reviewed and approved this analysis and waived the need for informed consent. This study followed the guideline for cohort studies of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE).

Variables

We used CPT and International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) codes to identify a sample 15 who (a) met the above inclusion criteria, (b) underwent a prostate biopsy, and (c) were subsequently diagnosed with prostate cancer. We also used ICD-9, ICD-10, and CPT codes to identify the types of treatment patients received. We considered patients to be undergoing robot-assisted radical prostatectomy if their claims included prostatectomy codes with a robotic modifier.16 Treatment costs were measured as the total amounts paid by Medicaid for all services related to their prostate cancer diagnosis for the first 12 months following diagnosis.17 In addition, Medicaid payments were grouped into costs for inpatient surgery, physician visits, and radiation-related imaging and procedures. Costs were converted to 2019 dollars using the Consumer Price Index.

Our key independent variable was regional access to robots, measured as (1) great-circle distance in miles between zip code centroids of patient’s residence and the nearest hospital that offered robot-assisted surgeries and (2) annual number of robotic prostatectomies conducted among Medicaid patients in the Health Referral Regions where patients resided. To adjust for differences in sample characteristics, our regression models included the following demographic and health-related variables (measured during the year before prostate biopsy) abstracted from our data set: age, mutually exclusive race/ethnicity categories (non-Hispanic White, non-Hispanic Black, Hispanic, other), Medicaid managed care enrollment, Medicaid-Medicare dual eligibility, Charlson comorbidity index, coagulopathy, general healthcare and preventive service utilization, and urologic symptoms (sexual dysfunction, urinary tract stricture, incontinence) diagnosed before biopsy. Census tract median household income was also controlled for based on the finding of a previous study that neighborhood socioeconomic factors are positively associated with the use of costly medical technology.18

Statistical Analysis

Initially, we described the average 1-year Medicaid-reimbursed prostate cancer–related treatment costs and total Medicaid cost for patients treated with open prostatectomy, robotic prostatectomy, radiation therapy, or no definitive treatment. We also measured the average time intervals from diagnosis to the receipt of initial definitive treatment by treatment type.

We then used a multivariate linear regression model to estimate (1) the association between regional access to robots and the choice of initial treatment each patient received and (2) the association between access to robots and Medicaid reimbursement for prostate cancer treatment. In addition to demographics and health-related variables described above, our regression models controlled for year fixed effects (adjusting for unobservable demographic and socioeconomic shocks, including changing technologies, common to all patients), census tract median household income, and zip code fixed effects. By including zip code fixed effects, we are identifying our results using within-region variation in access to robots over time generated by the diffusion of robotic surgical systems. In sensitivity checks, we relaxed this constraint by replicating our main analyses with HRR fixed effects (instead of zip code dummies) and without any geographic fixed effects. Preexisting urologic symptoms and healthcare utilization (which are not typically available in studies utilizing cancer registry data sets) and comorbidity measure were included to better capture the variation in treatment decision-making based on risks of conversion to an open approach if undergoing robotic surgery.19, 20

Given that all treatment decisions arise from a conditionally independent series of binary choices, we specified sequential logit models where the first decision node was whether a patient received any definitive treatment, and the second decision was which treatment was given among those who received definitive treatment. To confirm the robustness of our analysis of treatment costs, we also specified generalized linear models with a gamma variance and a log link and presented results expressed as average marginal effects. In addition, given the possibility that costs can decrease as physicians become more familiar with new technologies, we added results from linear regressions that included provider fixed effects and provider-specific linear trends, allowing us to see whether the association of access to robots with substitution across treatment type was driven by practitioner-specific factors and their temporal trends. We clustered standard errors at the zip code level.

Statistical analysis was performed between February and April 2020 using Stata statistical software version 16.0 (StataCorp). A two-sided P < 0.05 was deemed statistically significant.

RESULTS

Baseline Characteristics

Our sample included 9564 Medicaid beneficiaries (mean [SD] age, 57.8 [4.9] years) (Table 1) and consisted of 30% White, 26% Black, and 22% Hispanic. Eighty-two percent of the sample were enrolled in Medicaid Managed Care. Thirty-seven percent were dually eligible for Medicaid and Medicare. Forty-five percent had a non-zero Charlson comorbidity index score, and in the year prior to biopsy, 7% were diagnosed with sexual dysfunction, 2% were diagnosed with urinary tract stricture, and 4% were diagnosed with incontinence. Mean [SD] distance from zip code centroids of patient’s residence to the nearest hospital with robots was 7.6 [30.4] miles and mean [SD] annual robotic radical prostatectomies per HRR was 38.7 [32.8].

Table 1 Summary Statistics
Full size table

Medicaid Reimbursement by Prostate Cancer Treatment Type

Among the sample who were diagnosed with prostate cancer following biopsy, 51% underwent any definitive treatment (15% underwent robot-assisted radical prostatectomy, 9% underwent open radical prostatectomy, and 27% underwent radiation therapy) (Table 2). The average Medicaid reimbursement for care related to prostate cancer diagnosis was about 30% higher for radiation treatment than for other definitive treatment modalities; $18,868 (SD = 12082) for robot-assisted radical prostatectomy, $20,449 (SD = 14558) for open radical prostatectomy, $27,582 (SD = 25473) for radiation therapy, and $12,285 (SD = 10990) for those without definitive treatment (this last group included those who selected watchful waiting). Similarly, the average Medicaid reimbursement for all care incurred during the year following cancer diagnosis was higher for patients undergoing radiation therapy than for patients undergoing robot-assisted or open prostatectomy, consistent with the difference in total Medicaid reimbursement being largely attributable to the differential in reimbursements for prostate cancer–related care. Results from one-way ANOVA tests with Bonferroni correction indicated that both the average prostate cancer–related reimbursement and the average total Medicaid reimbursement were statistically different (P < 0.001) between the treatment types. By contrast, the time interval between diagnosis and treatment was not statistically different (P = 0.846) between the definitive treatment types, indicating that the differences in costs were not associated with the time difference between diagnosis and treatment.

Table 2 Average 1-Year Medicaid Reimbursements and Time Interval from Diagnosis to Initial Definitive Treatment for 40–64-Year-Old Men with Prostate Cancer by Treatment Type 2008–2017 During the Year Following Diagnosis
Full size table

Choice of Prostate Cancer Treatment Type

Doubling distance from zip code centroids of patient’s residence to the nearest hospital with robots was associated with a decline in the probability of undergoing robotic surgery (3.7%, 95% CI − 7.2 to − 0.3, P = 0.03) and an increase in the probability of undergoing radiation therapy (5.2%, 95% CI 1.2 to 9.8, P = 0.01) (Table 3). In addition, patients in a region where 10 additional robotic surgeries were performed had a lower probability of undergoing radiation therapy (− 0.6%, 95% CI − 0.1 to − 0.03, P = 0.04) and a higher probability of undergoing robotic surgery (1%, 95% CI 0.9 to 2, P < 0.001). Analyses excluding region fixed effects, replacing zip code dummies with HRR fixed effects, excluding those dually eligible for Medicaid and Medicare, and including provider fixed effects and provider-specific linear trends all generated similar results (eTable 1). Figure 1 (above two figures) presents fitted values taken from the regressions shown in Table 3. As the distance increased, the probability of undergoing robotic surgery increased while the propensity to undergo radiation therapy fell. The predicted probability of undergoing robotic prostatectomy was .2 if a hospital with robots were within 10 mi, while the predicted probability fell below .1 if no hospitals with robots were available within 80 mi. A sequential logic regression (eTable 2) showed that availability of robotic procedures was not associated with the likelihood that a patient received any definitive treatment (1st decision step) but was significantly associated with increased odds of receiving robotic prostatectomy conditional on definitive treatment (2nd step). Increased availability of robotic procedures was not associated with substitution between robotic surgeries and open surgeries but significantly associated with higher odds that a prostate cancer patient received robotic surgery rather than radiation treatment.

Table 3 Adjusted Association of Local Availability of Robotic Surgical Technology with Treatment Type—Patient Level
Full size table
Figure 1
figure 1

Predicted treatment type and treatment costs among Medicaid prostate cancer patients. Study sample included Medicaid prostate cancer patients aged 40–64 years in New York State who underwent prostate biopsy in 2008–2017 and were diagnosed with prostate cancer (N = 9564). Fitted values were taken from regressions in Table 3(predicted probability) and Table 4(predicted treatment cost). Cost values were converted to 2019 dollars using the Consumer Price Index.

Full size image

Prostate Cancer Treatment Costs

Doubling distance from zip code centroids of patient’s residence to the nearest hospital with robots was associated with a decline in reimbursement for inpatient surgery of $911 (95% CI − 1624 to – 198, P = 0.012), but this reduction was more than offset by an increase in radiation therapy–related imaging and procedure costs ($2167, 95% CI 933 to 3400, P = 0.001) (Table 4). Patients with prostate cancer in an HRR where 10 additional robotic surgeries were performed in a year averaged treatment costs that were $434 lower than otherwise (95% CI − 668 to – 199, P < 0.001). Sensitivity analyses replacing zip code fixed effects with HRR fixed effects, excluding those dually eligible for Medicaid and Medicare, including provider fixed effects and provider-specific linear trends, and specifying a generalized linear regression model all found similar results (eTable 3). Figure 1 (below two figures) presents fitted values from the regressions in Table 4 and shows that predicted costs were about $25,000 if distance from zip code centroids of patient’s residence to the nearest hospital with robots was 30 mi, while predicted costs increased to around $26,000 if the distance increased to 60 mi. In eTable 4, we present evidence that our regional access measures were significantly associated with a reduction in the number of prostate cancer–related hospitalization coupled with increased outpatient utilization, consistent with our findings of substitution of surgery for radiation treatment.

Table 4 Adjusted Association of Local Availability of Robotic Surgical Technology with Prostate Cancer Treatment Costs Within 1 Year After Diagnosis—Patient Level
Full size table

In addition, we analyzed the association between availability of robotic procedures and total Medicaid costs of care (not confined to prostate cancer–related care) and found that prostate cancer–related costs significantly decreased as availability of robotic procedures increased, but these differences were not statistically significant (eTable 5). Using nearest neighbor propensity score matching weights, we also compared Medicaid spending for robot-assisted prostatectomy to radiation treatment in eTable 6 by regressing spending on an indicator for robotic prostatectomy (as compared to radiation treatment). We found that prostate cancer patients who received robotic surgeries incurred $8643 lower treatment costs (95% CI − 10334 to – 6952, P < 0.001) than those who received radiation treatment.

DISCUSSION

The robotic approach has been assuming a greater role in surgical practice, but randomized controlled trials have offered limited evidence that robotic surgery is a more effective therapy for prostate cancer than alternatives.8 We found that the adoption of robotic surgical technology was associated with substitution across treatment types among nonelderly Medicaid prostate cancer patients. We also documented that increased availability of robotic procedures reduced Medicaid costs by leading to substitution of surgery for more costly radiation therapy.

Previous studies on robotic surgeries for cancer primarily focused on urban and affluent populations using Surveillance Epidemiology and End Results (SEER)-Medicare data21 and reported that the likelihood of undergoing procedures robotically was significantly lower among Medicaid patients than peers covered by private insurance or Medicare.22 Consistent with these findings, an analysis of case logs submitted by urologists estimated that around 70% of prostatectomies were performed robotically in 2010,9 while our estimate among Medicaid patients in NYS was 55% in that year.

In the absence of a consensus on the optimal treatment strategy for prostate cancer patients, this study highlighted changes in prostate cancer treatment patterns and spending that followed adoption of robotic surgical technology. Our result of substitution among therapies is consistent with a previous study using national cancer registry data reporting substitution of surgery for radiation at the state level.11 Notably, while Shen and Shih’s study11 reported that increased regional access to robots modestly decreased the rate of watchful waiting, our results indicated that increased availability of robotic procedures did not lead to substitution of robotic surgery for watchful waiting. Rather, the substitution between robotic surgeries and radiation therapy explains our finding of reductions in treatment costs.

Our results of lower treatment costs related to increased availability of robotic procedures shed a new light on the adoption of new medical technology and its consequences. Previous studies report higher costs for robotic prostatectomy relative to laparoscopic or open approaches.23, 24 NYS Medicaid, however, does not differentiate reimbursement for robotic versus open prostatectomies. And because radiation treatment encounters may involve multiple APG payments, in our sample, the weighted average of Medicaid payments for prostate cancer (taken from a regression on age, race/ethnicity, comorbidity indicators, and year fixed effects) during the year following diagnosis was 42% lower for surgical prostatectomies ($20,048) than for radiation therapy ($28,528). A study using SEER-Medicare data similarly reported that median reimbursement within a year after diagnosis for treating older prostate cancer patients was 59% higher for radiation therapy than for radical prostatectomy in 2008.17

A comprehensive analysis of the effect on costs of introducing a new technology, then, depends on the costs of the new technology and the costs of the direct replacement technology, as well as estimates of the extent to which the new technology replaces the old technology and related technologies.

Limitations

Our study has several limitations. First, our claims data do not contain clinical information such as PSA level, cancer stage and grade, or estimated life expectancy. These factors determine both treatment decision and outcomes, and their omission limits our ability to infer causal impacts of changes in access to robots on clinical outcomes. Second, our results are specific to Medicaid beneficiaries in NYS. Payment rates vary among payers and states and our results with respect to total cost would be different if payment rates were different. Moreover, the extent of substitution based on availability of robotic procedures may also vary among payers and states. Third, because we lack information from pathological reports, we were not able to distinguish those with rule-out diagnosis codes from patients without any definitive treatment.

Conclusions

The diffusion of the use of robotic prostatectomy across hospitals in NYS led to changes in the type of treatment provided to Medicaid beneficiaries 40–64 years old with prostate cancer. As the availability of robotic surgery increased, patients were more likely to be treated with surgery rather than radiation. This substitution effect reduced the cost of treating prostate cancer in NYS Medicaid.