The stability of physician-specific episode costs for urological cancer surgery: Implications for Urologists under the Merit-Based Incentive Program (MIPS)

Introduction

The Centers for Medicare and Medicaid Services (CMS) initiated the Merit Based Incentive Program (MIPS) to increase health care quality and decrease expenditures. MIPS was initiated on January 1, 2017. Nearly 1.5 million clinicians will take part in MIPS, receiving performance-based awards or penalties of up to 4% of Medicare Part B payments in 2019, and up to 9% in 2022. Clinicians who bill Medicare Part B > $30,000 per year and provide care for > 100 Medicare Part B beneficiaries are included in MIPS. Exclusion criteria include clinicians who are in non-MIPS eligible specialties, those who are in their first year of Medicare participation, and clinicians who participate in MACRA as a Qualifying Alternative Payment Model Participant or Partial Qualifying Participant. Participating clinicians include physicians, physician assistants, nurse practitioners, clinical nurse specialists, and certified registered nurse anesthetists (1, 2). The resource use measure is one of four categories (quality, Meaningful Use of Certified Electronic Health Records, clinical practice improvement activities, and resource use) that comprise the MIPS composite score; this metric specifically evaluates clinician spending for attributed Medicare beneficiaries. The goal of the resource use measure is to increase clinician awareness and accountability for health care expenditures (3). By 2021, this measure will account for 30% of the composite score. From a measurement perspective, CMS plans to evaluate the total health care expenditures for all patients or episodes of care attributed to an individual clinician.

Although the implementation details are still being finalized, the potential impact of the resource use measure is predicated on two key principles. The first principle, established in the literature for some diagnoses and procedures, is that wide variation exists in physician-specific episode payments (4-9). The second principle, not yet tested empirically, is that the care provided by some physicians is consistently more expensive. In other words, there is a group of providers that are at the high end of the overall spending measure both year-over-year and across many procedures and/or diagnoses. In fact, the impact of any resource use measure will arguably be greatest if these clinicians consistently have a large and consistent incentive to change behavior in a way that reduces resource utilization. Likewise, reliable and actionable payment data indicating that physicians are consistently more expensive than their peers may help identify areas of poor quality and/or inefficient care delivery (10, 11).

In contrast, the absence of stability in physician spending, defined in this manuscript, as similarities in physician payments over time and across procedures, has two important implications for the potential impact of the resource use measure. First, if physician spending signatures are not stable, the financial incentives associated with a resource use measure will be much less apparent. For instance, a clinician who is a high spender for one procedure, but not for another, will have a spending pattern that is similar to the average physician across procedures, and thus little imperative to change behavior. Second, although there are other measures of quality in the MIPS program, absence of stability in the resource use measure will greatly limit its ability to provide complementary information (to the MIPS quality metrics) that identifies, and motivates change, among physicians who are expensive due to concerns such as complications or decisions around the use of discretionary services.

In this context, evaluating surgical episode spending is a useful potential litmus test for measuring stability in physician-specific episode payments. Surgical admissions comprise one-third of all inpatient admissions and accounted for the greatest increase in acute inpatient spending in 2014 (12). Moreover, because a designated clinician performs a procedure, attribution of a surgical episodes, and therefore measuring physician resource utilization, may be more straightforward than attribution of a non-procedure-based inpatient hospitalization.

Accordingly, we used linked SEER-Medicare data to evaluate physician-specific episode payments for prostatectomy, nephrectomy, and cystectomy. We further examined whether procedure-specific spending was stable over time and if spending for one genitourinary (GU) cancer procedure was associated with high spending for the other procedures. We hypothesize that, while variation in spending exists for prostatectomy, nephrectomy, and cystectomy, there is little stability in patterns of clinician spending over time and/or across procedures. By virtue of this approach, our findings will inform physicians, payers, and policymakers on variation in physician-specific episode payments and the ability of the resource use measure to influence physician behavior, consistently differentiate physician spending, and provide reliable and actionable information to decrease expenditures while maintaining or improving quality.

Methods

Data

We used data from Surveillance, Epidemiology and End Results (SEER) registries linked with Medicare claims data from 2008 through 2012. SEER-Medicare is a patient-level dataset that links Medicare claims with clinical, demographic, and outcome data from SEER cancer registries covering 17 major regions across the United States. We used the following Medicare claims files for our analyses: 100% Medicare Provider and Analysis Review File, carrier, outpatient, home health and hospice.

Study Population

Our study cohort included patients aged 66-99 years who underwent a prostatectomy, nephrectomy, or cystectomy for cancer. We identified these patients using International Classification of Diseases, Ninth Edition (ICD-9) diagnosis and procedure codes and Current Procedural Terminology (CPT) codes (Appendix Table 1). We excluded cases that were missing an ICD-9 diagnosis or procedure code for the cancer of interest, a CPT code for the surgery of interest, and/or a cancer diagnosis listed in the SEER file. We further excluded patients who underwent multiple surgeries for different cancers, those that did not have continuous Medicare Parts A and B enrollment and/or those who participated in Medicare Health Maintenance Organizations (HMOs) during the study interval.

Surgeon attribution

In order to evaluate surgeon episode payments, we attributed each episode to a single surgeon. To do this, we first identified the National Provider Identifier (NPI) associated with the surgery claim. If more than one surgeon was associated with the claim, we attributed the episode to the surgeon with the HCFA specialty code for urology (34). If a urologist was not listed, the claim was attributed to 02 (general surgery) or 91(surgical oncology). If more than one provider with the appropriate HCFA specialty code was associated with the index surgery, the episode was attributed to the provider with the greatest payment amount.

Estimating 90-day episode payments

We first identified the index hospitalization for each patient meeting our inclusion criteria. We then aggregated all payments for claims that occurred two days prior to the initial hospital admission through 90 days after discharge. The 90-day episode is consistent with the global surgery period and other episode-based payment programs (13). To account for differences in Medicare payments based on disproportionate share payments, teaching status, and geography, we price standardized all payments using previously described methods, which were adapted from the Dartmouth Institute for Health Policy and Clinical Practice and the Medicare Payment Advisory Commission (MedPAC) (14). In addition, we used established methods to decrease the effect of outliers on spending, by excluding patients with episode payments below the first and above the 99^th percentile. This approach is consistent with CMS’s methods in performance programs, including the physician Quality and Resource Use Reports and the calculation for the Medicare Spending Per Beneficiary Measure (MSPB) (15-17).

Examining provider-level variation in 90-day episode payments

To evaluate provider-level variation in 90-day episode payments, we calculated the average episode payment per surgeon for each procedure. Similar to the proposed MIPS methodology which ranks physicians according to spending, we sorted each physician’s average episode payment into quartiles. We next evaluated the mean difference between the highest and lowest quartile physicians for each type of cancer surgery.

To assess the stability in spending patterns over time, we evaluated whether the top spenders in one year were also top spenders in the subsequent year. Specifically, we examined the spending quartile in 2011 for surgeons in the most expensive quartile in 2010. We further evaluated the correlation in provider spending rank for each year by procedure for 2008 to 2012 using the spearman rank correlation coefficient.

Finally, to evaluate the relationship between surgeon-specific expenditures across different procedures, we ranked each surgeon by his or her mean 90-day episode payment per procedure and evaluated the spearman rank correlation coefficient for surgeon rank across procedures (i.e., prostatectomy, nephrectomy, or cystectomy). All analyses were performed using SAS v 9.4 (Cary, NC) and STATA (College Station, TX). The University of Michigan’s Institutional Review Board deemed this study exempt from review.

Results

We identified 14,585 patients who underwent a major urological cancer surgery by one of 1,895 surgeons from 2008 through 2012. The distribution of cancer sites was as follows: 60% prostate (8,673 patients and 1,250 surgeons), 28% kidney (4,111 patients and 1,277 surgeons), and 12% bladder (1,801 patients and 673 surgeons).

We observed wide variation in episode payments for each type of cancer surgery. Differences in mean payments between the highest and lowest surgeon quartiles were $5,881, $17,714, and $40,288 for prostatectomy, nephrectomy, and cystectomy, respectively, reflecting relative differences of 40%, 62%, and 67% of the mean episode payments of the high quartile surgeons (Table 1).

Table 1:

Differences in mean 90-day surgical episode payments per physician, by spending rank quartile for prostate, kidney, and bladder cancer

Procedure	Low Payment Quartile	High Payment Quartile	Difference High and Low Payment	Difference as a % of High Payment Quartile
Prostatectomy	$8,829	$14,710	$5,881	40
Nephrectomy	$11,013	$28,727	$17,714	62
Cystectomy	$19,753	$60,041	$40,288	67

Surgeon-spending was only weakly correlated from one year to the next. Among the top quartile spenders in 2010, only 39%, 16% and 13% were also in the top spending quartile in 2011 for prostate, kidney, and bladder cancer, respectively (Figure 1). There was minimal correlation in year-over-year spending for prostate cancer, with Spearman rank correlation coefficients ranging from 0.108 (2008 to 2011, p=0.033) to 0.270 (2009 to 2011, p<0.001). Likewise, correlations between year-over-year payments were weak for surgeons performing cystectomy and nephrectomy, and none of these achieved statistical significance (Table 2A).

Figure 1:

2011 spending quartile for physicians that were in the top (i.e., most expensive) quartile of spending in 2010.

Table 2:

Correlation in mean 90-day surgical episode payments A) Over time and B) By Urologic procedure

A) Over time
Prostatectomy
		2008	2009	2010	2011
	2008	1
	2009	0.224 (<0.001)
	2010	0.234 (<0.001)	0.270 (<0.001)
	2011	0.108 (0.033)	0.165 (<0.001)	0.184 (<0.001)	1
Nephrectomy
		2008	2009	2010	2011
	2008	1
	2009	0.016 (0.850)
	2010	0.030 (0.729)	−0.038 (0.648)
	2011	0.280 (0.001)	0.081 (0.348)	−0.043 (0.615)	1
Cystectomy
		2008	2009	2010	2011
	2008	1
	2009	−0.042 (0.660)
	2010	0.122 (0.230)	−0.026 (0.802)
	2011	0.074 (0.482)	0.127 (0.224)	0.0620 (0.553)	1

B) By procedure
	Prostatectomy	Nephrectomy	Cystectomy
Prostatectomy	1
Nephrectomy	0.028 (0.447)
Cystectomy	0.068 (0.157)	0.098 (0.037)	1

Finally, high spending for surgical episodes with one urologic cancer was poorly correlated with high spending for another urologic cancer. Spearman rank correlation coefficients for surgeon-specific episode spending were 0.028 (p=0.447) for prostate vs. kidney, 0.068 (p=0.157) for prostate vs. bladder, and 0.098 (p=0.037) for kidney vs. bladder (Table 2B).

Discussion

We evaluated 90-day episode payments for Medicare beneficiaries undergoing major surgery for prostate, kidney, or bladder cancer to determine both variations in spending across providers, and the stability in spending rank over time and across procedures. We identified wide variation in physician-specific episode payments. In fact, for cystectomy, high spenders had average episode costs nearly 300% of those for less expensive surgeons. At the same time, however, there was little correlation between surgeon-specific episode payments over time and across procedures. Collectively, these findings suggest that, while providing physicians with payment data makes sense from the perspective of highlighting variation in resource use, metrics based on episode payments alone may be limited in their ability to change physician behavior and reliably distinguish those providing less efficient or lower quality care.

Our observation of wide variation in episode payments is consistent with, and expands upon, prior analyses demonstrating significant differences in physician-specific spending (4-9). An analysis of variation in Medicare Part B spending for patients hospitalized for non-elective conditions determined that payments varied more across physicians than hospitals, and that higher payments were not associated with lower 30-day readmissions or mortality rates (5). In an analysis of variation in 90-day payments for radical cystectomy using a nationally representative all-payer database of inpatient claims, the authors likewise found wide physician-specific variation in payments and that post-operative complications contributed the greatest to variation, followed by patient factors, including Charlson co-morbidity index, insurance type and race (6).

Moreover, our finding of little correlation between physician-specific spending over time and across procedures is also consistent with other reports calling into question the reliability of physician-specific measures of quality and cost (18-20). In an evaluation of physician spending for homogenous episodes of care for different clinical conditions, the authors determined that the majority of physicians had reliability scores well below the commonly established threshold of 0.70, with the surgical subspecialties having particularly low reliability scores. In addition, when the authors categorized physicians into 2 tiers, either low cost (i.e., the lowest 25%) or not low cost, using the previously calculated cost-based reliabilities, and compared the calculated assignment to physicians’ actual spending categories, 50% of orthopedic surgeons and 67% of vascular surgeons were misclassified as low cost providers, when they were in fact not low cost providers (19). Although sample size plays an important role in reliability, even substantial increases in sample size did not produce reliability results above the commonly accepted threshold for many specialties. Another analysis of surgeon-specific complications after colectomy, demonstrated that 86% of the variability in complication rates was explained by statistical noise, and that the vast number of surgeons do not perform enough colectomies to generate reliable complication rates (18).

Our study has several limitations. First, we evaluate stability in physician spending for 90-day episode payments so our findings cannot be directly generalized to the MIPS resource use measure, which has yet to finalized. We evaluated a 90-day episode window for this analysis because it is consistent with other episode-based payment programs (13). Second, our evaluation did not adjust for case-mix (e.g., age, comorbidity) or diagnosis related group payment differences, which may impact episode spending. Whether and how to risk-adjust performance measures remains a controversial topic, and the risk-adjustment methodology that CMS will implement for cost based measures in MIPS has not been finalized (3). Medicare has risk-adjusted in certain value-based payment programs (e.g., Hospital Readmission Reduction program), while not in others (e.g., proposed cardiac bundle payment program) (21). Accurate risk adjustment depends on documentation and coding accuracy; in addition, resource availability and has been found to be lower across public hospitals compared to private for-profit and non-profit, thus raising concern about using case-mix index as a marker of patient disease severity (22). Nonetheless, we are planning separate analyses to address this relevant issue. Third, we focus on episode spending for three inpatient surgical procedures and these findings may not be generalizable to other surgical or medical episodes. However, surgical episodes account for one third of all inpatient admissions (12) and the included procedures are all predominately performed by one specialty type (urology), making an evaluation of surgeons across procedures feasible. Fourth, we do not evaluate quality, which may vary across surgeons and impact the differences in spending reported herein. Put another way, our analysis does not answer the question as to whether higher spending surgeons achieve better outcomes for their patients.

These limitations notwithstanding, our results have significant implications for physicians, policymakers and payers. For physicians, our analyses highlight wide payment variation in physician-specific payments; the availability and dissemination of such data to clinicians is likely a necessary step toward improving efficiency in care delivery. At the same time, however, for policymakers and payers our analyses highlight some cautionary notes to be considered when developing payment programs based on physician-specific data. In particular, the absence of relative stability in surgeon-specific, episode-based resource use measures for individual surgeons over time and across procedures has two important implications. First, if payers are planning to leverage this measure from a quality perspective, it may be difficult to use it to identify consistent patterns of in hospital-based or post-acute spending that reflect complications, use of discretionary services, or other considerations that could be actionable from a quality or payment perspective. Furthermore, an individual surgeon may have little control over some components of episode spending (e.g., intensity of services provided at a post-acute care facility), further limiting the impact of the resource use measure on decreasing payments.

The second implication relates to the magnitude of the incentive required for a physician to spend the significant time, energy, and resources needed to make meaningful system-level changes to address increased resource utilization. If there really were consistently expensive surgeons across time and procedures, then the MIPS resource use measure would have the potential to provide a consistent signal and incentive to catalyzes the substantial efforts required to define and decrease resource utilization. However, if each physician ends up around the average, or if the measure varies widely year-over-year, it may be difficult to maintain the physician focus and energy required to implement some of these changes. Collectively, our findings may be viewed by many as significant limitations with the current MIPS program; the impact of these concerns will need to be examined as the program evolves. Moving forward, it will be important to evaluate 90-day episode payments for other surgical and medical conditions, assess how risk-adjustment impacts episode spending, and if there are any other measures of resource utilization that may be more reliable, beneficial, and actionable. It is only with accurate, reliable, and clinically meaningful measures that we can improve health care quality and decrease costs.

Appendix

Supplement Appendix Table 1:

ICD-9 diagnosis, procedure, and CPT codes for cancer cohort identification

Cancer	ICD-9 Diagnosis Codes	ICD-9 Procedure Codes	CPT codes (2014)	Procedure Description	SEER site codes
Prostate	185	60.3	55821 (subtotal)	Suprapubic prostatectomy	619
		60.4	55831 (subtotal) 55840, 55842, 55845 (radical open) 55866 (lap)	Retropubic prostatectomy
		60.5	55810, 55812, 55815 (perineal) 55840, 55842, 55845 (retropubic) 55866 (lap)	Radical prostatectomy
		60.61		Excision of lesion, prostate
		60.62	55801 (subtotal) 55810, 55812, 55815	Perineal prostatectomy
		60.69		Other prostatectomy
Lung	162 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629	32.2, 32.9, 32.29, 32.20	32500 32503, 32504, 32505	Excision of lesion, lung	C340-C343, C348, C349, C339
		32.30, 32.39	32484, 32486	Segmental resection
		32.4, 32.41, 32.49	32480, 32482	Lobectomy
		32.5, 32.50 32.59	32440, 32442, 32445, 32488	Pneumonectomy
Kidney	189 1890 18900 18901 1898 1891 23691	55.4	50240, 50543	Partial Nephrectomy	649
		55.51, 55.52, 55.54	50220, 50225, 50230, 50234, 50236, 50545, 50546, 50548	Complete Nephrectomy
		5534	50542	Lap renal ablation
		5532	50250	Open renal ablation
		555		Complete nephrectomy
Colorectal	1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541	45.7 (45.71-45.76, 45.79)	44140, 44141, 44143, 44144, 44145, 44146, 44147, 44160	Partial excision of large intestine	C18.0-18.9, C19.9, C20.9
		45.8 (4849 4581 4582 4583)	44150, 44151, 44155, 44156, 44157, 44158	Total intra-abdominal colectomy
		48.4 (4840 4841 4842 4843)	45112, 45119	Rectal pull through
		48.5 (4850 4851 4852 4859)	45110, 45111	APR
		48.6 (48.61-48.65, 48.69)	45113, 45114, 45116, 45123	Rectal resection
		1732 1733 1734 1735 1736 1739	44210, 44211, 44212, 45395, 44204, 44205, 44206, 44207, 44208	Lap colectomy