© The Author 2006. Published by Oxford University Press.
EDITORIAL |
Transcending the Volume–Outcome Relationship in Cancer Care
Correspondence to: Joseph Lipscomb, PhD, Department of Health Policy and Management, Rollins School of Public Health, Emory University, Atlanta, GA 30322 (e-mail: jlipsco{at}sph.emory.edu).
Setting the Context
Research on the volume–outcome relationship in health care has generated both a high "volume" (more than 200 Medline-listed publications to date) and a consistently recurring "outcome": the finding that, across a range of interventional procedures, hospitals and surgeons with higher case loads tend to have lower mortality rates (1).
The volume–outcome relationship has been most striking for certain low-frequency, high-risk operations, such as pancreatectomy and esophagectomy. In particular, U.S. hospitals that performed 11 or more of the latter procedure on Medicare patients between 1984 and 1993 had a fivefold lower 30-day mortality rate (3.4% versus 17.3%) than hospitals that performed between one and five such procedures during the same period (2). In a review commissioned by the National Cancer Policy Board (NCPB) of the Institute of Medicine (IOM), Halm et al. (3) found that in 17 of 20 studies that involved cancers ranging from esophageal and pancreatic to breast, lung, and colorectal, higher surgical case volume was associated with better outcomes (typically perioperative and 30-day mortality). In an earlier study also commissioned by the IOM, Hillner et al. (4) concluded there was compelling evidence of a positive volume–outcome relationship across a range of cancers and surgical procedures, notwithstanding substantial methodologic and data limitations in the literature reviewed. Using the Surveillance, Epidemiology, and End Results (SEER) cancer registry database linked to Medicare claims data, Bach et al. (5) and Schrag et al. (6) uncovered statistically significant positive associations between hospital volume and patient survival for lung and colorectal cancer surgery, respectively. A 2002 nationwide study of Medicare patients by Birkmeyer et al. (7) found that surgical mortality was lower in higher volume hospitals for all 14 procedures examined, including eight cancer operations. The volume-related mortality rate differential was most pronounced for technically difficult procedures like pancreatectomy and was much smaller for more commonly performed or less technically involved operations like colectomy.
By contrast, less is known about the relationship between surgeon volume and patient outcome and the interplay between this relationship and hospital volume. Perhaps the most striking findings to date were reported in 2003 by Birkmeyer et al. (8). They analyzed the relationship between surgeon volume and operative mortality for eight major procedures, including four for cancer, performed during 1998 and 1999 on a national sample of nearly 475 000 Medicare patients. From their statistical analyses, the authors concluded that surgeon volume was not only inversely related to operative mortality for all procedures but also that it "accounted for a large proportion of the apparent effect of hospital volume" on mortality: from 46% for esophagectomy to 24% for lung resection.
In the face of this accumulating evidence, what should be done? One response would be to launch efforts to "regionalize" cancer surgery—either selectively or broadly—to concentrate care among high-volume providers. This approach could have broad implications not only for the technical quality of cancer care but also for access to care and cost. The policy tools for such a transformation would include reformulated practice guidelines and new pay-for-performance standards. Urging such action on a selective basis, the National Cancer Policy Board has recommended that provider volume be incorporated as a quality indicator for a variety of purposes, "when a large and significant volume–outcome relationship is established firmly by the literature through consistent findings in multiple studies...." (1); this recommendation echoes statements from an earlier NCPB report on cancer care quality (9). Repeatedly cited examples of procedures meeting these requirements are esophagectomy and pancreatectomy. By 2003, the Leapfrog Group, a coalition of approximately 140 public and private purchasers of health care, had adopted two (and only two) specific cancer surgery volume standards as indicators of high-quality hospital care: 13 or more esophagectomies per year and 11 or more pancreatic resections per year (10).
However, some observers believe that current evidence on the volume–outcome relationship is so pervasive and persuasive that public and private payers should set minimum case volume requirements for providers as a condition for reimbursement (11).
An additional response to the volume–outcome evidence published to date is to conclude that it is still quite incomplete, in at least two respects. First, not all commonly performed surgical procedures have undergone detailed scrutiny, and there is also little information about whether a volume–outcome relationship exists for a wide range of nonsurgical cancer interventions. Second, for every candidate procedure and service, we need a deeper understanding of the causal mechanisms—the inner workings—of an apparent relationship between high volume and better outcome. How much of the effect is attributable to "learning by doing" and how much is due to higher-volume providers operating in better organized, more richly supported, better informed cancer care delivery environments? Similarly, how much of an observed effect of surgeon volume on outcome is attributable to a surgeon's specialty training and professional exposure rather than to technical skill enhancement through repetitive performance of the procedure itself?
Finally, virtually the entire volume–outcome literature is based on data from observational studies, not randomized clinical trials of high-volume versus low-volume providers. Only a few studies [most notably Birkmeyer et al. (8) and Panageas et al. (12)] have directly tackled such threats to valid inference as the clustering of patients within certain providers, and there has been little detailed analysis of whether patient referral patterns serve to create selection biases in the statistically estimated relationships between volume and outcome.
As Petitti, who coedited the NCPB report (1), has noted: "We need to get behind the volume–outcome relationship.... We need to understand when high-volume hospitals have good outcomes, and transfer those practices to low-volume hospitals so that you have a level playing field" (13). In fact, the NCPB report urges such ongoing inquiry.
Breaking New Ground
The stage is now set for us to consider the latest contributions to this burgeoning yet seriously incomplete literature: the companion articles by Schrag et al. (14) and Earle et al. (15) in this issue of the Journal. Taken together, the closely related analyses reported in these articles make several substantial contributions. They represent the first large-scale, population-based investigation in the United States of whether there are volume–outcome relationships in ovarian cancer surgery, focusing on potential effects at both the hospital level and the surgeon level. The study by Earle et al. (15), in particular, may be the first ever population-based study of the impact of provider specialty on surgical outcomes and care processes in cancer that controls for both surgeon and hospital volume.
Both studies focused on the approximately 3000 patients who had surgery for primary epithelial ovarian cancer diagnosed from 1992 through 1999, as identified in the SEER–Medicare database (which over this period included five states and six metropolitan areas, covering 14% of the U.S. population). From SEER-, Medicare-, or U.S. Census-based sources, the authors constructed case-mix control variables that included the patient's age, sex, race/ethnicity, cancer stage (with histology also available), date of diagnosis, date and cause of death, comorbid illnesses, marital status, and median income and population density in the patient's census tract. The first course of treatment and subsequent surgical and medical interventions were found in the SEER–Medicare database. The surgeon performing the ovarian resection was identified in the patient's Medicare claims data, and information on his/her specialty and other characteristics (e.g., age) was obtained by linking to AMA data. The volume of care rendered by each hospital and each surgeon in the geographic areas covered by SEER was measured by the actual number of these Medicare-eligible ovarian cases treated by the provider over this 8-year period.
Among the studies' main findings are the following six. First, neither the volume of Medicare-covered ovarian procedures performed at the hospital where the patient received care nor the volume of such procedures performed by her surgeon was statistically associated with the patient's chances of surviving 60 days beyond surgery (14). Second, higher hospital volume was associated with lower 2-year mortality, with or without case-mix adjustment—but the relationship lost statistical significance (by conventional standards) once surgeon volume was entered into the equation. There was no statistically significant relationship between surgeon volume and 2-year mortality, with or without case-mix adjustment (14). Third, a modest positive association between hospital volume and overall survival remained after case-mix adjustment, but the association was no longer statistically significant once surgeon volume was included and weakened even more after accounting for whether the patient had postoperative chemotherapy (14). Fourth, ovarian patients treated by gynecologic oncologists and general gynecologists had better case-mix–adjusted 30-day and 60-day mortality outcomes and longer median survival than patients treated by general surgeons. And this was the case whether or not one controlled for hospital volume or surgeon volume (15). Fifth, regarding key elements of ovarian cancer treatment generally considered to be quality enhancing and that may be associated with better survival (though these two studies were not designed to test this aspect), surgical specialists generally delivered superior care. Specifically, stage I and stage II patients treated by gynecologic oncologists were statistically significantly more likely to undergo lymph node dissection than those treated by general gynecologists or general surgeons. Likewise, advanced-stage patients treated by gynecologic oncologists were statistically significantly more likely to have a debulking procedure at time of first surgery (which is associated with a lower risk of tumor dissemination and spread, and also greater effectiveness of chemotherapy, due to reduced tumor volume) than those treated by general gynecologists or general surgeons. Patients treated by either gynecologic oncologists or general gynecologists were statistically significantly more likely to receive postoperative chemotherapy than those treated by general surgeons (15). Sixth, compared with other surgeons in the sample, gynecologic oncologists were younger, more likely to practice in a teaching hospital, and had a much higher mean volume of ovarian cases over the 8-year period (8.8 cases versus 2.2 cases for general gynecologists and 1.3 cases for general surgeons) (15).
On the basis of the array of statistics in Earle et al. describing the structural features, processes, and outcomes of care for ovarian cancer patients, gynecologic oncologists appeared to be delivering a different brand of cancer care from that of other types of surgeons. The authors believe their data offer support for current "professional societies' recommendations that ovarian cancer patients be operated on by gynecologic oncologists when possible."
Taking Stock
Two broad questions arise from these articles: How robust and plausible are the findings? What inferences might be drawn to guide subsequent studies of the determinants of the quality of cancer care, regardless of whether the focus is on surgical procedures or interventions in general?
The recurring theme in the studies by Schrag et al. (14) and Earle et al. (15)—that hospital volume and surgeon volume are not strong determinants of ovarian surgical outcomes but that surgeon specialty influences both the processes and outcomes of surgical care—is reinforced in several of their well-executed analyses. These findings generally hold regardless of outcome measure (whether 30-day, 60-day, or 2-year mortality, or overall survival). But as both papers point out, a major strength of these analyses—their use of the SEER–Medicare dataset—is also a source of some important limitations. Detailed clinical information, e.g., to inform predictions about women who are most likely to benefit from specialized care, or about whether there is evidence of residual cancer following surgery, was not available, thus limiting the number of covariates available to explain patterns of care and outcomes. Also, the sample excluded the 24% of ovarian cancer patients living in SEER areas who were enrolled in health maintenance organizations because these managed-care organizations do not submit detailed claims to Medicare.
Perhaps the most concerning limitation of the SEER–Medicare database is that it includes only Medicare beneficiaries aged 65 years or older, whereas it can be inferred from Earle et al. (15) that approximately 57% of all ovarian cancer cases reported to SEER from 1992 through 1999 involved women younger than 65 years. Couple this fact with the conclusion from Schrag et al. (14) that the strongest determinants of poor ovarian cancer outcomes were patient age at diagnosis and advanced clinical stage, and it is natural to wonder whether the papers' findings would be sustained if women of all ages had been included.
Consequently, it would be interesting to replicate these analyses on large-sample datasets that could provide greater clinical detail and encompass all ages and types of payers. This analysis could be accomplished by appropriately augmenting registry data as presently found in SEER or other sources, such as the National Program of Cancer Registries (supported by the Centers for Disease Control and Prevention [CDC]) or the National Cancer Data Base (cosupported by the American College of Surgeons and the American Cancer Society). Specifically, by reabstracting registry cases to add clinical and other patient-level and provider-level detail and then linking each case to insurance and other secondary data (as done now to create SEER–Medicare), we would be in a much stronger position to analyze volume–outcome and indeed an array of quality-of-care issues. All of this is technically feasible, and consistent with moving toward a "national cancer data system" as envisioned by the IOM (16). But rapid progress would require a substantial and sustained commitment by public and private decision makers, including cancer professional organizations. More immediately, such enhanced-data analyses of quality-of-care questions—including the associations between volume and outcome, and the role of specialists in care delivery—will undoubtedly be carried out for lung cancer and colorectal cancer by the Cancer Care Outcomes Research and Surveillance (CanCORS) Consortium, cosponsored by the NCI and the U.S. Department of Veterans Affairs, with additional support from the CDC (17).
Regardless of the particular set of data sources used, there are at least two vexing statistical problems that can arise in volume–outcome studies (and indeed in other types of quality-of-care investigations involving observational data). The first pertains to the potential clustering of patients within providers, with one particular concern being that a few unusually proficient providers will achieve higher-than-model predicted outcomes, which then effectively exaggerates the estimated difference in performance between the "typical" high-volume and low-volume provider. Building on previous work (12), Schrag et al. (14) tested whether there was additional hospital-to-hospital or surgeon-to-surgeon variation in 2-year mortality rates beyond that which would be predicted by the estimated multivariable models. Focusing on high-volume providers only because of sample-size considerations, they found that actual-versus-predicted differences in outcome were basically random, suggesting no patient clustering within providers. This diagnostic test should be applied routinely in volume–outcome and other quality-of-care studies in which the aim is to assess provider effects.
A second statistical issue was not addressed explicitly by either Schrag et al. (14) or Earle et al. (15), although both studies dealt with it implicitly, perhaps, through a careful and comprehensive selection of patient-level covariates. Specifically, any observational study of this type must cope with the possibility of selection biases, because patients are not randomly assigned to low-volume versus high-volume providers or to surgical specialists versus general surgeons. The particular concern is that factors beyond those reflected in the available covariates may partially account either for choice of provider, the outcome observed, or some interactive combination of these.
There are statistical approaches for coping (usually imperfectly) with these issues, including the use of instrumental variables and propensity scores; in fact, Earle et al. (18) have effectively executed and lucidly explained the application of both methods in an analysis of the effectiveness of chemotherapy among the elderly. The forthcoming analysis by Tsai et al. (19) features an experimental application of the instrumental variable method expressly to the volume–outcome question. Whereas sample size and other data constraints may have forestalled pursuit of these approaches in the ovarian cancer studies discussed here, future investigations should explore their feasibility and utility. After all, the choice of a high-volume versus low-volume provider, or the choice of a surgical specialist rather than a generalist, is essentially the choice of an "intervention" (which is a prelude to the choice of diagnostic and therapeutic interventions).
Beyond their implications for ovarian surgery quality and outcomes, the studies by Schrag et al. (14) and Earle et al. (15) convey a particularly important take-home message for future volume–outcome investigations: They should focus on a good deal more than just volume and outcome. Rather, building on the example set by these ovarian cancer analyses, there should be simultaneous consideration of institutional case volume, physician case volume, physician specialty, the processes of care selected, and the outcomes achieved, while controlling for case-mix and other exogenous factors.
The resulting analytic framework might take the form of a multiequation model that provides a rich depiction of posited causal relationships among at least the following: 1) the patient's choice of providers, both physician and hospital; 2) the (patient–provider) selection of diagnostic and treatment interventions, conditional on patient factors and provider characteristics (including case volume, specialty, and the potential interaction effect of volume and specialty); and 3) patient outcomes (survival, quality of life) as a function of patient factors and provider characteristics. In this way, we would deepen our understanding of the volume–outcome relationship while also transcending it to focus more broadly on the structure–process–outcome linkages that are critical to understanding cancer care quality and how to improve it.
NOTES
The author wishes to acknowledge Theresa W. Gillespie, PhD, whose comments on an earlier version of the editorial were of significant benefit.
REFERENCES
(1) Hewitt M, Petitti D, editors. Interpreting the volume-outcome relationship in the context of cancer care. National Cancer Policy Board, Institute of Medicine, and National Research Council. Washington (DC): National Academy Press; 2001.
(2) Begg CB, Cramer LD, Hoskins WJ, Brennan MF. Impact of hospital volume on operative mortality for major cancer surgery. JAMA 1998;280:1747–51.
(3) Halm EA, Lee C, Chassin MR. Is volume related to outcome in health care? A systematic review and methodologic critique of the literature. Ann Intern Med 2002;137:511–20.
(4) Hillner BE, Smith TJ, Desch CE. Hospital and physician volume or specialization and the outcomes of cancer treatment: Importance in quality of cancer care. J Clin Oncol 2000;18:2327–40.
(5) Bach PB, Cramer LD, Schrag D, Downey RJ, Gelfand SE, Begg CB. The influence of hospital volume on survival after resection for lung cancer. N Engl J Med 2001;345:181–8.
(6) Schrag D, Cramer LD, Bach PB, Cohen AM, Warren JL, Begg CB. Influence of hospital procedure volume on outcomes following surgery for colon cancer. JAMA 2000;284:3028–35.
(7) Birkmeyer JD, Siewers AE, Finlayson EV, Stuken TA, Lucas FL, Batista I, et al. Hospital volume and surgical mortality in the United States. N Engl J Med 2002;346:1128–37.
(8) Birkmeyer JD, Stukel TA, Siewers AE, Goodney PP, Wennberg DE, Lucas FL. Surgeon volume and operative mortality in the United States. N Engl J Med 2003;349:2117–27.
(9) Hewitt M, Simone JV, editors. Ensuring quality cancer care. National Cancer Policy Board, Institute of Medicine, and National Research Council. Washington (DC): National Academy Press; 1999.
(10) The Leapfrog Group. What Does the Leapfrog Group Ask Hospitals About? Available at: http://www.leapfroggroup.org. [Last accessed: December 17, 2005.]
(11) Smith TJ, Hillner BE, Bear HD. Taking action on the volume-quality relationship: how long can we hide our heads in the colostomy bag? J Natl Cancer Inst 2003;95:695–7.
(12) Panageas KS, Schrag D, Riedel E, Bach PB, Begg CB. The effect of clustering of outcomes on the association of procedure volume and surgical outcomes. Ann Intern Med 2003;139:658–65.
(13) Wang L. The volume-outcome relationship: busier hospitals are indeed better, but why? J Natl Cancer Inst 2003;95:700–2.
(14) Schrag D, Earle C, Xu F, Panageas KS, Yabroff KR, Bristow RE, et al. Associations between hospital and surgeon procedure volumes and patient outcomes after ovarian cancer resection. J Natl Cancer Inst 2006;98:163–71.
(15) Earle CC, Schrag D, Neville BA, Yabroff KR, Topor M, Fahey A, et al. Effect of surgeon specialty on processes of care and outcomes for ovarian cancer patients. J Natl Cancer Inst 2006;98:172–80.
(16) Hewitt M, Simone JV, editors. National Cancer Policy Board, Institute of Medicine. Enhancing data systems to improve the quality of cancer care. Washington (DC): National Academy Press; 2000.
(17) Ayanian JZ, Chrischilles EA, Wallace RB, Fletcher RH, Fouad MN, Kiefe CI, et al. Understanding cancer treatment and outcomes: the Cancer Care Outcomes Research and Surveillance Consortium. J Clin Oncol 2004;22:2992–6.
(18) Earle CC, Tsai JS, Gelber RD, Weinstein MC, Neumann PJ, Weeks JC. Effectiveness of chemotherapy for advanced lung cancer in the elderly: instrumental variable and propensity analysis. J Clin Oncol 2001;19:1064–70.
(19) Tsai AC, Votruba M, Bridges JFP, Cebul RD. Overcoming bias in estimating the volume–outcome relationship. HSR: Health Services Research. In press 2006.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  A. J. Vickers, F. J. Bianco, A. M. Serio, J. A. Eastham, D. Schrag, E. A. Klein, A. M. Reuther, M. W. Kattan, J. E. Pontes, and P. T. Scardino The Surgical Learning Curve for Prostate Cancer Control After Radical Prostatectomy J Natl Cancer Inst, August 1, 2007; 99(15): 1171 - 1177. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. D. Nathanson, J. K. Grogan, D. DeBruyn, A. Kapke, and K. Karvelis Breast Cancer Sentinel Lymph Node Identification Rates: The Influence of Radiocolloid Mapping, Case Volume, and the Place of the Procedure Ann. Surg. Oncol., May 1, 2007; 14(5): 1629 - 1637. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. K. Smith, J. T. McPhee, J. S. Hill, G. F. Whalen, M. E. Sullivan, D. E. Litwin, F. A. Anderson, and J. F. Tseng National Outcomes After Gastric Resection for Neoplasm Arch Surg, April 1, 2007; 142(4): 387 - 393. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||