|
 |
|
|||||||
|
|||||||||
© 2007 American Public Health Association
DOI: 10.2105/AJPH.2005.075663
RESEARCH AND PRACTICE |
Mary Ann Gilligan, Joan Neuner, and Ann B. Nattinger are with the Department of Medicine and Health Policy Institute, Medical College of Wisconsin, Milwaukee. At the time of the study, Xu Zhang was with the Division of Biostatistics, Medical College of Wisconsin, Milwaukee. Rodney Sparapani and Purushottam W. Laud are with the Division of Biostatistics, Medical College of Wisconsin, Milwaukee.
Correspondence: Requests for reprints should be sent to Mary Ann Gilligan, MD, MPH, Medical College of Wisconsin, Division of General Internal Medicine, FEOB Suite 4200, 9200 W Wisconsin Ave, Milwaukee, WI 53226 (e-mail: gilligan{at}mcw.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
Objectives. We examined the association between number of breast cancer operations performed in a hospital (hospital volume) and all-cause and breast cancer–specific mortality using a national database and statistical methods appropriate for clustering and reducing confounding.
Methods. In a retrospective cohort study, we linked Surveillance, Epidemiology, and End Results tumor registry data with Medicare claims data. The cohort included 11225 Medicare patients who had undergone surgery for early-stage breast cancer from 1994 to 1996 in 457 different hospitals. Primary outcomes were all-cause and breast cancer–specific survival rates at a mean follow-up time of 62.5 months.
Results. In comparison with treatment in a low-volume hospital, treatment in a high-volume hospital was associated with hazard ratios of 0.83 (95% confidence interval [CI]=0.75, 0.92) for all-cause mortality and 0.80 (CI=0.66, 0.97) for breast cancer–specific mortality.
Conclusions. An association between the volume of breast cancer operations performed in a hospital and 5-year survival rates was observed for both all-cause and breast cancer–specific mortality. Further work investigating the aspects of hospital volume that contribute to increased survival is warranted.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
Associations between number of operations performed in a hospital (hospital volume) and clinical outcomes have been reported for certain cancers such as esophageal and pancreatic cancer.1,2 Higher short-term mortality has been observed in hospitals performing lower numbers of the high-risk operations performed for these cancers. However, in the case of more-common, lower-risk operations, such as those performed for colorectal cancer and lung cancer, a hospital volume–clinical outcome (hereafter volume–outcome) relationship has been found less consistently.1,3,4
The initial operation performed for breast cancer is low risk, with short-term mortality rates typically below 1%.5 Therefore, any observed reduction in short-term mortality associated with hospital volume would be small in magnitude. However, 2 studies,6,7 although involving limited data, have supported a relationship between hospital volume and long-term mortality. These studies suggest that, among patients treated in New York State and California, treatment in a high-volume hospital was associated with better 5-year survival rates. Although population based, these studies were geographically limited. In addition, neither study explored clustering of patients by hospital or the possibility of selection bias based on patient socioeconomic status, 2 problems that can lead to overestimation of the volume–outcome relationship.8
We used a population-based, national database to examine the relationship between hospital volume of breast cancer cases and long-term survival rates. Our goal was to extend previous work by using a more geographically diverse sample to evaluate both overall and disease-specific mortality. Also, we used statistical methods appropriate for clustering of patients by hospital and for reducing confounding due to selection bias among patients choosing to use high-volume hospitals.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
Data
We linked data from the National Cancer Institute’s population-based Surveillance, Epidemiology, and End Results (SEER) tumor registry to Medicare claims data.9,10 SEER collects information from 11 population-based tumor registries covering approximately 14% of the US population. Data on incident cancer cases among individuals residing in coverage areas are gathered from hospitals, offices, and freestanding centers. For each patient with breast cancer, information is abstracted regarding demographic characteristics, extent of disease, and initial therapy. SEER information has been linked to Medicare claims for 94% of people aged 65 years and older.11 We used Medicare Provider Analysis and Review and outpatient Standard Analytic File data to determine the hospital at which the operation was performed.
Cohort
The patient cohort included women aged 66 years or older who were eligible for Medicare part A or B, were not enrolled in a Medicare health maintenance organization, had been diagnosed with microscopically confirmed stage 1 or 2 breast cancer between 1994 and 1996, and had undergone breast-conserving surgery or mastectomy as the initial therapy (N = 12 715). Cases in which the operation was performed in a hospital outside of the SEER region (n = 499) or the hospital was not identifiable (n = 991) were excluded, resulting in a final cohort of 11 225 women.
Hospital Volume and Patient Characteristics
We ascertained hospital volume by counting the number of operations for incident breast cancer of any stage performed on Medicare patients during the study period. Hospital volume was categorized into 3 groups (low: 0–19 cases per year; medium: 20–39 cases per year; high: 40 or more cases per year) with approximately equivalent numbers of patients in each.
Patient characteristics (age and race) and tumor characteristics (size, grade, nodal involvement, and hormone receptor status) were determined from the SEER database. Per capita income and educational level were estimated from US census data on the basis of the median per capita income and percentage of high school graduates among adults residing in the same zip code as the patient. This method is recognized as a valid approach to estimating socioeconomic status when individual-level data on income and education are not available.12,13 The size of the metropolitan standard area in which the patient resided was determined at the county level.
We calculated a comorbidity index for each patient according to the methods outlined by Charlson et al.,14 using the modifications described by Klabunde et al.15,16 The Charlson index, developed with inpatient claims data, comprises 15 noncancer conditions, each of which is weighted according to its impact on mortality.14 The Klabunde modifications incorporate the diagnostic and procedure data contained in Medicare part B (carrier) claims.15,16
Outcome Measures and Analysis
Two outcomes were analyzed: time until death by any cause and time until death by breast cancer. Both were measured from the time of breast cancer diagnosis. Cause of death was defined as that reported by SEER, which in turn, was based on death certificate data. We used Cox proportional hazards regression models for survival data,17 with modifications for clustering and propensity analyses. In assessments of survival time data, the problem of clustering is typically resolved through the use of a frailty model.17,18 We employed such a model to account for possible clustering by hospital.
We conducted a propensity analysis to attenuate the effects of potential selection bias caused by patients’ self-selection into low- vs high-volume hospitals.19,20 In such an analysis, often used in observational studies, subgroups with balanced covariates are created across the variable of interest—in our case hospital volume group. Then, in the primary analysis, the effect of the variable on the outcome is compared within each of these balanced subgroups. We developed a logistic regression model to classify patients as having a low, midlevel, or high propensity for treatment in a high-volume hospital on the basis of socioeconomic, race, comorbidity, and disease status variables (hormone receptor status, lymph node status, and tumor grade status).21
Such propensity analyses for variables with 2 categories have been cited frequently in the literature.20,22,23 However, our use of this method for analyzing 3 volume groups required an extension.21 Initially, we developed the propensity score model using a trichotomous logistic regression model, producing propensity scores for membership in each of the 3 volume groups. To construct propensity groups, we created a bivariate plot of the high- and low-volume group propensity scores. Then we constructed planar tertile groups by drawing lines at 45° angles with both axes. Thus, individuals showing both a high propensity to be in the low-volume category and a low propensity to be in the high-volume category were grouped together. The resulting 3 groups showed a balance of covariates among volume groups within each propensity tertile group.
We developed Cox proportional hazards survival models, with the modifications just described, for all-cause mortality and breast cancer–specific mortality. In all models, the patient was the unit of analysis and the following patient characteristics were controlled: age (with linear and quadratic components); zip code–level per capita income, educational level, and population density; comorbidity index; tumor characteristics; and propensity score. Tests of proportionality of hazards for the covariates revealed a lack of proportionality for (1) nodal involvement, (2) hormone receptor status, and (3) zip code–level per capita income.
In the case of the first 2 covariates, we addressed the hazard proportionality assumption by incorporating time points into the model that allowed differing effects before and after the particular point in time.17 The results presented for node status refer to deaths occurring 10 months or more after diagnosis, and the results presented for hormone receptor status refer to deaths occurring up to 56 months after diagnosis. For the third covariate, stratifying according to per capita income was sufficient to address the proportionality assumption. In addition, for all-cause mortality only, survival among patients in the medium-volume hospital group varied over time, requiring us to treat this group as a separate stratum in the model to retain proportionality of hazards in the other volume groups.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
A total of 11 225 patients underwent surgery for breast cancer in 457 different hospitals (Table 1
). Patients operated on in high-, medium-, and low-volume hospitals differed according to sociodemographic and tumor characteristics (Table 1). Patients treated in high-volume hospitals were slightly more likely to be White, resided in substantially more urban areas, and resided in areas with a higher per capita income than did patients treated in low-volume hospitals. Also, patients treated in high-volume hospitals had slightly more favorable tumor characteristics (tumor size, hormone receptor status; Table 1).
|
Mean follow-up time was 62.5 months. Numbers of deaths were 1036 among patients treated in low-volume hospitals (28.8%), 946 among patients treated in medium-volume hospitals (25.6%), and 919 among patients treated in high-volume hospitals (23.4%).
In the final multivariate Cox model, treatment in a high-volume (vs low-volume) hospital was associated with a hazard ratio of 0.83 for all-cause mortality (Table 2). Eliminating the propensity score from the final model did not have an effect on the results of the final analysis. Treatment in a medium-volume hospital was also associated with decreased all-cause mortality, although we were unable to compute a hazard ratio because of the violation of proportionality assumption, that is, because the hazard ratio changes over time. However, as can be seen in Figure 1, overall mortality for the medium-volume group was intermediate between the high- and low-volume groups at all times.
|
|
There were 710 deaths from breast cancer (6.3% of the cohort). Treatment in high-volume hospitals was associated with lower breast cancer–specific mortality (Table 2
). Extent of disease factors, such as tumor size and lymph node status, were stronger predictors of breast cancer–specific mortality than of all-cause mortality, as one would expect. Conversely, although still significantly associated with breast cancer–specific mortality, comorbidity was a stronger predictor of all-cause mortality. Even after we controlled for all available patient- and tumor-related factors, the effect of hospital volume remained significant.
Improved survival was observed among patients with both lymph-node-negative and lymph-node-positive disease. Figure 2 depicts adjusted survival probabilities for patients in these 2 groups stratified according to hospital volume. This figure provides a visual sense of the absolute improvement in survival associated with treatment in a high-volume hospital when all other patient and tumor factors were held constant.
|
Concern has been expressed in the literature that if assessment of tumor prognostic factors is less complete in low-volume hospitals, control for these factors in a regression model might cause a spurious association between volume and survival.24 To address this concern, we computed the regression models excluding these factors, and the results were essentially identical (data not shown).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
In this study, we found a relationship between hospital volume and all-cause mortality; that is, there were moderate reductions in both all-cause mortality and breast cancer–specific mortality among women treated in hospitals with annual volumes of 40 or more operations performed on Medicare breast cancer patients. These relationships were observed despite careful control for possible confounders such as patient characteristics and tumor prognostic characteristics. The effect of hospital volume remained measurable throughout the 5-year median follow-up time.
There are multiple plausible factors contributing to differences in survival (e.g., variable use of adjuvant therapies). We purposely did not include such treatment factors in our model, because one way a high-volume hospital can achieve better outcomes is through systems that facilitate follow-through with treatment. Control for such adjuvant treatment would be expected to obscure the relationship between volume of breast cancer cases and outcomes. Surgical technique may also play a role in the volume–outcome relationship, although not in the same way that it does with more high-risk procedures,25 given that breast cancer operations generally involve low short-term mortality. For example, in our study cohort, only 14 women (0.12%) died within 30 days of their operation. However, it is possible that the improved long-term survival results reported here for high-volume hospitals are attributable to aspects of surgical technique such as ensuring tumor-free margins of resection.
Although the hospital volume effect was significant, it is important to note that some patients operated on at low-volume hospitals did very well. We found that, in terms of 5-year survival, approximately 26% of low-volume hospitals and 37% of middle-volume hospitals outperformed the median high-volume hospital. Other studies have shown similar variations among patients of a hospital volume group.26 Hospital volume appears to be a significant, yet still imperfect, predictor of better outcomes.
Our study involved several limitations. For example, because it was an observational study, it was vulnerable to the biases inherent in all such studies. We used statistical techniques to address the most important of these biases, namely clustering of patients by hospital and selection bias among patients choosing to use high-volume hospitals. As mentioned, clustering was addressed through frailty methods, and propensity analysis methods were used to diminish the effect of selection bias. Although propensity analysis is an accepted method of controlling for observed factors, it cannot account for unmeasured effects. Hence, there could have been residual selection bias, a limitation of virtually all hospital volume–clinical outcome studies.27
Another limitation is that we included only women aged older than 65 years for whom information was available through Medicare claims. Because women in this age group account for almost half of incident cases of breast cancer,28 we can estimate that the number of operations across all ages would be approximately twice that found in our study. However, if the number of operations varied systematically according to age, the generalizability of our results to younger women would be limited.
Our findings are consistent with the effect of hospital volume seen in 2 previous US studies.6,7 Skinner et al.6 found that, in California, high hospital volume was associated with a 23% lower risk of death at 5 years than low hospital volume. Roohan et al.7 showed that New York State patients treated in very-low-volume hospitals had a 60% greater risk of all-cause mortality than patients treated in high-volume hospitals. A single study from the United Kingdom that evaluated hospital volume did not show a volume–outcome relationship,29 but initial breast cancer care is more regionalized in the United Kingdom than in the United States,29,30 which may account for the difference.
Taken in aggregate, 3 US studies now support the hypothesis that patients treated for breast cancer in high-volume hospitals have better survival outcomes. Future work should further evaluate possible mechanisms for this relationship and whether the effect is modified by surgeons’ case volumes. Further research investigating the aspects of hospital volume that contribute to increases in survival, with a view toward improving outcomes in low- and medium-volume hospitals, is warranted.
|
Acknowledgments |
|---|
This research was supported by the National Institutes of Health (grant R01-CA81379).
We acknowledge the efforts of the Applied Research Program of the National Cancer Institute; the Office of Research, Development, and Information of the Centers for Medicare and Medicaid Services; Information Management Services Inc; and the Surveillance, Epidemiology, and End Results (SEER) program tumor registries in the creation of the SEER–Medicare database.
Human Participant Protection
This study was approved by the Medical College of Wisconsin’s institutional review board.
|
Footnotes |
|---|
Peer Reviewed
Contributors
All of the authors were involved in the origination and design of the study, analysis and interpretation of data, and critical revision of the article for important intellectual content. M.A. Gilligan drafted the article. X. Zhang, R. Sparapani, and P.W. Laud were responsible for the statistical analysis.
Accepted for publication March 11, 2006.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
1. Hewitt M, Petitti D, eds. Interpreting the Volume-Outcome Relationship in the Context of Cancer Care. Washington, DC: National Academy Press; 2000.
2. Birkmeyer JD, Siewers AE, Finlayson EV, et al. Hospital volume and surgical mortality in the United States. N Engl J Med.2002;346:1128–1137.
3. Hodgson DC, Zhang W, Zaslavsky AM, Fuchs CS, Wright WE, Ayanian JZ. Relation of hospital volume to colostomy rates and survival for patients with rectal cancer. J Natl Cancer Inst.2003;95:708–716.
4. Meyerhardt JA, Catalano PJ, Schrag D, et al. Association of hospital procedure volume and outcomes in patients with colon cancer at high risk for recurrence. Ann Intern Med.2003;139:649–657.
5. Schneiderman MA, Axtell LM. Deaths among female patients with carcinoma of the breast treated by a surgical procedure only. Surg Gynecol Obstet.1979; 148:193–195.[Web of Science][Medline]
6. Skinner KA, Helsper JT, Deapen D, Ye W, Sposto R. Breast cancer: do specialists make a difference? Ann Surg Oncol.2003;10:606–615.[CrossRef][Web of Science][Medline]
7. Roohan P, Bickell N, Baptiste M, Therriault G, Ferrara E, Siu A. Hospital volume differences and five-year survival from breast cancer. Am J Public Health.1998;88:454–457.
8. 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;139:658–665.
9. Surveillance, Epidemiology, and End Results Program Public Use File [CD-ROM]. Bethesda, Md: National Cancer Institute; 1999.
10. Warren JL, Klabunde CN, Schrag D, Bach PB, Riley GF. Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care. 2002; 40(suppl):IV-3–IV-18.
11. Potosky A, Riley G, Lubitz J, Mentnech R, Kessler L. Potential for cancer related health services research using a linked Medicare-tumor registry database. Med Care.1993;31:732–748.[Web of Science][Medline]
12. Krieger N. Overcoming the absence of socioeconomic data in medical records: validation and application of a census-based methodology. Am J Public Health.1992;82:703–710.
13. Geronimus AT, Bound J. Use of census-based aggregate variables to proxy for socioeconomic group: evidence from a national sample. Am J Epidemiol.1998;148:475–486.
14. Charlson M, Szatrowski TP, Peterson J, Gold J. Validation of a combined comorbidity index. J Clin Epidemiol.1994;47:1245–1251.[CrossRef][Web of Science][Medline]
15. Klabunde CN, Warren JL, Legler JM. Assessing comorbidity using claims data: an overview. Med Care.2002;40(suppl 8):26–35.[CrossRef][Web of Science][Medline]
16. Klabunde CN, Potosky AL, Legler JM, Warren JL. Development of a comorbidity index using physician claims data. J Clin Epidemiol.2000;53:1258–1267.[CrossRef][Web of Science][Medline]
17. Klein J, Moeschberger M. Survival Analysis. New York, NY: Springer Publishing Co; 1997.
18. Hougaard P. Frailty models for survival data. Lifetime Data Anal.1995;1:255–273.[CrossRef][Medline]
19. Rosenbaum P, Rubin D. The central role of the propensity score in observational studies for causal effects. Biometrika.1983;70:41–55.
20. D’Agostino RJ. Propensity score method for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med.1998;17: 2265–2281.[CrossRef][Web of Science][Medline]
21. Sparapani R, Laud, PW. Multiple treatments and propensity scores. Available at: http://www.biostat.mcw.edu. Accessed December 6, 2006.
22. Normand S-L, Landrum M, Guadagnoli E. Validating recommendations for coronary angiography following acute myocardial infarction in the elderly: a matched analysis using propensity scores. J Clin Epidemiol.2001;54:387–398.[CrossRef][Web of Science][Medline]
23. Stone R, Obrosky S, Stringer D, et al. Propensity score adjustment for pretreatment differences between hospitalized and ambulatory patients with community-acquired pneumonia. Med Care. 1995;33:AS56–AS66.[Web of Science][Medline]
24. Gillis C, Hole D. Survival outcome of care by specialist surgeons in breast cancer: a study of 3786 patients in the west of Scotland. BMJ.1996;312: 145–148.
25. Epstein AM. Volume and outcome—it is time to move ahead. N Engl J Med.2002;346:1161–1164.
26. Begg CB, Riedel ER, Bach PB, et al. Variations in morbidity after radical prostatectomy. N Engl J Med.2002;346:1138–1144.
27. 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;137:511–520.
28. Cancer Facts and Figures 2004. Atlanta, Ga: American Cancer Society; 2004.
29. Mikeljevic JS, Haward R, Johnston C, et al. Trends in postoperative radiotherapy delay and the effect on survival in breast cancer patients treated with conservation surgery. Br J Cancer.2004;90:1343–1348.[CrossRef][Web of Science][Medline]
30. Sainsbury R, Haward B, Rider L, Johnston C, Round C. Influence of clinician workload and patterns of treatment on survival from breast cancer. Lancet.1995;345:1265–1270.[CrossRef][Web of Science][Medline]
This article has been cited by other articles:
|
|
 |
|
  C. Hery, J. Ferlay, M. Boniol, and P. Autier Quantification of changes in breast cancer incidence and mortality since 1990 in 35 countries with Caucasian-majority populations Ann. Onc., June 1, 2008; 19(6): 1187 - 1194. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. B. Nattinger, P. W. Laud, R. A. Sparapani, X. Zhang, J. M. Neuner, and M. A. Gilligan Exploring the Surgeon Volume Outcome Relationship Among Women With Breast Cancer Arch Intern Med, October 8, 2007; 167(18): 1958 - 1963. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |