Fax: (514) 597-1616
Investigating the correlation between hospital of primary treatment and the survival of women with breast cancer†
Article first published online: 3 AUG 2005
Copyright © 2005 American Cancer Society
Volume 104, Issue 7, pages 1343–1348, 1 October 2005
How to Cite
Hébert-Croteau, N., Brisson, J., Lemaire, J., Latreille, J. and Pineault, R. (2005), Investigating the correlation between hospital of primary treatment and the survival of women with breast cancer. Cancer, 104: 1343–1348. doi: 10.1002/cncr.21336
Presented in part at the fourth European Breast Cancer Conference, Hamburg, Germany, March 16–20, 2004.
- Issue published online: 17 SEP 2005
- Article first published online: 3 AUG 2005
- Manuscript Accepted: 20 APR 2005
- Manuscript Revised: 31 MAR 2005
- Manuscript Received: 21 DEC 2004
- Canadian Breast Cancer Research Alliance
- breast cancer;
- health services research;
- outcomes research;
To understand the relation between hospital of initial treatment and the survival of women with breast cancer, the authors investigated the characteristics of the treatment center that were related most to outcome.
The authors selected women from 5 regions of Quebec, Canada, who were diagnosed with lymph node-negative breast cancer between 1988 and 1994. Data were collected by chart review, queries to physicians, and linkage with administrative data bases. Overall survival to the end of 1999 was analyzed using the Kaplan–Meier method and Cox proportional hazards models.
The study population included 1727 women with a median follow-up of 6.8 years. The 7-year survival rate was 82% (95% confidence interval [95%CI], 80–84%). Compared with women who were treated in centers with ≥ 100 new cases per year, the hazard ratio (HR) of death from any cause was 1.80 (95%CI, 1.23–2.63), 1.44 (95%CI, 1.03–2.03), and 1.30 (95%CI, 0.96–1.76) among women who were treated in hospitals with < 25 new cases, 25–49 new cases, and 50–99 new cases per year after adjusting for case mix and characteristics of the attending physician. However, the significance of caseload disappeared after adjusting for the type of hospital. By contrast, women who were treated in centers with either on-site radiotherapy, research activity, or teaching status had significantly better outcomes, even after adjusting for caseload (HR, 0.68; 95%CI, 0.50–0.92). These associations were independent of primary treatment received, which was a strong determinant of outcome.
Primary treatment of early-stage breast cancer in larger hospitals was associated with improved survival. This relation was mediated by factors related to proficiency of care, which tended to cluster within institutions. Cancer 2005. © 2005 American Cancer Society.
Several studies have evaluated the relation between the characteristics of the source of care and the outcomes of patients with breast cancer.1–15 Two recent reviews16, 17 suggested that volume is a key determinant of performance. This association has been documented for several health conditions, but not for all,18, 19 and the underlying mechanisms are likely disease-specific.20 Critics of the volume-outcome relation have raised issues of data quality and adjustment for case mix.21 Nevertheless, the consistency of this association suggests that it is not an artifact for some health conditions.22
There is little information about how patient volume is related to other characteristics of the treatment center that are likely to influence the quality of care for patients with breast cancer. Larger hospitals often also are involved in research, teaching, or delivery of specific therapies for cancer, and these factors, which are typical of most academic centers, may explain the apparent volume-outcome relation. Indeed, a recent evaluation of patients who were treated within facilities designated by the National Cancer Institute (NCI) supports this notion.23 These institutions had lower surgical mortality rates than control hospitals with the highest volume of procedures for 6 major interventions, although 5-year overall survival was similar in both groups.
We previously reported that treatment according to practice guidelines was a significant determinant of survival in a population-based cohort of women with lymph node-negative breast cancer from Quebec, Canada.24 It has been shown that compliance with guidelines for systemic adjuvant treatment increases with the volume of breast cancer patients in hospitals that participate in collaborative clinical trials, whereas the same compliance decreased in institutions that were not involved in clinical research.25 We now report a survival analysis of this cohort by type of hospital (involved or not involved in either research, teaching, or delivery of radiation therapy) and caseload with the objective of evaluating whether these characteristics of the source of care are related independently to outcomes.
MATERIALS AND METHODS
The study population and sampling procedure have been described previously.24, 25 Briefly, approximately 50% of women from 5 health regions in Quebec who were newly diagnosed with pathologically confirmed, lymph node-negative breast cancer in 1988–1989, 1991–1992, and 1993–1994 had been selected randomly from the population tumor registry (Québec Tumor Registry [QTR]) and hospital discharge data base for a pattern-of-care study that was conducted between 1995 and 1997. Recurrences and vital status were assessed to the end of 1999. Individuals with multicentric, inflammatory tumors; multiple primary tumors; lesions with benign or uncertain behavior (e.g., lobular carcinoma in situ); or lesions that did not originate from the mammary gland (e.g., lymphomas) were excluded along with individuals who were lost to follow-up immediately after primary treatment. The final sample included 1727 women with in situ, Stage I, or Stage II breast carcinoma.
Data Collection Procedure
Information was collected by two consecutive reviews of medical charts and direct queries to physicians. Vital status was updated by linkage of our data with the mortality file, the QTR, and the data base on beneficiaries of the Quebec universal health insurance system. The Charlson comorbidity index, which was calculated as a weighted sum of selected health conditions that were predictive of mortality in longitudinal studies, was estimated from the discharge summary of the first hospital admission for primary treatment using the method described by Deyo et al.26 Treatment was assessed for its conformity with the National Institutes of Heath 1990 recommendations for locoregional management27 and the St.-Gallen 1992 consensus recommendations for systemic adjuvant therapy.28 Women under experimental protocol were classified as having received treatment consistent with guidelines. The hospital annual volume of patients with breast cancer was estimated from the QTR and was specific to incident cases, although ranking would be expected to be similar if new patients and patients with recurrent disease were used. Hospital research activity was defined as participation in collaborative clinical trials, but not in industry-funded research. The project had been approved by the Commission d'Accès à l'Information du Québec, the directors of professional services of all concerned institutions, and research ethics committees.
Data were analyzed by Kaplan–Meier and Cox proportional hazards analyses. Independent variables were grouped into strata related to individual patients, physicians, and hospitals. Patient-specific variables that were representative of case mix were age, comorbidity, tumor stage, grade, estrogen receptor status, and margin status. Years in practice and volume of patients with breast cancer were the characteristics of the attending physician that were taken into account. Finally, the annual breast cancer caseload, participation in clinical research, academic affiliation, and on-site radiation therapy as an indicator of resources for cancer treatment were the hospital-specific factors used in the analysis. The type of hospital was defined by involvement or no involvement in either research, teaching, or delivery of radiotherapy. To account for secular changes in breast cancer mortality, all models were adjusted for the year of diagnosis.
First, we investigated the independent association of survival with hospital caseload, successively adjusting for case mix, then adding characteristics of the attending physician, and finally type of hospital. Primary treatment, including locoregional therapy and systemic treatment consistent or not consistent with guidelines, was added to the model to evaluate potential effect modification and to assess the independent contribution of caseload beyond differences in patterns of treatment across facilities. To investigate further the mechanisms by which caseload is related to outcome, the analysis was repeated using the type of hospital as the main independent variable, with the same adjustment strategy. The association of the independent variables with survival was evaluated using likelihood ratio (LR) statistics and the Wald test with a 5% level of statistical significance. All tests were bilateral.
The patient sample is described in Table 1. About 90% of these women had no significant comorbidity. Tumors were positive for estrogen receptor in 53.9% of patients and were Grade 3 in 23.3% of patients. In addition, 8.9% of women had carcinoma in situ, 61.0% of women had Stage I disease (T1N0M0), and 28.3% of women had Stage II disease (T2N0M0). Overall, almost 85% of the cohort underwent conservative surgery, and 75.7% underwent dissection of the axilla. More than two-thirds of patients received systemic treatment consistent with guidelines. These women were treated by a total of 248 physicians in 57 different hospitals, 39 of which were involved in neither research, teaching, or delivery of radiation therapy.
|Characteristic||No. of patients||%|
|Yr of diagnosis|
|Age at diagnosis|
|< 50 yrs||445||25.8|
|≥ 70 yrs||446||25.8|
|Charlson comorbidity index|
|Estrogen receptor status|
|Tumor in situ||153||8.9|
|Unknown, no surgery, etc.||432||25.0|
|Conservative surgery with radiotherapy||1193||69.1|
|Conservative surgery without radiotherapy||271||15.7|
|Dissection of the axilla|
|Risk of recurrencea|
|Compliance with treatment guidelinesa|
Analysis of Survival
During a median follow-up of 6.8 years, 380 women died, including 143 deaths from breast cancer. The 7-year overall survival rate was 82% (95% confidence interval [95%CI], 80–84%).
Tables 2 and 3 summarize the correlation of survival with either caseload (Table 2) or type of hospital (Table 3) as the main exposure variable, using different adjustment covariates. The LR and P value for each model reflect the independent contribution of either caseload (Table 2) or type of hospital (Table 3) to survival when other variables listed in the table footnotes are taken into account.
|Hospital annual caseload||No. of women||Hazard ratio (95% confidence interval)|
|Model 1a||Model 2b||Model 3c||Model 4d|
|< 25 patients||204||1.46 (1.06–2.02)||1.80 (1.23–2.63)||1.29 (0.81–2.05)||1.38 (0.86–2.21)|
|25–49 patients||375||1.17 (0.89–1.55)||1.44 (1.03–2.03)||1.10 (0.73–1.65)||1.23 (0.81–1.85)|
|50–99 patients||473||1.12 (0.85–1.47)||1.30 (0.96–1.76)||1.16 (0.84–1.60)||1.26 (0.91–1.74)|
|≥ 100 patients||675||1.00||1.00||1.00||1.00|
|LR (P value)||5.215–0.157||9.474–0.024||1.607–0.658||2.397–0.494|
|Type of hospital||No. of women||Hazard ratio (95% confidence interval)|
|Model 1a||Model 2b||Model 3c||Model 4d|
|Without research, teaching, or radiotherapy||597||1.00||1.00||1.00||1.00|
|With research, teaching, or radiotherapy||1130||0.71 (0.57–0.88)||0.63 (0.49–0.80)||0.68 (0.50–0.92)||0.69 (0.50–0.95)|
|LR (P value)||9.729–0.002||14.035–0.0002||6.168–0.013||5.425–0.020|
Caseload was related inversely to the risk of death from any cause, but this was explained by its correlation with the type of hospital (Table 2). Fluctuations in the hazard ratios (HRs) across models reflect the confounding effect of specific categories of covariates. Using hospitals with ≥ 100 new breast cancer cases each year as referent, the HRs were 1.80 (95%CI, 1.23–2.63), 1.44 (95%CI, 1.03–2.03), and 1.30 (95%CI, 0.96–1.76) in facilities that admitted < 25 patients, 25–49 patients, and 50–99 patients, respectively, after adjustment for case mix and characteristics of the attending physician (Model 2). Caseload was associated significantly with survival in this model (LR, 9.474; P = 0.024), but statistical significance was lost when the type of hospital also was included (Model 3; LR, 1.607; P = 0.658). Point estimates varied little when primary treatment was taken into account, although this factor itself was a strong determinant of outcome (LR associated with caseload, 2.397; P = 0.494; LR associated with treatment variables, 69.176; P < 0.00001; Model 4).
The analysis was repeated using the type of facility as the main exposure variable (Table 3). Whether or not primary treatment was taken into account, admission in facilities involved in research, teaching, or delivery of radiotherapy services was associated with a significant reduction in the risk of death from any cause after adjustment for case mix, characteristics of the attending physician, and hospital caseload (HR, 0.68; 95%CI, 0.50–0.92; Model 3). The stability of the HR across models suggests that this correlation is not influenced substantially by confounding factors. Adding the type of hospital to a model that already included volume led to further improvement of the model (LR, 6.168; P = 0.013; Model 3).
The current data support the notion that primary treatment of breast cancer in large hospitals is associated with better survival when case mix and experience of the attending physician are taken into account. However, this correlation was not independent in the current analysis and lost statistical significance when other characteristics of the hospital, such as the involvement in research, teaching, or delivery of radiotherapy, also were considered. By contrast, the type of hospital, as defined by these features, was associated consistently with survival, regardless of the analytic model used.
Although several studies have reported on the association between the treatment center and outcomes of patients with breast cancer, their results are conflicting. Caseload was the usual focus of these investigations. Lee-Feldstein et al.,1 for example, studied 5892 patients from California who were diagnosed between 1984 and 1990 and were treated in 126 hospitals that were classified as small community, large community, health maintenance organization, and teaching. Survival was significantly better at large community hospitals after adjusting for case mix and locoregional treatment. Roohan et al.2 also showed better 5-year survival with treatment in large hospitals (≥ 151 surgeries per yr) in New York State after controlling for age, race, socioeconomic status, comorbidity, stage, distance to hospital, and type of surgery. Women who were treated in hospitals with ≤ 10 cases each year had a 60% greater 5-year risk of mortality from all causes. Hospital volume and teaching status were correlated, but this variable did not predict survival beyond the effect of volume. In 1 of the largest series to date that included 29,666 patients with breast cancer from Los Angeles County, Skinner et al.14 showed that both surgeon and hospital case volume, but not treatment in a specialty hospital designated by the NCI, were related to 5-year survival. Treatment received was not included in their multivariate models. These results contrast to some extent with the observation of Kingsmore et al.15 that surgical treatment in specialist units in the United Kingdom more often was adequate and resulted in lower local and regional recurrences and better survival. However, that study did not evaluate the effect of caseload as such.
In Canada, Chaudry et al.4 reported a 53% reduction in the risk of death among women with lymph node-negative tumors that measured ≤ 2 cm who were treated in teaching Ontario hospitals, but they observed no difference in survival among patients with larger tumors. Their estimates were adjusted for age, estrogen receptor status, and use of radiation therapy, but not for caseload. The same association of survival with hospital teaching status has been reported in Japan.7 Compared with patients who were treated in teaching hospitals, the HR of death, adjusted for age, stage, and surgical treatment, was 1.25 (95%CI, 1.01–1.58), 1.51 (95%CI, 1.20–1.89), and 1.65 (95%CI, 1.21–2.25) for patients who were treated in large hospitals (≥ 400 beds), medium-sized hospitals (150–399 beds), or small hospitals (20–149 beds), suggesting that volume itself is not influential.
Three other investigations6, 8, 9 also failed to show any relation between the hospital caseload and outcomes of patients with breast cancer, whereas a fourth investigation5 suggested that women who are treated in larger centers fare worse. Harcourt et al.6 found no correlation between the observed/expected survival ratio and hospital case volume among 2409 patients with breast cancer who were diagnosed between 1980 and 1995 in Washington State. No multivariate analysis was done in their study. In another study from Australia,8 no difference was observed in the survival of patients from the period 1980 to 1986 who were treated in large public hospitals, large private hospitals, and smaller hospitals after adjustment for age, tumor size, and lymph node status. Richards et al.9 also found no correlation between survival and the type of hospital in a study of young women with breast cancer in the South East Thames region of the United Kingdom, although there was substantial variation in patterns of care between teaching and nonteaching hospitals. Finally, a negative association of survival with hospital size was found among patients who were registered in the American College of Surgeons data base prior to 1978.5 Onsite radiation therapy also was associated negatively and significantly with survival. Adjustment for several characteristics of the patients and for treatment used aggregated data by hospital and, thus, may have been incomplete.
Overall, the literature on hospital volume and outcomes of patients with breast cancer, a condition with low operative mortality and short hospital stay, thus, is not as consistent as what has been reported for the medical treatment of acquired immunodeficiency syndrome, surgery for pancreatic and esophageal cancer, abdominal aortic aneurysm, or congenital hearth disease.29 This may be explained in part by methodological issues, including use of administrative data without risk adjustment in several instances, as well as failure to take into account critical factors, such as the experience or skills of the attending physician or treatment. The current analysis did avoid several of these pitfalls. However, residual confounding due to the imperfect measurement of some covariates, such as comorbidity, is plausible. Moreover, only primary treatment of breast cancer was taken into account, whereas the management of recurrences may extend survival to some extent.30 Finally, like several other investigators,1, 2, 4, 7, 9, 14 we used mortality from all causes as an endpoint; therefore, the management of comorbidity and complications of cancer treatment, as well as of breast cancer, may have been influential. However, treatment of these other health conditions was not measured. Nevertheless, we believe that this approach is justified from a comprehensive quality-of-care perspective.
In our data, the volume-outcome relation was explained by other factors, including the characteristics of facilities related to the proficiency of resources. The organization of health services itself is likely to vary with caseload, and this may influence the delivery of care. Although our indicator for classifying hospitals was rather crude, it reflects some mechanisms that have been proposed to explain the volume-outcome relation, including more intensive care resources in larger facilities, better nursing, or the availability of specialized services or highly technical equipment.29 In addition, hospital involvement in both research31 and teaching4, 7 have been linked with performance.
In recent years, several investigators have used a configurational approach to the analysis of health services.32, 33 This approach incorporates multiple dimensions of healthcare organizations into the development of integrated classifications.32 Given the complexity of health care,34 the configurational approach may be more appropriate on both theoretical and empirical grounds than focusing separately on individual features of organizations, such as caseload.32 This approach also is likely to provide information that is relevant more directly to healthcare decision makers and/or policy makers.
In conclusion, we observed a general trend toward improved survival associated with primary treatment for breast cancer in hospitals with increased patient volumes, an effect that was mediated by other factors, including features of the source of care associated with the quantity and quality of resources. Additional research should attempt to identify which characteristics of treatment centers are most predictive of outcome in patients with breast cancer.
The authors thank Nadia Abdelaziz, Christine Beaulieu, Sylvie Bérubé, Fatima Bouharaoui, Magali Girard, Monika Lessard, Michéle Perron, and Brigitte Simard for assistance in data collection and analysis.
- 20Concentration and choice in the provision of hospital services. CRD Report 8 (summary). NHS Centre for Reviews and Dissemination, York: University of York, 1997., , .
- 30The impact of new chemotherapeutic and hormonal agents on the survival of women with metastatic breast cancer (MBC) in a population based cohort [abstract 22]. Proc Am Soc Clin Oncol. 2003; 22: 6., , , et al.
- 33Choices for change: the path for restructuring primary healthcare services in Canada. CHSRF report, November 2003. Ottawa: Canadian Health Services Research Foundation, 2003., , , , , .