Page 13

Appendix A

Volume and Outcome in Cancer Surgery

Excerpted from “How Is Volume Related to Quality in Health Care? A Systematic Review of the Research Literature,” by Ethan A. Halm, MD, MPH, Clara Lee, MD, MPP, and Mark R. Chassin, MD, MPP, MPH, Department of Health Policy, Mount Sinai School of Medicine

Prepared for:
Institute of Medicine, National Academy of Sciences
Division of Health Care Services,
Committee on Quality of Care in America
National Cancer Policy Board

Workshop
Interpreting the Volume–Outcome Relationship in the Context of Health Care Quality
May 1, 2000

VOLUME AND OUTCOME IN CANCER SURGERY

We examined a total of 38 studies on cancer (criteria for rating the quality of published studies and literature review methods are described on pages 22–25). All of the eight studies of medical treatment of cancer were excluded because none of them looked at volume as an independent variable. Of the 30 studies of surgical treatment, 10 were excluded. The most common reason for exclusion was a sample that was not community- or population-based (7 studies). Two studies did not evaluate volume as an independent variable (Gordon, 1998; Whittle, 1998). One paper was a review article, not primary research (Steele, 1996).

Thus, 20 papers, all about cancer surgery, were included in the systematic review. Three of these studies looked at more than one procedure (Hannan, 2000; Gordon, 1999; Begg, 1998). To analyze these articles, we examined the data for each procedure separately. In total, 11 papers studied pancreatic resection, five studied colorectal resection, three studied esophagectomy, three studied lung resection, and two studied breast surgery (see attached summaries). The three articles that looked at other cancer procedures are summarized separately in a table called “Cancer Miscellaneous.”

We did not include other papers that studied these operations for benign as well as malignant disease, with the exception of Gordon (1999). We included



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 13
Page 13 Appendix A Volume and Outcome in Cancer Surgery Excerpted from “How Is Volume Related to Quality in Health Care? A Systematic Review of the Research Literature,” by Ethan A. Halm, MD, MPH, Clara Lee, MD, MPP, and Mark R. Chassin, MD, MPP, MPH, Department of Health Policy, Mount Sinai School of Medicine Prepared for: Institute of Medicine, National Academy of Sciences Division of Health Care Services, Committee on Quality of Care in America National Cancer Policy Board Workshop Interpreting the Volume–Outcome Relationship in the Context of Health Care Quality May 1, 2000 VOLUME AND OUTCOME IN CANCER SURGERY We examined a total of 38 studies on cancer (criteria for rating the quality of published studies and literature review methods are described on pages 22–25). All of the eight studies of medical treatment of cancer were excluded because none of them looked at volume as an independent variable. Of the 30 studies of surgical treatment, 10 were excluded. The most common reason for exclusion was a sample that was not community- or population-based (7 studies). Two studies did not evaluate volume as an independent variable (Gordon, 1998; Whittle, 1998). One paper was a review article, not primary research (Steele, 1996). Thus, 20 papers, all about cancer surgery, were included in the systematic review. Three of these studies looked at more than one procedure (Hannan, 2000; Gordon, 1999; Begg, 1998). To analyze these articles, we examined the data for each procedure separately. In total, 11 papers studied pancreatic resection, five studied colorectal resection, three studied esophagectomy, three studied lung resection, and two studied breast surgery (see attached summaries). The three articles that looked at other cancer procedures are summarized separately in a table called “Cancer Miscellaneous.” We did not include other papers that studied these operations for benign as well as malignant disease, with the exception of Gordon (1999). We included

OCR for page 13
Page 14 Gordon (1999) because it studied pancreaticoduodenectomy and esophagectomy, both of which are rarely performed for benign disease. Pancreatic Resection Eleven studies evaluated pancreatic resection. The quality scores varied greatly, ranging from 3 to 10, with a median of 7. The study with the lowest quality score had a small sample that was not representative of the entire population and did not perform any risk adjustment (Wade, 1996). The study with the highest quality score had a large, representative sample, and it examined physician volume, hospital volume, and the interaction between the two (Lieberman, 1995). The unit of analysis was the hospital for all studies, except for two that looked at both surgeon and hospital volume (Lieberman, 1995; Sosa, 1998). No study evaluated appropriateness of patient selection. The definition of low hospital volume ranged from less than 1 to less than 9 procedures per year. Begg et al. defined volume as the annual volume of procedures done on Medicare patients. Two studies of Maryland had only one high-volume hospital (Gordon, 1995; Gordon, 1999). In Lieberman and colleagues' study of New York State, two hospitals were high-volume, and four surgeons were high-volume. The two analyses of surgeon and hospital volume interaction were limited by the fact that most of the high-volume surgeons practiced only in high-volume hospitals. No study effectively addressed the question of “volume of what.” Gordon et al. studied the association between the total volume of 6 “complex gastrointestinal” procedures (total colectomy, esophagectomy, total gastrectomy, hepatic lobectomy, biliary tract anastomosis, and pancreaticoduodenectomy) and individual procedure mortality. They did not also study, however, the association between individual procedure volumes and mortality (Gordon, 1999). No study evaluated the appropriateness of patient selection. Risk adjustment was based almost exclusively on administrative data. Only Begg et al. used some clinical data (cancer staging from the Survival, Epidemiology, and End Results database). None of the studies examined clinical processes. Inpatient death was the primary outcome of interest. Three studies looked at death beyond the inpatient stay (Simunovic, 1999; Birkmeyer, 1999a; Birkmeyer, 1999b), and one measured rates of complications, specifically infection and hemorrhage (Glasgow, 1996). Other complications such as pancreatic or biliary leak, gastric dysmotility, pneumonia, and other outcomes such as recurrence and quality of life were not examined. Of the nine studies that looked at hospital volume only, all but one (Wade, 1996) found a significant relationship between volume and outcomes. The highest quality score of 8 was achieved by a study of 1705 pancreatectomies at 298 hospitals in California from 1990 to 1994 (Glasgow, 1996). In this study, the risk-adjusted mortality at high-volume hospitals (> 50 cases per year) was 3.5%, compared to 14% at low-volume hospitals (< 5 cases per year).

OCR for page 13
Page 15 Lieberman et al. (1995) analyzed both physician and hospital volumes; 1,972 procedures were performed by 748 surgeons in 184 hospitals in New York State from 1984 to 1991. In separate analyses of surgeon volume and hospital volume, high-volume surgeons (> 41 cases per year) had lower risk-adjusted mortality rates than low-volume surgeons (< 9 cases per year)—6% versus 13%, and high-volume hospitals (> 8 cases per year) had lower risk-adjusted mortality rates than low-volume hospitals (< 10 cases per year)—5% versus 19%. When surgeon volume and hospital volume were analyzed together, however, only hospital volume was significant. Sosa et al. (1998) analyzed both physician and hospital volumes for 1,236 procedures by 373 surgeons at 48 hospitals in Maryland. They found that the relative risk of death at low-volume hospitals (< 5 cases per year) was 19 times that at high-volume hospitals (> 20 cases per year). Analyzing physician and hospital volume together, they found hospital volume to be significant regardless of physician volume. Although the studies on pancreatic resection had a great deal of methodological heterogeneity, they suggested that outcomes were related to provider volume and to hospital volume in particular. The magnitude of this volume effect was relatively large compared to most of the other procedures we studied. This is a function of both the high absolute mortality rate for pancreatic cancer as well as a very strong volume and outcome relationship. The number needed to be treated by a high-volume provider to prevent one inpatient death attributable to low volume was only 10 to 15 for most higher-quality studies. Esophagectomy The three studies of esophagectomy had low quality scores (6, 6, and 8). The two lower-scoring studies had relatively small sample sizes—518 patients in one (Gordon, 1999) and 503 patients in another (Begg, 1998). The unit of analysis was the hospital in all three studies. The definition of low volume was relatively similar across studies, ranging from less than 6 to less than 10 procedures per year. Begg et al. measured volume of Medicare cases only. All studies performed some risk adjustment, and only one utilized clinical data (Begg, 1998). No study evaluated clinical processes such as operative approach (abdominal versus thoracoabdominal) and method of reconstruction. The only outcome evaluated was inpatient mortality. No study examined long-term survival, recurrence, or quality of life. Complications such as anastomotic leak, respiratory failure, pneumonia, and digestive dysfunction were not measured. All three studies found large differences in mortality between low-volume and high-volume hospitals. Gordon and colleagues found that the relative risk of death at a low-volume hospital was 3.8 times that at a very-high-volume hospital, although there was only one institution in this latter category (Gordon, 1999).

OCR for page 13
Page 16 Begg et al. found that the risk-adjusted mortality at high-volume hospitals was 3.4%, compared to 17.3% at low-volume hospitals. Patti et al. (1998) found similar mortality rates—6% at high-volume hospitals and 17% at low-volume hospitals. This study had the highest quality score of 8, in part because of its large size. Overall, the magnitude of the volume and outcome relationship for esophagectomy was striking. The number needed to treat by a high-volume provider to prevent one inpatient death attributable to low volume was seven to nine patients. Breast Cancer Surgery The two studies of breast cancer surgery had relatively high quality scores (10 and 11) because they had large numbers of patients, surgeons, hospitals, and adverse events, and because they utilized clinical data from cancer registries in their risk adjustment models. The unit of analysis was the hospital in one study (Roohan, 1998) and the surgeon in the other (Sainsbury, 1995). Neither study looked at the appropriateness of patient selection. Roohan et al. defined “very low” hospital volume as fewer than 10 cases per year. Sainsbury et al. defined low surgeon volume as fewer than 30 cases per year. Sainsbury et al. attempted to include extent of disease and tumor grade in their risk-adjustment model, though this information was missing for 50% of patients. The two studies were noteworthy for their measurement of clinical processes. Roohan et al. included the type of operation (mastectomy or breast-conserving surgery) as an independent variable in the multivariate analysis. Sainsbury et al. included the percentage of patients treated by mastectomy (versus local excision), chemotherapy, hormone therapy, radiation therapy, or surgery alone for each surgeon. These two studies were unique in that they both selected a long-term outcome (5-year survival) as their dependent variable. Neither study measured other outcomes such as recurrence, complications of surgery, or complications of adjuvant therapy. Roohan et al. looked at 47,890 cases of breast cancer surgery performed in 266 hospitals in New York State from 1984 to 1989. In a multivariate regression model, they found volume to be related to 5-year mortality, with a clear “dose-response” relationship. The increased risk of death was 19% in moderate-volume versus high-volume hospitals, 30% in low-volume versus high-volume hospitals, and 60% in very-low-volume versus high-volume hospitals. The authors conjectured that since breast surgery has negligible operative and inpatient mortality, the volume–outcome relationship might be caused by higher-volume hospitals providing more effective adjuvant treatment. Sainsbury et al. studied 12,861 cases of breast cancer surgery performed by 180 surgeons in the Yorkshire Regional Health Authority area from 1979 to 1988. Risk adjustment included age, extent of disease, tumor grade, socioeconomic status, date of treatment, and type of therapy (surgery, radiation, chemotherapy, hormone therapy, surgery alone). They found that the risk of death was

OCR for page 13
Page 17 significantly lower for patients of high-volume surgeons (greater than 29 cases per year) compared to low-volume surgeons (fewer than 10 cases per year). There was no difference in survival between moderate-volume (10 to 29 cases per year) and low-volume surgeons. The volume effect was slightly smaller after risk adjustment (risk ratio of 0.86 versus 0.82 before adjustment). Variation among surgeons in use of mastectomy, radiation, chemotherapy, hormone therapy, and surgery alone accounted for 8% of the variation in survival. Surgeon volume and use of chemotherapy accounted for 20 to 25% of the variation in survival. Lung Resection The quality scores of the three studies of lung resection were relatively high (8, 8, and 10). The numbers of patients, physicians, hospitals, and adverse events were all high. The unit of analysis was the hospital in two studies (Begg, 1998; Romano, 1992) and both hospital and physician in one study (Hannan, 2000). No study evaluated the appropriateness of patient selection. The three studies looked at different types of lung resection—lobectomies (Hannan, 2000), pneumonectomies (Begg, 1998), and all resections (Romano, 1992). The definitions of low hospital volume were heterogeneous, ranging from less than 6 to less than 38 procedures per year. Risk adjustment was based on administrative data in two of the studies (Hannan, 2000; Romano, 1992) and clinical data in one (Begg, 1998). No study looked at clinical processes of care. The outcome of interest was inpatient death in all three studies. Complications such as bronchopleural fistula, respiratory failure, and pneumonia were not measured. In addition, no study evaluated other outcomes such as long-term survival, recurrence, or quality of life. In the study with the highest quality score of 10, Hannan et al. (2000) looked at 6,954 lobectomies by 373 surgeons at 178 hospitals. The risk-adjusted mortality rate at low-volume hospitals (>37 cases per year) was 1.65% higher than at high-volume hospitals (>169 cases per year). There was no difference between medium-volume and high-volume hospitals. The vast majority of hospitals were low-volume (133 hospitals). Only 4 hospitals were high-volume. No significant relationship between surgeon volume and outcome was found Begg and colleagues examined 1,375 pneumonectomies performed on Medicare patients at 313 hospitals in the United States. They utilized clinical data for risk adjustment. No difference in outcomes existed between high-volume and low-volume hospitals. Romano and colleagues found 40% lower risk of death after pneumonectomy at high-volume hospitals compared to low-volume hospitals. They also found a similar volume–outcome relationship for lesser resections.

OCR for page 13
Page 18 Colorectal Resection The five studies of colorectal cancer resection had quality scores ranging from 7 to 10, with a median of 9. The studies were very heterogeneous. Three studies evaluated resections of all types of colorectal cancer (Hannan, 2000; Harmon, 1999; Parry, 1999), one looked at total colectomy for benign and malignant disease (Gordon, 1999), and one looked at resections for rectal cancer (Porter, 1998). The unit of analysis was the hospital in one study (Gordon, 1999), the physician in one study (Porter, 1998), and both hospitals and physicians in three (Hannan, 2000; Harmon, 1999; Parry, 1999). The definition of low volume was variable, even among the three studies that looked at volume of all colorectal resections. Among these three studies, the definition of low surgeon volume ranged from less than 6 to less than 12 procedures per year. The definition of low hospital volume ranged from less than 40 to less than 84 per year. Gordon et al. looked at the relationship between 6 complex gastrointestinal procedures including total colectomy and the outcomes of total colectomy. All studies performed risk adjustment, and two studies (Porter, 1998; Parry, 1999) used clinical data. Two studies examined clinical processes, but neither incorporated the processes into their risk adjustment model. Parry et al. measured whether or not an abdominoperineal resection was performed, use of ultrasound or CT scan, and operating “after hours.” Porter et al. looked at the type of operation (low anterior resection versus abdominoperineal resection) and the use of adjuvant therapy. The outcome studied was primarily inpatient mortality. One study (Parry, 1999) measured local recurrence rates as well as disease-specific survival. No study measured complications such as anastomotic leak, intraabdominal abscess, wound infection, or genitourinary dysfunction. Three of the four studies that assessed hospital volume did not find a significant relationship to outcomes. Harmon et al. studied all resections in Maryland and found a trend toward lower mortality at high-volume hospitals, but this was not significant (odds ratio 0.78, p < 0.10). Parry et al. studied all resections in the northwestern United Kingdom and found no relationship between volume and outcomes. Gordon et al. found no relationship between volume of complex gastrointestinal surgery and outcome of total colectomy. The only study to find a significant relationship for hospital volume found that the risk-adjusted mortality rate at low-volume hospitals was 1.9% higher than at high-volume hospitals (Hannan, 2000). Of the four studies that measured physician volume, three found a significant volume–outcome relationship. Only Parry et al. found no relationship between physician volume and outcomes. Porter et al. found that patients of low-volume surgeons had worse disease-specific survival than patients of high-volume surgeons (hazard ratio = 1.40) and a higher risk of local recurrence (hazard ratio = 1.80). High-volume surgeons were more likely to perform a low ante-

OCR for page 13
Page 19 rior resection as might be expected. They were no more likely, however, to use adjuvant therapy than low-volume surgeons were. Three studies analyzed physician volume and hospital volume together. The physician effect found by Hannan et al. disappeared when hospital volume was controlled for in the analysis. Harmon et al. found that surgeon volume was related to volume regardless of hospital volume. The studies of volume and outcome in colorectal surgery do not uniformly find a significant relationship. The magnitude of the volume effect on mortality is relatively modest—an absolute difference in inpatient mortality of 1% to 2% corresponding to a number needed to treat of 50–100. SUMMARY The 20 studies of cancer surgery suggest that a significant relationship between volume and outcomes does exist. The largest differences between low-and high-volume providers were found for the most complicated operations in rare cancers—pancreatectomy and esophagectomy. For colorectal resection and lung resection, two operations for more common cancers, the relationship between volume and outcome is not as clear. The common methodological issues for these studies point to a need for more clinical data. Information about the type of tumor and cancer stage would be highly desirable, particularly in studies that look at long-term survival. An examination of the different clinical processes being employed and how they vary with provider volume might elucidate the differences in outcomes. For example, the use of adjuvant therapies is particularly important but has not been well-studied with respect to volume. The roles of other providers besides the surgeon have also not been examined. Particularly when long-term survival is being evaluated, characteristics of other providers who care for the patient years after surgery, such as the medical oncologist and radiation oncologist, would be relevant. More appropriate referral to these providers or better coordination of the many elements of cancer care, such as diagnostic testing, adjuvant therapy, and follow-up surveillance, may underlie the hospital volume effects that have been found. It is worth noting that the literature on volume and outcomes in cancer has disproportionately focused on rare operations for rare cancers. For the most common cancer operations—breast cancer surgery, colon resection, and lung resection—we found 10 studies that met our inclusion criteria. By contrast, the most rare operations—esophagectomy and pancreatectomy—had 13 publications. In addition, we found no studies of medical treatment of cancers.

OCR for page 13
Page 20 CRITERIA FOR RATING THE QUALITY OF PUBLISHED STUDIES We developed a scoring system to assess the quality of the research studies included in our systematic review. The full list of criteria is described on page 23. Our aim was to create a quantitative method of assessing the research design of the studies we reviewed such that higher scores would reflect increasing likelihood of the study's ability to discern generalizable conclusions about the nature and magnitude of the relationship between volume and outcome. The first four criteria assess various aspects of the patient sample used in the research. We assigned one point if the sample was representative of the general population of all patients who might receive the treatments under study. Thus, studies of managed care plan enrollees or Medicare beneficiaries were not considered representative. We assigned two points if the study included patients of 50 or more physicians and 20 or more hospitals. If only one of these criteria was met, we assigned one point. No points were assigned if neither criterion was met. In some studies authors reported the number of hospitals in their sample but not the number of treating physicians. In these cases we estimated the number of physicians by assuming it would be at least equal to the number of hospitals. The vast majority of these studies included hundreds of hospitals from administrative databases, so we estimated the number of physicians as > 50 for scoring this criterion. If the total sample size was 1,000 patients or more, we assigned one point. Because statistical power to detect significant relationships in logistic regression models depends more on the total number of adverse events represented in the sample than on total sample size (and because the various conditions and procedures in this literature have widely varying adverse event rates), we assigned 2 points if the total number of adverse events was greater than 100, one point if it was 21–100, and no points if it was 20 or less. We assigned no points if the study assessed the relationship between out-come and either hospital or physician volume. If both were assessed separately, we assigned one point. If the joint relationships of hospital and physician volume were assessed independently in a multivariate analysis, we assigned 2 points. And if a study examined both of these and the volume of another important component of the care process, we assigned 3 points. If the appropriateness of patient selection was not addressed, we assigned no points. If appropriateness was measured, we assigned 1 point. If it was measured and taken into account in the analysis of the volume–outcome relationship, we assigned 2 points. If volume was analyzed in only 2 categories, we assigned no points. If more than 2 categories were assessed or if volume was treated as a continuous variable, we assigned 1 point to credit a more sophisticated assessment of a possible dose-response relationship. In considering the various ways in which outcomes might be risk-adjusted, we assigned no points if no risk-adjustment at all was done. If data from insurance claims, hospital discharge abstract databases, or

OCR for page 13
Page 21 other sources of administrative data were used, we assigned 1 point. If data from clinical sources (e.g., medical records or prospectively designed clinical registries) were used for risk-adjustment, we assigned 2 points. If clinical data were used in a logistic regression model that demonstrated good calibration by the Hosmer-Lemeshow test and good discrimination (by a C-statistic of 0.75 or greater), we assigned 3 points. If specific clinical processes of care were not measured, we assigned no points. If a single process was measured and its impact on risk-adjusted out-comes assessed, we assigned 1 point. If 2 or more such processes were measured and evaluated, we assigned 2 points. Finally, if death was the only outcome evaluated, we assigned no points. If other adverse outcomes in addition to mortality were assessed, we assigned 2 points. Quality scores were summed across all 10 criteria for each study. The maximum possible total score was 18. Literature Review Methods We performed two electronic subject-based searches of the literature on MEDLINE (1966–1999). A professional reference librarian assisted us in the development of our search strategy We developed a list of search terms based on subject headings from articles known to be highly relevant to our topic and from the official indexing terms of the MEDLINE database. We performed multiple searches with combinations of these terms and evaluated the results of those searches for sensitivity and specificity, with respect to our topic of volume and outcomes. The search algorithm that yielded the greatest number of highly relevant articles combined the conditions with the terms volume, utilization, frequency, statistics, and outcomes. In order to broaden our search to include articles on regionalization of care, we added another search that combined the conditions with the term regionalization. We also performed MEDLINE searches on authors known to have published widely on the study topic, and we searched the Cochrane Collaboration Database for systematic reviews. In addition to performing electronic database searches, we consulted experts in the field for further references. Finally, we reviewed the references cited by each article that was ultimately included. We did not hand-search any journals. This review was limited to the English-language research literature. This paper includes the findings of our review of cancer-related procedures and conditions. Study inclusion criteria were: 1. Time: patient cohorts treated from 1980 forward. 2. Sample: community- or population-based sample—case series or convenience samples were excluded. 3. Multiple publications from the same database excluded; only the most recent or most complete publication was included.

OCR for page 13
Page 22 4. Health outcome(s) must be assessed as the dependent variable(s). 5. Volume must be an independent variable. We limited the review to studies of patients treated from 1980 to the present, because of the rapidity of changes in hospital care, available treatments, and surgical techniques. In our view, data from patient cohorts prior to 1980 would have questionable relevance to today's policy issues. In a few instances, we included studies if part of their patient sample included patients treated in 1978 or 1979, but most of the sample comprised patients from the 1980s. We excluded studies from single institutions, from voluntary registries, or other convenience samples because of the weak generalizability of such studies. We excluded a few studies in which the only dependent variable was a composite of deaths or long lengths of stay, because, formulated in this way, the dependent variable was not purely a health outcome. We also excluded a few studies in which the only dependent variable was a composite of death or complications, with the latter determined solely by secondary diagnosis codes in administrative databases. These studies were excluded because of the notorious unreliability of using such data to identify complications. In general, we excluded multiple publications from the same set of data, selecting only the most recent or complete, unless different publications reported substantially different analyses (e.g., one reported the relationship of hospital volume to outcome and another analyzed physician volume and outcome). Three reviewers assessed the articles for inclusion or exclusion, with at least two reviewers independently examining each article and applying the criteria. Discrepancies in the application of the criteria were resolved by discussion between the reviewers. Our final criteria for quality assessment and the scoring system were described earlier and are listed in on page 23. The same pair of reviewers who assessed each article for inclusion or exclusion then independently evaluated each article and assigned quality scores. Discrepancies were resolved by discussion between the two reviewers.

OCR for page 13
Page 23 RATING THE QUALITY OF RESEARCH ON VOLUME AND OUTCOME Objective of Scoring System: designed to measure the degree to which the study design is likely to reveal generalizable conclusions about the magnitude and nature of the relationship between volume and outcome. ~ enlarge ~ TOTAL POSSIBLE POINTS = 18

OCR for page 13
Page 24 PANCREAS Study Population Time period Patient # MD # Hospital # Unit of analysis Primary outcome Risk adjustment data source Definition of low volume Volume: Outcomes results   Score Gordon 1999 All Maryland Benign & malignant 1989–1997 1092 NS 51 Hosp Inpt death Admin Hosp: < 10/yr Vol <10 11–20 21–50 > 200 RR 12.5 10.4 6.3 1 7 Birkmeyer 1999a Medicare US Benign & malignant 1992–1995 7229 NS 1772 Hosp 3yr death Admin Hosp Very low: < 1 Low: 1–2 High: > 5 OR = 0.69   7 Birkmeyer 1999b Medicare US Benign & malignant 1992–1995 7229 NS 1772 Hosp Inpt death 30d death Admin Hosp Very low: < 1 Low: 1–2 High: > 5 Inpt death: 16% vs. 4.1% (very high 1.7%) 30d death: 12.9 vs. 3.0%   7 Sosa 1998a All Maryland 1990–1995 1236 373 48 MD Hosp Both Inpt death Admin MD: Low: < 5, High: > 50 Hosp: Low: < 5 High: > 20 LVH vs. HVH: RR = 19.3 HVH better, regardless of MD volume   9

OCR for page 13
Page 25 Begg 1998 Medicare US 1984–1993 742 NS 252 Hosp Inpt death Clinical Low: < 6 high: > 10/yr Mortality: 12.9 vs. 5.8% 6 Simunovic 1999 All Ontario 1988/89 or 1994/95 842 NS 68 Hosp Inpt death 64d death Admin < 22 LVH: OR = 5.1 MVH: OR = 4.5 6 Glasgow 1996 All CA 1990–1994 1705 NS 298 Hosp Inpt death Bleeding Infection Admin Low: 1–5 High: > 50 RAMR: 14 vs. 3.5% 8 Imperato 1996 Medicare NY 1991–1994 579 NS 117 Hosp Inpt death Admin Low: 1–5/yr high: > 25/yr Mortality: 14.3 vs. 2.2% (RR 6.87) 5 Wade 1996 Dept of Defense 1989–1994 130 NS 111 Hosp Inpt death None <1 Mortality < 1: 6% 1–2: 9% > 2: 9% (no p value given) 3 Lieberman 1995 All NY 1984–1991 1972 748 184 MD Hosp Both Inpt death Admin MD: < 9 Hosp: < 10 MD: 6 vs. 13%; Hosp: 5 vs. 18.9%; Both: Only hospital volume is important 10 Gordon 1995 All Maryland 1988–1993 501 NS 39 Hosp Inpt death Admin Low: < 1–5/yr high: > 20/yr Mortality: 19vs. 2.2% (RR = 8.7) 6 OR: odds ratio RR: relative risk NS: not specified LVH: low-volume hospital

OCR for page 13
Page 26 ESOPHAGUS Study Population Time period Patient # MD # Hospital # Unit of analysis Primary outcome Risk adjustment data source Definition of low volume Volume: Outcomes results Score Gordon 1999 All Maryland (Benign and malignant) 1989–1997 518 NS 51 Hosp Inpt death Admin Hosp: < 10/yr Volume of 6 complex GI procedures Vol < 10 11–20 21–50 > 200 RR 3.8 4.0 2.4 1.0 6 Begg 1998 Medicare US 1984–1993 503 NS 190 Hosp Inpt death Clinical Hosp: Low: < 5/yr high: > 11/yr Mortality 17.3 vs. 3.4%   6 Patti 1998 All CA 1990–1994 1561 NS 273 Hosp Inpt death Admin Hosp: Low: < 5/yr High: > 30/yr Mortality 17 vs. 6%   8 RR: Relative Risk

OCR for page 13
Page 27 BREAST Study Population Time period Patient # MD # Hospital # Unit of analysis Primary outcome Risk adjustment data source Definition of low volume Volume: Outcomes results Score Roohan 1998 All women NY 1984–1989 47890 NS 266 Hosp 5yr survival Clinical Hosp: Low: <10/yr high: >149/yr OR = 1.6 10 Sainsbury 1995 All women Yorkshire, UK 1979–1988 12861 180 NS MD 5yr survival Clinical MD: <30/yr Vol <10 10–29 30–49 > = 50 Adjusted RR Ratio 1.0 0.97 0.85 0.86 11   Abbreviations: OR: odds ratio RR: relative risk NS: not specified

OCR for page 13
Page 28 LUNG Study Population Time period Patient # MD # Hospital # Unit of analysis Primary outcome Risk adjustment Data source Definition of low volume Volume: Outcomes Results Score Hannan in press All NY Lobectomies 1994–1997 6954 373 178 MD Hosp Both Inpt death Admin MD: < 23/yr Hosp: < 38/yr Hosp: RAMR for LVH 1.65% > HVH MD: no relationship 10 Begg 1998 Medicare US Pneumo-nectomies 1984–1993 1375 NS 313 Hosp 30 day mortality Clinical Hosp: < 6/yr No relationship 8 Romano 1992 All CA All resections 1983–1986 12439 NS 389 Hosp Inpt death Admin Hosp: < 9/yr Lesser resections (high- relative to low-volume): OR = 0.6 Pneumonectomy: OR = 0.6 8 Abbreviations: LVP: low-volume physician LVH: low-volume hospital HVP: high-volume physician HVH: high-volume hospital RAMR: risk-adjusted mortality rate OR: odds ratio

OCR for page 13
Page 29 COLORECTAL Study Population Time period Patient # MD # Hospital # Unit of analysis Primary outcome Risk adjustment data source Definition of low volume Volume: Outcomes results Score Hannan in press All NY 1994–1997 22128 2052 229 MD Hosp Both Inpt death Admin MD: low: < 12 high: > 34 Hosp: low: < 84 high: > 253 RAMR for LVH 1.93% > HVH; No MD effect when hosp volume controlled 10 Harmon 1999 All Maryland 1992–1996 9739 812 50 MD Hosp Both Inpt death Admin MD: < 6/yr Hosp: < 40/yr MD: HVS vs. LVS; OR = .64; Hosp: HVH vs. LVH; OR = .78; MVS at HVH/MVP equiv to HVS; HVS better at any hosp 10 Parry 1999 All NW UK 1993 (6 mos) 927 123 39 MD Hosp 30 day death; 3 year survival Clinical MD: < 7 in 6 mos Hosp: < 30 in 6 mos No relationship 9 Gordon 1999 All Maryland Total colectomy 1989–1997 1015 NS 51 Hosp Inpt death Admin Hosp: < 10/yr No relationship 8 Porter 1998 All Edmonton Rectal cancer 1983–1990 683 52 5 MD Local recurrence Disease-specific survival Clinical MD: < 21/yr Local recurrence HR = 1.8; DSS: HR = 1.4 HVP no more likely to give adjuvant Rx; HVP more likely to do LAR 7 RAMR: risk-adjusted mortality rate LVH: low-volume hospital HVH: high-volume hospital DSS: disease-specific survival HR: hazards ratio LVP: low-volume physician HVP: high-volume physician MRP: medium-volume physician LAR: low anterior resection NS: not specified

OCR for page 13
Page 30 CANCER MISCELLANEOUS Study Population Time period Patient # MD # Hospital # Unit of analysis Primary outcome Risk adjustment data source Definition of low volume Volume: Outcomes results Score Hannan in press All NY Gastrectomy for cancer 1994–1997 3711 1114 207 MD Hosp Both Inpt death Admin MD: 1–2 Hosp: 1–15 Risk-adjusted increase in rate for lowest- relative to highestv olume quartile; Hosp: 7.1% Surgeon: 5.7%; No MD effect when hosp volume controlled 10 Glasgow 1999 All CA Hepatic resections for cancer 1990–1994 507 NS 138 Hosp Inpt death Admin Low: < 2 high: > 16 Risk-adjusted mortality rate: Low: 22.7 High: 9.4% 6 Gordon 1999 All Maryland Biliary tract anastomosis, gastrectomy, hepatic lobectomy (benign and malignant) 1989–1997 938; 705; 293 NS 51 Hosp Inpt death Admin < 11 Measured vol of 6 complex GI procedures Biliary tract anastomosis: adjusted RR = 5.3 Gastrectomy: no relationship; Hepatic lobectomy: adjusted RR = 4.7; 6 GI procedures: Benign: no relationship Malignant: adjusted RR = 5.2 6 Begg 1998 Medicare/US Pelvic exenteration, hepatic resection 1984–1993 1592; 801 NS 250+ Hosp 30 day death Clinical Low: < 1–5 high: > 11 Unadjusted 30 day mortality: Pelvic: 3.7 vs. 1.5% Hepatic: 5.4 vs. 1.7% 7 Abbreviations LVP: low-volume physician HVP: high-volume physician NS: not specified LVH: low-volume hospital HVH: high-volume hospital

OCR for page 13
Page 31 REFERENCES Begg CB, Cramer LD, Hoskins WJ, Brennan MF. Impact of hospital volume on operative mortality for major cancer surgery. JAMA. 1998 ; 280(20): 1747–51. Birkmeyer JD, Finlayson SR, Tosteson AN, Sharp SM, Warshaw AL, Fisher ES. Effect of hospital volume on in-hospital mortality with pancreaticoduodenectomy. Surgery. 1999b ; 125(3): 250–6. Birkmeyer JD, Warshaw AL, Finlayson SR, Grove MR, Tosteson AN. Relationship between hospital volume and late survival after pancreaticoduodenectomy. Surgery. 1999a ; 126(2): 178–83. Glasgow RE, Mulvihill SJ. Hospital volume influences outcome in patients undergoing pancreatic resection for cancer. Western Journal of Medicine. 1996 ; 165(5): 294–300. Glasgow RE, Showstack JA, Katz PP, Corvera CU, Warren RS, Mulvihill SJ. The relationship between hospital volume and outcomes of hepatic resection for hepatocellular carcinoma. Arch Surg. 1999 ; 134(1): 30–5. Gordon TA, Bowman HM, Bass EB, et al. Complex gastrointestinal surgery: Impact of provider experience on clinical and economic outcomes. J Am Coll Surg. 1999 ; 189(1): 46–56. Gordon TA, Burleyson GP, Tielsch JM, Cameron JL. The effects of regionalization on cost and outcome for one general high-risk surgical procedure. Annals of Surgery. 1995 ; 221(1): 43–9. Hannan E. The influence of hospital and surgeon volume on mortality for common cancer procedures. Surgery, in press. Harmon JW, Tang DG, Gordon TA, et al. Hospital volume can serve as a surrogate for surgeon volume for achieving excellent outcomes in colorectal resection. Ann Surg. 1999 ; 230(3): 404–11; discussion 411–3. Imperato PJ, Nenner RP, Starr HA, Will TO, Rosenberg CR, Dearie MB. The effects of regionalization on clinical outcomes for a high risk surgical procedure: A study of the Whipple procedure in New York State. American Journal of Medical Quality. 1996 ; 11(4): 193–7. Lieberman MD, Kilburn H, Lindsey M, Brennan MF. Relation of perioperative deaths to hospital volume among patients undergoing pancreatic resection for malignancy. Annals of Surgery. 1995 ; 222(5): 638–45. Parry JM, Collins S, Mathers J, Scott NA, Woodman CB. Influence of volume of work on the outcome of treatment for patients with colorectal cancer. Br J Surg. 1999 ; 86(4): 475–81. Patti M, Corvera CU, Glasgow RE, Way LW. A hospital's annual rate of esophagectomy influences the operative mortality rate. Journal of Gastrointestinal Surgery. 1998 ; 2: 186–92. Porter GA, Soskolne CL, Yakimets WW, Newman SC. Surgeon-related factors and outcome in rectal cancer. Ann Surg. 1998 ; 227(2): 157–67. Romano PS, Mark DH. Patient and hospital characteristics related to in-hospital mortality after lung cancer resection. Chest. 1992 ; 101(5): 1332–7. Roohan PJ, Bickell NA, Baptiste MS, Therriault GD, Ferrara EP, Siu AL. Hospital volume differences and five-year survival from breast cancer. American Journal of Public Health. 1998 ; 88(3): 454–7.

OCR for page 13
Page 32 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(8960): 1265–70. Simunovic M, To T, Theriault M, Langer B. Relation between hospital surgical volume and outcome for pancreatic resection for neoplasm in a publicly funded health care system. CMAJ. 1999 ; 160(5): 643–8. Sosa JA, Bowman HM, Gordon TA, et al. Importance of hospital volume in the overall management of pancreatic cancer. Annals of Surgery. 1998 ; 228(3): 429–38. Wade TP, Halaby IA, Stapleton DR, Virgo KS, Johnson FE. Population-based analysis of treatment of pancreatic cancer and Whipple resection: Department of Defense hospitals, 1989–1994. Surgery. 1996 ; 120(4): 680–5; discussion 686–7.