The Relationship Between Hospital Volume and Outcomes of Hepatic Resection for Hepatocellular Carcinoma
Total Page:16
File Type:pdf, Size:1020Kb
ARTICLE The Relationship Between Hospital Volume and Outcomes of Hepatic Resection for Hepatocellular Carcinoma Robert E. Glasgow, MD; Jonathan A. Showstack, PhD, MPH; Patricia P. Katz, PhD; Carlos U. Corvera, MD; Robert S. Warren, MD; Sean J. Mulvihill, MD Background: Volume-outcome relations have been es- Regression analyses were used to adjust for differences tablished for several complex therapies. However, few in patient mix. studies have examined volume-outcome relations for high- risk procedures in general surgery, such as hepatec- Results: Five hundred seven patients underwent hepa- tomy for hepatocellular carcinoma (HCC). tectomy for HCC during the study. Hepatic resections were performed in 138 hospitals, with an overall in- Objective: To evaluate the relation between hospital vol- hospital mortality rate of 14.8%. Three quarters of pa- ume and outcome for patients undergoing hepatectomy tients were treated at hospitals that average 3 or fewer for HCC. hepatic resections for HCC per year. These low-volume providers represent 97.1% of all hospitals treating pa- Design: Retrospective cohort study. tients with HCC statewide. Significant reductions in risk- adjusted operative mortality rates (22.7%-9.4%; P = .002, Setting: All acute-care hospitals in California. multiple logistic regression) and risk-adjusted length of stay (14.3-11.3 days; P = .03, multiple linear regression) Patients: Hospital discharge data were analyzed for each were observed as hospital volume increased. patient in California who underwent major hepatic resec- tion for HCC from January 1, 1990, through December 31, Conclusions: Low operative mortality and length of stay 1994. Hospitals were grouped according to number of hepa- were associated with high-volume centers. These data sup- tectomies performed at each center during the 5-year study. port regionalization of high-risk procedures in general surgery, such as hepatectomy for HCC. Main Outcome Measures: Outcome measures in- cluded operative mortality and length of hospital stay. Arch Surg. 1999;134:30-35 ODAY’S CHANGING health gionalization is beginning to occur as pay- care environment is being ers selectively contract with providers for driven, in part, by external these services. However, this is not the case pressures on providers to de- withothercomplextherapies.Ingeneralsur- liver economical, high- gical practice, standards for the minimum Tquality care. For some medical therapies, of experience necessary to perform highly quality of care varies little among provid- ers, making cost a primary focus.1,2 For other See Invited Critique treatments, however, quality of care is not at end of article uniform. Such is the case with coronary angioplasty, coronary surgery, and bone complex and risky procedures, ie, major he- marrow and solid organ transplantation. patic, pancreatic, or esophageal resection for For these complex therapies, a volume- neoplasia, do not exist. The number of these outcome relationship exists where poor pa- complex operations performed each year is From the Departments of tient outcome, such as in-hospital mortal- insufficient for all surgeons and hospitals to Surgery (Drs Glasgow, ity, is related to low provider volume and have experience. Most of these operations Corvera, Warren, and inexperience.1,3-6 These volume-outcome re- are performed on an elective rather than Mulvihill) and Medicine (Drs Showstack and Katz) and lations serve as the basis for the argument emergent basis. Thus, if centers with supe- the Institute for Health Policy that high-risk procedures should be region- rior patient outcomes could be identified, 3,7,8 Studies (Drs Showstack and alized to centers of excellence. these procedures could be regionalized as Katz), University of California, In the case of coronary angioplasty, a means of providing the most efficacious San Francisco. coronary surgery, and transplantation, re- and cost-effective care. ARCH SURG/ VOL 134, JAN 1999 30 ©1999 American Medical Association. All rights reserved. Downloaded From: https://jamanetwork.com/ on 09/29/2021 MATERIALS AND METHODS sures of postoperative complications were not directly avail- able within this database.4 To characterize a profile of hos- pitals within each volume group, the distribution of the vari- DATA SOURCES ous hospital characteristics was analyzed. Regression modeling was used to evaluate the indepen- We retrospectively analyzed standardized patient dis- dent associations of patient and hospital characteristics with charge abstracts obtained from the California Office of State- the primary outcomes of interest (ie, operative mortality and wide Health Planning and Development (OSHPD), Sacra- length of hospital stay). The patient was the unit of analysis, mento. This database contains discharge data abstracts for with hospital volume group defined as a patient character- every patient hospitalization from every acute-care facil- istic. This allowed for a volume group effect to be assessed ity in the state of California. Each abstract includes a va- while controlling for the characteristics of individual pa- riety of demographic, clinical, and hospitalization data that tients. Multiple logistic regression was used to model the di- characterize a specific hospitalization. Each patient is as- chotomous outcome, in-hospital mortality, and multiple lin- signed a principal diagnosis and procedure and up to 16 ear regression was used to model length of hospital stay. secondary diagnoses and procedures. The OSHPD data- The independent variables in these analyses included base uses diagnostic and procedural codes derived from the hospital volume, age group, sex, year of surgery, source of International Classification of Diseases, Ninth Revision, Clini- admission, type of resection (hepatic lobectomy or partial cal Modification (4th ed) (ICD-9-CM), issued by the US De- hepatectomy), presence of chronic liver disease, and pres- partment of Health and Human Services.9 ence of other preoperative comorbid illnesses. Age was en- All discharge abstracts from January 1, 1990, through tered into the regression equations as the following sets of December 31, 1994, were included in the initial search of the dummy variables: 45 to 60 years, 60 to 75 years, and older OSHPD database. From these abstracts, all patients who un- than 75 years, with younger than 45 years as the reference derwent hepatic lobectomy (ICD-9-CM code 50.3) or partial group. Significant preoperative comorbid illnesses within hepatectomy (ICD-9-CM code 50.22) were examined. From a given patient abstract were grouped into 1 dichotomous this group, a subset of patients undergoing hepatic resection variable to minimize potential colinearities among the vari- for HCC was selected (ICD-9-CM code 155.0). Hospitals were ous comorbidities. For example, patients with a history of characterized with regard to the number of acute and inten- congestive heart failure are likely to also have coronary ar- sive care beds, discharges and patient hospital days per year, tery disease. We believed the following comorbidities to have yearly overall surgical volume and number of hepatic resections a significant influence on operative risk: coronary artery for benign and malignant neoplasia, presence of a liver trans- disease (ICD-9-CM codes 412-414), chronic obstructive pul- plantation program and general surgery residency program, monary disease (ICD-9-CM codes 490-496), diabetes melli- university affiliation, and capability for other complex surgery, tus (ICD-9-CM codes 250), congestive heart failure (ICD- asdeterminedbythepresenceofcardiacsurgeryservices.These 9-CM code 428), nutritional deficiencies (ICD-9-CM codes data were derived, in part, from the Licensed Services and Uti- 260-263), and preoperative intra-abdominal hemorrhage lization Profiles: Annual Report of Hospitals for January 11, 1991, (ICD-9-CM code 459). The presence of chronic liver dis- through December 31, 1991.10 Frequency distributions for ease, including cirrhosis (ICD-9-CM code 571), was treated the individual patient characteristics within the data set and as a separate dichotomous variable, as it represents an in- hospital characteristics listed above were computed. dependent factor associated with poor operative risk. The dependent variables for these analyses were operative mor- DATA ANALYSIS tality or death before discharge and length of hospital stay. Adjusted means for operative mortality rate and length Patients were grouped according to hospital identification of hospital stay were calculated from regression equations number. Hospitals were then classified into quartile groups that included all of the independent variables. A complete based on the number of hepatic resections performed in description of the process of adjustment is provided by Co- the study period. Crude operative mortality rate and length hen and Cohen.11 An adjusted mean is an estimate based of hospital stay were calculated for each volume range. Op- on the hypothetical situation that all hospital volume groups erative mortality in this study was defined as patient death had the same mean values on each of the independent vari- before hospital discharge. Because length of hospital stay ables that were entered into the equation. In other words, is directly related to events within the postoperative course, the adjusted mean represents the estimated operative mor- patients with long hospital stays are most likely patients tality rate or length of stay if each of the volume groups in whom significant perioperative