Specificity of Procedure Volume and In-hospital Mortality Association

Abstract

Objective:

Several studies have examined the association between procedure-specific volume and in-hospital mortality and concluded that high-volume hospitals have lower mortality rates when compared with low-volume hospitals. There is a paucity of studies examining the association between unrelated procedure volume and in-hospital mortality. The objective of our study is to examine the procedure-specific volume–outcome association as well as unrelated procedure volume–outcome association for 5 procedures: coronary artery bypass graft (CABG), percutaneous coronary interventions (PCI), elective abdominal aortic aneurysm repair (AAA), pancreatectomy (PAN), and esophagectomy (ESO).

Methods:

Nationwide Inpatient Sample for years 2000 through 2003 was used. All discharges with primary procedure codes for CABG, PCI, AAA, PAN, and ESO were selected. The average number of procedures performed by the hospitals per year during the study period was computed, and hospitals were categorized as having met or not met the Leapfrog Group-recommended volume thresholds. Procedure specific and unrelated procedure volume–in-hospital mortality association was examined by using multivariable logistic regression analysis. Procedure volume–in-hospital mortality association was adjusted for patient and hospital characteristics.

Results:

For all 5 procedures, hospitals that did not meet Leapfrog Group volume thresholds were associated with significantly higher odds for in-hospital mortality when compared with hospitals that met Leapfrog Group volume thresholds (P < 0.05). Hospital volume levels for PAN or ESO did not influence outcomes following CABG, PCI, and AAA. Similarly, hospital volumes for CABG, PCI, and AAA did not influence the outcomes for PAN or ESO.

Conclusions:

Hospital volume–in-hospital mortality association appears largely to be specific to the procedure being studied.

Specificity of procedure volume–outcome association was examined following coronary artery bypass graft, percutaneous coronary interventions, elective abdominal aortic aneurysm repair, pancreatectomy, and esophagectomy. Hospital volume-in-hospital mortality association is largely specific to the procedure being studied.

There is considerable evidence to conclude that high-volume hospitals have lower mortality rates compared with low-volume hospitals following complex surgical procedures.^1,2 While several studies have examined the association between procedure-specific volume and in-hospital mortality,^3–7 there are only a few studies that have examined the association between unrelated procedure volume and in-hospital mortality.^8,9 Urbach and Baxter examined the association between procedure-specific/unrelated procedure volume and 30-day mortality following esophagectomy (ESO), major lung resection for cancer, repair of unruptured abdominal aortic aneurysm (AAA), and pancreaticoduodenectomy.⁹ Konety et al examined the influence of meeting Leapfrog Group-recommended volume thresholds for complex surgical procedures (coronary artery bypass grafts [CABG], percutaneous coronary interventions [PCIs], AAA repair, pancreatectomy [PAN], and ESO) on in-hospital mortality following radical cystectomy, nephrectomy, and radical prostatectomy.⁸

The Leapfrog Group, a private entity formed by representatives from 155 major health insurance purchasers, recommends that members contract for selected complex surgical procedures (CABG, PCI, elective AAA repair, PAN, and ESO) only with hospitals that meet minimum volume thresholds to achieve better outcomes.¹⁰ There is lack of published data examining the effect of meeting any Leapfrog Group-recommended volume thresholds on mortality after other unrelated Leapfrog Group-specified procedures. We examined the association between procedure volume and in-hospital mortality after CABG, PCI, elective AAA repair, PAN, and ESO. We examined the procedure-specific volume–outcome association as well as unrelated procedure volume–outcome association.

METHODS

Patient Selection

We performed retrospective analyses of the Nationwide Inpatient Sample (NIS) of the Healthcare Cost and Utilization Project for the years 2000 through 2003. The NIS is a 20% stratified sample of community hospitals in the United States. Hospitals are chosen based on the strata (geographic region, hospital location, hospital teaching status, hospital bed-size, and hospital ownership) to which they belong.¹¹ All patients aged ≥18 years and who underwent CABG (ICD-9-CM procedure codes of 36.1x), PCI (36.01, 36.02, 36.05, 36.06, and 36.07), AAA (38.34, 38.44, 38.64, 39.71, and 39.25), PAN (52.51, 52.53, 52.6, and 52.7), or ESO (42.4, 42.4x, 42.5, 42.5x, 42.6, 42.6x) as the primary procedure during the hospitalization were considered for analysis. For AAA, only those patients who underwent repair for AAA on an elective basis were considered. For all procedures, patients who were transferred to another short-term hospital or whose discharge status was not known were not considered.

Outcome and Predictor Variables

In-hospital mortality was the outcome variable of interest. The predictor variable of interest for this study is the effect of meeting Leapfrog Group volume thresholds (met vs. unmet). The Leapfrog Group has established minimum volume criteria for individual hospitals for each of 5 complex surgical procedures (≥450 for CABG, ≥400 for PCI, ≥50 for AAA, ≥11 for PAN, and ≥13 for ESO) to designate a hospital as meeting the volume thresholds for those procedures. For computing number of cases performed in a hospital, we selected all patients who were ≥18 years of age and who underwent the procedure of interest during their hospitalization (either as primary procedure or any of the secondary procedures). For computing number of cases in a hospital for AAA, both elective and nonelective patients were included. The average number of procedures performed by the hospitals per year during the study period was computed, and hospitals were designated as either meeting or not meeting Leapfrog Group-recommended volume thresholds (Leapfrog Group evidence-based hospital referral, version 3.0).¹⁰

Statistical Analyses

Multivariable logistic regression models were built to examine the association between hospital volume (meeting or not meeting Leapfrog Group volume thresholds) and in-hospital mortality. The confounding effects of age, sex, admission type (elective vs. nonelective), comorbid severity, primary diagnosis, extent/type of primary procedure, year of procedure, hospital teaching status, and hospital bed size were controlled in the analyses. We used the Charlson Comorbid Severity Index to control for comorbid conditions.¹² Patients were categorized into 4 groups based on the Charlson Comorbid Severity Index (severity index of 0, 1, 2, ≥3). A score of 0 refers to no comorbid conditions (as defined by Charlson's Comorbid Severity Index). A score of 1, 2, or ≥3 refers to the severity scores. We used the severity index as a categorical variable and 0 was used as the reference in the analysis.

We first examined the association between procedure-specific volume and in-hospital morality for each procedure and then examined the association between unrelated procedure volume and in-hospital mortality. For example, we first examined the association between procedure volumes for CABG and in-hospital mortality following CABG. Next, we examined the association between procedure volumes for PCI and in-hospital mortality following CABG (how do hospitals not meeting the Leapfrog Group volume thresholds for PCI and hospitals that did not perform PCI but performed CABG fare when compared with hospitals that met Leapfrog Group volume thresholds for PCI). The association between procedure volumes (for elective AAA repair, PAN, and ESO) and in-hospital mortality following CABG was similarly examined. This method was repeated for all 5 procedures.

All multivariable regression models were fit using the Generalized Estimating Equation method to correct for possible clustering of similar outcomes within hospitals.¹³ An exchangeable correlation matrix was specified and empirical standard errors were used to compute 95% confidence intervals of the estimates. Two-sided P values were calculated for all analyses, and 0.05 was set as the statistical significance level.

RESULTS

The baseline characteristics of patients and results of the multivariable analyses examining the association between hospital volume and in-hospital mortality are summarized in Tables 1 and 2, respectively. In this sample, discharge information was available for 261551 CABG, 573072 PCI, 35104 elective AAA repair, 4931 PAN, and 2473 ESO cases. The overall in-hospital mortality rates for CABG, PCI, elective AAA repair, PAN, and ESO were 2.42%, 0.82%, 3.01%, 6.21%, and 7.80%, respectively. Each cell in Table 2 represents results from a separate logistic regression model. In general, the benefits of high volume were transferable across procedures to some extent. This spill-over effect was, however, restricted within broad procedure families within the Leapfrog Group-specified procedures (gastrointestinal, PAN and ESO; vs. cardiovascular, PCI, CABG, elective AAA repair). For ESO, only procedure-specific volume was significantly associated with in-hospital mortality. Hospitals not meeting the Leapfrog Group volume thresholds for ESO were associated with higher odds for in-hospital mortality when compared with hospitals that met Leapfrog Group volume thresholds for ESO (P < 0.01). For PAN, hospitals that did not meet Leapfrog Group volume thresholds for PAN as well as hospitals that did not meet Leapfrog Group volume thresholds for ESO were associated with higher odds for in-hospital mortality when compared with hospitals that met Leapfrog Group volume thresholds for PAN and ESO, respectively (P < 0.001). When examining the cardiovascular procedures as a group, there was a volume driven spill-over effect among the procedures. For CABG, PCI, and elective AAA repair, hospitals that did not meet Leapfrog Group volume thresholds for any of these 3 procedures were associated with significantly higher odds for in-hospital mortality when compared with hospitals that met Leapfrog Group volume thresholds (P < 0.05). Hospital volume levels for PAN or ESO did not significantly influence the outcomes following CABG, PCI, and elective AAA repair. Similarly, hospitals volumes for CABG, PCI, and elective AAA repair did not significantly influence the outcomes for PAN or ESO.

TABLE 1. Baseline Characteristics of Patients

TABLE 2. Association Between Procedure Volume and In-Hospital Mortality (Multivariable Analyses)

DISCUSSION

Our study provides an insight into the specificity of procedure volume–outcome associations. The study results suggest that the benefits of high volume are restricted to the specific procedure or family of procedures that affect the same organ system. Thus, more than the general structural components of care, it is the procedure-specific processes of care that are most important determinants of short-term outcomes. It is possible that procedure-specific volume is correlated with processes of care factors, such as implementing procedure-specific clinical practice guidelines or familiarity with treating postoperative complications. Another possible explanation for procedure-specific volume–outcome association could be the influence of surgeon volume in determining outcomes. It is possible that high-volume hospitals for a particular procedure are likely to have high-volume surgeons for that procedure and, consequently, experience better outcomes. Previous studies have demonstrated that high-volume surgeons have better outcomes in terms of lower mortality rates when compared with low-volume surgeons after complex surgical procedures.^14,15 The surgeon effect was more pronounced when high-volume surgeons performed procedures in high-volume hospitals.^14,15 We did not examine the influence of surgeon volumes on outcomes in this study because in the NIS sample there are variations within states and hospitals with regards to reporting surgeon identifiers.¹¹

Urbach and Baxter have previously examined the specificity of hospital volume–outcome associations for surgical procedures in Canada.⁹ They examined the association between hospital volumes and 30-day mortality following ESO, major lung resection for cancer, repair of unruptured AAA, and PAN. They found that, with the exception of colorectal resection, 30-day mortality rates were lower not only in high-volume hospitals performing the same procedure but also in high-volume hospitals of other procedures and concluded that the inverse relationship between high hospital volume and mortality is not specific to the volume of the procedure being studied.⁹ This suggests a primacy of structural components of care in determining medium-term outcomes following major surgical procedures. This also implies that the baseline level of outcomes can be boosted by greater uniformity in structural components of care across institutions. In contrast to the study by Urbach and Baxter, we found that hospital volume–outcome association is specific to the procedure being studied. However, it should be noted that our study examined different procedures and assessed in-hospital mortality while the study by Urbach and Baxter assessed 30-day mortality. Hence the difference between results our study and that of Urbach and Baxter may relate to type of procedures studied as well as the endpoint at which outcome was evaluated. Medium-term outcome may be more structure driven while short-term outcomes such as in-hospital mortality may be less sensitive to structure of care variation and more sensitive to process of care. Since we had no medium-term outcomes, our sample may have been skewed toward those who had an adverse outcome mainly due to failure of process measures such as surgical technique, postoperative complications, disease-related management, etc.

Our study results are consistent with our earlier observations suggesting that, for urologic cancer surgery, procedure-specific hospital volume is most important for short-term outcomes.⁸ We found that in-hospital mortality after radical cystectomy and radical prostatectomy is affected only by procedure-specific volumes and is independent of unrelated procedure volume. We concluded that generalized process measures existing in hospitals performing a high volume of general urologic procedures or unrelated complex procedures may be less important determinants of procedure-specific outcomes in patients.⁸

The current study has several limitations. The study uses administrative discharge data. The dataset does not have information regarding severity of the primary diagnosis or severity of comorbid conditions, thus precluding us from conducting a more robust risk adjustment. We did not adjust for race in the multivariable models because close to 30% of discharges did not provide information about race. Eleven states in the NIS sample do not release race information.¹¹ This study used in-hospital mortality as a proxy for assessing outcomes. An alternate approach is to examine postdischarge outcomes, such as 30-, 60-, or 90-day mortality. The NIS sample does not capture long-term mortality or follow-up clinical information, thus precluding us from examining these. What we are unable to ascertain from these data is whether outcomes at 30 or 90 days are just as sensitive to procedure-specific volumes or family of procedures relating to the same organ system. Finally, it should be noted that we used a large secondary database for the current study, and the statistically significant results that we demonstrate could be due to the large sample size.

CONCLUSION

The association between in-hospital mortality and hospital volume appears to be largely procedure specific. However, there could be other outcome variables, such as postoperative complications, which may not be procedure specific and demonstrate more complex interchangeable relationships across procedures.