After adjusting for differences in health status and disease severity, patients treated at low volume hospitals were nearly 50% more likely to die in the postoperative period. We observed substantial disparities in the use of processes of care by hospital volume for all phases of patient care. With regard to most processes, high volume hospitals had substantially higher rates of utilization, exhibiting a more intensive practice style overall. Cumulatively, these processes had a moderate effect in attenuating the volume-mortality effect among those undergoing cystectomy, accounting for 23% of the variation in operative mortality.
The processes of care that varied most by volume included those related to the extent of the resection (e.g., lymphadenectomy) and invasive monitoring (e.g., central venous catheter). The beneficial effect of some of these processes is supported in the literature. For example, extensive pelvic lymphadenectomy at the time of radical cystectomy improves long-term survival.8
Preoperative urinary diversion (among patients with acute renal failure) and epidural anesthesia (at the time of cystectomy) appear to reduce postoperative morbidity.15
Thus, the more frequent use of these processes at high volume hospitals may reflect better quality of patient care. While other processes characteristic of high volume hospitals lack high level evidence supporting their effectiveness, many have face validity in elderly patients undergoing high risk surgery. These include processes related to preoperative risk stratification (e.g., cardiac stress testing), and perioperative invasive monitoring (e.g., pulmonary artery catheter).
However, it is unclear whether the observed differences in process are part of the causal pathway for the volume-outcome relationship or are merely proxies for factors that more directly improve patient outcomes. Under the former scenario, exportation of some processes of care to lower volume hospitals could lead to downstream improvements in operative mortality. If the latter scenario holds and the greater use of invasive monitoring simply reflects, for instance, more “careful” physicians, then disseminating these processes of care would not likely translate into quality improvement.
These findings should be interpreted with a few caveats. First, this study involved Medicare patients only, and the generalizability of our findings to a younger population is uncertain. However, bladder cancer is generally a geriatric disease, with approximately three-quarters of cases occurring within the U.S. Medicare population.16
Our recent analysis of data from the Nationwide Inpatient Sample suggests that Medicare patients represent approximately half of U.S. patients undergoing cystectomy (unpublished data). Second, because we estimate hospital volume status using Medicare data instead of all-payer data, our classification of hospital volume may be imprecise.11
However, our previous work has demonstrated that volume estimates based on Medicare and all-payer data are highly correlated for most procedures. Further, since errors in volume categorization are likely random, any misclassification would likely bias our findings toward the null hypothesis.
The third and perhaps most important limitation of this study relates to the ability of administrative data to adequately adjust for differences in case-mix.17
To address this limitation, we used well-described methods to account for differences in comorbidity13
and controlled for other important predictors of survival after cancer surgery, including age, race, gender, socioeconomic status, cancer stage and grade, and admission acuity. While more detailed measures of patients’ health status and demographics may improve our ability to risk adjust, such would require a large cohort study or clinical trial which is not possible for practical reasons (e.g., cost, sample size). At this time, the direction of the bias related to imperfect risk adjustment is unclear in the bladder cancer population.
Although the processes of care measured in this study play a moderate role in explaining the hospital volume effect among patients undergoing cystectomy for bladder cancer, a considerable component remains undefined. Differences in operative mortality are likely explained by processes of care not readily captured by administrative data. These may include those related to surgeon proficiency and expertise, above and beyond that measured by volume.5
In addition, other aspects of perioperative care not reflected in administrative data, such as the use of beta blockers or other medications, may play a role as well. For this reason, efforts dedicated to measuring what happens in the operating room and those focused on widening the scope of the measured clinical data would be of considerable utility.