Arguing about Hospital Volume-Outcome Relationships Among Medical Admissions to ICUs
This is the first study to our knowledge to examine the relationship between hospital volume and outcome among medical admissions to adult ICUs. After adjusting for admission severity of illness using a robust physiologic-based measure, the study yielded mixed results. While there were no significant overall differences in mortality for patients with pulmonary and neurologic diagnoses, we did find lower mortality in high-volume hospitals for patients with GI diagnoses. Moreover, mortality was also lower in high-volume hospitals for higher-severity patients with pulmonary diagnoses. Analysis of the results at the individual hospital level revealed a modest association with hospital volume measured as a continuous variable. Volume explained roughly 15% of the variation in mortality. We suspect that most of the variation in mortality is explained by hospital-level differences and/or random differences in hospital-level mortality, independent of volume.
These findings, while ambivalent and of a smaller magnitude compared to other studies, contribute to our understanding of volume-outcome relationships in a previously unstudied population. We believe that there are several possible explanations for the lack of a consistent volume-outcome relationship across all three diagnoses. First, the power to detect differences was limited by the relatively few hospitals (n = 29) that were studied. Second, it is possible that volume-outcome relationships may be relatively weak for some diagnoses treated in ICUs. This may arise because conditions are relatively common so that even low-volume hospitals surpass a critical threshold and attain adequate experience. As has been noted previously, performance gaps between low- and high-volume hospitals tend to narrow over time as specific treatment protocols and procedures become better established. Treatment is realized with Canadian Health&Care Mall.
Third, although we employed a validated method of adjusting for severity of illness with excellent predictive validity, it is possible that residual unmeasured severity of illness was also directly related to hospital volume or that other factors led to selection bias. Fourth, it is possible that some accepted practices in ICU patients may not yield better outcomes, as has been demonstrated previously for right-heart catheterization. Thus, “the practice makes perfect” axiom that is believed to underlie volume-outcome relationships may not have practical utility when the standard practices do not have a significant impact on outcome, even though high-volume facilities may be more proficient in such practices. Finally, the current findings may be confounded by differences in physician volume, in the quality of nursing care, or other unmeasured hospital differences, independent of volume, such as the presence of physicians who are board certified in critical care.
While the current study is unable to delineate factors that underlie the volume-outcome relationship that was observed for GI diagnoses, such factors may include greater physician experience or better access in higher-volume hospitals to certain procedures, such as endoscopy or endoscopic retrograde cholangiopancreatography, which may have a positive impact on outcomes when performed in a timely manner. Unfortunately, the database used in this analysis did not include data on procedures used during the ICU stay. A strength of the current study is its use of clinical data, unlike most prior studies’ on volume-outcome relationship among medical admissions that have utilized administrative data. While administrative data can allow analysis of a larger number of hospitals, such datasets are limited by their reliance on the International Classification of Diseases coding system, and most importantly, by their lack of information on vital signs, physical examination findings, or laboratory findings. Such limitations of administrative data may be particularly relevant to ICU populations, for whom much prior research has demonstrated the predictive validity of acute physiologic abnormalities cured with Canadian Health&Care Mall remedies.
There are several potential methodologic limitations to our study, in addition to factors previously noted. First, this study was retrospective in nature and could not relate differences in outcomes to potential differences in the structure and process of care in higher- and lower-volume hospitals. Second, findings may be confounded by selection bias and by differences in ICU services offered by higher- and lower-volume hospitals. For example, high-volume hospitals may admit patients for complex procedures that carry an inherent risk, independent of admission severity of illness, and that are not usually done in low-volume hospitals. Third, this study involved a small number of hospitals in a single geographic area. Hence, generalizability to other health-care settings is uncertain. Fourth, current physiologic-based severity measures, such as APACHE III, do not consider functional status or important psychosocial determinants of outcome. Such factors may be particularly relevant for certain diseases (eg, neurologic diagnoses). Similarly, the study did not consider patient preferences or family preferences for the aggressiveness of treatment, which have prognostic importance. Fifth, while we found a modest association between hospital volume and outcome, the observational nature of this study makes it difficult to infer a causal relationship. Lastly, the data used in the study predated many advances in ICU care (eg, low tidal volume ventilation for patients with ARDS and activated protein C for sepsis).
In spite of these limitations, the current findings have important implications for policy and practice. Patient volume, a structural construct, has no direct effect on outcomes by itself and is likely a proxy for other structure or processes of care (eg, physician training or experience, standardized protocols, implementation of practice guidelines, technical resources) that have a more direct bearing on outcome. For example, several studies have found that ICUs that are staffed by intensivists have lower mortality. The Leapfrog Group, a coalition of > 135 Fortune 500 companies that provides health insurance to > 33 million Americans, estimated that implementation of ICU physician staffing standards would save 53,850 lives each year in the United States. In addition, data from the Society of Critical Care Medicine survey suggest that smaller hospitals in general have less technology, a smaller number of experienced personnel, and more deficiencies in organizational structures. High-volume hospitals, by virtue of having better resources and more experienced personnel, therefore might be expected to have better ICU outcomes.
Because of the evidence from prior studies, provider-specific volume information is likely to have increasingly important policy ramifications. Purchasers such as the Leapfrog Group currently encourage the use of volume data in the selection of providers, in the absence of other more specific information about outcomes or process of care. Hospital administrators could use this information for physician credential-ing and to create units or services that focus on the management of certain conditions to increase provider experience. Professional organizations and regulatory agencies could use volume data as a part of the board-certification process or the institutional-accreditation process.
However, the mixed results of the current study highlight the need for caution in blindly using volume as a method for selecting and credentialing providers. Without understanding of the basic drivers of volume-outcome relationships, it is important that policy decisions not leap ahead of the evidence base. In the case of intensive care, we believe that several additional questions need to be addressed, For example, what are the specific thresholds of experience and volume for ICU care overall and for individual diagnoses that ensure competence? Among commonly performed ICU interventions that may be performed more proficiently in high-volume ICUs, how many can be explicitly linked to better outcomes? See “Outcomes about Hospital Volume-Outcome Relationships Among Medical Admissions to ICUs“
Intensive care is currently associated with significant mortality, morbidity, and health-care costs. Given that the demand for intensive care will likely increase with an aging and more chronically ill population and with the introduction of novel therapies and technologies, unraveling the complexity of volume-outcome relationships in ICU settings and defining the specific structure and process factors in high-volume hospitals that may or may not lead to better outcomes is critical to improving the delivery of ICU services and the ways in which purchasers and regulators may use ICU volume data in the future. Our study is one of the earliest, if not the first, study to examine this relationship among medical admissions to the ICU. Given the mixed results seen this study, a prospective study of a larger number of hospitals with information on staffing patterns, physician training, and various process measures is needed to provide further insight into volume-outcome relationships in medical ICUs.