CORR Insights®: Are Case Volume and Facility Complexity Level Associated With Postoperative Complications After Hip Fracture Surgery in the Veterans Affairs Healthcare System?

Where Are We Now? The current study by Wong and colleagues [11] has broad implications for health policy as we continue to look for more-effective and efficient models of care. Evaluating how case volume and facility complexity impacts postoperative hip-fracture surgery has drawn considerable attention, with studies finding that increased surgical volume at the surgeon level (as opposed to the facility level) reduces morbidity and mortality following orthopaedic surgery [6, 8-10]. However, studies focusing on the impact of hospital/surgeon volume and facility complexity on hip fractures often disagree with one another, with some indicating that higher volume/higher complexity facilities produce better results, and others finding the opposite [2, 3, 7]. This is important because hip-fracture surgery is among the most-common nonelective orthopaedic operations, with 300,000 performed each year in the United States [1, 4]. In networked care, complex surgery can be echeloned. If increased surgical volume and facility complexity indeed influence the safety of these operations, patients then could be transferred to higher-volume centers. However, if surgical volume is not an important factor, then healthcare systems might reasonably decide not to deploy resources in the transfer of care for what would be a large number of patients. That gap in knowledge left the door open for the current study [11], which reviewed 11 years of data from the Veterans Affairs Surgical Quality Improvement Program (VASQIP) database from VA facilities on the West coast of the United States. The authors found no correlation between surgical volume or facility complexity level and early complications following hip-fracture surgery. This was used to support their conclusion that patients who are treated with hip fracture surgery at high- and low-volume/complexity facilities in the VA should expect similar risk of 30-day adverse outcomes. Adding some external validation to this work, Okike and colleagues [7] came to the same conclusion after reviewing results within the Kaiser-Permanente system from 14,294 patients treated for hip fracture. Where Do We Need To Go? The primary limitation of large-database research is that it lacks the ability to answer why an association (if any) exists. Merely finding an association (or lack thereof, as was the case in the current study) does not, and more importantly should not, define best practice. Good health policy requires more than associations; it depends on thoughtful and convincing explanations. For example, in this study, the incidence of tracked complications was not different between higher- and lower-volume or lower-complexity facilities [11]. This finding alone can be interpreted in multiple ways. It could mean that experience does not matter, and all facilities perform hip fracture surgery with a comparable level of safety, or it could mean that higher-volume or higher-complexity facilities care for patients who have more medical comorbidities and by virtue of the fact that there is no increase in the frequency of complications, these sites actually may be providing considerably better care. One way to adjudicate this would be to review the charts of all cases with complications. However, this is almost never possible with administrative databases, which typically deidentify records. Using retrospectively applied comorbidity scores to overcome this limitation relies on the adequacy of comorbidity recording, which can be unreliable [5]. As such, we are left without an answer to the “why?” Turning to the full medical record and imaging could enable an explanation for this finding. The “why” has and will always be locked in the full dataset. Administrative database studies are complimentary to, but cannot replace, hypothesis-driven research. In addition, many large-database studies are limited in terms of the endpoints they can evaluate; by definition, they can only explore those endpoints that are available, which typically are limited to basic demographics, a limited number of coded diagnoses and procedures (which themselves can be subject to coding error), and adverse outcomes incurred over a prespecified time period [5]. These elements alone can identify research questions, but rarely can they answer them. By contrast, prospective, hypothesis-driven research is limited in its ability to deal with infrequent events, such as rare complications; it would take an extraordinarily large randomized study, for example, to determine whether mixing antibiotics into bone cement during hip hemiarthroplasty surgery causes more help than harm. Our best evidence is likely to come from a combination of large-database, registry, and high-quality prospective research. How Do We Get There? Given the prevalence of hip fracture surgery in the VA and general population, prospective, validation of these findings is needed. For events that occur with sufficient frequency, such as 1-year mortality, researchers should try to use prospective methods and formal hypothesis testing, rather than large-database research. To keep sample size manageable, rare events—those in the range of 1% to 2% in terms of frequency—could be aggregated as “any major adverse event” to help generate greater expected occurrences. A cost of this aggregation is that it may pool complications or mask correlations that exist between a specific treatment approach and complication. For these reasons, while there is a clear hierarchy between prospective hypothesis-driven clinical research and large-database studies, the two can be complimentary. Independently assessing rare events could be too onerous for hypothesis-driven methods. As such, registries or prospectively maintained healthcare data, akin to the VASQIP, may be the only pragmatic option. However, in these circumstances, it is essential to assess the impact of confounders, such as comorbidities, demographics, and facility-specific/surgery-specific factors. Ideally, the database would have a means for identifying patients who sustain a rare occurrence, which would then allow direct full chart and imaging review of these unique patients. This enhanced method overcomes the limitations of studies that rely solely on data prerecorded in the database. Administrative database studies have a place in our evidence base, but their findings must be understood in the context of their limitations.