eTable. Surgical Outcomes by Hospital Care Intensity Stratified by Procedure
eFigure. Scatterplot of Hospital Care Intensity (HCI) and Mortality, Morbidity, and Failure to Rescue for All Hospitals Represented
Sheetz KH, Dimick JB, Ghaferi AA. The Association Between Hospital Care Intensity and Surgical Outcomes in Medicare Patients. JAMA Surg. 2014;149(12):1254-1259. doi:10.1001/jamasurg.2014.552
Hospitals’ care intensity varies widely across the United States. Payers and policy makers have become focused on promoting quality, low-cost, efficient health care.
To evaluate whether increased hospital care intensity (HCI) is associated with improved outcomes following major surgery.
Design, Setting, and Participants
Using national Medicare data in this retrospective cohort study, we identified 706 520 patients at 2544 hospitals who underwent 1 of 7 major cardiovascular, orthopedic, or general surgical operations.
The HCI Index, which is validated and publicly available through the Dartmouth Atlas of Healthcare.
Main Outcomes and Measures
Risk- and reliability-adjusted mortality, major complication, and failure-to-rescue rates.
Hospital care intensity varied 10-fold. High-HCI hospitals had greater rates of major complications when compared with low-HCI centers (risk ratio, 1.04; 95% CI, 1.03-1.05). There was a decrease in failure to rescue at high compared with low-HCI hospitals (risk ratio, 0.95; 95% CI, 0.94-0.97). Using multilevel-models, HCI reduced the variation in failure-to-rescue rates between hospitals by 2.7% after accounting for patient comorbidities and hospital resources. Patients treated at high-HCI hospitals had longer hospitalizations, more inpatient deaths, and lower hospice use during the last 2 years of life.
Conclusions and Relevance
Failure-to-rescue rates were lower at high–care intensity hospitals. Conversely, care intensity explains a very small proportion of variation in failure-to-rescue rates across hospitals.
The overall intensity of medical care varies widely across the United States.1,2 Intensity, generally synonymous with aggressive treatment style, is implicated in rising health care costs. This observation is relevant during the end-of-life period, where we accrue a high proportion of overall care expenditures.3 Inpatient surgical care also represents a substantial cost burden to our health care system.4 Some posit that innovative strategies to reduce variation in health care intensity and its associated inefficiencies may reduce the cost of care we provide to our sickest patients.5 Along these lines, provisions included in the Patient Protection and Affordable Care Act provide an impetus for bundling payments for health care services in an effort to reduce variation in episode costs.6 To our knowledge, most research on aggressive treatment style to date has focused on highlighting variation in the medical management of chronic disease.3,7
However, little is known about the relationship between hospital quality and care intensity for surgical patients. This may be particularly relevant given the acuity of surgical care and the importance of immediate decisions on patient outcomes. Previous work showed a modest outcome benefit for patients treated at high care intensity centers after general, vascular, and orthopedic surgical procedures.8 The extent to which addressing hospitals’ care intensity will modify postsurgical outcomes is unknown. Studies set in the more immediate health care reform climate indicate that high-intensity care may reduce amputation rates in patients with peripheral vascular disease.9 Variation in care intensity holds important policy and financial implications. Whether efforts to increase care intensity, particularly in the management of postoperative complications, will result in measureable outcome benefits remains unclear.10
We studied the relationship between a hospital’s intensity of care based on a validated index and their outcomes following 7 common major surgical procedures in Medicare beneficiaries.3 We also sought to characterize the relationship between indicators of intensive treatment style and a hospital’s access to resources used for the care of surgical patients. These data have the potential to improve our understanding of the relationship between intensity of care and surgical outcomes, with the possibility to better inform new payment structures for surgical episodes of care.
We used data from the Medicare Provider Analysis and Review files from 2010. The Centers for Medicare and Medicaid Services maintains this database using claims submitted by hospitals where Medicare beneficiaries receive care. Patient-level data included age, sex, race/ethnicity, comorbidities (principal and secondary diagnosis codes), procedural codes, 30-day morbidity and mortality, and information on length of hospital stay. We selected patients who underwent 7 common major surgical operations using International Classification of Diseases, Ninth Revision, Clinical Modification codes. Quiz Ref IDFor this analysis, we included the following procedures: colectomy, lower extremity revascularization, lower extremity amputation, abdominal aortic aneurysm repair, coronary artery bypass graft, aortic valve repair, and hip fracture repair. We excluded patients younger than 65 years or those with incomplete clinical data in the registry.
This study was approved by the University of Michigan institutional review board; patient consent was waived.
The Dartmouth Atlas of Healthcare maintains an extensive data repository focusing on defining and understanding the wide variation of health care resource use in the United States. Using Medicare data from patients with 1 of 9 chronic conditions (malignant cancer/leukemia, dementia, diabetes mellitus with end-organ damage, congestive heart failure, chronic pulmonary disease, peripheral vascular disease, severe chronic liver disease, coronary artery disease, and chronic renal failure), the Dartmouth Atlas generates metrics of health care intensity for beneficiaries in their last 2 years of life. Quiz Ref IDFor this analysis, we used the Hospital Care Intensity (HCI) Index as our primary exposure variable.3 The HCI Index is an age-, sex-, race/ethnicity–, and illness-standardized ratio of inpatient days and physician encounters. Adjustment is accomplished using ordinary least squares regression. This ratio is then normalized to the national average to provide a relative comparison of each hospital’s intensity of care. Additional metrics of care intensity were obtained using the hospital-specific data contained within the Dartmouth Atlas’ Care of Chronically Ill Patients During the Last Two Years of Life data registry for 2010. Data on hospital structure and resources were derived from the American Hospital Association Annual Survey Database. These data included hospital bed size and occupancy, annual surgical volumes, and staffing (full-time–equivalent nursing and technician support).
Our primary outcomes for this analysis were 30-day mortality, major complications, and failure to rescue. Quiz Ref IDMajor complications were identified by International Classification of Diseases, Ninth Revision, Clinical Modification codes for the following occurrence categories: pulmonary failure (518.81, 518.4, 518.5, and 518.8), pneumonia (481, 482.0-482.9, 483, 484, 485, and 507.0), myocardial infarction (410.00-410.91), deep venous thrombosis/pulmonary embolism (415.1, 451.11, 451.19, 451.2, 451.81, and 453.8), renal failure (584), surgical site infection (958.3, 998.3, 998.5, 998.59, and 998.51), gastrointestinal bleeding (530.82, 531.00-531.21, 531.40, 531.41, 531.60, 531.61, 532.00-532.21, 532.40, 532.41, 532.60, 532.61, 533.00-533.21, 533.40, 533.41, 533.60, 533.61, 534.00-534.21, 534.40, 534.41, 534.60, 534.61, 535.01, 535.11, 535.21, 535.31, 535.41, 535.51, 535.61, and 578.9), and hemorrhage (998.1). Overall complication rates were consistent with previously published work using similar patient populations and data sets. We defined failure to rescue as mortality in patients with at least 1 major complication (ie, the case fatality rate for patients sustaining a major complication), as has been previously described.10
We compared demographic, comorbidity, and operative differences between hospitals with the t test, χ2 test, and Fisher exact test, as appropriate. We compared differences between HCI as a continuous variable and postoperative outcomes in bivariate analysis using Pearson correlation coefficient. Hospitals were compiled into 3 groups based on HCI: low, average, and high intensity. We stratified reporting of all adjusted outcomes by these categories to provide a generalizable comparison of hospitals based on their relative care intensity.
We constructed 3 separate logistic regression models using patient demographics, comorbidities, urgency of operation, and procedural factors to generate risk-adjusted rates of mortality, major complications, and failure to rescue for each hospital. Next, we used hierarchical logistic regression modeling to adjust all outcome rates for reliability to account for hospital-level random effects.11 Reliability adjustment reduces statistical noise that can result from hospitals with lower surgical case volumes. The c statistic for all models ranged between 0.73 and 0.88, with good discriminatory power on the basis of the Hosmer-Lemeshow test. Outcome rates for low–, average–, and high–care intensity hospitals were calculated using the individual hospital’s risk- and reliability-adjusted rates.11 We then calculated risk ratios (RRs) and 95% CIs for mortality, major complications, and failure to rescue using low care intensity hospitals as a reference category.
For each outcome, we conducted model testing to determine the relative contribution of patient-level covariates, hospital structural factors, and hospital care intensity to the variation observed among hospitals. We first quantified the variance ascribed to hospital-level random effects using an empty mixed-effects logistic regression model (xtmelogit in Stata version 12.1). We subsequently generated linear predictors of the outcome in question using patient-level covariates first. We sequentially added hospital structural factors and HCI, each time generating a new linear predictor. The relative decrease in variance attributed to hospital-level random effects was then calculated to determine each parameter’s influence on observed variation.
Finally, we compared alternative metrics for HCI across low-, average-, and high-intensity centers, as identified by the HCI Index. We used the t test and Mann-Whitney U test to compare differences between high– and low–care intensity hospitals. We also conducted a sensitivity analysis using 2 alternative proxies for intensity of care (average inpatient Medicare spending and percentage of deaths occurring in the hospital). These metrics are compiled for Medicare beneficiaries with chronic illness, as just defined, in the last year of life. Hospitals were similarly stratified into categories of low, average, and high care intensity and outcomes were compared in an identical fashion, as just described using our primary exposure, HCI.
A significance level of α = .05 was used. All statistical analyses were performed using Stata statistical software version 12.1.
Hospital care intensity varied widely across hospitals from 0.35 to 3.41. Consistent with its characteristic as a normalized ratio, the mean (1.04) and median (0.98) HCI for the entire cohort of 2544 hospitals contributing to this analysis was near 1.0. As expected, low–care intensity hospitals had a significantly lower HCI than high–care intensity hospitals (0.74 vs 1.41; P < .001). We identified 706 520 patients who underwent 1 of 7 designated procedures. Patient characteristics did not differ significantly across low–, average–, and high–care intensity hospitals with these 7 operations (Table 1). The single exception to this was nonwhite race, which was nearly twice as high at high- compared with low-intensity hospitals (P < .001). In contrast, hospital structural factors were significantly different across categories of care intensity. For example, high–care intensity hospitals had significantly greater average daily censuses, inpatient surgical case volumes, and full-time–equivalent technical support (P < .001 for all). Procedural volumes for the entire cohort were as follows: colectomy (n = 121 560), lower extremity revascularization (n = 169 930), lower extremity amputation (n = 48 658), abdominal aortic aneurysm repair (n = 41 327), coronary artery bypass graft (n = 128 510), aortic valve repair (n = 46 388), and hip fracture repair (n = 150 147). We found no differences in the overall procedural mix across low–, average–, and high–care intensity hospitals. The median length of stay for all patients was 6 days (interquartile range, 7 days).
The unadjusted 30-day mortality rate for the entire patient cohort was 6.4%. We observed the lowest unadjusted mortality after coronary artery bypass grafting (3.8%) and the highest after lower extremity amputation (9.8%). The unadjusted major complication rate for all patients was 31.2%. The lowest morbidity rates were observed after coronary artery bypass grafting (18.8%), whereas the highest rates were observed after hip fracture repair (44.8%).
We first assessed the relationship between HCI and postoperative outcomes in bivariate analysis. With HCI treated as a continuous variable, we assessed the linear correlation between HCI and mortality (r = 0.024; P = .22), major complications (r = 0.163; P < .001), and failure to rescue (r = −0.052; P = .008). These results are graphically displayed in scatterplot format (eFigure in the Supplement). Hospital care intensity was significantly associated with major complications and failure to rescue. However, the strength of this relationship was weakly positive for major complications and very weakly negative for failure to rescue.
Next, we assessed the relationship between HCI and outcomes in multivariate analysis. Risk- and reliability-adjusted mortality rates for low–, average–, and high–care intensity hospitals are reported in Table 2. There were no differences in postoperative mortality across low–, average–, or high–care intensity hospitals. Quiz Ref IDWe observed a small, but statistically significant, increase in major complication rates for patients who underwent surgery at high– vs low–care intensity hospitals (RR, 1.04; 95% CI, 1.03-1.05). In contrast, patients who underwent operations at high–care intensity centers were 5% less likely to die in the setting of a major complication (failure to rescue) (RR, 0.95; 95% CI, 0.94-0.97). We repeated all analyses within each of the 7 distinct procedural categories and obtained similar outcomes (eTable in the Supplement). Furthermore, we conducted a sensitivity analysis using average inpatient Medicare spending and percentage of deaths occurring in the hospital as alternative proxies for HCI. When stratifying centers by these variables, we obtained nearly identical results, indicating a high level of correlation between the HCI summary measure and other putative surrogates for intensive treatment styles. The results of model testing indicated that the addition of HCI to the multivariate model minimally reduced the magnitude of between-hospital variation in outcomes (Table 3). For example, the addition of HCI to the multivariate model reduced the magnitude of variation in failure-to-rescue rates between hospitals by 18.2% (patient and hospital structural factors) to 20.9% (including HCI).
Finally, we evaluated the relationship between HCI and other measures of care intensity and resource use. We compared overall differences in several metrics across low–, average–, and high–intensity hospitals as designated by HCI (Table 4). All outcomes were calculated for Medicare beneficiaries in the last 2 years of life. In general, we observed significantly higher overall and inpatient Medicare spending. Quiz Ref IDPatients treated at high–care intensity hospitals had more physician contact and, on average, spent more days in the hospital and intensive care unit. In contrast, patients treated at high–care intensity hospitals were less likely to be enrolled in hospice and spent fewer days in hospice when compared with low–care intensity hospitals.
While previous studies have shown wide variation in the intensity of medical care provided by hospitals, to our knowledge, few have explicitly addressed the relationship between aggressive treatment style and patient outcomes with surgery. We investigated the relationship between HCI and outcomes in Medicare beneficiaries after 7 common operations. We observed a small, but statistically significant, increase in the rates of major complications for patients treated at high– compared with low–care intensity centers. In contrast, failure-to-rescue rates were lower at high–care intensity hospitals, potentially indicating differences in complication management compared with low–care intensity centers. Despite this, HCI explained a small proportion of the overall variation in failure-to-rescue rates across hospitals. We also showed that HCI, defined by the HCI Index, is highly correlated with per-patient health care expenditures, inpatient care, and hospice use practices.
These data have significant implications for surgeons given the increasing age and preexisting disease burden of today’s surgical patients.12 Managing these patients and their complications imposes substantial demands on care teams and the financiers of their care.13 Previous work has also shown that many elderly decedents undergo operations in their last year of life, suggesting that aggressive treatment styles are not tempered by manifestations of advanced disease.14 Several studies in the medical oncology literature supply additional evidence for this observation.7,15 Furthermore, work specifically addressing this question in surgical patients has rendered differing results as to the benefits of high-intensity care.8,9 To date, comparability of these studies has been limited by heterogeneity in the definitions of high-intensity care. This is likely owing to differences in patient populations, available data sources, and no consensus metric for a hospital’s care intensity. As health care reform attempts to modify and streamline payment structures, it will be important for surgical care teams to understand the benefits and drawbacks of their efforts to treat patients with surgical problems.
Medicare payments for many common inpatient surgical procedures already vary widely by region.16 At present, it is unclear how much of this variation is influenced by differences in the intensity of care. As provisions of the Patient Protection and Affordable Care Act move toward bundled payments for many surgical procedures, there may be a growing need for surgeons and hospitals to understand what aspects of their care truly impact patient outcomes.6 This is particularly relevant given estimations that inpatient surgical care represents 11% to 19% of our total health care expenditures.4 Further movement toward patient-sharing networks of physicians will only increase the impetus for understanding necessary and appropriate levels of care for surgical patients.17 However, some posit that current fee-for-service payment structures will remain a prominent entity in reimbursement for surgical procedures.18 Our results conveyed a moderate outcome benefit to high care intensity in managing surgical patients. An intuitive next step is to study specific aspects of practice that differ between high- and low-intensity-of-care hospitals. It is expected that some practice patterns more prevalent at high-intensity hospitals could promote effective management or rescue from major complications. At the same time, there are also practices that render no outcome benefit for patients, increase inefficiency, and drive higher costs. The distinction between beneficial vs unnecessary practices is critical and will likely require the expertise and collaboration of clinical surgeons, palliative care specialists, nurses, and ethicists. Although these are logical next steps, surgeons must also consider that HCI underlies a relatively small amount of variation in postoperative outcomes. Thus, enthusiasm for care intensity as a failure-to-rescue countermeasure should be tempered.
This study had several limitations and it is important to note that this work does not suggest causation. The use of Medicare data restricts this study to a particular patient population, which may reduce generalizability. Furthermore, the use of administrative data imposes some limitation on adequate risk adjustment.19 However, confounding from unmeasured patient factors would generally bias our results further toward the null hypothesis. We also attempted to address possible inaccuracies in coding of complications by restricting our analysis to a validated subset of events known to have high sensitivity and specificity.20 We did observe differences in patient race/ethnicity between high and low care intensity centers. It is possible that these differences could alter our results given that disparities in outcomes exist for certain minority groups. Furthermore, it is unclear why more nonwhite patients were treated at high HCI centers. This represents a potentially important observation for future investigation. Because HCI is calculated for patients admitted to the hospital with chronic medical conditions, it is plausible that these care practices may not be reflected in surgical patients. However, we deliberately selected operations performed predominately for end-stage management of chronic conditions (eg, bypass for coronary artery disease). Additionally, we have no reason to believe that a hospital’s resource use would be significantly different for medical vs surgical patients. Finally, it is possible that the HCI is not an appropriate marker for true care intensity. We attempted to address this in comparing this measure to other surrogates for aggressive treatment styles. At present, to our knowledge, there are no detailed means by which care intensity can be profiled on the national scale targeted in this analysis.
Our findings have important implications for surgeons, in addition to payers and policy makers. Hospital care intensity has an independent influence on established quality metrics for surgical care, although its ability to improve quality through direct augmentation appears limited. The wide differences in care-intensity metrics is in many ways analogous to the variation in objective outcomes (ie, complications or failure to rescue) that continue to be the target of quality-improvement programs on the local and national scales. As health care reform pushes for more efficient care, it will be critically important for surgeons to understand when and how resources and effort are met with tangible benefits to the patients. Changes in hospital culture (eg, unconscious caregiver attitudes and behaviors) and climate (eg, conscious perceptions of hospital leadership and authority) may provide greater leverage to reduce failure-to-rescue rates within hospitals. Measuring and modifying these elements will require collaboration with engineering and systems-analysis professionals as these skill sets are uncommon in the medical field. Ultimately, payment-reform strategies should be informed by the clinical evidence for and against the specific care practices that underlie the vast differences in treatment styles observed across the United States.
Corresponding Author: Kyle H. Sheetz, MD, MS, Center for Healthcare Outcomes and Policy, 2800 Plymouth Rd, Bldg 16, Floor 1, Ann Arbor, MI 48109 (firstname.lastname@example.org).
Accepted for Publication: February 19, 2014.
Published Online: October 1, 2014. doi:10.1001/jamasurg.2014.552.
Author Contributions: Drs Sheetz and Ghaferi had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: All authors.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: All authors.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Sheetz, Dimick.
Obtained funding: Dimick.
Administrative, technical, or material support: Dimick.
Study supervision: Dimick, Ghaferi
Conflicts of Interest Disclosures: Dr Dimick serves as a consultant for and has an equity interest in ArborMetrix Inc, a venture capital–backed company that provides software and analytics for measuring and improving hospital quality and efficiency. The company had no role in the study herein. No other disclosures were reported.
Disclaimer: The views expressed herein are those of the authors and do not necessarily represent the views of the US government. Dr Dimick is on the JAMA Surgery editorial board but was not involved in the review process or the acceptance of the manuscript.