Comparative Effectiveness of Regional versus General Anesthesia for Hip Fracture Surgery in Adults

Background: Hip fracture is a common, morbid, and costly event among older adults. Data are inconclusive as to whether epidural or spinal (regional) anesthesia improves outcomes after hip fracture surgery. Methods: The authors examined a retrospective cohort of patients undergoing surgery for hip fracture in 126 hospitals in New York in 2007 and 2008. They tested the association of a record indicating receipt of regional versus general anesthesia with a primary outcome of inpatient mortality and with secondary outcomes of pulmonary and cardiovascular complications using hospital fixed-effects logistic regressions. Subgroup analyses tested the association of anesthesia type and outcomes according to fracture anatomy. Results: Of 18,158 patients, 5,254 (29%) received regional anesthesia. In-hospital mortality occurred in 435 (2.4%). Unadjusted rates of mortality and cardiovascular complications did not differ by anesthesia type. Patients receiving regional anesthesia experienced fewer pulmonary complications (359 [6.8%] vs. 1,040 [8.1%], P < 0.005). Regional anesthesia was associated with a lower adjusted odds of mortality (odds ratio: 0.710, 95% CI 0.541, 0.932, P = 0.014) and pulmonary complications (odds ratio: 0.752, 95% CI 0.637, 0.887, P < 0.0001) relative to general anesthesia. In subgroup analyses, regional anesthesia was associated with improved survival and fewer pulmonary complications among patients with intertrochanteric fractures but not among patients with femoral neck fractures. Conclusions: Regional anesthesia is associated with a lower odds of inpatient mortality and pulmonary complications among all hip fracture patients compared with general anesthesia; this finding may be driven by a trend toward improved outcomes with regional anesthesia among patients with intertrochanteric fractures.

logistic regressions. Subgroup analyses tested the association of anesthesia type and outcomes according to fracture anatomy. Results: Of 18,158 patients, 5,254 (29%) received regional anesthesia. In-hospital mortality occurred in 435 (2.4%). Unadjusted rates of mortality and cardiovascular complications did not differ by anesthesia type. Patients receiving regional anesthesia experienced fewer pulmonary complications (359 [6.8%] vs. 1,040 [8.1%], P Ͻ 0.005). Regional anesthesia was associated with a lower adjusted odds of mortality (odds ratio: 0.710, 95% CI 0.541, 0.932, P ϭ 0.014) and pulmonary complications (odds ratio: 0.752, 95% CI 0.637, 0.887, P Ͻ 0.0001) relative to general anesthesia. In subgroup analyses, regional anesthesia was associated with improved survival and fewer pulmonary complications among patients with intertrochanteric fractures but not among patients with femoral neck fractures. Conclusions: Regional anesthesia is associated with a lower odds of inpatient mortality and pulmonary complications among all hip fracture patients compared with general anesthesia; this finding may be driven by a trend toward improved outcomes with regional anesthesia among patients with intertrochanteric fractures. F RAGILITY fractures of the hip present an urgent need for better information to guide perioperative care. Hip fractures are a global public health problem, occurring 1.6 million times worldwide each year, and their incidence is anticipated to grow rapidly during the next three decades because of the aging of the population. 1-5 After indicated surgical treatment, hip fracture patients experience high rates of morbidity, mortality, and disability, with approximately 5% dying during hospitalization and 10% dying within 30 days because of high rates of pulmonary and cardiovascular complications. 6 -9 Few interventions exist to reduce mortality after hip fracture, 10,11 so identifying new opportunities to improve outcomes in this population is of urgent public health importance. Use of regional anesthesia, via epidural, spinal, or peripheral neural blockade, as a principal anesthetic technique has been hypothesized to reduce the risks of postoperative complications among patients undergoing surgery for hip fracture. 10,12 Proposed reasons for improved outcomes with regional anesthesia include the avoidance of intubation and mechanical ventilation, decreased blood loss, and improved postoperative analgesia. 13 Conversely, general anesthesia may offer improved hemodynamic stability relative to regional anesthesia. 13 Past clinical trials comparing outcomes of regional versus general anesthesia for hip fracture offer insufficient insights to guide current practice because of small sample sizes 14 -19 and exclusion of key patient groups, including those with delirium or dementia, 14,20 and those undergoing hemiarthroplasty or total hip arthroplasty. 21 In this context, research using observational data may offer unique strengths for comparing the effectiveness of anesthesia type for hip fracture surgery. 22,23 Nonetheless, past observational studies have produced conflicting results regarding the association between anesthesia type and mortality after hip fracture, alternately showing no difference in outcomes according to anesthesia type 24 and decreased mortality with regional anesthesia. 8 In addition, little is known about how associations between anesthesia type and outcomes may differ among clinically relevant subgroups of hip fracture patients.
To compare the effectiveness of regional versus general anesthesia for preventing mortality and major complications after hip fracture surgery, we undertook an analysis of a large, population-based cohort. Specifically, we aimed to compare the probability of in-hospital death and selected pulmonary and cardiovascular complications among hip fracture patients receiving regional versus general anesthesia, and to evaluate whether the association between anesthesia type and outcome varied according to fracture anatomy. Our overall hypothesis was that the probability of inpatient mortality, major pulmonary complications, and major cardiovascular complications would be lower among patients receiving regional techniques as their principal anesthetic modality; we also hypothesized that the association of anesthesia type with outcomes would differ between patients with femoral neck fractures and those with intertrochanteric fractures.

Data Sources and Study Sample
This study was exempted from review by the Institutional Review Board of the University of Pennsylvania, Philadelphia, Pennsylvania. We examined data on adults aged 50 yr and older undergoing hip fracture repair at hospitals in New York state between January 1, 2007, andDecember 31, 2008, made publically available through the U.S. Agency for Healthcare Quality's Health Care Utilization Project as the New York State Inpatient Database. This database includes discharge records for all patients treated in general acute-care hospitals in the state of New York and contains information on patient demographics, discharge diagnoses, inpatient procedures, anesthesia type, discharge status, and hospital identifiers.
To create a cohort of adults undergoing surgery for hip fracture, we selected all discharges with a principal or secondary diagnosis code for hip fracture, including pathologic fractures of the femoral neck (International Classification of Diseases-9-Clinical Modification [ICD-9-CM] diagnosis codes 820.00 -820.9, 733.14); within this group, we excluded patients without a principal procedure code corresponding to open reduction, internal fixation, hemiarthroplasty, or total hip arthroplasty (ICD-9-CM codes 00. 70 -7; 79.15, 79.25, 79.35, 81.40, 81.51-3). We excluded patients undergoing closed reduction without internal fixation (79.05). To restrict our sample to a group of patients with low-energy fragility fractures, we excluded patients younger than 50 yr and those presenting with a diagnosis-related group code indicating multiple trauma (diagnosis-related groups 280, 418, 444 -5, 484 -7, 506, 508, or 510).

Validation of Exposure Variable
The New York State Inpatient Database has made data on anesthesia type available since 1994, as collected and reported by individual hospitals. Anesthesia type is reported as a categorical variable with values corresponding to general, regional, local, other, or no anesthesia. Each discharge record in the New York State Inpatient Database contains a maximum of one value for anesthesia type; anesthesia care for patients receiving more than one type of anesthesia during their hospital stay is reported in the following hierarchical order: general, regional, other, and local. Thus, patients receiving both general and regional anesthesia for a single procedure or across multiple procedures within one hospitalization are listed as having received general anesthesia. For the purposes of this analysis, we confined our comparisons of outcomes to patients treated in hospitals where at least one hip fracture discharge record included a valid anesthesia type code; we also examined only those patients listed as receiving either general or regional anesthesia, excluding patients coded as receiving other, local, or no anesthesia. We also excluded patients who underwent a secondary surgical procedure likely to require general anesthesia during their hospital stay (appendix 1) because the anesthesia type received for hip fracture surgery could not be determined reliably for these patients because of the coding structure of the study data set.
The anesthesia type variables in the New York State Inpatient Database have been used in previous efforts to define predictors of postoperative outcomes. 25 However, we are not aware of previous research validating these variables, so our initial analyses explored the validity of anesthesia coding within the study data set. Notably, the New York State Inpatient Database lacks patient-specific identifiers or calendar dates, precluding validation by comparison with external chart review. Thus, we initially assessed the face validity of the database's anesthesia type codes by qualitatively examining those ICD-9-CM procedure codes most frequently associated with regional and general anesthesia within the data set.
Next, we examined the distribution of missing and invalid anesthesia type codes for patients meeting our inclusion criteria across all hospitals in our set. We compared hospital characteristics, as recorded in the 2006 American Hospital Association member survey, annual hip fracture volume, selected patient characteristics, and clinical outcomes, between hospitals in which all anesthesia codes were missing or invalid and all remaining hospitals in our sample.

Outcome Variables
The primary outcome for the current study was in-hospital mortality as recorded in the study database. Secondary outcomes included major pulmonary and cardiovascular complications, which represent common causes of postoperative mortality among hip fracture patients 6 with previously hypothesized mechanistic associations with anesthesia type. 13 We examined the following complications: pneumonia/ empyema, aspiration, respiratory failure, acute myocardial infarction, congestive heart failure, and cardiac arrest. In addition, we created indicator variables for occurrence of any pulmonary complication (i.e., pneumonia/empyema, aspiration, or respiratory failure) and occurrence of any cardiovascular complication (i.e., acute myocardial infarction, congestive heart failure, or cardiac arrest). These variables were equal to one if one or more complications within the relevant organ system (i.e., cardiac or pulmonary) occurred, and zero otherwise. We identified complications using ICD-9-CM diagnosis and procedure codes based on algorithms originally outlined by Romano et al. 26 and Silber et al. 27 (appendix 2). The New York State Inpatient Database contains a presenton-admission indicator to distinguish preexisting conditions from complications developing during hospitalization. Past research has demonstrated that use of present-on-admission indicators can minimize bias in assessments of hospital outcomes by distinguishing between preexisting conditions and complications 28 -30 ; thus, we deemed a complication to have occurred if a discharge record contained a relevant secondary ICD-9-CM code not identified as being present on admission.

Control Variables
Control variables for patient sex, age in years, and race as reported by the hospital were taken directly from the New York State Inpatient Database. Race was categorized as black, white, or other 31 and examined as a potential confounder because of known differences in hip fracture treatment and outcomes occurring as a function of race. 32,33 Based on ICD-9-CM diagnosis and procedure codes, we developed variables indicating fracture location (femoral neck, intertrochanteric, subtrochanteric, and other location/multiple fractures), pathologic fracture, and surgery type (hemiarthroplasty, total hip arthroplasty, or internal fixation). Comorbidities were defined according to algorithms outlined by Elixhauser et al., 34 as adapted for ICD-9-CM diagnoses by Quan et al.,35 as well as algorithms outlined by Silber et al. 27 Comorbidities were considered present if the relevant ICD-9-CM diagnosis was present in a secondary diagnosis field and indicated as being present on admission.

Statistical Analyses
Initial analyses used the Wilcoxon rank sum test and the chi-square test to compare characteristics of patients receiving regional and general anesthesia. Chi-square tests were used to compare unadjusted rates of inpatient mortality and respiratory and cardiovascular complications according to anesthesia type. Next, we developed hospital fixed-effects logistic regression models to measure the association of anesthesia type with our study outcomes while controlling for potential patient-and hospital-level confounders. To select variables for inclusion in models predicting each of our three outcomes (mortality, pulmonary complications, and cardiovascular complications), we first constructed three separate logistic regression models without hospital fixed-effects. These models used backward elimination 36 with a threshold for variable removal of P Ͼ 0.2. Variables considered for inclusion were: age, sex, fracture location, type of surgery, presence of pathologic fracture, race, and 35 indicator variables for comorbidities; an indicator variable that equaled one for patients receiving regional anesthesia was forced into the model. Model discrimination was evaluated using the c-statistic.
We constructed additional control variables for inclusion in our regression models based on a propensity score to improve balance on observed covariates between patients receiving regional versus general anesthesia. 37,38 This score indicated the probability of receiving regional anesthesia, based on a logistic regression model that included all patient-level covariates considered in development of our initial regression models, along with a quadratic term for age. We stratified patients into five groups of equal size by values of the propensity score and developed indicator variables for each propensity score quintile. For each of our outcomes, we developed hospital fixed-effects logistic regressions using Stata's xtlogit and clogit commands, setting the New York State Inpatient Database's unique hospital identifier as the cluster-ing variable; these models included the regional anesthesia indicator, all variables selected by the relevant backward elimination procedure, and indicators for propensity score stratum. We anticipated that outcomes for patients treated at a single hospital would be potentially correlated, so we adjusted all standard errors for clustering at the hospital level. 39 To assess the robustness of our results to model specification and to the variables included in the regression model, we replicated all analyses in logistic regression models that treated hospital as a random factor rather than a fixed factor and in fixed and random effects models that omitted the indicator for propensity score quintile.
We conducted subgroup analyses to test the association of anesthesia type with outcomes among patients with femoral neck and intertrochanteric fractures by reestimating each fixed-effects regression model within the relevant patient subgroup. All analyses used a threshold of P Ͻ 0.05 for statistical significance and were conducted using Stata 10.0 (StataCorp, College Station, TX) and SAS 9.2 (SAS Institute, Cary, NC).

Validation of Exposure Variable
We initially made qualitative assessments of the distribution of anesthesia codes across procedures listed in the New York State Inpatient Database as a whole. We found codes for regional anesthesia to be most common among the procedures in which our clinical experience suggested regional anesthesia would be likely to be used. Among all discharges for 2007 and 2008, the three most common procedures associated with regional anesthesia were cesarean section (ICD-9-CM procedure code 74.1), manually assisted delivery (code 73.59), and repair of obstetrical laceration (code 75.69). The three procedures most frequently associated with a code for general anesthesia were laparoscopic cholecystectomy (code 51.23), total knee replacement (code 81.54), and laparoscopic appendectomy (code 47.01).
We next examined the frequency of missing or invalid values for anesthesia type among patients meeting our inclusion criteria. Beginning with the universe of discharges from New York hospitals in 2007 and 2008 (5,237,998 discharges), we identified 38,605 patients with a relevant ICD-9-CM diagnosis code for hip fracture; of these, we excluded 8,746 discharges without a relevant listed surgical procedure, 1,048 discharges among patients younger than 50 yr or with a missing value for age, 155 hospitalizations for care of multiple trauma, and 1 discharge without a valid hospital identifier. This resulted in a sample of 28,655 discharges from 173 hospitals; of these hospitals, all anesthesia type codes were missing or invalid for 47, accounting for 5,612 discharges. Within the remaining 126 hospitals, anesthesia type was missing or invalid for 11% of discharges. Comparison of the 47 hospitals not reporting anesthesia type to the remaining facilities in the sample (table 1) showed nonreporting facilities to be less frequently located in metropolitan areas and to have lower annual hip fracture volumes. Hip fracture patients receiving care at reporting versus nonreporting facilities did not differ in terms of age, gender, or unadjusted in-hospital mortality.

Unadjusted Analyses
From our sample of 28,655 discharges, we excluded 10,122 patients with missing or invalid anesthesia codes, or an anes- thesia type of local, none, or other. We also excluded 375 patients with a listed major secondary surgical procedure, obtaining a final study cohort of 18,158 patients. A diagram depicting our process of defining our study sample appears in figure 1. Of 18,158 patients in our sample, 5,254 (28.9%) received regional anesthesia (table 2). The median percentage of patients receiving regional anesthesia at a given hospital was 25.7%, and the interquartile range was from 4.4% to 53.3%. The median number of cases discharged per facility during the period was 119.5 (range: 1-739). Patients receiving regional anesthesia were older, more frequently of white race, and less often had pathologic fractures. Chronic obstructive pulmonary disease and dementia were more common among those receiving regional anesthesia; malignancy, diabetes, and chronic renal disease were more often present among patients receiving general anesthesia.
Unadjusted comparisons of outcomes (table 3) showed no significant differences in mortality according to anesthesia type. Although we found no differences in rates of aspiration or infectious pneumonia by anesthesia type, patients receiving regional anesthesia experienced fewer episodes of respiratory failure (3.4% compared with 5.0% for patients with general anesthesia, P Ͻ 0.0001) and had a lower rate of any pulmonary complication (6.8% vs. 8.1%, P ϭ 0.005). Unadjusted rates of cardiovascular complications did not vary according to anesthesia type.

Adjusted Regression Analyses
We developed hospital fixed-effects logistic regressions to predict in-hospital mortality, occurrence of any pulmonary complication, and occurrence of any cardiac complication. Selected model results are shown in table 4 and discussed here; refer to appendices 3 through 11 for full model results.
For our mortality model, we began by developing a logistic regression model without adjustment for hospital effects. This model included an indicator for regional anesthesia and 21 control variables selected by backward elimination. These were: sex, age, lymphoma, psychosis, metastatic cancer, pathologic fracture, neurologic disorder, alcohol abuse, arrhythmia, hypothyroidism, congestive heart failure, weight loss, chronic obstructive pulmonary disease, valvular disease, depression, diabetes (uncomplicated), renal disease, liver disease, electrolyte abnormality, hypertension (uncomplicated), and hypertension (complicated); the model c-statistic was 0.77. We then developed a logistic regression model to predict mortality based on these covariates, an indicator for the propensity score quintile, and hospital fixed effects. Using this model, we found regional anesthesia to be associated with a lower odds of in-hospital mortality relative to general anesthesia (odds ratio [OR]: 0.710, 95% CI 0.541, 0.932, P ϭ 0.014).
We followed the same procedure to determine the adjusted odds of a pulmonary complication with regional versus general anesthesia. We began by developing a logistic model, without adjustment for hospital effects, to predict occurrence of a pulmonary complication. This model included the regional anesthesia indicator and the following control variables: sex, age, liver disease, fracture location, surgery type, paralysis, hypothyroidism, renal disease, neurologic disorder, peptic ulcer disease, congestive heart failure, hypertension (complicated), hypertension (uncomplicated), chronic obstructive pulmonary disease, weight loss, and diabetes (uncomplicated); the model c-statistic was 0.67. A hospital fixed-effects logistic regression model including these covariates and an indicator for propensity score quintile showed regional anesthesia to be associated with a lower odds of pulmonary complications relative to general anesthesia (OR: 0.752, 95% CI 0.637, 0.887, P Ͻ 0.0001).
To predict cardiovascular complications, we began by developing a logistic regression model, without adjustment for hospital effects. In addition to the regional anesthesia in- dicator, this model included control variables for sex, age, race, fracture location, surgery type, rheumatoid arthritis, hypertension, neurologic disorder, liver disease, electrolyte abnormality, congestive heart failure, coagulopathy, chronic obstructive pulmonary disease, dementia, depression, hypothyroidism, hypertension (complicated), and valvular disease. The model c-statistic was 0.66. After including these covariates and an indicator for propensity score quintile in a hospital fixed-effects logistic regression model, we found no difference in cardiovascular complications according to anesthesia type (OR: 0.877, 95% CI 0.748, 1.029, P ϭ 0.107).
We obtained similar results when we replicated these analyses in hierarchical logistic models that used identical sets of control variables but treated hospital as a random factor, rather than a fixed factor; these models are reported in full in appendices 12 through 20. Finally, we repeated all regressions in fixed-effects and random-effects logistic mod-

Subgroup Analyses
We repeated all of our regression analyses in subgroups restricted to patients with femoral neck fractures and patients with intertrochanteric fractures of the femur. Among patients with femoral neck fractures, our full hospital fixedeffects models (table 5) indicated no difference in inpatient mortality, pulmonary complications, or cardiovascular complications by anesthesia type. In contrast, regional anesthesia was associated with significantly lower odds of mortality and pulmonary complications for patients with intertrochanteric fractures.

Discussion
Among patients undergoing hip fracture surgery, we found a 29% lower adjusted odds of mortality among patients receiving a regional technique as their principal anesthetic modality relative to patients receiving general anesthesia. We found a 24% decrease in the adjusted odds of any inpatient pulmonary complication among patients receiving regional anesthesia, and both of these findings were consistent across regression models that used a variety of approaches to risk adjustment. In contrast, we did not observe a difference in the odds of a major inpatient cardiovascular complication according to anesthesia type. Finally, we found the association of anesthesia type and outcomes to vary according to fracture location; regional anesthesia was consistently associated with a lower odds of inpatient mortality and pulmonary complications among patients with intertrochanteric fractures, but we found no similar association between anesthesia type and outcomes among patients with femoral neck fractures. Our findings regarding mortality and complications are similar in direction and magnitude to those reported by Radcliff et al. 8 in research on hip fracture outcomes among community-dwelling male veterans in the United States. This previous study, using data from the Veterans Affairs-National Surgical Quality Improvement Program, found general anesthesia to be associated with a 27% increased odds of mortality and a 33% increased odds of any complication at 30 days relative to regional anesthesia. Although our analyses are confined to inpatient outcomes, we extend on this previous work in four key ways: first, by examining a populationbased sample of patients, we present data that may be more generalizable to populations not included in the previous study, such as female patients and non-community-dwelling individuals. Second, our use of fixed-effect regression allows us to compare outcomes among patients receiving regional or general anesthesia within a given hospital. As such, our findings regarding differences in mortality and respiratory complications are unlikely to be explained by variations in the quality of care between hospitals that differ in their use of regional anesthesia. Third, our main regression models incorporated control variables based on a propensity score that indicated the probability of receiving regional anesthesia. Propensity score adjustment potentially decreases bias in regression models by balancing the distributions of observed covariates across patient groups receiving different treatments 37,38 ; thus, our use of a propensity score for risk adjustment reduces the likelihood that our findings can be attributed to imbalances in observed covariates between patients receiving regional versus general anesthesia. Fourth, we performed subgroup analyses to understand how the association between anesthesia care and outcome might differ according to fracture characteristics. Mortality and functional impairment are greater after intertrochanteric fractures than after femoral neck fractures, 40,41 although these differences may be attributable to baseline differences in functional status and illness severity between patients experiencing each fracture type. 42,43 Although it is important to note that our subgroup analyses should be regarded as preliminary until replicated in future work, the observation that certain groups of hip fracture patients may have more to gain from the use of regional anesthesia highlights the need for additional research to define subsets of the hip fracture population in which this technique may be beneficial; these subsets may include the oldest-old, patients with advanced functional disability or frailty, and patients with baseline risk factors for pulmonary complications.
The current work must be interpreted in the context of multiple limitations. First, because of the observational nature of the study, we cannot fully exclude the possibility that unobserved differences may have existed between the groups we compare here. Specifically, if sicker patients were more likely to receive general anesthesia, our findings of a lower odds of mortality and complications with regional anesthesia may reflect selection bias. Although our statistical models controlled for a range of observed confounders, the possibility of confounding attributable to unobserved differences between patients receiving regional or general anesthesia precludes determination of a causal effect of anesthesia type on outcome from the current data. In addition, our data do not offer insight into why the prevalence of selected conditions, such as malignancy and renal dysfunction, varied according to anesthesia type. Additional research is needed to characterize the decision-making processes that underlie these variations.
Second, our findings are limited by the nature of anesthesia variable coding in the study data set, which lists patients receiving anesthetics that combine regional and general techniques as having received general anesthesia alone. As a result, our findings are best interpreted as a comparison between patients receiving regional anesthesia, without general anesthesia, to patients undergoing general anesthesia, with or without an accompanying regional technique. Thus, we cannot comment on the potential effects on outcomes of adding a regional technique to a general anesthetic for hip fracture care or the relative advantages of different types of regional anesthesia. At the same time, it is important to note that this pattern of coding was likely to have biased the comparisons we present here toward the null hypothesis by diluting any direct effect of regional anesthesia on outcome. As such, our reported measures of association may underestimate the true effect of regional anesthesia on mortality and major complications after hip fracture. Because the coding structure of the study database prevented accurate identification of anesthesia type among patients undergoing multiple procedures during a single hospital stay, we chose to exclude these patients from our analysis. Nonetheless, because these patients are likely to be sicker than the overall hip fracture population, our exclusion of them may limit the generalizability of our results.
Lastly, we note that 47 of the 173 hospitals we examined initially did not provide information on anesthesia type, with lower-volume hospitals and hospitals located outside of metropolitan areas being less likely to report anesthesia type. Thus, our findings may not be fully generalizable to these types of hospital settings or to facilities located outside of New York state. These considerations indicate that future research using alternate data sources remains necessary to confirm our findings.