Coronary artery bypass grafting and cardiac valve surgery have dramatically different risk-adjusted outcomes over the 90 days following discharge, and demonstrate the opportunity for care improvement.
Published Online: March 14, 2016
Donald E. Fry, MD; Michael Pine, MD, MBA; Susan M. Nedza, MD, MBA; David G. Locke, BS; Agnes M. Reband, BS; and Gregory Pine, BA
Objectives: To develop metrics for objective and risk-adjusted adverse outcomes for cardiac surgical care and to compare hospital performances to define opportunities for care improvement.
Study Design: To develop predictive risk models for adverse events that occur during inpatient and postdischarge care, and then apply those models to define comparative hospital performance in cardiac surgery.
Methods: The population was elective coronary artery bypass and cardiac valve surgery patients in the Medicare Limited Data Set for 2010-2012 in the United States. Logistic prediction models for inpatient deaths, inpatient prolonged length-of-stay outliers, 90-day postdischarge deaths without readmission, and 90-day readmissions among cardiac surgery patients were designed. Observed versus predicted differences for risk-adjusted adverse outcomes (AOs) were then performed among all hospitals that met minimal volume criteria.
Results: A total of 1031 hospitals for coronary artery bypass surgery averaged 27.2% AOs for all 4 measurements of interest; 71 hospitals had observed outcomes that were ≥2 standard deviations (SDs) greater than expected. A total of 794 hospitals for cardiac valve surgery averaged 32.3% AOs; 57 hospitals had observed outcomes that were ≥2 SDs greater than expected. Median risk-adjusted AO rates were 17% for coronary artery bypass and 20.4% for valve surgery in the best performing decile, but were 38.8% and 45.8%, respectively, in the poorest performing deciles.
Conclusions: The wide range of risk-adjusted AOs in cardiac surgery indicates a genuine opportunity for care improvement.
New technology, refined surgical techniques, and greatly enhanced critical care have translated into progressive improvement in cardiac surgery over the last several decades. A portion of this progressive improvement is secondary to the efforts of the Society of Thoracic Surgeons (STS) to provide risk-adjusted outcomes using clinical data and to provide feedback to providers.1 Even in the Medicare population, elective coronary artery bypass grafting (CABG) and cardiac valve surgery (CVS) have 30-day death rates of less than 3%. Mortality rates at this level cannot be used alone to evaluate outcomes of care except in extreme circumstances.2
Complications of care have been studied for comparative performance evaluation. Although severe complications are easily recognized, many complications (eg, surgical site infection) have definitions that require institutional interpretation and standardized surveillance for recognition. In an era of short inpatient length of stay (LOS), many such complications are not recognized until after discharge and may not be reported. For cardiac surgery, the full array of potential complications has a varying spectrum of severity, and reporting all of them may yield very high observed rates that confound understanding of significant events. In previous studies, we identified that liberally coded complications (eg, hypopotassemia) may result in overall complication rates that approach 70% in elective CABG surgery.3
Recent attention has focused on readmissions as a metric for outcomes of inpatient care. Jencks et al4 identified that all-cause 30-day readmission was 19% and 90-day readmission was 34%, in all Medicare patients. The Readmission Reduction Program (RRP) by Medicare, which began in 2014, is assessing financial penalties on hospitals that exceed an arbitrarily defined rate.5 The initial penalties affected over 2600 hospitals,6 and the second year of the RRP demonstrated little improvement.7 Other medical and surgical conditions—likely including cardiac surgery—will be added to the RRP. For providers of care, readmissions may have legitimate and unrelated causes to the original hospitalization, and many patients are readmitted to hospitals other than the initial index facility, which makes monitoring of readmission patterns difficult.8,9 The current Medicare RRP is for 30 days following discharge, but early indications are that 90 days may be the target in the program for bundled payments.10
In the current study, we examined elective CABG and CVS patients to identify the full spectrum of adverse outcomes (AOs) from inpatient care through 90 days of postdischarge care. Separate risk-adjusted models have been designed for 4 AOs: inpatient deaths (IpDs), risk-adjusted prolonged LOS (prLOS) outliers as surrogates for severe inpatient complications, 90-day postdischarge deaths without readmission (PD-90), and 90-day readmissions (RA-90), except for specific exclusions. These prediction models were used to compare hospital outcomes to identify opportunities for care improvement.
We designed 4 unique prediction models each for CABG and CVS using hospitals that met quality-coding criteria.11 These patients and hospitals will be referred to as the developmental database. The models from the developmental database were then used for the evaluation of outcomes among all hospitals—the study database—that met minimum volume criteria for each procedural group.
We used the Medicare Limited Data Set for 2010-2012. Patients with an International Classification of Diseases, 9th Revision, Clinical Modification procedure code of 36.10-36.19 for CABG, or 35.12, 35.21, 35.22, 35.23, 35.24, or 35.33 for CVS, were identified. CABG cases were required to have a principal diagnosis of 414.00-414.05, and CVS cases were required to have a principal diagnosis of 394.0-395.2, 424.0, or 424.1. Tricuspid and pulmonic valve surgeries were not included. Inclusion criteria required that the procedure was performed on inpatient day 0, 1, or 2; only hospitals with 20 or more eligible cases were included in order to meet the minimum number of cases required for hospital-specific control charts for each procedural group. Cases were excluded if patients were aged under 65 years, had missing patient/hospital identifiers, were transfers from another facility, or were discharged against medical advice.
With the developmental database, models for the 4 AOs were created using forward stepwise logistic regression. Candidate risk factors reflected comorbid conditions that were present upon admission. Cases with simultaneous CABG and CVS were classified as CVS, with CABG identified as a risk factor. Hospital effects upon final coefficients were addressed by using hospital-specific dummy variables. Schwarz criterion was used to avoid overfitting final models.12 C statistics were used to evaluate discrimination of final models. SAS version 9.4 (SAS Institute, Cary, North Carolina) was used for all analyses.
Logistic models of IpD and prLOS were designed individually for CABG and CVS. The prLOS model estimates serious inpatient complications of care by first developing a linear model of inpatient LOS in cases without any coded complications. The predicted LOS was subtracted from the observed LOS of all cases in the developmental data set. Cases were aligned in temporal order, and a moving-range control chart was constructed to identify cases that were 3-sigma outliers above the upper control limit by methods previously reported.13-15 PrLOS then became a dependent variable in the development of postdischarge models for CABG and CVS. Previous studies have associated prLOS with severe complications and increased costs16 and its role as a predictor of postdischarge deaths and readmissions.17
Postdischarge AOs were observed across 90 days following discharge separately for CABG and CVS. The PD-90 model was designed using all deaths without readmission as the dependent variable. To study RA-90 cases, we identified all RA-90 and removed cases that were from Medical Diagnostic Categories (MDCs) of the Medicare Severity Diagnosis-Related Groups (MS-DRGs) system: 2 (Eye Diseases), 17 (Myeloproliferative Diseases), 22/24 (Burns/Major Trauma), and all MS-DRGs related to the management of trauma or cancer, regardless of MDC. All valid readmissions were divided into those occurring within 30, 60, and 90 days of discharge. The MS-DRGs were then identified for each readmission. An RA-90 prediction model was then designed with the total RA-90 for CABG and for CVS as the dependent variable. PrLOS events of the index hospitalization were used in PD-90 and RA-90 models to identify the relationship of inpatient complications with postdischarge AOs, but also to remove effects of inpatient complications on coefficients of postdischarge models.
Comparative Hospital Performance
To compare hospital performance, the 4 AO prediction models from the developmental database were applied to all hospitals in the study database with 20 or more qualifying cases. All hospitals had a minimum of 4.6 total predicted AOs for each surgical group. The IpD model was applied to all cases. The prLOS model was used to predict LOS outliers among live discharges. Live discharges without prLOS were used to predict PD-90, and finally, live discharges without prLOS or PD-90 were used to predict readmissions. Among all hospitals, total predicted AOs was set equal to total observed AOs to give an adjusted predicted total. For each hospital, a z score was computed by [observed AOs – predicted AOs]/standard deviation (SD), where SD = √(N × p × (1– p)). Positive and negative z scores represent outcomes that were poorer or better than predicted, respectively. Z scores were stratified to permit evaluations of outcomes and case volumes at the hospital level. The risk-adjusted AO rate for each hospital was computed by this formula: [overall observed AO rate] × [hospital-observed AOs/hospital-predicted AOs]. Risk-adjusted AO rates were then categorized into deciles for comparison. Cases were also subdivided into deciles of case volume to assess effects upon outcomes.
Table 1 summarizes the 4 AO models from the developmental database for CABG and CVS. C statistics declined when the hospital dummy variables were dropped from the model, although this effect was less in the RA-90 models. Final models between CABG and CVS had similar C statistics. The significant variables with odds ratios for all 4 prediction models are detailed in eAppendix Tables 1 through 4 (eAppendix available at www.ajmc.com). Among all models for both groups of operations, female gender, age groups over 75 years, and the usual chronic diseases (eg, chronic renal failure) were significant variables. Patients who were prLOS after inpatient care had significant odds ratios (ORs) for both PD-90 (CABG: OR, 11.34; CVS: OR, 13.94) and RA-90 (CABG: OR, 2.07; CVS: OR, 2.10).
The MS-DRGs of readmissions after exclusions are identified in the developmental dataset in Tables 2 and 3 for days 1 to 30, 31 to 60, and 61 to 90 following discharge. For CABG patients, cardiac events (39.1%), infections (17.4%), and pulmonary events (13.9%) were the most common causes of readmissions. Among all readmissions, 57% occurred during the first 30 days following discharge, 24.8% occurred between days 31 to 60, and 18.2% occurred between days 61 to 90, for a total of 43% of associated readmissions occurring between days 31 to 90.
PDF is available on the last page.