Currently Viewing:
The American Journal of Accountable Care June 2017
Evolving Health Workforce Roles in Accountable Care Organizations
Shana F. Sandberg, PhD; Clese Erikson, MPAff; and Emily D. Yunker, MPA, PMP
Case Study: How Does an ACO Generate Savings Three Years in a Row?
Peter A. Gross, MD; Morey Menacker, DO; Mitchel Easton, BS; Edward Gold, MD; Linda Naughton, RN; James A. Colbert, MD; Juliana Hart, BSN, MPH; Denise Patriaco, APN; and Mark Sparta, MPA
Creating Clarity: Distinguishing Between Community and Population Health
Amber M. Maraccini, PhD; Panagis Galiatsatos, MD; Mitch Harper, BS; and Anthony D. Slonim, MD, DrPH
The Promises and Pitfalls of Telemedicine
Dawn Sherling, MD, and Michael Sherling, MD, MBA
Defining Value for Better Decision Making: Lessons From the Spring ACO Coalition Meeting
Mary K. Caffrey and Laura Joszt, MA
Currently Reading
Medicare Inpatient and Postdischarge Outcomes of Elective Percutaneous Coronary Interventions
Michael Pine, MD, MBA; Donald E. Fry, MD; Susan M. Nedza, MD, MBA; David G. Locke, BS; Agnes M. Reband, BS; and Gregory Pine, BA

Medicare Inpatient and Postdischarge Outcomes of Elective Percutaneous Coronary Interventions

Michael Pine, MD, MBA; Donald E. Fry, MD; Susan M. Nedza, MD, MBA; David G. Locke, BS; Agnes M. Reband, BS; and Gregory Pine, BA
An understanding of risk-adjusted outcomes for percutaneous coronary interventions for both inpatient and 90-day postdischarge events is necessary for the redesign of care outcomes.

Objectives: To compare hospital-level, risk-adjusted inpatient and 90-day postdischarge adverse outcomes in Medicare patients undergoing elective percutaneous coronary interventions (PCIs).

Study Design: Develop predictive risk models for adverse outcomes during inpatient and 90-day postdischarge care, and use those models to compare hospital performance in elective PCI.

Methods: Elective PCIs in the 2010 to 2012 Medicare Limited Data Set were used to develop logistic prediction models for inpatient deaths, inpatient prolonged-length-of-stay outliers, 90-day postdischarge deaths without readmission, and 90-day readmissions. Observed versus predicted differences for risk-adjusted adverse outcomes were then performed among all hospitals with 50 or more cases during the study period.

Results: There were 978 hospitals with 168,518 patients that qualified for this study; 25.9% of all patients had 1 or more adverse outcome. There were 67 hospitals with adverse outcome rates that were 2 or more standard deviations (SDs) better than predicted and 81 hospitals with rates that were more than 2 SDs worse than predicted. The best- and worst-performing deciles of hospitals had median risk-adjusted adverse outcome rates of 17.6% and 35.5%, respectively. Hospital case volume was not independently associated with better outcomes.

Conclusions: Comparison of risk-adjusted adverse outcome rates demonstrated the existence of opportunities for substantial improvements in quality among suboptimal-performing hospitals.  

The American Journal of Accountable Care. 2017;5(2): e7-e15
Inpatient percutaneous coronary intervention (PCI) was performed on an estimated 515,000 patients in 2013, 50% of whom were 65 years or older.1 The results of this procedure have progressively improved from the early efforts with coronary angioplasty to the current use of stents. The inpatient mortality rate with PCI is less than 1%, and the rate of inpatient complications, which include technical problems with coronary vessels, hematoma at the vascular entry site, stroke, and contrast-associated nephropathy, is less than 7%.2,3

In a study of Medicare coronary artery bypass grafting (CABG) and valve replacement surgery,4 we found that inpatient adverse outcomes of these major cardiac procedures constituted less than 35% of total adverse events (AEs) when 90-day postdischarge deaths and readmissions were included. Furthermore, there were dramatic differences among hospitals in their rates of overall inpatient and postdischarge adverse outcomes. These data on comparative hospital risk-adjusted outcomes among the best and poorest performances strongly suggested that many of these adverse outcomes were preventable.

In the current study, we similarly evaluated risk-adjusted outcomes among inpatient Medicare patients who underwent PCI to identify comparative hospital performance in both inpatient and 90-day postdischarge events. Differences between top and suboptimal performances among all hospitals should define the margin of preventable AEs. Because it was anticipated that readmissions would be the predominant adverse outcome, we have examined the causes of readmission to better define opportunities for improvement.


The Medicare Limited Data Set for 2010 to 2012 was used in this study. Patients with an International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) procedure code of 00.66, 36.06, 36.07, 36.09, and 37.34 for PCI were identified. All patients included in the study were required to have a principal diagnosis code of 414.00-414.05. Patients who had had an acute myocardial infarction were excluded. Cases were also excluded if patients were younger than 65 years, had missing patient or hospital identifiers, were transferred from another facility, or were discharged against medical advice.

Two overlapping databases were designed. Risk-adjustment models were derived using a developmental database that consisted of patients from hospitals that satisfied quality coding criteria and had the 20 or more qualifying cases needed to construct hospital-level moving-range control charts.5 Final models were used to compare the overall adverse outcome rates of all hospitals that met the minimum volume criteria of 50 evaluable cases required for inclusion in a study database. Hospitals with fewer than 50 qualifying cases were excluded from the study database to avoid skewing final comparisons of hospitals’ performance by including hospitals with few qualifying cases and very low predicted numbers of adverse outcomes.

Model Derivation

Using the developmental database, prediction models were designed for: 1) inpatient deaths (IpD), 2) prolonged risk-adjusted lengths-of-stay (prLOS), 3) 90-day postdischarge deaths without readmission (PD-90), and 4) 90-day readmission (RA-90). Models were derived using stepwise logistic regression on more than 500 candidate risk factors, including patient age and coded comorbid conditions that were present at admission. Hospital variables were employed to account for hospital effects. The Schwarz criterion was used to avoid overfitting final models,6 and the discrimination of final models was measured with C statistics. All analyses were performed with SAS version 9.4 (SAS Institute; Cary, North Carolina).

The IpD model included only inpatient deaths. PrLOS was used as a surrogate for severe complications of inpatient care among live discharges. Cases with prLOS were identified by first developing a linear model for inpatient length of stay (LOS) among cases without any coded complications. Predicted LOS for all cases in the developmental database were subtracted from observed LOS, differences were temporally aligned within each hospital, and moving-range control charts were created to identify outliers that exceed the 3 sigma upper control limit.7-9 Previous studies including cardiac procedures have associated prLOS with severe complications, increased costs,10 and higher rates of postdischarge deaths and readmissions.11

The PD-90 and RA-90 models included AEs occurring within 90 days of discharge. We used 90 days because our previous work demonstrated that following surgical procedures, more than 40% of relevant deaths and readmissions occur between 31 and 90 days after discharge.12 To study RA-90, we excluded all 90-day readmissions for Medical Diagnostic Categories (MDCs) 2 (Eye Diseases), 17 (Myeloproliferative Diseases), and 22/24 (Burns/Major Trauma), as well as all Medicare-Severity, Diagnosis-Related Groups (MS-DRGs) related to the management of trauma or cancer regardless of MDC. PrLOS events during the index hospitalization were included in PD-90 and RA-90 models to clarify the relationship between inpatient complications and postdischarge AEs and to remove the effects of inpatient complications on coefficients in postdischarge models. MS-DRGs were identified for each readmission. All valid readmissions were divided into those occurring within 30, 60, and 90 days of discharge, but only 90-day readmissions were used in RA-90 model development.

Comparative Hospital Performance

The 4 PCI prediction models for adverse outcomes from the developmental database were applied to all patients in hospitals in the study database. All hospitals had more than 4.5 predicted total adverse outcomes and more than 4.5 predicted 90-day readmissions; 90% had more than 4.5 predicted prolonged LOS outliers. The IpD model was applied to all cases. The prLOS model was applied to all live inpatient discharges. Live discharges without prLOS were used to predict PD-90, and live discharges without prLOS or PD-90 were used to predict readmissions.

Among all hospitals, total predicted adverse outcomes were adjusted to equal total observed adverse outcomes. For each hospital, a z score was computed as [observed adverse outcomes – predicted adverse outcomes] ÷ standard deviation (SD), where the SD equals and N equals the number of qualifying cases at that hospital. Negative z scores indicated that outcomes were better than predicted; positive z scores indicated that outcomes were poorer than predicted. Each hospital’s risk-adjusted adverse outcome (RAAO) rate was computed as [overall observed adverse outcomes rate] × [hospital-observed adverse outcomes ÷ hospital-predicted adverse outcomes]. RAAO rates were grouped into deciles for comparison. To better define the contribution of inpatient and postdischarge adverse outcomes to overall performance, each hospital’s total adverse outcomes were subdivided into observed and predicted prLOS and observed and predicted RA-90. To evaluate the influence of case volume on RAAO rates, hospitals were grouped into deciles based on their case volume during the study period.


Prediction Models

Of the 150,903 patients in the developmental database, 503 (0.33%) died during their index admission; 10,341 (6.8%) had prLOS; 730 (0.5%) died within 90 days of discharge without a readmission; and 31,124 (20.6%) were readmitted within 90 days of discharge. The significant risk factors (P <.001) in each of the predictive models for these outcomes are summarized in eAppendix Table 1 (eAppendices available at The IpD model had 10 significant risk factors and a C statistic of 0.748 after removal of hospital variables. The prLOS model had 66 significant risk factors and a final C statistic of 0.747. The PD-90 model had 13 significant risk factors and a final c-statistic of 0.768. The RA-90 model had 43 risk factors and a final C statistic of 0.644. PrLOS was a significant predictor of 90-day readmissions (odds ratio [OR], 1.62) and 90-day postdischarge mortality without readmission (OR, 3.99).

The Table details the MS-DRGs of qualifying readmissions in the developmental dataset. There were 31,124 patients readmitted 41,274 times, with 18,696 (45.3%) readmissions occurring during the first 30 days following discharge; 12,411 (30.1%) between days 31 and 60; and 10,167 (24.6%) between days 61 and 90. The majority of readmissions (58%) had cardiovascular principal diagnoses.

Hospital Comparisons

The study database consisted of 168,518 patients from 978 hospitals, with each having 50 or more qualifying cases (average of 172 cases per hospital; median of 119 cases). There were 571 (0.34%) inpatient deaths; 11,603 (6.9%) patients with prLOS; 1613 (1.0%) postdischarge deaths without readmission; and 34,841 (20.7%) patients with 1 or more readmissions. There were 820 deaths after readmissions. Inpatient and 90-day postdischarge deaths totaled 3004 (1.8%). There were 43,613 (25.9%) patients who had 1 or more adverse outcomes after PCI.

Figure 1 demonstrates the distribution of z scores for observed versus expected adverse outcomes rates. Z scores for study hospitals ranged from –6.2 for the best performing hospital to +6.3 for the poorest performing hospital. There were 67 hospitals with differences more than 2 SDs better than the average and 81 hospitals with differences more than 2 SDs poorer than average.

Figure 2 demonstrates the distribution of RAAO rates among hospitals by decile of performance. The best performing decile had a median adverse outcome rate of 17.6%, while the poorest performing decile had a median adverse outcome rate of 35.5%. Error bars represent the interquartile range within each decile.

Figure 3 demonstrates the relationship between risk-adjusted rates of prLOS and comparable rates of RA-90. The ratio of observed to predicted length of stay outliers are strongly correlated with the ratio of observed to predicted 90-day readmissions. Hospitals that have low risk-adjusted rates of inpatient complications of care have comparable low risk-adjusted rates of postdischarge readmissions. Hospitals with higher risk-adjusted inpatient complication rates have higher rates of risk-adjusted 90-day postdischarge readmissions.

Figure 4 illustrates the relationship of adverse outcomes and hospitals’ case volumes. There is very little variation in outcomes across decile groups organized by volume. Each decile had a standard error of the mean of 0.5%. Analysis of variance demonstrated no significant relationship between the case volume of each hospital and its RAAO rates (P = .86).


In this study, 4 objectively defined, risk-adjusted adverse outcomes were used to compare hospital performance in PCI. These adverse outcomes were treated as no-fault events, each of which may or may not have been preventable. However, when risk-adjusted outcomes at hospitals were compared, remarkable differences in measured performance suggested that there were substantial opportunities for quality improvement at many facilities.

Copyright AJMC 2006-2019 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
Welcome the the new and improved, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up