Statistical Methods for Profiling Providers of Medical Care: Issues and Applications

Comparing a multivariate response Bayesian random effects logistic regression model with a latent variable item response theory model for provider profiling on multiple binary indicators simultaneously

Provider profiling entails comparing the performance of hospitals on indicators of quality of care. Many common indicators of healthcare quality are binary (eg, short‐term mortality, use of appropriate medications). Typically, provider profiling examines the variation in each indicator in isolation across hospitals. We developed Bayesian multivariate response random effects logistic regression models that allow one to simultaneously examine variation and covariation in multiple binary indicators across hospitals. Use of this model allows for (i) determining the probability that a hospital has poor performance on a single indicator; (ii) determining the probability that a hospital has poor performance on multiple indicators simultaneously; (iii) determining, by using the Mahalanobis distance, how far the performance of a given hospital is from that of an average hospital. We illustrate the utility of the method by applying it to 10 881 patients hospitalized with acute myocardial infarction at 102 hospitals. We considered six binary patient‐level indicators of quality of care: use of reperfusion, assessment of left ventricular ejection fraction, measurement of cardiac troponins, use of acetylsalicylic acid within 6 hours of hospital arrival, use of beta‐blockers within 12 hours of hospital arrival, and survival to 30 days after hospital admission. When considering the five measures evaluating processes of care, we found that there was a strong correlation between a hospital's performance on one indicator and its performance on a second indicator for five of the 10 possible comparisons. We compared inferences made using this approach with those obtained using a latent variable item response theory model.

Risk-Standardizing Rates of Return of Spontaneous Circulation for In-Hospital Cardiac Arrest to Facilitate Hospital Comparisons

Background

Sustained return of spontaneous circulation (ROSC) is the most proximal and direct assessment of acute resuscitation quality in hospitals. However, validated tools to benchmark hospital rates for ROSC after in‐hospital cardiac arrest currently do not exist.

Methods and Results

Within the national Get With The Guidelines‐Resuscitation registry, we identified 83 206 patients admitted from 335 hospitals from 2014 to 2017 with in‐hospital cardiac arrest. Using hierarchical logistic regression, we derived and validated a model for ROSC, defined as spontaneous and sustained ROSC for ≥20 consecutive minutes, from 24 pre‐arrest variables and calculated rates of risk‐standardized ROSC for in‐hospital cardiac arrest for each hospital. Overall, rates of ROSC were 72.0% and 72.7% for the derivation and validation cohorts, respectively. The model in the derivation cohort had moderate discrimination (C‐statistic 0.643) and excellent calibration (R² of 0.996). Seventeen variables were associated with ROSC, and a parsimonious model retained 10 variables. Before risk‐adjustment, the median hospital ROSC rate was 70.5% (interquartile range: 64.7–76.9%; range: 33.3–89.6%). After adjustment, the distribution of risk‐standardized ROSC rates was narrower: median of 71.9% (interquartile range: 68.2–76.4%; range: 42.2–84.6%). Overall, 56 (16.7%) of 335 hospitals had at least a 10% absolute change in percentile rank after risk standardization: 27 (8.0%) with a ≥10% negative percentile change and 29 (8.7%) with a ≥10% positive percentile change.

Conclusions

We have derived and validated a model to risk‐standardize hospital rates of ROSC for in‐hospital cardiac arrest. Use of this model can support efforts to compare acute resuscitation survival across hospitals to facilitate quality improvement.

Coronary artery bypass grafting in Spain. Influence of procedural volume on outcomes

INTRODUCTION AND OBJECTIVES

To analyze the association between volume and outcomes in coronary artery bypass grafting (CABG) in the Spanish National Health System.

METHODS

We analyzed CABG episodes from 2013 to 2015. The selected outcome variables were in-hospital mortality in the index episode, 30-day cardiac-related readmissions, and mortality during readmission. Risk-adjusted rates of in-hospital mortality (RAMR) and 30-day readmissions (RARR) were calculated using multilevel logistic regression. High- and low-volume hospitals for CABG were identified by a nonconditioned analysis (k-means) and by compliance with the volume recommendation of clinical practice guidelines.

RESULTS

A total of 17 335 CABG index episodes were included, with a crude in-hospital mortality rate of 5.0%. Episodes attended in low-volume centers for CABG (< 155 CABG per year) showed 17% higher RAMR (5.81%±2.07% vs 4.96%±1.76%; P <.001) and a negative linear correlation between volume and RARR (r=-0.318; P=.029), as well as a higher percentage of complications during the episode. The same association between volume and more favorable outcomes was found in isolated CABG.

CONCLUSIONS

The mean CABG volume is low in Spanish National Health System hospitals. Higher volume was associated with better outcomes in CABG, both total and isolated. The findings of this study indicate the need for a higher concentration of CABG programs, as well as the publication of risk-adjusted outcomes of coronary intervention.

Temporal trends and prognostic impact of length of hospital stay in uncomplicated ST-segment elevation myocardial infarction in Spain

INTRODUCTION AND OBJECTIVES

There are few data on the safety of length of stay in uncomplicated ST-segment elevation myocardial infarction. We studied trends in hospital stay and the safety of short (≤ 3 days) vs long hospital stay in Spain.

METHODS

Using data from the Minimum Basic Data Set, we identified patients with uncomplicated ST-segment elevation myocardial infarction who underwent primary percutaneous coronary intervention and who were discharged alive between 2003 and 2015. The mean length of stay was adjusted by multilevel Poisson regression with mixed effects. The effect of short length of stay on 30-day readmission for cardiac diseases was evaluated in episodes from 2012 to 2014 by propensity score matching and multilevel logistic regression. We also compared risk-standardized readmissions for cardiac diseases and mortality rates.

RESULTS

The adjusted length of stay decreased significantly (incidence rate ratio <1; P <.001) for each year after 2003. Short length of stay was not an independent predictor of 30-day readmission (OR, 1.10; 95%CI, 0.92-1.32) or mortality (OR, 1.94; 95%CI, 0.93-14.03). After propensity score matching, no significant differences were observed between short and long hospital stay (OR, 1.26; 95%CI, 0.98-1.62; and OR, 1.50; 95%CI, 0.48-5.13), respectively. These results were confirmed by comparisons between risk-standardized readmissions for cardiac disease and mortality rates, except for the 30-day mortality rate, which was significantly higher, although probably without clinical significance, in short hospital stays (0.103% vs 0.109%; P <.001).

CONCLUSIONS

In Spain, hospital stay ≤ 3 days significantly increased from 2003 to 2015 and seems a safe option in patients with uncomplicated ST-segment elevation myocardial infarction.

Trends in cardiogenic shock management and prognostic impact of type of treating center

INTRODUCTION AND OBJECTIVES

Current guidelines recommend centralizing the care of patients with cardiogenic shock in high-volume centers. The aim of this study was to assess the association between hospital characteristics, including the availability of an intensive cardiac care unit, and outcomes in patients with ST-segment elevation myocardial infarction (STEMI)-related cardiogenic shock (CS).

METHODS

Discharge episodes with a diagnosis of STEMI-related CS between 2003 and 2015 were selected from the Minimum Data Set of the Spanish National Health System. Centers were classified according to the availability of a cardiology department, catheterization laboratory, cardiac surgery department, and intensive cardiac care unit. The main outcome measured was in-hospital mortality.

RESULTS

A total of 19 963 episodes were identified. The mean age was 73.4±11.8 years. The proportion of patients with CS treated at hospitals with a catheterization laboratory and cardiac surgery department increased from 38.4% in 2005 to 52.9% in 2015 (P <.005). Crude- and risk-adjusted mortality rates decreased over time, from 82% to 67.1%, and from 82.7% to 66.8%, respectively (both P <.001). Coronary revascularization, either percutaneous or coronary artery bypass grafting, was independently associated with a lower mortality risk (OR, 0.29 and 0.25; both P <.001, respectively). Intensive cardiac care unit availability was associated with lower adjusted mortality rates (65.3%±7.9 vs 72±11.7; P <.001).

CONCLUSIONS

The proportion of patients with STEMI-related CS treated at highly specialized centers increased while mortality decreased during the study period. Better outcomes were associated with the increased performance of revascularization procedures and access to intensive cardiac care units over time.

The profile inter-unit reliability

To assess the quality of health care, patient outcomes associated with medical providers (eg, dialysis facilities) are routinely monitored in order to identify poor (or excellent) provider performance. Given the high stakes of such evaluations for payment as well as public reporting of quality, it is important to assess the reliability of quality measures. A commonly used metric is the inter-unit reliability (IUR), which is the proportion of variation in the measure that comes from inter-provider differences. Despite its wide use, however, the size of the IUR has little to do with the usefulness of the measure for profiling extreme outcomes. A large IUR can signal the need for further risk adjustment to account for differences between patients treated by different providers, while even measures with an IUR close to zero can be useful for identifying extreme providers. To address these limitations, we propose an alternative measure of reliability, which assesses more directly the value of a quality measure in identifying (or profiling) providers with extreme outcomes. The resulting metric reflects the extent to which the profiling status is consistent over repeated measurements. We use national dialysis data to examine this approach on various measures of dialysis facilities.

Trends on Diabetes Mellitus's healthcare management in Spain 2007-2015

AIMS

To analyze the trends on diabetes mellitus (DM) healthcare management in Spain.

METHODS

Retrospective observational study between January 1st 2007 and 31th December 2015 with DM as the principal diagnosis. The main clinical outcome measures were all-cause, in-hospital mortality and 30-day readmissions. We also analyze three Prevention Quality Indicators (PQI) for DM.

RESULTS

The number of hospitalization episodes decreased significantly as well as the frequentation rate and average length of stay (Incidence Rate Ratio [IRR] = 0.963, p < 0.001; 0.91, p < 0.001 and 0.986, p < 0.001, respectively). Crude in-hospital mortality and readmissions rates and risk-standardized in-hospital mortality rates (RSMR), however, remained stable (IRR = 0.988, p = 0.073; IRR = 1.003, p = 0.334 and IRR = 0.997, p = 0.116, respectively). A relevant variability in RSMR, both at hospital (Median Odds Ratio 1.49) and regional level, was found. High volume hospitals (≥105 DM discharges at year) showed better outcomes. High variability was also found in PQI indicators al regional level.

CONCLUSION

The present analysis shows an improvement in hospitalizations related to DM in Spain in the period 2007-2015. There was also a decrease in the frequentation rate and in the average length of stay. These findings are probably explained by quality improvements in the healthcare management of the DM at the ambulatory level. However, there were important differences in the management of diabetic inpatients both at the hospital and the regional level.

Causal Mediation Analysis for Standardized Mortality Ratios

Indirectly standardized mortality ratios (SMR) are often used to compare patient outcomes between health care providers as indicators of quality of care. Observed differences in the outcomes raise the question of whether these could be causally attributable to earlier processes or outcomes in the pathway of care that the patients received. Such pathways can be naturally addressed in a causal mediation analysis framework. Adopting causal mediation models allows the total provider effect on outcome to be decomposed into direct and indirect (mediated) effects. This in turn enables quantification of the improvement in patient outcomes due to a hypothetical intervention on the mediator. We formulate the effect decomposition for the indirectly standardized SMR when comparing to a health care system-wide average performance, propose novel model-based and semiparametric estimators for the decomposition, study the properties of these through simulations, and demonstrate their use through application to Ontario kidney cancer data.

Profiling Hospital Performance on the Basis of Readmission After Transcatheter Aortic Valve Replacement in Ontario, Canada

Background

Readmission rates are a widely accepted quality indicator. Our objective was to develop models for calculating case‐mixed adjusted readmission rates after transcatheter aortic valve replacement for the purpose of profiling hospitals.

Methods and Results

In this population‐based study in Ontario, Canada, we identified all transcatheter aortic valve replacement procedures between April 1, 2012, and March 31, 2016. For each hospital, we first calculated 30‐day and 1‐year risk‐standardized (predicted versus expected) readmission rates, using 2‐level hierarchical logistic regression models, including clustering of patients within hospitals. We also calculated the risk‐adjusted (observed versus expected) readmission rates, accounting for the competing risk of death using a Fine‐Gray competing risk model. We categorized hospitals into 3 groups: those performing worse than expected, those performing better than expected, or those performing as expected, on the basis of whether the 95% CI was above, below, or included the provincial average readmission rate respectively. Our cohort consisted of 2129 transcatheter aortic valve replacement procedures performed at 10 hospitals. The observed readmission rate was 15.4% at 30 days and 44.2% at 1 year, with a range of 10.9% to 21.7% and 38.8% to 55.0%, respectively, across hospitals. Incorporating the competing risk of death translated into meaningful different results between models; as such, we concluded that the risk‐adjusted readmission rate was the preferred metric. On the basis of the 30‐day risk‐adjusted readmission rate, all hospitals performed as expected, with a 95% CI that included the provincial average. However, we found that there was significant variation in 1‐year risk‐adjusted readmission rate.

Conclusions

There is significant interhospital variation in 1‐year adjusted readmission rates among hospitals, suggesting that this should be a focus for quality improvement efforts in transcatheter aortic valve replacement.

An administrative model for benchmarking hospitals on their 30-day sepsis mortality

Background

Given the increased attention to sepsis at the population level there is a need to assess hospital performance in the care of sepsis patients using widely-available administrative data. The goal of this study was to develop an administrative risk-adjustment model suitable for profiling hospitals on their 30-day mortality rates for patients with sepsis.

Methods

We conducted a retrospective cohort study using hospital discharge data from general acute care hospitals in Pennsylvania in 2012 and 2013. We identified adult patients with sepsis as determined by validated diagnosis and procedure codes. We developed an administrative risk-adjustment model in 2012 data. We then validated this model in two ways: by examining the stability of performance assessments over time between 2012 and 2013, and by examining the stability of performance assessments in 2012 after the addition of laboratory variables measured on day one of hospital admission.

Results

In 2012 there were 115,213 sepsis encounters in 152 hospitals. The overall unadjusted mortality rate was 18.5%. The final risk-adjustment model had good discrimination (C-statistic = 0.78) and calibration (slope and intercept of the calibration curve = 0.960 and 0.007, respectively). Based on this model, hospital-specific risk-standardized mortality rates ranged from 12.2 to 24.5%. Comparing performance assessments between years, correlation in risk-adjusted mortality rates was good (Pearson’s correlation = 0.53) and only 19.7% of hospitals changed by more than one quintile in performance rankings. Comparing performance assessments after the addition of laboratory variables, correlation in risk-adjusted mortality rates was excellent (Pearson’s correlation = 0.93) and only 2.6% of hospitals changed by more than one quintile in performance rankings.

Conclusions

A novel claims-based risk-adjustment model demonstrated wide variation in risk-standardized 30-day sepsis mortality rates across hospitals. Individual hospitals’ performance rankings were stable across years and after the addition of laboratory data. This model provides a robust way to rank hospitals on sepsis mortality while adjusting for patient risk.

Electronic supplementary material

The online version of this article (10.1186/s12913-019-4037-x) contains supplementary material, which is available to authorized users.

In-hospital Mortality and Readmissions for Heart Failure in Spain. A Study of Index Episodes and 30-Day and 1-year Cardiac Readmissions

INTRODUCTION AND OBJECTIVES

Heart failure (HF) is a major health care problem in Spain. Epidemiological data from hospitalized patients are scarce and the association between hospital characteristics and patient outcomes is largely unknown. The aim of this study was to identify the factors associated with in-hospital mortality and readmissions and to analyze the relationship between hospital characteristics and outcomes.

METHODS

A retrospective analysis of discharges with HF as the principal diagnosis at hospitals of the Spanish National Health System in 2012 was performed using the Minimum Basic Data Set. We calculated risk-standardized mortality rates (RSMR) at the index episode and risk-standardized cardiac diseases readmissions rates (RSRR) and in-hospital mortality at 30 days and 1 year after discharge by using a multivariate mixed model.

RESULTS

We included 77 652 HF patients. Mean age was 79.2±9.9 years and 55.3% were women. In-hospital mortality during the index episode was 9.2%, rising to 14.5% throughout the year of follow-up. The 1-year cardiovascular readmissions rate was 32.6%. RSMR were lower among patients discharged from high-volume hospitals (> 340 HF discharges) (in-hospital RSMR, 10.3±5.6%; 8.6±2.2%); P <.001). High-volume hospitals had higher 1-year RSRR (32.3±3.7%; 33.7±4.5%; P=.006). The availability of a cardiology department at the hospital was associated with better outcomes (in-hospital RSMR, 9.9±3.8%; 9.2±2.4%; P <.001).

CONCLUSIONS

High-volume hospitals and the availability of a cardiology department were associated with lower in-hospital mortality.

Performance Characteristics of Profiling Methods and the Impact of Inadequate Case-mix Adjustment

Profiling or evaluation of health care providers involves the application of statistical models to compare each provider's performance with respect to a patient outcome, such as unplanned 30-day hospital readmission, adjusted for patient case-mix characteristics. The nationally adopted method is based on random effects (RE) hierarchical logistic regression models. Although RE models are sensible for modeling hierarchical data, novel high dimensional fixed effects (FE) models have been proposed which may be well-suited for the objective of identifying sub-standard performance. However, there are limited comparative studies. Thus, we examine their relative performance, including the impact of inadequate case-mix adjustment.

Double robust estimation for multiple unordered treatments and clustered observations: Evaluating drug-eluting coronary artery stents

Postmarket comparative effectiveness and safety analyses of therapeutic treatments typically involve large observational cohorts. We propose double robust machine learning estimation techniques for implantable medical device evaluations where there are more than two unordered treatments and patients are clustered in hospitals. This flexible approach also accommodates high-dimensional covariates drawn from clinical databases. The Massachusetts Data Analysis Center percutaneous coronary intervention cohort is used to assess the composite outcome of 10 drug-eluting stents among adults implanted with at least one drug-eluting stent in Massachusetts. We find remarkable discrimination between stents. A simulation study designed to mimic this coronary intervention cohort is also presented and produced similar results.

Center Variation in Medicare Spending for Durable Left Ventricular Assist Device Implant Hospitalizations

Importance

Hospitalizations for durable left ventricular assist device (LVAD) implants are expensive and increasingly common. Insights into center-level variation in Medicare spending for these hospitalizations are needed to inform value improvement efforts.

Objective

To examine center-level variation in Medicare spending for durable LVAD implant hospitalizations and its association with clinical outcomes.

Design, Setting, and Participants

Retrospective cohort study of linked Medicare administrative claims and Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS) clinical data comprising 106 centers in the United States providing durable LVAD implant. Centers were grouped into quartiles based on the mean price-standardized Medicare spending of their patients. The study included Medicare beneficiaries receiving primary durable LVAD implant between January 2008 and December 2014. Data were analyzed between November 2017 and October 2018.

Main Outcomes and Measures

Price-standardized Medicare payments and clinical outcomes. Overall and component (facility diagnosis-related group payments, outlier payments, physician services) payments and clinical outcomes (postimplant length of stay and adverse events) were compared across payment quartiles.

Results

The study sample included 4442 hospitalized patients, with mean (SD) age of 63.0 (10.8) years, 18.7% female, 27.2% nonwhite, and 6.1% Hispanic ethnicity. Among 4442 hospitalizations, the mean (SD) price-standardized Medicare payment was $176 825 ($60 286) and ranged from $122 953 to $271 472 across 106 centers. The difference in price-standardized payments between lowest and highest spending quartiles was $55 446 ($152 714 vs $208 160; 36%; P < .001), with outlier payments making up most of the difference ($42 742; 77%), followed by DRG ($6929; 13%) and physician services ($5774; 10%). After risk standardization, there was a modest decline in the difference in payments between quartiles ($53 221; 35%), with outlier payments accounting for a larger proportion of the difference (84%). After adjusting for patient characteristics, higher price-standardized payment quartiles were associated with longer postimplant length of stay but were not associated with any adverse events.

Conclusions and Relevance

Medicare payments for durable LVAD implant hospitalizations vary widely across centers; this was not well explained by prices or case mix. While associated with longer postimplant length of stay, increased spending was not associated with adverse events. As the supply and demand for durable LVAD therapy continues to rise, identifying opportunities to reduce variation in spending from both explained and unexplained sources will ensure high-value use.

Profiling Hospital Performance Based on Mortality After Transcatheter Aortic Valve Replacement in Ontario, Canada

BACKGROUND

Public reporting of hospital-level outcomes is increasingly common as a means to target quality improvement strategies to ensure the delivery of optimal care. Despite the rapid dissemination of transcatheter aortic valve replacement (TAVR), there is a paucity of reliable case-mix adjustment models for hospital profiling in TAVR. Our objective was to develop and evaluate different models for calculating risk-standardized all-cause mortality rates (RSMRs) post-TAVR.

METHODS AND RESULTS

In this population-based study in Ontario, Canada, we identified all patients who underwent a TAVR procedure between April 1, 2012, and March 31, 2016. For each hospital, we calculated 30-day and 1-year RSMR, using 2-level hierarchical logistic regression models that accounted for patient-specific demographic and clinical characteristics, as well as the clustering of patients within the same hospital using a hospital-specific random effects. We classified each hospital into one of 3 groups: performing worse than expected, better than expected, or performing as expected, based on whether the 95% CI of the RSMR was above, below, or included the provincial average mortality rate, respectively. Our cohort consisted of 2129 TAVR procedures performed at 10 hospitals. The observed mortality was 7.0% at 30 days and 16.4% at 1 year, with a range of 4% to 10% and 8% to 22%, respectively, across hospitals. We developed case-mix adjustment models using 28 clinically relevant variables. Using 30-day and 1-year RSMR to profile each hospital, we found that all hospitals performed as expected, with 95% CI that included the provincial average.

CONCLUSIONS

We found no significant interhospital variation in RSMR among hospitals, suggesting that quality improvement efforts should be directed at aspects other than the variation in observed mortality.

Impact of Case-Mix Measurement Error on Estimation and Inference in Profiling of Health Care Providers

Profiling analysis aims to evaluate health care providers by modeling each provider's performance with respect to a patient outcome, such as unplanned hospital readmission. High-dimensional regression models are used in profiling to risk-adjust for patient case-mix covariates. Case-mix covariates typically ascertained from administrative databases are inherently error-prone. We examine the impact of case-mix measurement error (ME) on profiling models. The results show that even though the models' coefficient estimates are biased, this does not affect the estimation of standardized readmission ratio (SRR). However, ME leads to increased variation in SRR estimates and degrades the ability to identify under-performing providers.

Contemporary use of intra-aortic balloon pumps during percutaneous coronary intervention: insights from the Veterans Affairs Clinical Assessment, Reporting, and Tracking program

BACKGROUND

Intra-aortic balloon pumps (IABP) support nonemergent and emergent percutaneous coronary intervention (PCI). Recent studies have not showed a routine benefit to this practice. We sought to evaluate the temporal trends in balloon pump utilization and site-level variation within a large integrated healthcare system.

PATIENTS AND METHODS

We identified all patients that underwent PCI in the Veterans Affairs Healthcare System between 1 January 2008 and 31 December 2015. Procedural information was ascertained from the medical record and stratified by the concomitant use of an IABP. Site-specific variation was determined with mixed logistic regression models and reported as a median odds ratio.

RESULTS

There were 88 851 interventions performed on 71 529 patients across 71 hospitals with 1289 (1.5%) of these utilizing an IABP. Patients that underwent an intervention with this device had more medical comorbidities, as reflected by an increase in the median National Cardiovascular Data Registry CathPCI mortality score (34 vs. 15, P<0.001). The overall utilization of balloon pumps was constant throughout the study period (P=0.446). However, there was a significant decline (P=0.027) in its use during emergent cases with a significant increase (P=0.009) during nonemergent cases. Furthermore, there was site variation in use independent of patient or procedural characteristics (median odds ratio: 1.82, 95% confidence interval: 1.58-2.16).

CONCLUSION

In the largest integrated healthcare system in the USA, there was a significant decline in IABP use among emergent cases and a significant increase during nonemergent cases. Residual site variation suggests an opportunity to standardize a procedural approach consistent with currently available data.

A Quantitative Observational Study of Physician Influence on Hospital Costs

Physicians serve as the nexus of treatment decision-making in hospitalized patients; however, little empirical evidence describes the influence of individual physicians on hospital costs. In this study, we examine the extent to which hospital costs vary across physicians and physician characteristics. We used all-payer data from 2 states representing 15 237 physicians and 2.5 million hospital visits. Regression analysis and propensity score matching were used to understand the role of observable provider characteristics on hospital costs controlling for patient demographics, socioeconomic characteristics, clinical risk, and hospital characteristics. We used hierarchical models to estimate the amount of variation attributable to physicians. We found that the average cost of hospital inpatient stays registered to female physicians was consistently lower across all empirical specifications when compared with male physicians. We also found a negative association between physicians’ years of experience and the average costs. The average cost of hospital inpatient stays registered to foreign-trained physicians was lower than US-trained physicians. We observed sizable variation in average costs of hospital inpatient stays across medical specialties. In addition, we used hierarchical methods and estimated the amount of remaining variation attributable to physicians and found that it was nonnegligible (intraclass correlation coefficient [ICC]: 0.33 in the full sample). Historically, most physicians have been reimbursed separately from hospitals, and our study shows that physicians play a role in influencing hospital costs. Future policies and practices should acknowledge these important dependencies. This study lends further support for alignment of physician and hospital incentives to control costs and improve outcomes.

Patient- and Hospital-Level Factors Associated With Readmission for Malignant Pleural Effusion

Purpose:

Readmission after hospitalization for malignant pleural effusion (MPE) may represent gaps in the quality of health care delivery. The goal of this study was to determine the frequency of 30-day readmission for MPE and identify clinical factors associated with rehospitalization.

Patients and Methods:

A retrospective cohort of adults hospitalized for MPE from 2009 to 2011 was analyzed using an administrative database. The primary outcome was all-cause 30-day readmission rate. Hierarchic mixed-effects logistic regression models were used to examine associations between patient- and hospital-level factors and 30-day readmission and assess variation in readmission rates across hospitals.

Results:

The 7-, 14-, 30-, 60-, and 90-day readmission rates for MPE were 16.1%, 25.9%, 38.3%, 52.5%, and 63.8%, respectively. The most common primary diagnoses for 30-day readmission were MPE (69.5%) and other clinical issues related to malignancy (21.1%). Clinical factors associated with 30-day readmission were female sex (odds ratio [OR], 0.78; 95% CI, 0.63 to 0.95), greater number of medical comorbidities (OR, 1.51; 95% CI, 1.15 to 1.99), and having a do-not-resuscitate order (OR, 1.37; 95% CI, 1.03 to 1.84). Hospitals in the 90th percentile were only 1.1 times more likely to have a 30-day readmission for MPE than those in the lowest 10th percentile (40.9% v 37%).

Conclusion:

Readmission for MPE is common and frequently results from progression of malignancy. Readmission rates were similar across all hospitals, suggesting they are unlikely to be mutable using conventional approaches to reduce rehospitalizations. Instead, interventions may need to focus on addressing care planning at the end of life.

Optimal Bayesian point estimates and credible intervals for ranking with application to county health indices

It is fairly common to rank different geographic units, e.g. counties in the USA, based on health indices. In a typical application, point estimates of the health indices are obtained for each county, and the indices are then simply ranked as if they were known constants. Several authors have considered optimal rank estimators under squared error loss on the rank scale as a default method for general purpose ranking, e.g. situations where ranking units across the full spectrum of performance (low, medium, high) is important. While computationally convenient, squared error loss on the rank scale may not represent the true inferential goals of rank consumers. We construct alternative loss functions based on three components: (1) the inferential goal (rank position or pairwise comparisons), (2) the scale (original, log-transformed or rank) and (3) the (positional or pairwise) loss function (0/1, squared error or absolute error). We can obtain optimal ranks for loss functions based on rank positions and nearly optimal ranks for loss functions based on pairwise comparisons paired with highest posterior density (HPD) credible intervals. We compare inferences produced by the various ranking methods, both optimal and heuristic, using low birth weight data for counties in the Midwestern United States, from 2006 to 2012.

Mortality Measures to Profile Hospital Performance for Patients With Septic Shock

OBJECTIVES

Sepsis care is becoming a more common target for hospital performance measurement, but few studies have evaluated the acceptability of sepsis or septic shock mortality as a potential performance measure. In the absence of a gold standard to identify septic shock in claims data, we assessed agreement and stability of hospital mortality performance under different case definitions.

DESIGN

Retrospective cohort study.

SETTING

U.S. acute care hospitals.

PATIENTS

Hospitalized with septic shock at admission, identified by either implicit diagnosis criteria (charges for antibiotics, cultures, and vasopressors) or by explicit International Classification of Diseases, 9th revision, codes.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

We used hierarchical logistic regression models to determine hospital risk-standardized mortality rates and hospital performance outliers. We assessed agreement in hospital mortality rankings when septic shock cases were identified by either explicit International Classification of Diseases, 9th revision, codes or implicit diagnosis criteria. Kappa statistics and intraclass correlation coefficients were used to assess agreement in hospital risk-standardized mortality and hospital outlier status, respectively. Fifty-six thousand six-hundred seventy-three patients in 308 hospitals fulfilled at least one case definition for septic shock, whereas 19,136 (33.8%) met both the explicit International Classification of Diseases, 9th revision, and implicit septic shock definition. Hospitals varied widely in risk-standardized septic shock mortality (interquartile range of implicit diagnosis mortality: 25.4-33.5%; International Classification of Diseases, 9th revision, diagnosis: 30.2-38.0%). The median absolute difference in hospital ranking between septic shock cohorts defined by International Classification of Diseases, 9th revision, versus implicit criteria was 37 places (interquartile range, 16-70), with an intraclass correlation coefficient of 0.72, p value of less than 0.001; agreement between case definitions for identification of outlier hospitals was moderate (kappa, 0.44 [95% CI, 0.30-0.58]).

CONCLUSIONS

Risk-standardized septic shock mortality rates varied considerably between hospitals, suggesting that septic shock is an important performance target. However, efforts to profile hospital performance were sensitive to septic shock case definitions, suggesting that septic shock mortality is not currently ready for widespread use as a hospital quality measure.

Emergency department provider and facility variation in opioid prescriptions for discharged patients

STUDY OBJECTIVE

To study the variation in opioid prescribing among emergency physicians and facilities for discharged adult ED patients.

METHODS

We conducted a retrospective analysis of ED visits from five U.S. hospitals between January and May 2014 using records from Data to Intelligence (D2i). We examined physician- and facility-level variation in opioid prescription rates for discharged ED patients. We calculated unadjusted opioid prescription rates at the physician and facility levels and used a multivariable mixed-effect logistic regression model to examine within-facility physician variation in opioid prescription adjusting for patient and situational factors including time of presentation, ED census, and physician workload.

RESULTS

In 47,304 visits across five EDs, median patient age was 40 years old (IQR 28,55), and 89% had some form of insurance. There were 17,098 (36%) ED discharges with at least one opioid prescription. The unadjusted facility-level opioid prescription rate ranged from 24%-46%. Among 253 ED physicians, the adjusted opioid prescription rate varied from 22%-76%. Increased physician workload is related to decreased odds of opioid prescription at ED discharge for the lowest (<3 patients) and moderate (6-9 patients) physician workload levels, while the association weakened with increasing levels of workload.

CONCLUSION

There was substantial physician and facility variation in opioid prescription for discharged adult ED patients. Emergency physicians were less likely to prescribe opioids when their workload was lower, and this effect diminished at high workload levels. Understanding situational and other factors that explain this variation is important given the rising U.S. opioid epidemic and the need for urgent intervention.

Time-dynamic profiling with application to hospital readmission among patients on dialysis

Standard profiling analysis aims to evaluate medical providers, such as hospitals, nursing homes, or dialysis facilities, with respect to a patient outcome. The outcome, for instance, may be mortality, medical complications, or 30-day (unplanned) hospital readmission. Profiling analysis involves regression modeling of a patient outcome, adjusting for patient health status at baseline, and comparing each provider's outcome rate (e.g., 30-day readmission rate) to a normative standard (e.g., national "average"). Profiling methods exist mostly for non time-varying patient outcomes. However, for patients on dialysis, a unique population which requires continuous medical care, methodologies to monitor patient outcomes continuously over time are particularly relevant. Thus, we introduce a novel time-dynamic profiling (TDP) approach to assess the time-varying 30-day readmission rate. TDP is used to estimate, for the first time, the risk-standardized time-dynamic 30-day hospital readmission rate, throughout the time period that patients are on dialysis. We develop the framework for TDP by introducing the standardized dynamic readmission ratio as a function of time and a multilevel varying coefficient model with facility-specific time-varying effects. We propose estimation and inference procedures tailored to the problem of TDP and to overcome the challenge of high-dimensional parameters when examining thousands of dialysis facilities.

Sex, race, and insurance status differences in hospital treatment and outcomes following out-of-hospital cardiac arrest

BACKGROUND

Sex, race, and insurance status are associated with treatment and outcomes in several cardiovascular diseases. These disparities, however, have not been well-studied in out-of-hospital cardiac arrest (OHCA).

OBJECTIVE

Our objective was to evaluate the association of patient sex, race, and insurance status with hospital treatments and outcomes following OHCA.

METHODS

We studied adult patients in the 2011-2015 California Office of Statewide Health Planning and Development (OSHPD) Patient Discharge Database with a "present on admission" diagnosis of cardiac arrest (ICD-9-CM 427.5). Insurance status was classified as private, Medicare, and Medi-Cal/government/self-pay. Our primary outcome was good neurologic recovery at hospital discharge, which was determined by discharge disposition. Secondary outcomes were survival to hospital discharge, treatment at a 24/7 percutaneous coronary intervention (PCI) center, "do not resuscitate" orders within 24 h of admission, and cardiac catheterization during hospitalization. Data were analyzed with hierarchical multiple logistic regression models.

RESULTS

We studied 38,163 patients in the OSHPD database. Female sex, non-white race, and Medicare insurance status were independently associated with worse neurologic recovery [OR 0.94 (0.89-0.98), 0.93 (0.88-0.98), and 0.85 (0.79-0.91), respectively], lower rates of treatment at a 24/7 PCI center [OR 0.89 (0.85-0.93), 0.88 (0.85-0.93), and 0.87 (0.82-0.94), respectively], and lower rates of cardiac catheterization [OR 0.61 (0.57-0.65), 0.90 (0.84-0.97), and 0.44 (0.40-0.48), respectively]. Female sex, white race, and Medicare insurance were associated with DNR orders within 24 h of admission [OR 1.16 (1.10-1.23), 1.14 (1.07-1.21), and 1.25 (1.15-1.36), respectively].

CONCLUSIONS

Sex, race, and insurance status were independently associated with post-arrest care interventions, patient outcomes and treatment at a 24/7 PCI center. More studies are needed to fully understand the causes and implications of these disparities.

Hospital Variation in Risk-Adjusted Pediatric Sepsis Mortality

OBJECTIVES

With continued attention to pediatric sepsis at both the clinical and policy levels, it is important to understand the quality of hospitals in terms of their pediatric sepsis mortality. We sought to develop a method to evaluate hospital pediatric sepsis performance using 30-day risk-adjusted mortality and to assess hospital variation in risk-adjusted sepsis mortality in a large state-wide sample.

DESIGN

Retrospective cohort study using administrative claims data.

SETTINGS

Acute care hospitals in the state of Pennsylvania from 2011 to 2013.

PATIENTS

Patients between the ages of 0-19 years admitted to a hospital with sepsis defined using validated International Classification of Diseases, Ninth revision, Clinical Modification, diagnosis and procedure codes.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

During the study period, there were 9,013 pediatric sepsis encounters in 153 hospitals. After excluding repeat visits and hospitals with annual patient volumes too small to reliably assess hospital performance, there were 6,468 unique encounters in 24 hospitals. The overall unadjusted mortality rate was 6.5% (range across all hospitals: 1.5-11.9%). The median number of pediatric sepsis cases per hospital was 67 (range across all hospitals: 30-1,858). A hierarchical logistic regression model for 30-day risk-adjusted mortality controlling for patient age, gender, emergency department admission, infection source, presence of organ dysfunction at admission, and presence of chronic complex conditions showed good discrimination (C-statistic = 0.80) and calibration (slope and intercept of calibration plot: 0.95 and -0.01, respectively). The hospital-specific risk-adjusted mortality rates calculated from this model varied minimally, ranging from 6.0% to 7.4%.

CONCLUSIONS

Although a risk-adjustment model for 30-day pediatric sepsis mortality had good performance characteristics, the use of risk-adjusted mortality rates as a hospital quality measure in pediatric sepsis is not useful due to the low volume of cases at most hospitals. Novel metrics to evaluate the quality of pediatric sepsis care are needed.

Observed to expected or logistic regression to identify hospitals with high or low 30-day mortality?

INTRODUCTION

A common quality indicator for monitoring and comparing hospitals is based on death within 30 days of admission. An important use is to determine whether a hospital has higher or lower mortality than other hospitals. Thus, the ability to identify such outliers correctly is essential. Two approaches for detection are: 1) calculating the ratio of observed to expected number of deaths (OE) per hospital and 2) including all hospitals in a logistic regression (LR) comparing each hospital to a form of average over all hospitals. The aim of this study was to compare OE and LR with respect to correctly identifying 30-day mortality outliers. Modifications of the methods, i.e., variance corrected approach of OE (OE-Faris), bias corrected LR (LR-Firth), and trimmed mean variants of LR and LR-Firth were also studied.

MATERIALS AND METHODS

To study the properties of OE and LR and their variants, we performed a simulation study by generating patient data from hospitals with known outlier status (low mortality, high mortality, non-outlier). Data from simulated scenarios with varying number of hospitals, hospital volume, and mortality outlier status, were analysed by the different methods and compared by level of significance (ability to falsely claim an outlier) and power (ability to reveal an outlier). Moreover, administrative data for patients with acute myocardial infarction (AMI), stroke, and hip fracture from Norwegian hospitals for 2012-2014 were analysed.

RESULTS

None of the methods achieved the nominal (test) level of significance for both low and high mortality outliers. For low mortality outliers, the levels of significance were increased four- to fivefold for OE and OE-Faris. For high mortality outliers, OE and OE-Faris, LR 25% trimmed and LR-Firth 10% and 25% trimmed maintained approximately the nominal level. The methods agreed with respect to outlier status for 94.1% of the AMI hospitals, 98.0% of the stroke, and 97.8% of the hip fracture hospitals.

CONCLUSION

We recommend, on the balance, LR-Firth 10% or 25% trimmed for detection of both low and high mortality outliers.

Variation in mortality rates after admission to long-term acute care hospitals for ventilator weaning

PURPOSE

We sought to examine variation in long-term acute care hospital (LTACH) quality based on 90-day in-hospital mortality for patients admitted for weaning from mechanical ventilation.

METHODS

We developed an administrative risk-adjustment model using data from Medicare claims. We validated the administrative model against a clinical model using data from LTACHs participating in a 2002 to 2003 clinical registry. We then used our validated administrative model to assess national variation in 90-day in-hospital mortality rates in LTACHs from 2013.

RESULTS

The administrative risk-adjustment model was derived using data from 9447 patients admitted to 221 LTACHs in 2003. The model had good discrimination (C statistic=0.72) and calibration. Compared to a clinically derived model using data from 1163 patients admitted to 14 LTACHs, the administrative model generated similar performance estimates. National variation in risk-adjusted mortality was assessed using data from 20,453 patients admitted to 380 LTACHs in 2013. LTACH-specific risk-adjusted mortality rates varied from 8.4% to 48.1% (median: 24.2%, interquartile range: 19.7%-30.7%).

CONCLUSIONS

LTACHs vary widely in mortality rates, underscoring the need to better understand the sources of this variation and improve the quality of care for patients requiring long-term ventilator weaning.

Separating variability in healthcare practice patterns from random error

Improving the quality of care that patients receive is a major focus of clinical research, particularly in the setting of cardiovascular hospitalization. Quality improvement studies seek to estimate and visualize the degree of variability in dichotomous treatment patterns and outcomes across different providers, whereby naive techniques either over-estimate or under-estimate the actual degree of variation. Various statistical methods have been proposed for similar applications including (1) the Gaussian hierarchical model, (2) the semi-parametric Bayesian hierarchical model with a Dirichlet process prior and (3) the non-parametric empirical Bayes approach of smoothing by roughening. Alternatively, we propose that a recently developed method for density estimation in the presence of measurement error, moment-adjusted imputation, can be adapted for this problem. The methods are compared by an extensive simulation study. In the present context, we find that the Bayesian methods are sensitive to the choice of prior and tuning parameters, whereas moment-adjusted imputation performs well with modest sample size requirements. The alternative approaches are applied to identify disparities in the receipt of early physician follow-up after myocardial infarction across 225 hospitals in the CRUSADE registry.

The Effect of Intensive Care Unit Admission Patterns on Mortality-based Critical Care Performance Measures

RATIONALE

Current mortality-based critical care performance measurement focuses on intensive care unit (ICU) admissions as a single group, conflating low-severity and high-severity ICU patients for whom performance may differ and neglecting severely ill patients treated solely on hospital wards.

OBJECTIVES

To assess the relationship between hospital performance as measured by risk-standardized mortality for severely ill ICU patients, less severely ill ICU patients, and severely ill patients outside the ICU.

METHODS

Using a statewide, all-payer dataset from the Pennsylvania Healthcare Cost Containment Council, we analyzed discharge data for patients with nine clinical conditions with frequent ICU use. Using a validated severity-of-illness measure, we categorized hospitalized patients as either high severity (predicted probability of in-hospital death in top quartile) or low severity (all others). We then created three mutually exclusive groups: high-severity ICU admissions, low-severity ICU admissions, and high-severity ward patients. We used hierarchical logistic regression to generate hospital-specific 30-day risk-standardized mortality rates for each group and then compared hospital performance across groups using Spearman's rank correlation.

MEASUREMENTS AND MAIN RESULTS

We analyzed 87 hospitals with 22,734 low-severity ICU admissions (mean per hospital, 261 ± 187), 10,991 high-severity ICU admissions (mean per hospital, 126 ± 105), and 6,636 high-severity ward patients (mean per hospital, 76 ± 48). We found little correlation between hospital performance for high-severity ICU patients versus low-severity ICU patients (ρ = 0.15; P = 0.17). There were 29 hospitals (33%) that moved up or down at least two quartiles of performance across the ICU groups. There was weak correlation between hospital performance for high-severity ICU patients versus high-severity ward patients (ρ = 0.25; P = 0.02). There were 24 hospitals (28%) that moved up or down at least two quartiles of performance across the high-severity groups.

CONCLUSIONS

Hospitals that perform well in caring for high-severity ICU patients do not necessarily also perform well in caring for low-severity ICU patients or high-severity ward patients, indicating that risk-standardized mortality rates for ICU admissions as a whole offer only a narrow window on a hospital's overall performance for critically ill patients.

RECALAD. Patient care at National Health System Digestive Care Units - A pilot study, 2015

OBJECTIVES

To reach a situation diagnosis on the status of patient management at digestive care units (DCUs) in Spain.

MATERIAL AND METHODS

A cross-sectional descriptive study across DCUs in general acute care hospitals within the Spanish National Health System (data referred to 2015). The study variables were collected with a questionnaire including items on structure, services portfolio, activity, education, research, and good practice. Hospital discharge rates for digestive diseases were also assessed using the minimum basic data set (2005-2014).

RESULTS

Two hundred and nine hospitals invited, 55 responders (26.3%). Average discharges from hospital were 1,139 ± 653 per DCU/year, and 100 ± 66 per year per dedicated gastroenterologist. In 2014, admission rate to DCUs per 1,000 population and year was 280, with a mean stay of 7.4 days. The analysis of the MBDS for 2005-2014 reveals a progressive increase in the number of discharges (37% more in 2014 versus 2005), with a 28% decrease in hospital gross mortality rate (3.7% in 2014) and a slightly reduced (14%) mean stay (7.6 days in 2014). Considerable variability may be seen in structure, activity, and results indicators. Mortality and readmission rates, as well as mean stay, vary more than 100% amongst DCUs, and major dispersions also exist in frequentation and results amongst autonomous communities.

CONCLUSIONS

The RECALAD 2015 survey unveiled relevant aspects related to DCUs organization, structure, and management. The notable variability encountered likely reflects relevant differences in efficiency and productivity, and thus points out there is ample room for improvement.

The role of the clinical departments for understanding patient heterogeneity in one-year mortality after a diagnosis of heart failure: A multilevel analysis of individual heterogeneity for profiling provider outcomes

Purpose

To evaluate the general contextual effect (GCE) of the hospital department on one-year mortality in Swedish and Danish patients with heart failure (HF) by applying a multilevel analysis of individual heterogeneity.

Methods

Using the Swedish patient register, we obtained data on 36,943 patients who were 45–80 years old and admitted for HF to the hospital between 2007 and 2009. From the Danish Heart Failure Database (DHFD), we obtained data on 12,001 patients with incident HF who were 18 years or older and treated at hospitals between June 2010 and June2013. For each year, we applied two-step single and multilevel logistic regression models. We evaluated the general effects of the department by quantifying the intra-class correlation coefficient (ICC) and the increment in the area under the receiver operating characteristic curve (AUC) obtained by adding the random effects of the department in a multilevel logistic regression analysis.

Results

One-year mortality for Danish incident HF patients was low in the three audit years (around 11.1% -13.1%) and departments performed homogeneously (ICC ≈1.5% - 3.5%). The discriminatory accuracy of a model including age and gender was rather high (AUC≈ 0.71–0.73) but the increment in AUC after adding the department random effects into these models was only about 0.011–0.022 units in the three years.

One-year mortality in Swedish patients with first hospitalization for heart failure, was relatively higher for 2007–2009 (≈21.3% - 22%) and departments performed homogeneously (ICC ≈ 1.5% - 3%). The discriminatory accuracy of a model including age, gender and patient risk score was rather high (AUC≈ 0.726–0.728) but the increment in AUC after adding the department random effects was only about 0.010–0.017 units in the three years.

Conclusion

Using the DHFD standard benchmark for one-year mortality, Danish departments had a good, homogeneous performance. In reference to literature, Swedish departments had a homogeneous performance and the mortality rates for patients with first hospitalization for heart failure were similar to those reported since 2000. Considering this, if health authorities decide to further reduce mortality rates, a comprehensive quality strategy should focus on all Swedish hospitals. Yet, a complementary assessment for the period after the study period is required to confirm whether department performance is still homogeneous or not to determine the most appropriate action.

Regional Medicare Expenditures and Survival Among Older Women With Localized Breast Cancer

BACKGROUND

Despite evidence on large variation in breast cancer expenditures across geographic regions, there is little understanding about the association between expenditures and patient outcomes.

OBJECTIVES

To examine whether Medicare beneficiaries with nonmetastatic breast cancer living in regions with higher cancer-related expenditures had better survival.

RESEARCH DESIGN

A retrospective cohort study of women with localized breast cancer from the Surveillance, Epidemiology, and End Results-Medicare linked database. Hospital referral regions (HRR) were categorized into quintiles based on risk-standardized per patient Medicare expenditures on initial phase of breast cancer care. Hierarchical generalized linear models were estimated to examine the association between patients' HRR quintile and survival.

SUBJECTS

In total, 12,610 Medicare beneficiaries diagnosed with stage II-III breast cancer during 2005-2008 who underwent surgery.

MEASURES

Outcome measures for our analysis were 3- and 5-year overall survival.

RESULTS

Risk-standardized per patient Medicare expenditures on initial phase of breast cancer care ranged from $13,338 to $26,831 across the HRRs. Unadjusted 3- and 5-year survival varied from 66.7% to 92.2% and 50.0% to 84.0%, respectively, across the HRRs, but there was no significant association between HRR quintile and survival in bivariate analysis (P=0.08 and 0.28, respectively). After adjustment for sociodemographic and clinical characteristics, quintiles of regional cancer expenditures remained unassociated with patients' 3-year (P=0.35) and 5-year survival (P=0.20). Further analysis adjusting for treatment factors (surgery type and receipt of radiation and systemic therapy) and stratifying by cancer stage showed similar results.

CONCLUSIONS

For Medicare beneficiaries with nonmetastatic breast cancer, residence in regions with higher breast cancer-related expenditures was not associated with better survival. More attention to value in breast cancer care is warranted.

Variation in Management of Patients With Obstructive Coronary Artery Disease: Insights From the Veterans Affairs Clinical Assessment and Reporting Tool (VA CART) Program

BACKGROUND

Little is known about facility-level variation in the use of revascularization procedures for the management of stable obstructive coronary artery disease. Furthermore, it is unknown if variation in the use of coronary revascularization is associated with use of other cardiovascular procedures.

METHODS AND RESULTS

We evaluated all elective coronary angiograms performed in the Veterans Affairs system between September 1, 2007, and December 31, 2011, using the Clinical Assessment and Reporting Tool and identified patients with obstructive coronary artery disease. Patients were considered managed with revascularization if they received percutaneous coronary intervention (PCI) or coronary artery bypass grafting within 30 days of diagnosis. We calculated risk-adjusted facility-level rates of overall revascularization, PCI, and coronary artery bypass grafting. In addition, we determined the association between facility-level rates of revascularization and post-PCI stress testing. Among 15 650 patients at 51 Veterans Affairs sites who met inclusion criteria, the median rate of revascularization was 59.6% (interquartile range, 55.7%-66.7%). Across all facilities, risk-adjusted rates of overall revascularization varied from 41.5% to 88.1%, rate of PCI varied from 23.2% to 80.6%, and rate of coronary artery bypass graftingvariedfrom 7.5% to 36.5%. Of 6179 patients who underwent elective PCI, the median rate of stress testing in the 2 years after PCI was 33.7% (interquartile range, 30.7%-47.1%). There was no evidence of correlation between facility-level rate of revascularization and follow-up stress testing.

CONCLUSIONS

Within the Veterans Affairs system, we observed large facility-level variation in rates of revascularization for obstructive coronary artery disease, with variation driven primarily by PCI. There was no association between facility-level use of revascularization and follow-up stress testing, suggesting use rates are specific to a particular procedure and not a marker of overall facility-level use.

RECALMIN II. Eight years of hospitalisation in Internal Medicine Units (2007-2014). What has changed?

OBJECTIVES

To analyse the evolution of care provided by the internal medicine units (IMU) of the Spanish National Health System from 2007 to 2014.

MATERIAL AND METHODS

We analysed all discharges from the IMU of the Spanish National Health System in 2007 and 2014, using the Minimum Basic Data Set. We compared the risk factors by episode, mortality and readmissions between the two periods. We prepared specific fits for the risk for mortality and readmissions in heart failure, pneumonia and chronic obstructive pulmonary disease, as well as the Charlson index for all activity.

RESULTS

Discharges from the IMU between the two periods increased 14%. The average patient age increased by 2.8 years (71.2±17.1 vs. 74±16.2; p<.001), with a marked increase in comorbidity (Charlson index, 4±3.7 vs. 4.7±3.9; p<.001; 24% increase in risk factors per episode). The adjusted mortality rates decreased slight but significantly, with a slight increase in readmissions.

CONCLUSIONS

During the analysed period, there was an increase of almost 3 years in the mean age of patients treated in the IMU of the Spanish National Health System, with a marked increase in comorbidity. These results should lead to a more appropriate assignment of nurse workloads and an increased implementation of good practices in clinical management.

Interhospital Comparison of Surgical Site Infection Rates in Orthopedic Surgery

OBJECTIVE To investigate whether comparison by deep or adjusted deep surgical site infection (SSI) rates in orthopedic surgeries are a better basis for feedback to Finnish hospitals than overall SSI rates DESIGN Retrospective cohort study SETTING Hospitals conducting surveillance of hip arthroplasties (HPROs) and knee arthroplasties (KPROs) in the Finnish Hospital Infection Program METHODS We analyzed surveillance data for 73,227 HPROs and 56,860 KPROs performed in 18 hospitals during 1999-2014. For each hospital, the overall, deep, and adjusted deep SSI rates with 95% confidence intervals (CIs) were calculated, and the hospital ranks were simulated in the Bayesian framework. Adjustments were performed using relevant patient and hospital characteristics. The correlation between the median expected hospital ranks in overall versus deep SSI rates and deep vs adjusted deep SSI rates were assessed using Spearman's correlation coefficient ρ. RESULTS For HPRO, the overall SSI rates ranged from 0.92 to 6.83, the deep SSI rates ranged from 0.34 to 1.86, and the adjusted deep hospital-specific SSI rates ranged from 0.37 to 1.85. For KPRO, the overall SSI rates ranged from 0.71 to 5.03, the deep SSI rates ranged from 0.42 to 1.60, and the adjusted deep hospital-specific SSI rates ranged from 0.56 to 1.55. For both procedures, the 95% CIs of the rates between hospitals largely overlapped; only single outliers were detected. Hospital rank did not correlate between overall and deep SSI rates (HPRO, ρ=0.03; KPRO, ρ=0.40), but a correlation was observed in hospital rank for deep and adjusted deep SSI rates (HPRO, ρ=0.85; KPRO, ρ=0.94). CONCLUSION Deep SSI rates may be a better tool for interhospital comparisons than overall SSI rates. Although the adjustment could lead to fairer hospital ranking, it is not always necessary for feedback. Infect Control Hosp Epidemiol 2017;38:423-429.

Assessing Hospital Performance After Percutaneous Coronary Intervention Using Big Data

BACKGROUND

Although risk adjustment remains a cornerstone for comparing outcomes across hospitals, optimal strategies continue to evolve in the presence of many confounders. We compared conventional regression-based model to approaches particularly suited to leveraging big data.

METHODS AND RESULTS

We assessed hospital all-cause 30-day excess mortality risk among 8952 adults undergoing percutaneous coronary intervention between October 1, 2011, and September 30, 2012, in 24 Massachusetts hospitals using clinical registry data linked with billing data. We compared conventional logistic regression models with augmented inverse probability weighted estimators and targeted maximum likelihood estimators to generate more efficient and unbiased estimates of hospital effects. We also compared a clinically informed and a machine-learning approach to confounder selection, using elastic net penalized regression in the latter case. Hospital excess risk estimates range from -1.4% to 2.0% across methods and confounder sets. Some hospitals were consistently classified as low or as high excess mortality outliers; others changed classification depending on the method and confounder set used. Switching from the clinically selected list of 11 confounders to a full set of 225 confounders increased the estimation uncertainty by an average of 62% across methods as measured by confidence interval length. Agreement among methods ranged from fair, with a κ statistic of 0.39 (SE: 0.16), to perfect, with a κ of 1 (SE: 0.0).

CONCLUSIONS

Modern causal inference techniques should be more frequently adopted to leverage big data while minimizing bias in hospital performance assessments.

Reliability of 30-Day Readmission Measures Used in the Hospital Readmission Reduction Program

OBJECTIVE

To assess the reliability of risk-standardized readmission rates (RSRRs) for medical conditions and surgical procedures used in the Hospital Readmission Reduction Program (HRRP).

DATA SOURCES

State Inpatient Databases for six states from 2011 to 2013 were used to identify patient cohorts for the six conditions used in the HRRP, which was augmented with hospital characteristic and HRRP penalty data.

STUDY DESIGN

Hierarchical logistic regression models estimated hospital-level RSRRs for each condition, the reliability of each RSRR, and the extent to which socioeconomic and hospital factors further explain RSRR variation. We used publicly available data to estimate payments for excess readmissions in hospitals with reliable and unreliable RSRRs.

PRINCIPAL FINDINGS

Only RSRRs for surgical procedures exceeded the reliability benchmark for most hospitals, whereas RSRRs for medical conditions were typically below the benchmark. Additional adjustment for socioeconomic and hospital factors modestly explained variation in RSRRs. Approximately 25 percent of payments for excess readmissions were tied to unreliable RSRRs.

CONCLUSIONS

Many of the RSRRs employed by the HRRP are unreliable, and one quarter of payments for excess readmissions are associated with unreliable RSRRs. Unreliable measures blur the connection between hospital performance and incentives, and threaten the success of the HRRP.

Hierarchical models for semi-competing risks data with application to quality of end-of-life care for pancreatic cancer

Readmission following discharge from an initial hospitalization is a key marker of quality of health care in the United States. For the most part, readmission has been studied among patients with 'acute' health conditions, such as pneumonia and heart failure, with analyses based on a logistic-Normal generalized linear mixed model (Normand et al., 1997). Naïve application of this model to the study of readmission among patients with 'advanced' health conditions such as pancreatic cancer, however, is problematic because it ignores death as a competing risk. A more appropriate analysis is to imbed such a study within the semi-competing risks framework. To our knowledge, however, no comprehensive statistical methods have been developed for cluster-correlated semi-competing risks data. To resolve this gap in the literature we propose a novel hierarchical modeling framework for the analysis of cluster-correlated semi-competing risks data that permits parametric or non-parametric specifications for a range of components giving analysts substantial flexibility as they consider their own analyses. Estimation and inference is performed within the Bayesian paradigm since it facilitates the straightforward characterization of (posterior) uncertainty for all model parameters, including hospital-specific random effects. Model comparison and choice is performed via the deviance information criterion and the log-pseudo marginal likelihood statistic, both of which are based on a partially marginalized likelihood. An efficient computational scheme, based on the Metropolis-Hastings-Green algorithm, is developed and had been implemented in the SemiCompRisks R package. A comprehensive simulation study shows that the proposed framework performs very well in a range of data scenarios, and outperforms competitor analysis strategies. The proposed framework is motivated by and illustrated with an on-going study of the risk of readmission among Medicare beneficiaries diagnosed with pancreatic cancer. Using data on n=5,298 patients at J=112 hospitals in the six New England states between 2000-2009, key scientific questions we consider include the role of patient-level risk factors on the risk of readmission and the extent of variation in risk across hospitals not explained by differences in patient case-mix.

Making the cut: improved ranking and selection for large-scale inference

Identifying leading measurement units from a large collection is a common inference task in various domains of large-scale inference. Testing approaches, which measure evidence against a null hypothesis rather than effect magnitude, tend to overpopulate lists of leading units with those associated with low measurement error. By contrast, local maximum likelihood (ML) approaches tend to favor units with high measurement error. Available Bayesian and empirical Bayesian approaches rely on specialized loss functions that result in similar deficiencies. We describe and evaluate a generic empirical Bayesian ranking procedure that populates the list of top units in a way that maximizes the expected overlap between the true and reported top lists for all list sizes. The procedure relates unit-specific posterior upper tail probabilities with their empirical distribution to yield a ranking variable. It discounts high-variance units less than popular non-ML methods and thus achieves improved operating characteristics in the models considered.