Hospital Phenotypes in the Management of Patients Admitted for Acute Myocardial Infarction.

OBJECTIVES: To characterize hospital phenotypes by their combined utilization pattern of percutaneous coronary interventions (PCI), coronary artery bypass grafting (CABG) procedures, and intensive care unit (ICU) admissions for patients hospitalized for acute myocardial infarction (AMI). RESEARCH DESIGN: Using the Premier Analytical Database, we identified 129,138 hospitalizations for AMI from 246 hospitals with the capacity for performing open-heart surgery during 2010-2013. We calculated year-specific, risk-standardized estimates of PCI procedure rates, CABG procedure rates, and ICU admission rates for each hospital, adjusting for patient clinical characteristics and within-hospital correlation of patients. We used a mixture modeling approach to identify groups of hospitals (ie, hospital phenotypes) that exhibit distinct longitudinal patterns of risk-standardized PCI, CABG, and ICU admission rates. RESULTS: We identified 3 distinct phenotypes among the 246 hospitals: (1) high PCI-low CABG-high ICU admission (39.2% of the hospitals), (2) high PCI-low CABG-low ICU admission (30.5%), and (3) low PCI-high CABG-moderate ICU admission (30.4%). Hospitals in the high PCI-low CABG-high ICU admission phenotype had significantly higher risk-standardized in-hospital costs and 30-day risk-standardized payment yet similar risk-standardized mortality and readmission rates compared with hospitals in the low PCI-high CABG-moderate ICU admission phenotype. Hospitals in these phenotypes differed by geographic region. CONCLUSIONS: Hospitals differ in how they manage patients hospitalized for AMI. Their distinctive practice patterns suggest that some hospital phenotypes may be more successful in producing good outcomes at lower cost.

Measuring Adverse Events in Hospitalized Patients: An Administrative Method for Measuring Harm.

CONTEXT: Current methods for tracking harm either require costly full manual chart review (FMCR) or rely on proxy methods that have questionable accuracy. We propose an administrative measure of harm detection that uses electronically captured data. OBJECTIVE: Determine the level of agreement on harm event occurrence when harm is detected based on an administrative harm measurement tool (AHMT) compared with FMCR. DESIGN: A retrospective chart review was used to measure the level of agreement in harm detection between an AHMT that uses electronically captured data and a FMCR. SETTING: The inpatient hospital setting was used. PATIENTS: Approximately 771 medical records from 5 hospitals were reviewed. MAIN OUTCOME MEASURES: Measures of positive predictive value, negative predictive value, weighted sensitivity, weighted specificity, and concordance were used to evaluate agreement between the 2 methods. RESULTS: Although there was agreement at the harm-event level, the results were not all as high as desired: adjusted sensitivity 65%, adjusted specificity 85%, positive predictive value (PPV) 59%, negative predictive value (NPV) 88%, and concordance 75%. The patient-level results show greater agreement: adjusted sensitivity 95%, adjusted specificity 86%, PPV 61%, NPV 99%, and concordance 81%. CONCLUSION: The AHMT is sufficiently accurate for use as a within hospital tool to reliably detect and track harm. Nevertheless, it is not recommended as a tool to make comparisons across institutions, which has policy and payment implications. Further research using administrative harm detection, including the use of a broader set of measures and electronic health records, is needed.

"Phenotyping" hospital value of care for patients with heart failure.

OBJECTIVE: To characterize hospitals based on patterns of their combined financial and clinical outcomes for heart failure hospitalizations longitudinally. DATA SOURCE: Detailed cost and administrative data on hospitalizations for heart failure from 424 hospitals in the 2005-2011 Premier database. STUDY DESIGN: Using a mixture modeling approach, we identified groups of hospitals with distinct joint trajectories of risk-standardized cost (RSC) per hospitalization and risk-standardized in-hospital mortality rate (RSMR), and assessed hospital characteristics associated with the distinct patterns using multinomial logistic regression. PRINCIPAL FINDINGS: During 2005-2011, mean hospital RSC decreased from $12,003 to $10,782, while mean hospital RSMR declined from 3.9 to 3.2 percent. We identified five distinct hospital patterns: highest cost and low mortality (3.2 percent of the hospitals), high cost and low mortality (20.4 percent), medium cost and low mortality (34.6 percent), medium cost and high mortality (6.2 percent), and low cost and low mortality (35.6 percent). Longer hospital stay and greater use of intensive care unit and surgical procedures were associated with phenotypes with higher costs or greater mortality. CONCLUSIONS: Hospitals vary substantially in the joint longitudinal patterns of cost and mortality, suggesting marked difference in value of care. Understanding determinants of the variation will inform strategies for improving the value of hospital care.

Making hospital mortality measurement more meaningful: incorporating advance directives and palliative care designations.

Accounting for patients admitted to hospitals at the end of a terminal disease process is key to signaling care quality and identifying opportunities for improvement. This study evaluates the benefits and caveats of incorporating care-limiting orders, such as do not resuscitate (DNR) and palliative care (PC) information, in a general multivariate model of mortality risk, wherein the unit of observation is the patient hospital encounter. In a model of the mortality gap (observed - expected from the baseline model), DNR explains 8% to 24% of the gap variation. PC provides additional explanatory power to some disease groupings, especially heart and digestive diseases. One caveat is that DNR information, especially if associated with the later stages of hospital care, may mask opportunities to improve care for certain types of patients. But that is not a danger for PC, which is unequivocally valuable in accounting for patient risk, especially for certain subpopulations and disease groupings.