Faculty Bibliography

BACKGROUND: Interventions to reduce readmission following hospitalization for acute decompensated heart failure (ADHF) require early identification of patients. Electronic health record (EHR)-based approaches to identify patients can range in complexity from simple algorithms based on few data elements to machine learning algorithms using big data. The purpose of this study was to compare performance of a range of algorithms to identify patients with ADHF. Given the emphasis that hospitals currently place on patients with ADHF, we developed models with high sensitivity to avoid missed opportunities for care improvement; this approach assumed secondary chart review would be necessary to confirm a diagnosis. We therefore estimated the time needed for secondary review by providers following initial screening with each algorithm. METHODS: We performed a retrospective study of hospitalizations for patients age > 18 at an academicmedical center in 2013-2015. Using a random 75% development set, we developed four algorithms to identify hospitalizations with a principal diagnosis of ADHF using EHR data through the second midnight of hospitalization: 1) one of three characteristics: heart failure on the problem list, loop diuretic use, or brain natriuretic peptide > 500 pg/ml; 2) logistic regression of 31 clinically-relevant data elements; 3) machine learning approach using L1-regularization logistic regression and unstructured data, including provider notes and radiology reports; 4) machine learning with both structured and unstructured data. We assessed performance of each algorithm in the 25% validation set. We also conducted a brief survey of providers who perform chart review for ADHF to estimate time needed for secondary screening following primary screening with each algorithm. RESULTS: We included 37,229 hospitalizations in the study, of which 1,294 (3.5%) carried a principal diagnosis of ADHF. Algorithm 1 had a sensitivity of 0.98 and positive predictive value (PPV) of 0.14 for ADHF. Algorithm 2 had an area under the receiver operating characteristic curve (AUC) of 0.96 and a PPV of 0.15 when setting the sensitivity at 0.98. Both machine learning algorithms had AUCs of 0.99; with a sensitivity of 0.98, algorithms 3 and 4 had PPVs of 0.30 and 0.34, respectively. Based on survey of three providers, we estimated providers spent 8.6 min per chart review; using this this parameter, providers would spend 61.4, 57.3, 28.7, and 25.3 min on secondary chart review for each case of ADHF if initial screening was done with algorithms 1, 2, 3, and 4, respectively. CONCLUSIONS: Traditional approaches with structured data can accurately identify nearly all patients with ADHF but are limited by a large number of false positives. Machine learning algorithms with unstructured notes and radiology reports can retain this sensitivity while significantly reducing false positives, thereby improving provider efficiency for delivery of quality improvement interventions

BACKGROUND: Electronic health record (EHR)-based clinical decision support tools (CDS) incorporate individualized data to produce patient-specific recommendations at the point-of-care. However, these tools are often limited in their effectiveness, which may be due to poor consideration of usability. The purpose of this study was to evaluate the utilization of a CDS intervention to increase prescription of Angiotensin Converting Enzyme inhibitor (ACEi) or Angiotensin Receptor Blocker (ARB) for patients with heart failure. METHODS: We performed a retrospective study of hospitalized patients with heart failure from the time of CDS implementation, 7/10/13, through 11/30/15. The CDS that we investigated offers providers an opportunity to prescribe an ACEi or ARB or report a contraindication to therapy for patients with documented heart failure. All patients with an EF < 40% who were not on an ACEi or ARB at time of discharge were included in the study. We identified the number of patients for whom the CDS triggered; of those, we categorized provider response as: dismissed, ordered an ACEi/ARB, or contraindication reported. We then performed manual chart review to identify the CDS reported contraindication and structured chart abstraction with standard guidelines to identify gold standard contraindications. We compared each CDS contraindication to gold standard contraindications to determine their accuracy. RESULTS: Out of the 618 subjects who had an EF < 40% but no ACEi or ARB at the time or discharge, 435/618 (70%) had a triggered CDS. Of these 435 subjects for who a CDS was triggered, 180 (41%) were dismissed, 225 (52%) had a contraindication response and 30 (7%) had a prescription for an ACEi/ARB therapy. Overall the accuracy of the documented CDS was 42% (Table 1). CONCLUSIONS: The CDS that we reviewed was poorly utilized and contraindications documented in the tool poorly correlated with patient clinical status reflected elsewhere in the EHR. These findings identify this CDS as a possible impedance to user workflow. One way to improve CDS tools at the point of care is through thorough usability testing and consideration of physician workflow prior to implementation. (Table Presented)

BACKGROUND: Many hospitalized patients have at least 1 chronic disease that is not optimally controlled. The purpose of this study was to explore inpatient provider attitudes about chronic disease management and, in particular, barriers and facilitators of chronic disease management in the hospital. METHODS: We conducted a qualitative study of semi-structured interviews of 31 inpatient providers from an academic medical center. We interviewed attending physicians, resident physicians, physician assistants, and nurse practitioners from various specialties about attitudes, experiences with, and barriers and facilitators towards chronic disease management in the hospital. Qualitative data were analyzed using constant comparative analysis. RESULTS: Providers perceived that hospitalizations offer an opportunity to improve chronic disease management, as patients are evaluated by a new care team and observed in a controlled environment. Providers perceived clinical benefits to in-hospital chronic care, including improvements in readmission and length of stay, but expressed concerns for risks related to adverse events and distraction from the acute problem. Barriers included provider lack of comfort with managing chronic diseases, poor communication between inpatient and outpatient providers, and hospital-system focus on patient discharge. A strong relationship with the outpatient provider and involvement of specialists were facilitators of inpatient chronic disease management. CONCLUSIONS: Providers perceived benefits to in-hospital chronic disease management for both processes of care and clinical outcomes. Efforts to increase inpatient chronic disease management will need to overcome barriers in multiple domains. Journal of Hospital Medicine 2017;12:162-167.

Importance: Accurate, real-time case identification is needed to target interventions to improve quality and outcomes for hospitalized patients with heart failure. Problem lists may be useful for case identification but are often inaccurate or incomplete. Machine-learning approaches may improve accuracy of identification but can be limited by complexity of implementation. Objective: To develop algorithms that use readily available clinical data to identify patients with heart failure while in the hospital. Design, Setting, and Participants: We performed a retrospective study of hospitalizations at an academic medical center. Hospitalizations for patients 18 years or older who were admitted after January 1, 2013, and discharged before February 28, 2015, were included. From a random 75% sample of hospitalizations, we developed 5 algorithms for heart failure identification using electronic health record data: (1) heart failure on problem list; (2) presence of at least 1 of 3 characteristics: heart failure on problem list, inpatient loop diuretic, or brain natriuretic peptide level of 500 pg/mL or higher; (3) logistic regression of 30 clinically relevant structured data elements; (4) machine-learning approach using unstructured notes; and (5) machine-learning approach using structured and unstructured data. Main Outcomes and Measures: Heart failure diagnosis based on discharge diagnosis and physician review of sampled medical records. Results: A total of 47119 hospitalizations were included in this study (mean [SD] age, 60.9 [18.15] years; 23 952 female [50.8%], 5258 black/African American [11.2%], and 3667 Hispanic/Latino [7.8%] patients). Of these hospitalizations, 6549 (13.9%) had a discharge diagnosis of heart failure. Inclusion of heart failure on the problem list (algorithm 1) had a sensitivity of 0.40 and a positive predictive value (PPV) of 0.96 for heart failure identification. Algorithm 2 improved sensitivity to 0.77 at the expense of a PPV of 0.64. Algorithms 3, 4, and 5 had areas under the receiver operating characteristic curves of 0.953, 0.969, and 0.974, respectively. With a PPV of 0.9, these algorithms had associated sensitivities of 0.68, 0.77, and 0.83, respectively. Conclusions and Relevance: The problem list is insufficient for real-time identification of hospitalized patients with heart failure. The high predictive accuracy of machine learning using free text demonstrates that support of such analytics in future electronic health record systems can improve cohort identification.

OBJECTIVES: This study sought to determine whether racial and ethnic differences exist among patients with similar access to care. We examined outcomes after heart failure hospitalization within a large municipal health system. BACKGROUND: Racial and ethnic disparities in heart failure outcomes are present in administrative data, and one explanation is differential access to care. METHODS: We performed a retrospective cohort study of 8,532 hospitalizations of adults with heart failure at 11 hospitals in New York City from 2007 to 2010. Primary exposure was ethnicity and race, and outcomes were 30- and 90-day readmission and 30-day and 1-year mortality rates. Generalized estimating equations were used to test for associations between ethnicity and race and outcomes with covariate adjustment. RESULTS: Of the number of hospitalizations included, 4,305 (51%) were for blacks, 2,449 (29%) were for Hispanics, 1,494 (18%) were for whites, and 284 (3%) were for Asians. Compared to whites, blacks and Asians had lower 1-year mortality, with adjusted odds ratios (aORs) of 0.75 (95% confidence interval [CI]: 0.59 to 0.94) and 0.57 (95% CI: 0.38 to 0.85), respectively, and rates for Hispanics were not significantly different (aOR: 0.81; 95% CI: 0.64 to 1.03). Hispanics had higher odds of readmission than whites (aOR: 1.27; 95% CI: 1.03 to 1.57) at 30 (aOR: 1.40; 95% CI: 1.15 to 1.70) and 90 days. Blacks had higher odds of readmission than whites at 90 days (aOR:1.21; 95% CI: 1.01 to 1.47). CONCLUSIONS: Racial and ethnic differences in outcomes after heart failure hospitalization were present within a large municipal health system. Access to a municipal health system may not be sufficient to eliminate disparities in heart failure outcomes.

Importance: Reduced rates of cardiac catheterization, percutaneous coronary intervention (PCI), and coronary artery bypass graft (CABG) are an unintended consequence of public reporting of cardiogenic shock outcomes in New York. Objectives: To evaluate whether the referral rates for cardiac catheterization, PCI, or CABG have improved in New York since cardiogenic shock was excluded from public reporting in 2008 and compare them with corresponding rates in Michigan, New Jersey, and California. Design, Setting, and Participants: Patients with cardiogenic shock complicating acute myocardial infarction from 2002 to 2011 were identified using the National Inpatient Sample. Propensity score matching was used to assemble a cohort of patients with cardiogenic shock with similar baseline characteristics in New York and Michigan. Main Outcomes and Measures: Percutaneous coronary intervention (primary outcome), invasive management (cardiac catheterization, PCI, or CABG), revascularization (PCI or CABG), and CABG were evaluated with reference to 3 calendar year periods: 2002-2005 (time 1: cardiogenic shock included in publicly reported outcomes), 2006-2007 (time 2: cardiogenic shock excluded on a trial basis), and 2008 and thereafter (time 3: cardiogenic shock excluded permanently) in New York and compared with Michigan. Results: Among 2126 propensity score-matched patients representing 10795 (weighted) patients with myocardial infarction complicated by cardiogenic shock in New York and Michigan, 905 (42.6%) were women and mean (SE) age was 69.5 (0.3) years. A significantly higher proportion of the patients underwent PCI (time 1 vs 2 vs 3: 31.1% vs 39.8% vs 40.7% [OR, 1.50; 95% CI, 1.12-2.01; P = .005 for time 3 vs 1]), invasive management (time 1 vs 2 vs 3: 59.7% vs 70.9% vs 73.8% [OR, 1.84; 95% CI, 1.37-2.47; P < .001 for time 3 vs 1]), or revascularization (43.1% vs 55.9% vs 56.3% [OR, 1.66; 95% CI, 1.26-2.20; P < .001 for time 3 vs 1]) after the exclusion of cardiogenic shock from public reporting in New York. However, during the same periods, a greater proportion of patients underwent PCI (time 1 vs 2 vs 3: 41.2% vs 52.6% vs 57.8% [OR, 1.93; 95% CI, 1.45-2.56; P < .001 for time 3 vs 1]), invasive management (time 1 vs 2 vs 3: 64.4% vs 80.5% vs 78.6% [OR, 2.01; 95% CI, 1.47-2.74; P < .001 for time 3 vs 1]), or revascularization (51.2% vs 65.8% vs 68.0% [OR, 2.00; 95% CI, 1.50-2.66; P < .001 for times 3 vs 1]) in Michigan. Results were largely similar in several sensitivity analyses comparing New York with New Jersey or California. Conclusions and Relevance: Although the rates of PCI, invasive management, and revascularization have increased substantially after the exclusion of cardiogenic shock from public reporting in New York, these rates remain consistently lower than those observed in other states without public reporting.

STUDY OBJECTIVE: Observation unit admissions have been increasing, a trend that will likely continue because of recent changes in reimbursement policies. The purpose of this study is to determine the effect of the availability of observation units on hospitalizations and discharges to home for emergency department (ED) patients. METHODS: We studied ED visits with a final diagnosis of chest pain in the National Hospital Ambulatory Medical Care Survey from 2007 to 2010. ED visits that resulted in an observation unit admission were propensity-score matched to visits at hospitals without an observation unit. We used logistic regression to develop a prediction model for hospitalization versus discharge home for matched patients treated at nonobservation hospitals. The model was applied to matched observation unit patients to determine the likely alternative disposition had the observation unit not been available. RESULTS: There were 1,325 eligible visits that represented 5,079,154 visits in the United States. Two hundred twenty-seven visits resulted in an observation unit admission. The predictive model for hospitalization had a c statistic of 0.91; variables significantly associated with subsequent hospitalization included age, history of coronary atherosclerosis, systolic blood pressure less than 115 beats/min, and administration of antianginal medications. When the model was applied to matched observation unit patients, 49.9% of them were categorized as discharge home likely. CONCLUSION: In this study, we estimated that half of ED visits for chest pain that resulted in an observation unit admission were made by patients who may have been discharged home had the observation unit not been available. Increased availability of observation units may result in both decreased hospitalizations and decreased discharges to home.