Faculty Bibliography

BACKGROUND:Mineralocorticoid receptor antagonists (MRA) are an underutilized therapy for heart failure with a reduced ejection fraction (HFrEF), but the current impact of hospitalization on MRA use is not well characterized. The objective of this study was to describe contemporary MRA prescription for heart failure patients before and after the full scope of hospitalizations and the association between MRA discharge prescription and post-hospitalization outcomes. METHODS:We conducted a retrospective cohort study at an academic hospital system in 2013-2016. Among 1500 included hospitalizations of 1009 unique patients with HFrEF and without MRA contraindication, the mean age was 71.9 ± 13.6 years and 443 (29.5%) were female. We compared MRA prescription before and after hospitalizations with McNemar's test and between patients with principal and secondary diagnoses of HFrEF with the chi-square test, and association of MRA discharge prescription with 30-day and 180-day mortality and readmissions using generalized estimating equations. RESULTS:MRA prescriptions increased from 303 (20.2%) to 375 (25.0%) at discharge (+4.8%, p < 0.0001). More patients with principal diagnosis of HFrEF compared to those hospitalized for other reasons received MRA (34.9% versus 21.3%, p < 0.0001) and had them initiated (21.8% versus 9.7%, p < 0.0001). MRA prescription at discharge was not associated with mortality or readmission at 30 and 180 days, and there was no interaction with principal/secondary diagnosis. CONCLUSIONS:Among hospitalized HFrEF patients, 75% did not receive MRA before or after hospitalization, and nearly 90% of eligible patients did not have MRA initiated. As we found no signal for short-term harm after discharge, hospitalization may represent an opportunity to initiate guideline-directed heart failure therapy.

Background: Clinical decision support (CDS) systems can be valuable resources in chronic disease management, but provider utilization of these tools and their integration into workflow remains suboptimal. We used a user-centered design approach to build a CDS recommending evidence-based therapy for heart failure in an inpatient setting. We implemented two versions of the CDS: an interruptive (pop-up) alert and a non-interruptive alert displayed in a provider checklist activity. In a prior study, we found that the interruptive alert was more effective than the non-interruptive alert but suffered from a high dismissal rate. The purpose of this study was to understand provider's perceptions of factors impacting CDS utilization following its implementation.
Method(s): We performed a qualitative study following implementation of two versions of a CDS at an academic medical center. We recruited providers who had either version of the CDS triggered in the prior 24 hours and obtained feedback through semi-structured interviews. Interviews were recorded, transcribed, and double-coded. We performed a constant comparative analysis of the transcripts and developed a coding scheme informed by the Five Rights of CDS combined with Proctor's outcomes for implementation research framework. We recruited participants until thematic saturation was achieved.
Result(s): Fourteen providers participated in interviews. In general, providers found the CDS triggered for appropriate patients, provided the right information to determine appropriateness of recommendations, and had good usability. At least four providers believed they were not the right person to receive the alert, reporting that they primarily participated in cross-coverage, worked in a setting where these medications were typically contraindicated, or were already fully compliant with evidence-based medications. Providers complained that the interruptive alert led to workflow disruption and that frequently they would "just need to click through all of this." Nonetheless, many providers still preferred the interruptive version of the alert because they were either: 1) unaware of the non-interruptive alert in the checklist; 2) "don't use the provider checklist" where the non-interruptive alert resided; or 3) were unaware of the provider checklist.
Conclusion(s): We found that CDS was generally found to be acceptable, although the interruptive version of the alert was limited by disruptions in workflow. The interruptive alert was ultimately more successful as providers were unaware of the existence of the non-interruptive alert. Furthermore, they infrequently used the provider checklist, a native EHR feature for non-interruptive alerts and messages. Our findings suggest that incorporating user-centered design features can lead to success of an interruptive alert. Furthermore, future efforts to implement non-interruptive alerts must incorporate approaches to increase awareness of the alert and encourage changes in workflow to monitor alert lists

Background: Older adults with diabetes continue to be overtreated despite current guidelines recommending less aggressive target A1c levels based on life expectancy. The suboptimal management of this vulnerable population could be due to physicians having conflicting beliefs regarding this guideline or simply lacking awareness, and changing these behaviors is challenging. Clinical decision support (CDS) within the electronic health record (EHR) has the potential to address this issue, but effectiveness is undermined by alert fatigue and poor workflow integration. Incorporating behavioral economics into CDS tools is an innovative approach to improve adherence to these guidelines while reducing physician burden, and offers the promise of improving care in this population.
Method(s): We applied a systematic, user-centered approach to incorporate behavioral economic " nudges" into a CDS module and performed user testing in six pilot primary care practices in a large academic medical center. To build the nudges, we conducted: (1) semi-structured interviews with key informants (n=8); (2) a two-hour design thinking workshop to derive and refine initial module ideas; and (3) semi-structured group interviews at each site with clinic leaders and clinicians to elicit feedback on the module components. Clinicians were observed using the module in practice; detailed field notes were collected and summarized by module idea and usability theme for rapid iteration and refinement. Frequency of firing and user action taken were assessed in the first month of implementation via EHR reporting to confirm that module components and reporting were working as expected, and to assess utilization.
Result(s): Insights from key stakeholder and clinician group interviews identified the refill protocol, inbasket lab result, and medication preference list as candidate EHR CDS targets for the module. A new EHR navigator section notification and peer comparison message, derived from the design workshop, were also prototyped and produced. User feedback from site visits confirmed compatibility with clinical workflows, and contributed to refinement of design and content. The initial prototypes were first piloted at two sites, refined, and then activated at an additional four additional sites. Preliminary Results for the six clinics indicate that over approximately 31 weeks: 1) the navigator alert fired 1047 times for 53 unique clinicians, and 2) the refill protocol alert fired 421 times for 53 unique clinicians. Reports for the other " nudges" are in development.
Conclusion(s): Integrating behavioral economic nudges into the EHR is a promising approach to enhancing guideline awareness and adherence for older adults with diabetes. This novel pilot will demonstrate the initial feasibility and preliminary efficacy of this strategy and determine if a full-scale effectiveness trial is warranted

BACKGROUND:Clinical decision support (CDS) has been shown to improve compliance with evidence-based care, but its impact is often diminished because of issues such as poor usability, insufficient integration into workflow, and alert fatigue. Noninterruptive CDS may be less subject to alert fatigue, but there has been little assessment of its usability. OBJECTIVE:This study aimed to study the usability of interruptive and noninterruptive versions of a CDS. METHODS:We conducted a usability study of a CDS tool that recommended prescribing an angiotensin-converting enzyme inhibitor for inpatients with heart failure. We developed 2 versions of the CDS: an interruptive alert triggered at order entry and a noninterruptive alert listed in the sidebar of the electronic health record screen. Inpatient providers were recruited and randomly assigned to use the interruptive alert followed by the noninterruptive alert or vice versa in a laboratory setting. We asked providers to "think aloud" while using the CDS and then conducted a brief semistructured interview about usability. We used a constant comparative analysis informed by the CDS Five Rights framework to analyze usability testing. RESULTS:A total of 12 providers participated in usability testing. Providers noted that the interruptive alert was readily noticed but generally impeded workflow. The noninterruptive alert was felt to be less annoying but had lower visibility, which might reduce engagement. Provider role seemed to influence preferences; for instance, some providers who had more global responsibility for patients seemed to prefer the noninterruptive alert, whereas more task-oriented providers generally preferred the interruptive alert. CONCLUSIONS:Providers expressed trade-offs between impeding workflow and improving visibility with interruptive and noninterruptive versions of a CDS. In addition, 2 potential approaches to effective CDS may include targeting alerts by provider role or supplementing a noninterruptive alert with an occasional, well-timed interruptive alert.

BACKGROUND:The Medicare Hospital Readmissions Reduction Program has led to fewer readmissions following hospitalizations with a principal diagnosis of heart failure (HF). Patients with HF are frequently hospitalized for other causes. OBJECTIVES/OBJECTIVE:This study sought to compare trends in Medicare risk-adjusted, 30-day readmissions following principal HF hospitalizations and other hospitalizations with HF. METHODS:This was a retrospective study of 12,973,853 Medicare hospitalizations with a principal or secondary diagnosis of HF between January 2008 and June 2015. Hospitalizations were categorized as follows: principal HF hospitalizations; principal acute myocardial infarction or pneumonia hospitalizations with secondary HF; and other hospitalizations with secondary HF. The study examined trends in risk-adjusted, 30-day, all-cause readmission rates for each cohort and trends in differences in readmission rates among cohorts by using linear spline regression models. RESULTS:Before passage of the Affordable Care Act in March 2010, risk-adjusted, 30-day readmission rates were stable for all 3 cohorts, with mean monthly rates of 26.1%, 24.9%, and 24.4%, respectively. Risk-adjusted readmission rates started declining after passage of the Affordable Care Act by 1.09% (95% confidence interval [CI]: 0.51% to 1.68%), 1.24% (95% CI: 0.92% to 1.57%), and 1.05% (95% CI: 0.52% to 1.58%) per year, respectively, until implementation of the Hospital Readmissions Reduction Program in October 2012 and then stabilized for all 3 cohorts. CONCLUSIONS:Patients with HF are often hospitalized for other causes, and these hospitalizations have high readmission rates. Policy changes led to decreases in readmission rates for both principal and secondary HF hospitalizations. Readmission rates in both groups remain high, suggesting that initiatives targeting all hospitalized patients with HF continue to be warranted.

Background/UNASSIGNED:Current guidelines recommend less aggressive target hemoglobin A1c (HbA1c) levels based on older age and lower life expectancy for older adults with diabetes. The effectiveness of electronic health record (EHR) clinical decision support (CDS) in promoting guideline adherence is undermined by alert fatigue and poor workflow integration. Integrating behavioral economics (BE) and CDS tools is a novel approach to improving adherence to guidelines while minimizing clinician burden. Methods/UNASSIGNED: = 8), (2) a 2-h, design-thinking workshop to derive and refine initial module ideas, and (3) semi-structured group interviews at each site with clinic leaders and clinicians to elicit feedback on three proposed nudge module components (navigator section, inbasket refill protocol, medication preference list). Detailed field notes will be summarized by module idea and usability theme for rapid iteration. Frequency of firing and user action taken will be assessed in the first month of implementation via EHR reporting to confirm that module components and related reporting are working as expected as well as assess utilization. To assess the utilization and feasibility of the new tools and generate estimates of clinician compliance with the Choosing Wisely guideline for diabetes management in older adults, a 6-month, single-arm pilot study of the BE-EHR module will be conducted in six outpatient primary care clinics. Discussion/UNASSIGNED:We hypothesize that a low burden, user-centered approach to design will yield a BE-driven, CDS module with relatively high utilization by clinicians. The resulting module will establish a platform for exploring the ability of BE concepts embedded within the EHR to affect guideline adherence for other use cases.

BACKGROUND:Hospitalization and readmission rates have decreased in recent years, with the possible consequence that hospitals are increasingly filled with high-risk patients. OBJECTIVE:We studied whether readmission reduction has affected the risk profile of hospitalized patients and whether readmission reduction was similarly realized among hospitalizations with low, medium, and high risk of readmissions. DESIGN/METHODS:Retrospective study of hospitalizations between January 2009 and June 2015. PATIENTS/METHODS:Hospitalized fee-for-service Medicare beneficiaries, categorized into 1 of 5 specialty cohorts used for the publicly reported hospital-wide readmission measure. MEASUREMENTS/METHODS:Each hospitalization was assigned a predicted risk of 30-day, unplanned readmission using a risk-adjusted model similar to publicly reported measures. Trends in monthly mean predicted risk for each cohort and trends in monthly observed to expected readmission for hospitalizations in the lowest 20%, middle 60%, and highest 20% of risk of readmission were assessed using time series models. RESULTS:Of 47,288,961 hospitalizations, 16.2% (n = 7,642,161) were followed by an unplanned readmission within 30 days. We found that predicted risk of readmission increased by 0.24% (P = .03) and 0.13% (P = .004) per year for hospitalizations in the surgery/ gynecology and neurology cohorts, respectively. We found no significant increase in predicted risk for hospitalizations in the medicine (0.12%, P = .12), cardiovascular (0.32%, P = .07), or cardiorespiratory (0.03%, P = .55) cohorts. In each cohort, observed to expected readmission rates steadily declined, and at similar rates for patients at low, medium, and high risk of readmission. CONCLUSIONS:Hospitals have been effective at reducing readmissions across a range of patient risk strata and clinical conditions. The risk of readmission for hospitalized patients has increased for 2 of 5 clinical cohorts.

BACKGROUND: Interventions to reduce readmissions following acute heart failure hospitalization require early identification of patients. The purpose of this study was to develop and test accuracies of various approaches to identify patients with acute decompensated heart failure (ADHF) using data derived from the electronic health record. METHODS AND RESULTS: We included 37,229 hospitalizations of adult patients at a single hospital in 2013-2015. We developed four algorithms to identify hospitalization with a principal discharge diagnosis of ADHF: 1) presence of one of three clinical characteristics; 2) logistic regression of 31 structured data elements; 3) machine learning with unstructured data; 4) machine learning with both structured and unstructured data. In data validation, 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, while both machine learning algorithms had AUCs of 0.99. Based on a brief survey of three providers who perform chart review for ADHF, we estimated providers spent 8.6 minutes per chart review; using this this parameter, we estimated providers would spend 61.4, 57.3, 28.7, and 25.3 minutes on secondary chart review for each case of ADHF if initial screening was done with algorithms 1, 2, 3, and 4, respectively. CONCLUSION: Machine learning algorithms with unstructured notes had best performance for identification of ADHF and can improve provider efficiency for delivery of quality improvement interventions.