Six million older adults every year are hospitalized and then transition to post-acute care services.
Despite years of substantial policy intervention, these transitions remain poorly coordinated and disruptive.
Nearly one in four patients with common conditions like heart failure and pneumonia continue to experience
post-discharge destabilization severe enough that they end up back in the hospital. Robust discharge
planning is critical to transitional care quality - clinicians need to prepare and communicate high-quality
information (e.g. summary of pending results, changes in medication and therapy needs) that supports
follow-up care. Unfortunately, the quality of discharge documentation produced by discharge planning
actions is known to be highly variable and error-prone and puts patients at increased risk of gaps and
errors in care. Health systems need actionable data to assess where the discharge planning process is
breaking down if they are to focus improvements that effect and sustain stronger transitional care practices.
Our long-term objective is to provide health systems with the tools to monitor and strengthen
specific discharge workflow behaviors that optimize post-acute transitions. In this proposal, we analyze
EHR metadata to help health systems identify inconsistencies and hone best practices in preparing
patients for discharge. EHR metadata are the digital “fingerprints” generated through clinicians’ interactions
with the EHR, such as logging in and out, clicks, and time spent viewing or modifying patient data. These
data have been described as a potential goldmine for research. With metadata, we can reconstruct and
characterize important process variation in discharge planning activities, ultimately evaluating whether
specific behaviors are associated with avoided readmissions and other transitional care outcomes.
Using a sample of patients with high-volume, high-readmission risk conditions (e.g. heart failure
and pneumonia), we first use mixed methods process mining and pattern analysis techniques with EHR
metadata to define, measure, and assess the extent of variation in key discharge planning tasks and
workflows. Examples might include how often discharging providers reference social work notes during
discharge orders, or the timing of when medications are reconciled relative to the time of discharge. We
then descriptively analyze key patient and contextual factors that drive this variation. Next, we use
generalized linear models to assess which discharge planning tasks and workflows are associated with
process (e.g. timely discharge and follow-up care) and outcome-based (e.g. readmissions) measures of
transitional care quality. Our work tests novel methods of implementation evaluation that support health
system learning and improvement, and aligns directly with AHRQ health IT priorities to develop data-driven
solutions to help providers organizations refine and advance impactful, scalable changes in care delivery.