(888) Flying Blind? Delays in Documentation Limit Clinical Decision Support Tools for Complex Problems

Introduction: Sepsis is a dynamic, complex disease with tension between the “true” essence of sepsis and the tools we use to identify it. Early recognition of sepsis through electronic health record (EHR)-based clinical decision support (CDS) tools has led to improved treatment of sepsis. Most CDS tools depend on the availability of data elements to properly function. Delays in data entry hinder a CDS tool’s function and introduces lost opportunities to inform decision making. We hypothesized that delays in data documentation create blind windows within a hospital’s sepsis CDS tool.

Methods: We performed a retrospective cohort analysis of patients admitted to a pediatric intensive care unit (PICU) in 2021 who triggered the sepsis CDS tool. Documented-time (data entry time) was described as either early, on-time, or late in relation to the real-time (estimated data occurrence time). Blind windows were identified as the accumulated time between the value’s real-time and the documented-time. Quantitative analysis included comparison of periods of time-misalignment between documented-times and real-times. We performed summary statistics to quantify the relative frequencies of these elements.

Results: The cohort consisted of 1.5 thousand (K) patients and 1.7K encounters. Flowsheet value data was available for 520 patients with 13.7 million (M) flowsheet values related to the CDS tool. There were 1.2M (9%) documented values and 12.5M (91%) tool-automated values. Of the 1.2M values, 81.1K (6.8%) were excluded due to outlier status or incompleteness, leaving 1.1M values analyzed. The vast majority of documented values were late (90.1%; 993K), with 7.4% (81.3K) of values being early and 2.4% (26.8K) being on-time. Within the late documentation group, the tool spent most of the time (62.4%; 57.2M out of 91.6M minutes) within a blind window.

Conclusions: Delays in scoring documentation contribute to creation of blind windows with risk of wasted effort and potential delays in problem identification and resolution. Blind windows are a potential limitation in CDS models deployed to predict complex problems. Attempts to minimize blind windows may improve timeliness of sepsis identification.