Learning effect on door-to-needle time for telemedicine-administered thrombolytics in acute ischemic stroke: A frequentist analysis from a rural stroke network in the Midwest

Abstract

Background and Purpose

Telemedicine technology in acute stroke care is utilized in underserved areas to allow remote neurology specialists to rapidly evaluate acute stroke patients and make clinical decisions regarding thrombolytic therapy. This work aims to describe the learning curve of door-to-needle time (DTN) in telemedicine-administered thrombolytics in a rural stroke network of the Midwest.

Methods

This is a retrospective analysis of thrombolytics administered via telemedicine for acute ischemic stroke at three rural hospitals within the Southern Illinois Healthcare network between July 2017 and August 2024. Demographics, medical history, clinical presentation, time of evaluation, severity of stroke, neuroimaging, and laboratory testing were reviewed. Major technology upgrades were noted. Providers entering the telestroke system did not receive formalized training; however, DTN performance was included in physicians’ quality metrics, which contributed to gradual learning and improvement. Outcome measure was DTN. A linear regression model was fitted using the ordinary least squares (OLS) method to estimate the effect of sequential telemedicine encounters and preselected co-variables on DTN. A risk-adjusted cumulative sum chart (RA-CUSUM) was plotted to demonstrate the cumulative sum of probability of achieving DTN within 60 min by sequence of encounters. P value was set at 0.05.

Results

This study included 170 telemedicine-administered thrombolytics. Once adjusted for age, systolic blood pressure, NIHSS, and time of evaluation, the sequence of telemedicine encounters was independently associated with DTN (B: −0.086, 95 %CI: −0.169, −0.03, p = 0.04). RA-CUSUM demonstrated a learning curve that was irregularly associated, but not invariably, with new providers beginning to use telemedicine and major technology upgrades. The learning curve followed cycles of steep learning, followed by semi-plateau phases, each lasting approximately 40 encounters. Probabilistically, every 10 encounters lowered the DTN by 42 s. Statistically, 98 encounters were needed to predict DTN to be within 60 min.

Conclusions

In our stroke network, we identified a learning effect on the DTN in telemedicine-administered thrombolytics. The learning curve occurred at three levels: 1) system level of the network, 2) provider experience, and 3) improvement in the technology. Our experience suggests that provider learning and system enhancements may drive improvements in DTN, but additional studies are needed to assess generalizability to other telestroke networks.