Improving Prehospital Stroke Diagnosis Using Natural Language Processing of Paramedic Reports.

Background and Purpose: Accurate prehospital diagnosis of stroke by emergency medical services (EMS) can increase treatments rates, mitigate disability, and reduce stroke deaths. We aimed to develop a model that utilizes natural language processing of EMS reports and machine learning to improve prehospital stroke identification. Methods: We conducted a retrospective study of patients transported by the Chicago EMS to 17 regional primary and comprehensive stroke centers. Patients who were suspected of stroke by the EMS or had hospital-diagnosed stroke were included in our cohort. Text within EMS reports were converted to unigram features, which were given as input to a support-vector machine classifier that was trained on 70% of the cohort and tested on the remaining 30%. Outcomes included final diagnosis of stroke versus nonstroke, large vessel occlusion, severe stroke (National Institutes of Health Stroke Scale score >5), and comprehensive stroke center-eligible stroke (large vessel occlusion or hemorrhagic stroke). Results: Of 965 patients, 580 (60%) had confirmed acute stroke. In a test set of 289 patients, the text-based model predicted stroke nominally better than models based on the Cincinnati Prehospital Stroke Scale (c-statistic: 0.73 versus 0.67, P=0.165) and was superior to the 3-Item Stroke Scale (c-statistic: 0.73 versus 0.53, P<0.001) scores. Improvements in discrimination were also observed for the other outcomes. Conclusions: We derived a model that utilizes clinical text from paramedic reports to identify stroke. Our results require validation but have the potential of improving prehospital routing protocols.

Use of a Novel Patient-Flow Model to Optimize Hospital Bed Capacity for Medical Patients.

BACKGROUND: There is no known method for determining the minimum number of beds in hospital inpatient units (IPs) to achieve patient waiting-time targets. This study aims to determine the relationship between patient waiting time-related performance measures and bed utilization, so as to optimize IP capacity decisions. METHODS: The researchers simulated a novel queueing model specifically developed for the IPs. The model takes into account salient features of patient-flow dynamics and was validated against hospital census data. The team used the model to evaluate inpatient capacity decisions against multiple waiting time outcomes: (1) daily average, peak-hour average, and daily maximum waiting times; and (2) proportion of patients waiting strictly more than 0, 1, and 2 hours. The results were published in a simple Microsoft Excel toolbox to allow administrators to conduct sensitivity analysis. RESULTS: To achieve the hospital's goal of rooming patients within 30 to 60 minutes of IP bed requests, the model predicted that the optimal daily average occupancy levels should be 89%-92% (182-188 beds) in the Medicine cohort, 74%-79% (41-43 beds) in the Cardiology cohort, and 72%-78% (23-25 beds) in the Observation cohort. Larger IP cohorts can achieve the same queueing-related performance measure as smaller ones, while tolerating a higher occupancy level. Moreover, patient waiting time increases rapidly as the occupancy level approaches 100%. CONCLUSION: No universal optimal IP occupancy level exists. Capacity decisions should therefore be made on a cohort-by-cohort basis, incorporating the comprehensive patient-flow characteristics of each cohort. To this end, patient-flow queueing models tailored to the IPs are needed.