UCSF: Predicting Patient Outcomes Using OR Data

Healthcare Non-Profit Clinical Improve Health Outcomes
UCSF-BASIC uses DataRobot and operating room data to predict the outcomes of patients with traumatic spinal cord injuries.
Build with Free Trial


Business Problem

Acute spinal cord injuries (SCI) are rare but fundamentally life-altering events, most often caused by sudden, traumatic accidents such as vehicle crashes or falls. Less than 1% of SCI patients experience complete neurological recovery prior to hospital discharge. Because this type of injury is so rare, there have been limited opportunities for empirical, data-driven research to inform treatment.

UCSF Brain and Spinal Injury Center (UCSF-BASIC) is partnered with the Zuckerberg San Francisco General Hospital to form the Transforming Research and Clinical Knowledge in Spinal Cord Injury (TRACK-SCI) team. TRACK-SCI has accumulated the largest dataset so far on acute SCI.

In its partnership with DataRobot through the AI for Good program, UCSF-BASIC’S first area of focus is blood pressure management in the OR. Existing guidelines dictate that patient neurological outcomes are improved by maintaining the patient’s blood pressure over 85 mmHg for the first seven days after injury, including during any surgeries in that time period.

Intelligent Solution

TRACK-SCI collected data on the vital signs of patients undergoing spinal stabilization surgery after injury. Among that dataset of 74 patients, recovery progress was measured by comparing neurological scores at admission and discharge. Neurological function is measured by using the American Spinal Injury Association (ASIA) impairment scale. This scale (grades A-F) indicates how much sensation a person feels after light touch and a pin prick at multiple points on the body and tests key motions on both sides of the body. A grade of A is the most severe. The data included time-series readings of systolic and diastolic blood pressure and heart rate taken every five minutes, in addition to data on the nature and location of the injury.

Using characteristics of the patient’s injury and derived variables from the patient’s blood pressure and heart-rate measurements in the OR, UCSF-BASIC built a predictive model using the ASIA scale to determine if a patient would improve. Those derived variables included the skew and kurtosis of the blood pressure and heart-rate variables, and a suite of variables that described how long a patient was beyond upper and lower thresholds of mean arterial pressure. The goal is to validate whether the existing lower bound of 85 mmHg is clinically relevant and to probe if an upper bound is of significance, too.

The resulting models are undergoing feature reduction and stability analysis, in addition to testing on a prospective dataset, to confirm and validate their findings before publication. However, the data suggests that time spent in high blood-pressure regimes has significant predictive impact on whether a patient improves. Clinicians and neurosurgeons on the UCSF-BASIC team are studying these results and their implications for improving patient care.

banner purple waves bg

Experience the DataRobot AI Platform

Less Friction, More AI. Get Started Today With a Free 30-Day Trial.

Sign Up for Free
Explore More Healthcare Use Cases
Healthcare companies are using machine learning and AI to increase top and bottom line through gaining competitive advantages, reducing expenses, and improving efficiencies. They are optimizing all areas of their business from readmission risk and occupancy rates to marketing, in order to make data-driven decisions that lead to increased profitability.