BACKGROUND
The ultimate goal of our stroke risk software is to both risk stratify patients and personalize the management of stroke risk factors in order to prevent stroke - with the exception of blood pressure management which is so complex that it needs a dedicated system.
Current stroke risk algorithms primarily look at non-modifiable risk factors which offer little guidance to neurologists.
The goal is not to be the best predictive algorithm. Predicting a stroke will occur in March of 2027 does not help a clinician do anything meaningful to prevent that stroke. The goal is to identify which modifiable stroke risk factors are not being adequately managed and monitor the progress as management of those risk factors improves. This is an important distinction. This algorithm will inherently be predictive but the focus is on personalized preventative care.
This is one of the functions of a neurologist that we plan to not only automate but improve beyond what a neurologist can do.
We’ve seen that one of the key elements driving outcomes is identifying one or two risk factors that are not being adequately managed in our patients and this is the software that supports that aspect of our value proposition.
CURRENT VERSION:
STROKE RISK SOFTWARE V1
A rule-based expert system designed by neurologists to map an individual’s modifiable risk factors to determine their overall level of risk and identify gaps in management of specific risk factors.
Stroke Risk Score
Individual risk factors are weighted based on impact on future stroke and degree of possible treatment optimization
Stroke risk score represents the aggregate of the modifiable risk factors that need to be addressed for a particular patient
Precision Care Algorithm
Evaluates ~40M combinations of potential interventions to find the greatest potential reduction in the stroke risk score
Reviewed by neurologists during active patient care. Exists in prototype form and was never fully deployed as it became apparent changes would be necessary to meet the needs of the business.
INSIGHTS:
Didn’t capture all modifiable risk factors we were seeing in our patients
Needs to deal with missing data in a meaningful way
Most risk factor adjustments are straightforward (and even built into the stroke risk score) and can be performed in parallel. The prioritization and sequencing component of the Precision Care algorithm had limited value for our clinicians.
Blood pressure management is very complex and needs dedicated system for personalized management
Value is in the tracking and notification capabilities and less so in intervention guidance
NEXT VERSION:
STROKE RISK SOFTWARE V2 (IN DEVELOPMENT)
Adapted for primary and secondary prevention (V1 was limited to stroke survivors)
More risk factors
More sophisticated weighting system (design by our biostatistician)
Development of stroke priority score to be calculated in parallel to stroke risk score
The stroke priority score is derived from the stroke risk score but defaults missing data to highest risk parameter and gives a higher weight for risk factors our team is more focused on and assessing as part of our clinical study (BP, physical inactivity, med noncompliance, etc).
Plan to optimize weights using large CMS dataset (will likely use machine learning)
Functions:
Monitor when data is missing or when risk factors change in a meaningful way
Guide clinicians on risk factor management outside of blood pressure
Determine which patients are higher risk and need more touchpoints
Provides key metrics for judging the overall performance of service (in essence the goal of the service is to reduce a patient’s stroke risk score)
Regulated as clinical decision support
FUTURE VERSIONS BEYOND V2
Will use additional large external datasets to refine algorithm beyond the CMS dataset
Accuracy of prediction will be tested prospectively in our growing dataset
Future versions may be regulated as either clinical decision support or software as a medical device