Looking for Collaborators
I am trying to find an opportunity to keep working on some health risk models that I have developed. These models seem to show promise, and could provide new directions for healthcare. One model that seems to have a high potential is a rising risk ranking model that predicts which patients will have an increase in health problems in the next year.
This rising risk model has a high degree of accuracy and has the potential to help lower health costs if suitable interventions can be developed. The model was developed in a healthcare system using Epic electronic health records. If anyone is interested in partnering on this project, please contact me.
Some Background
In 2017, I developed a Probit Health Risk Ranking model that seems to provide a substantial breakthrough in health risk prediction. This model predicts population health rank in the next year with a 39% R-Squared, which seems to be exceptional accuracy.
The best health risk prediction model that I have seen documented anywhere is a commercial health risk model developed by Milliman that has an R-Squared of 37.7%. The Probit model has an R-Squared that is 42% higher than the Milliman model. This appears to be a substantial improvement.
Some Problems
There were some problems when the Probit Health Risk Ranking model was rolled out to providers.
- First, when the model worked well, providers would remark that it was obvious that high risk score patients were unhealthy. (We don’t need this.)
- Second, when the model provided a risk score that did not conform to their opinion, providers assumed that the model was not working. (We don’t trust this.)
- Finally, since the total score was a black box (the model accuracy was due in part to a regression model with over two hundred predictor variables) providers were uncomfortable with not knowing how the scores were calculated. (We don’t understand this.)
A Proposed Solution (Rising Risk Prediction)
One way to overcome some of these problems is to use “rising risk” as the dependent variable. Rising risk is a measure of the likelihood of future health problems. A person with a high rising risk is a seemingly healthy patient who is likely to have substantial health problems in the next year. In tests, it appears that rising risk can also be predicted with the same 39% R-Squared that is obtained with the Probit Health Risk Ranking model.
In theory, if we could identify the patients with hidden health problems who are most likely to have substantial health problems in the next year, something might be done to change their health trajectory. Further work will need to be done to see what factors cause high levels of rising risk and whether the potential risk is treatable.
Interested?
If this seems interesting, please contact me. I would love to show you how the model is built and discuss the possibility of some type of collaboration.
Thomas Arnold, (320) 252-4993.
