Appeal Overturn Predictor

This tool can support patients and case workers who want to make an informed decision about appealing an inappropriate health insurance denial. Many patients forgo appeals assuming they have little to no chance of success. This is actually far from true.

If you have a denial and are considering forgoing an appeal, you should know that seeking the advice of a professional is your best option. If that is not an option accessible to you, another good option is to just submit an appeal if you can afford the time. Appeals are extremely successful on average, and it usually doesn't hurt to try aside from the time committment. If you are concerned that the chance of success in your particular case is too low to be worth your time, consider using our tool to see if it will convince you otherwise!

Just enter a brief description of your situation, and you'll be given a prediction for the likelihood that your denial would get overturned if you sought all levels of appeal available to you.

Absolutely nothing! We take privacy and opt-out-by-default extremely seriously. In fact, this tool is almost purely local, and offline first.

This means that it primarily functions on your computer, without communicating externally. There are exactly two types of data transfer involved when you use this site.

First, when you load the page and enter a case or select a pre-built one, one of our models is actually downloaded to your computer to use in your browser. You will see a loading widget while this download occurs.

Second, our site collects usage statistics telling us how many people view the site each day, and what countries their IPs originate from.

That is it. There is no more data transfer. In particular, 0% of the text you enter into this tool ever leaves your device. You can test that the functionality does not rely on us collecting your data as follows. Select 'Manual Entry' from the dropdown, which will trigger the model download. Then leave the webpage up on your browser, but turn off your internet connection. You will see that the tool still works, because it relies on no extra data transfer to the internet.

The model we use to predict case overturn outcomes was trained on data we curated from historical case adjudications in numerous markets. These adjudications correspond to independent medical reviews. We intend to open source the curated dataset with the release of an associated research paper, in late fall 2024.

Yes, it most certainly is. You should be aware of this bias to inform your use. While we lack the information necessary to completely understand all bias baked into this tool, one important type of bias worth noting is that this tool was trained on historical outcomes, and has been trained to predict what will happen in reality, not what should happen in an ideal reality.

While we aim to predict the expected outcome well, this expected outcome in itself reflects bias! You should never use this tool in any way that assumes it is instead predicting whether an appeal should or ought to be overturned. This is not problem the model was designed to address, and using this model for that problem runs the grave risk of propagating harmful, existing bias.

We recommend using this tool with an informed view of its bias, limitations, and risks, and making assessments of its claims with an eye towards qualified trust. That means that you should qualify the level of trust you place in its outputs by the risk of harm of inaccurate outputs.

For example, if the worst case scenario for your use case is that you waste hours submitting an appeal because the model convinces you to appeal, and you were planning on not appealing otherwise, that relatively inconsequential risk may mean you can trust the model to inform such use rather blindly.

On the other hand, if the worst case scenario for your use case is that you will advise a patient, in an automated way, to not appeal based on the model outputs, potentially jeopardizing their financial or physical wellbeing compared to if they did appeal, you ought to place very little trust in the model! Doing so poses too grave a risk to be done without extremely thorough model performance evaluations, which we have not yet developed.

There are many resources that can help you understand how to appeal a denial. Typically the process involves access to an appeal process that your insurer administers, and a subsequent one that an independent review entity administers. As a starting point, you can take a look at this primer.

We also help people navigate these processes, for free. Feel free to reach out to info@persius.org.

Persius is an organization that builds AI to help people resolve inappropriate health insurance coverage denials, and provides human support in such cases for free. In helping to resolve over $275,000 In inappropriate denials at zero cost since our formation, we've learned a thing or two about some of the most problematic insurance related barriers jeopardizing people's access to care.

One of those barriers is simply a lack of knowledge about the recourse one can seek when facing a denial. Many people either do not know that they have appeal rights, or believe that the chance of success is extremely low. This is a self-help tool that can empower patients, case workers, and advocates to get a rough sense for the likelihood that their denial could be overturned, if it were appealed to the level of an external review.

It can also help case workers managing large volumes of requests for help triage cases, and focus on denials which seem very likely to be inappropriate. This is a problem we have some experience with :)