A Prognosis Predictor for Covid-19
Considering pre-existing conditions, shot status (time dependent), outcomes for those that get the disease, including long Covid, as well as all adverse effects of the shot
I propose to develop, using Artificial Intelligence technology, a physician assistant that will provide a prognosis for patients who
1. Are considering a Covid-19 shot, whether initial series or booster and/or
2. Have tested positive for the SARS-CoV-2 virus.
The outcomes will range from the general, such as recovered with no symptoms after 1 week to death, to the specific, including all the important adverse effects (+ other) or long Covid (if I can find a medically explicit definition of a set of one or more symptoms, plus severity.
Of course, the difficulty here will not be in building the training and execution machines, but in finding adequate data to use for training and testing.
I know that the data are going to be sloppy, with outcomes or conditions like hospitalized with Covid or hospitalized because of Covid. Therefore, I intend for my outcomes to be in the form of distribution functions. I expect that I will build on Bayesian statistics, with some stochastic Petri nets to model the processes. This will be a stateful model!
In general, I would expect the observations-to be in the form of distribution functions as well, and I would expect that there may be costs associated with each outcome.
The math is going to be “difficult”, since statistical conformity will need to be maintained through the stochastic Petri net. This means than convolutions of distribution functions will need to be computed, a resource intensive task.
At this time, I don’t see either decision trees or neural networks to be adequate for the job. However, as I find a reasonable set of data for even a part of this process, I will test those techniques.
I would like the Covid-19 Prognosis Predictor to be as adaptive as possible, so, as new data become available, additional training will be able to incorporate those data.
I am aware that Kaggle has a number of datasets they provide that might be useful. I’m also aware that some models have been created that touch on this topic. I will evaluate those and publish any relevant observations on this substack.
So, for the next number of contributions to this substack, I will be developing a semantic model of Covid-19, its effectiveness and its adverse effects, which should make translation to the stochastic Petri net representation feasible. Alternately, I will be evaluating the published set of models and results available in the literature that touch on this topic.
Feel free to contribute your thoughts and recommendations on this project in the Comments section.
Testing for Nutrient deficiencies that contribute to vulnerability to pathogens makes more sense to me .