| CPC G16H 50/20 (2018.01) [G06N 5/01 (2023.01); G16H 10/60 (2018.01); G16H 50/70 (2018.01); G16H 50/80 (2018.01)] | 12 Claims |
|
1. A computer-implemented method comprising:
receiving input corresponding to a selection of an identifier of a subject record associated with a subject, the identifier of the subject record being selected using an interface;
retrieving the subject record from a data store, the subject record including a set of subject attributes that medically characterize the subject, the subject attributes including at least a medical diagnosis;
generating an array representation for the subject, the array representation being generated by transforming the set of subject attributes using singular value decomposition (SVD) to generate one or more numerical representations for each attribute and combine the numerical representations into a vector representation as the array representation;
inputting the array representation for the subject into a trained machine-learning model, the trained machine-learning model comprising:
a set of parameters that were learned using a set of other subject records stored in a data registry, each other subject record of the set of other subject records being associated with another subject who was infected with COVID-19 and subsequently treated using a treatment, and each other subject record of the set of other subject records including a respective set of other subject attributes that medically characterize the associated another subject; and
one or more functions configured to transform array-representation input into size-reduced output using the set of parameters, wherein the one or more functions are configured to repeatedly monitor and aggregate large-scale data from across a plurality of institutions;
outputting, from the trained machine-learning model, an output that includes a classification for the subject based upon the input and the set of learned parameters;
determining, based on the output, that the subject record corresponds to criteria of a COVID-19 diagnosis and/or factors indicative of suitability of a particular COVID-19 treatment; and
in response to determining that the subject record corresponds to the criteria of a COVID-19 diagnosis:
determining, using a random forest model trained to iterate through the set of other subject records, one or more segmenting thresholds for segmenting the set of other subject records into one or more subject groups, each subject group corresponding to a subject outcome, where one of the subject groups corresponds to subjects who have been discharged after recovering from COVID-19;
determining a set of treatments associated with the subject group corresponding to subjects who have been discharged after recovering from COVID-19;
determining, using the random forest model and based on the one or more segmenting thresholds, a set of characteristics determined to contribute to the discharging of the subjects who have been discharged after recovering from COVID-19;
determining a respective similarity metric between the subject record and each characteristic from the set of characteristics by automatically combining numerical representations of the SVD-transformed attributes in a domain space;
determining, based on the similarity metrics, a confidence score for each treatment in the set of treatments; and
presenting, on the interface and based on the confidence scores, the set of treatments as proposed treatments for treating the subject.
|