Systems and methods for continuous cancer treatment and prognostics

A system for determining a cancer prognosis for an individual, comprising:

• one or more processors;
• a memory readable by the one or more processors;
• a network interface;
• instructions in the memory that when read by a processor configures the processor to:
• obtain medical data for an individual from a medical records database, wherein the medical data comprises physician notes;
• perform natural language processing (NLP) on the medical data by:
• extracting a plurality of tokens from the physician notes using a feature extraction model that breaks the physician notes into tokens of sub-word units, wherein each of the plurality of tokens comprises one or more of:
• a token embedding, a segment embedding that identifies a sequence the token embedding associates with, and a positional embedding comprising a position of the token in the physician notes;
• applying term frequency inverse document frequency (TF-IDF) to the extracted tokens;
• in response to applying the TF-IDF, identify a set of key terms that predict cancer survivability, wherein each of the set of key terms includes a set of one or more extracted tokens;
• generating a conditional latent space as a multi-dimensional vector from a final hidden layer of the feature extraction model using the sets of one or more extracted tokens from the identified set of key terms;
• determining at least one prognostic for the individual by applying a machine learning model on the generated conditional latent space to determine a prognostic for the individual;
• identifying at least one treatment option for the individual based on the prognostic for the individual, wherein the at least one treatment option is selected from immunotherapy, targeted therapy, radiation therapy, chemotherapy, and surgical intervention;
• transmitting the at least one treatment option to one or more devices associated with a physician;
• and
• in response to the transmitting, administering the individual with the at least one treatment option.