Artificial intelligence predictive interactivity based on recursive data analysis
An artificial intelligence (AI) agent interacts with patients via a user interface presented on a computing device. The AI agent may be trained by generating, based on session logs of a cohort of subjects interacting with the user interface, one or more training datasets comprising data on user interactions with the user interface by subjects in the cohort, and data based on health records of the subjects in the cohort of subjects. Machine-learning techniques may be applied to the training datasets to train the AI agent. Based on interactions of a patient with the AI agent, the AI agent may generate a recommendation for a medical screening, a medical procedure, etc. The recommendation may additionally be based on data in an electronic health record of the patient. Data on interactions with additional patients may be used to update models employed through the AI agent.
1 . A method comprising:
providing an artificial intelligence (AI) agent to a patient to assess the patient via a user interface presented on a computing device, wherein the user interface dynamically alters a sequence of assessment questions presented to the patient based on at least one answer received from the patient, wherein the AI agent was trained by:
generating, based on session logs of a cohort of subjects interacting with the user interface, one or more training datasets comprising (i) data on user interactions with the user interface by subjects in the cohort, and (ii) data based on health records of the subjects in the cohort of subjects;
the AI agent by:
processing the one or more training datasets to extract one or more features indicative of user interaction patterns and health characteristics,
inputting the one or more features into the AI agent,
iteratively training the AI agent by adjusting one or more trainable parameters of the AI agent to reduce an error between predicted and actual healthcare recommendations indicated in the health records, and
validating the AI agent using a portion of the one or more training datasets until a predetermined proficiency or accuracy threshold is satisfied, wherein training the AI agent comprises applying a regression model to identify causal factors for one or more medical screenings or procedures based on interactions of the subjects with the user interface;
generating, based on interactions of the patient with the AI agent and on data in an electronic health record of the patient, a recommendation for at least one of a medical screening or a medical procedure; and
presenting the recommendation to at least one of the patient or a healthcare provider of the patient.
2 . The method of claim 1 , further comprising administering the recommended medical screening or the recommended medical procedure to the patient.
3 . The method of claim 1 , further comprising updating the AI agent based on the interactions of the patient with the AI agent.
4 . The method of claim 3 , wherein updating the AI agent comprises generating one or more updated training datasets and applying one or more machine-learning techniques to the updated training datasets.
5 . The method of claim 1 , wherein training the AI agent further comprises applying a pattern recognition model or a classification model to recognize normal or abnormal patterns of interaction with the user interface.
6 . The method of claim 1 , wherein training the AI agent further comprises applying a decisioning model to identify interactions suited to obtaining certain information on subjects through the user interface.
7 . The method of claim 1 , wherein training the AI further comprises applying natural language processing (NLP) to responses by the subjects in the cohort.
8 . The method of claim 1 , wherein the user interface digitally presents a plurality of assessment questions to the patient using an interactive display on a computing device, wherein at least one assessment question comprises an assessment question suited to having multiple simultaneous answers.
9 . The method of claim 6 , wherein the AI agent, in response to receiving one answer to an assessment question suited to having multiple simultaneous answers, updates assessment data about the patient in an assessment database and recursively presents the assessment question suited to having multiple simultaneous answers to the patient again.
10 . The method of claim 9 , wherein the AI agent iteratively analyzes answers stored in the assessment database after each update to determine whether the stored answers indicate that the recommendation should be triggered.
11 . The method of claim 10 , wherein presenting the recommendation comprises electronically communicating assessment data and the recommendation to the healthcare provider.
12 . The method of claim 1 , wherein the data on user interactions indicates selections made through the user interface.
13 . The method of claim 1 , wherein the data on user interactions includes text entries submitted through the user interface.
14 . The method of claim 1 , wherein the data based on health records identifies at least one of medical screenings administered to the subjects in the cohort, medical procedures administered to subjects the subjects in the cohort, or medical conditions of the subjects in the cohort.
15 . A computing system comprising one or more processors configured to:
provide an artificial intelligence (AI) agent to a patient to assess the patient via a user interface presented on a computing device, wherein the user interface dynamically alters a sequence of assessment questions presented to the patient based on at least one answer received from the patient, wherein the AI agent was trained by:
generating, based on session logs of a cohort of subjects interacting with the user interface, one or more training datasets comprising (i) data on user interactions with the user interface by subjects in the cohort, and (ii) data based on health records of the subjects in the cohort of subjects;
training the AI agent by:
processing the one or more training datasets to extract one or more features indicative of user interaction patterns and health characteristics,
inputting the one or more features into the AI agent,
iteratively training the AI agent by adjusting one or more trainable parameters of the AI agent to reduce an error between predicted and actual healthcare recommendations indicated in the health records, and
validating the AI agent using a portion of the one or more training datasets until a predetermined proficiency or accuracy threshold is satisfied, wherein training the AI agent comprises applying a regression model to identify causal factors for one or more medical screenings or procedures based on interactions of the subjects with the user interface;
generate, based on interactions of the patient with the AI agent and on data in an electronic health record of the patient, a recommendation for at least one of a medical screening or a medical procedure; and
present the recommendation to at least one of the patient or a healthcare provider of the patient.
16 . The computing system of claim 15 , wherein training the AI agent further comprises at least one of applying:
a pattern recognition model or a classification model to recognize normal or abnormal patterns of interaction with the user interface;
a decisioning model to identify interactions suited to obtaining certain information on subjects through the user interface; or
natural language processing (NLP) to responses submitted by the subjects in the cohort.
17 . The computing system of claim 15 , wherein:
the user interface digitally presents a plurality of assessment questions to the patient using an interactive display on the computing device, wherein at least one assessment question comprises an assessment question suited to having multiple simultaneous answers;
the AI agent, in response to receiving one answer to an assessment question suited to having multiple simultaneous answers, updates assessment data about the patient in an assessment database and recursively presents the assessment question suited to having multiple simultaneous answers to the patient again; and
the AI agent iteratively analyzes answers stored in the assessment database after each update to determine whether the stored answers indicate that the recommendation should be triggered.
18 . The computing system of claim 15 , wherein:
the data on user interactions includes selections made through the user interface and text entries submitted through the user interface; and
the data based on health records identifies at least one of medical screenings administered to the subjects in the cohort, medical procedures administered to the subjects in the cohort, or medical conditions of the subjects in the cohort.
19 . The method of claim 1 , wherein altering the user interface comprises re-presenting at least one assessment question configured to accept multiple simultaneous answers to solicit additional different answers from the patient.
20 . The computing system of claim 15 , wherein altering the user interface comprises re-presenting at least one assessment question configured to accept multiple simultaneous answers to solicit additional different answers from the patient.