Tool for predicting prognosis and improving survival of patients
A diagnostic and decision support technology is provided for determining the likely prognosis and potential treatment for patients experiencing a condition, such as COVID-19, for example. In particular, a mechanism is provided for receiving a historical patient dataset comprising one or more historical health parameters associated with a plurality of historical patients. Additionally, a patient dataset is received comprising one or more patient health parameters associated with a patient. A cluster is identified based on the similarity of the patient dataset and a plurality of historical patient datasets. From the cluster, a set of treatable features are identified and evaluated for their potential impact on the patient's successful recovery from the condition. A recommendation is generated based on the evaluation as to what feature should be treated first to decrease the mortality of the patient.
1 . A computer-implemented method performed by at least one device including a hardware processor, the computer-implemented method comprising:
training a neural network, implemented using an unsupervised machine learning algorithm, to identify one or more treatable health parameters, wherein training the neural network comprises:
accessing, via a network connection by a multi-agent system comprising multiple artificial intelligence agents (multiple AI agents), a database cluster that comprises an electronic health record system (EHR system) that is remote from the multi-agent system, the EHR system storing patient medical records in disparate data sources in a variety of locations that are remote from one another;
receiving, by the multi-agent system from the disparate data sources of the EHR system via the network connection, a historical patient dataset associated with a plurality of historical patients, wherein the historical patient dataset comprises one or more historical time-dependent health parameters and one or more historical time-independent health parameters, the one or more historical time-dependent health parameters comprising multiple historical blood test measurements obtained over a period of time;
converting, by the multi-agent system, the one or more historical time-dependent health parameters of the historical patient dataset into one or more converted historical time-independent health parameters using a transformation algorithm and a fuzzy logic algorithm;
generating, by the multi-agent system, multiple thousands of versions of an analytical dataset using the historical patient dataset, the analytical dataset including the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters, wherein generating the multiple thousands of versions of the analytical dataset comprises using different combinations of unsupervised learning algorithms to define different relationships between data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset;
populating, by the multi-agent system, a probabilistic graph database with the different relationships between the data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset;
generating, by the unsupervised machine learning algorithm of the neural network, a plurality of clusters from the multiple thousands of versions of the analytical dataset using the different relationships between the data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset from the probabilistic graph database,
wherein populating the probabilistic graph database and generating the plurality of clusters comprise modifying a graph-based data structure that encodes probabilistic relationships between health parameters across the multiple thousands of versions of the analytical dataset;
identifying, using the unsupervised machine learning algorithm of the neural network, sets of treatable historical health parameters that correspond respectively to the plurality of clusters; and
ranking, using the unsupervised machine learning algorithm of the neural network, the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters relative to one another respectively within the corresponding sets of treatable historical health parameters based on each treatable historical health parameter's effect on a mortality of each of the plurality of historical patients corresponding to the respective clusters of the corresponding sets of treatable historical health parameters;
receiving, by the multi-agent system from the EHR system via the network connection, a particular patient dataset associated with a particular patient, wherein the particular patient dataset comprises one or more particular time-dependent health parameters and one or more particular time-independent health parameters, the one or more particular time-dependent health parameters comprising multiple particular blood test measurements obtained over a period of time;
identifying, using the unsupervised machine learning algorithm of the neural network, a particular cluster in the plurality of clusters that is associated with the particular patient based on a measure of similarity between the health parameters of the particular patient and the health parameters of the particular cluster; and
initiating a treatment for the particular patient based on at least one treatable historical health parameter in the particular cluster, wherein initiating the treatment comprises:
generating a recommendation to treat the at least one treatable historical health parameter in the particular cluster for the particular patient based on the ranking of the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters; and
communicating the recommendation to a computing device for display on the computing device.
2 . The computer-implemented method of claim 1 , further comprising identifying one or more treatable health parameters from the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters within the particular cluster.
3 . The computer-implemented method of claim 1 , further comprising generating a survival percentage of each one or more time-independent health parameters and the one or more converted historical time-independent health parameters, the survival percentage being generated based on an effect treatment of each of the health parameters is predicted to have on the particular patient associated with the particular patient dataset.
4 . The computer-implemented method of claim 1 , further comprising modeling the one or more historical time-dependent historical health parameters and the one or more particular time-dependent health parameters using sine and cosine functions.
5 . The computer-implemented method of claim 1 , wherein:
the multiple AI agents of the multi-agent system comprise a first agent solver, a second agent solver, a third agent solver, and a fourth agent solver;
wherein the first agent solver comprises a fuzzy neural network, converting the one or more historical time-dependent health parameters of the historical patient dataset into one or more converted historical time-independent health parameters comprises:
converting the one or more historical time-dependent health parameters from time-dependent data in a time domain to time-independent data in a frequency domain using a Fourier transform; and
the first agent solver executing the fuzzy logic algorithm using the fuzzy neural network; and
wherein the second agent solver generates the plurality of clusters from the multiple thousands of versions of the analytical dataset using a clustering algorithm;
wherein the ranking the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters comprises:
the third agent solver creating a self-organizing map for the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters; and
the fourth agent solver running a K-nearest neighbor algorithm on the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters within the self-organizing map to perform an evaluation of the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters.
6 . The computer-implemented method of claim 1 , wherein the at least one treatable historical health parameter comprises pneumonia and further comprising treating the particular patient for pneumonia.
7 . The computer-implemented method of claim 1 , wherein converting the one or more historical time-dependent health parameters of the historical patient dataset into the one or more converted historical time-independent health parameters further comprises:
converting the one or more historical time-dependent health parameters from time-dependent data in a time domain to time-independent data in a frequency domain using a Fourier transform algorithm;
computing one or more statistical measures of the time-independent data in the frequency domain; and
generating the one or more converted historical time-independent health parameters based at least on the one or more statistical measures of the time-independent data using the fuzzy logic algorithm.
8 . One or more non-transitory computer-readable storage media storing computer-useable instructions that, when used by one or more computing devices, cause the one or more computing devices to perform a method, the method comprising:
training a neural network, implemented using an unsupervised machine learning algorithm, to identify one or more treatable health parameters, wherein training the neural network comprises:
accessing, via a network connection by a multi-agent system comprising multiple artificial intelligence (AI) agents, a database cluster that comprises an EHR system that is remote from the multi-agent system, the EHR system storing patient medical records in disparate data sources in a variety of locations that are remote from one another;
receiving, by the multi-agent system from the disparate data sources of the EHR system via the network connection, a historical patient dataset associated with a plurality of historical patients, wherein the historical patient dataset comprises one or more historical time-dependent health parameters and one or more historical time-independent health parameters, the one or more historical time-dependent health parameters comprising multiple historical blood test measurements obtained over a period of time;
converting, by the multi-agent system, the one or more historical time-dependent health parameters of the historical patient dataset into one or more converted historical time-independent health parameters using a transformation algorithm and a fuzzy logic algorithm;
generating, by the multi-agent system, multiple thousands of versions of an analytical dataset using the historical patient dataset, the analytical dataset including the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters, wherein generating the multiple thousands of versions of the analytical dataset comprises using different combinations of unsupervised learning algorithms to define different relationships between data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset;
populating, by the multi-agent system, a probabilistic graph database with the different relationships between the data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset;
generating, by the unsupervised machine learning algorithm of the neural network, a plurality of clusters from the multiple thousands of versions of the analytical dataset using the different relationships between the data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset from the probabilistic graph database,
wherein populating the probabilistic graph database and generating the plurality of clusters comprise modifying a graph-based data structure that encodes probabilistic relationships between health parameters across the multiple thousands of versions of the analytical dataset;
identifying, using the unsupervised machine learning algorithm of the neural network, sets of treatable historical health parameters that correspond respectively to the plurality of clusters; and
ranking, using the unsupervised machine learning algorithm of the neural network, the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters relative to one another respectively within the corresponding sets of treatable historical health parameters based on each treatable historical health parameter's effect on a mortality of each of the plurality of historical patients corresponding to the respective clusters of the corresponding sets of treatable historical health parameters;
receiving, by the multi-agent system from the EHR system via the network connection, a particular patient dataset associated with a particular patient, wherein the particular patient dataset comprises one or more particular time-dependent health parameters and one or more particular time-independent health parameters, the one or more particular time-dependent health parameters comprising multiple particular blood test measurements obtained over a period of time;
identifying, using the unsupervised machine learning algorithm of the neural network, a particular cluster in the plurality of clusters that is associated with the particular patient based on a measure of similarity between the health parameters of the particular patient and the health parameters of the particular cluster; and
initiating a treatment for the particular patient based on at least one treatable historical health parameter in the particular cluster, wherein initiating the treatment comprises:
generating a recommendation to treat the at least one treatable historical health parameter in the particular cluster for the particular patient based on the ranking of the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters; and
communicating the recommendation to a computing device for display on the computing device.
9 . The one or more non-transitory computer-readable storage media of claim 8 , further comprising identifying one or more treatable health parameters from one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters within the particular cluster.
10 . The one or more non-transitory computer-readable storage media of claim 8 , further comprising generating a survival percentage of each one or more time-independent health parameters and the one or more converted historical time-independent health parameters, the survival percentage being generated based on an effect treatment of each of the health parameters is predicted to have on the particular patient associated with the particular patient dataset.
11 . The one or more non-transitory computer-readable storage media of claim 8 , further comprising modeling the one or more historical time-dependent health parameters and the one or more particular time-dependent health parameters using sine and cosine functions.
12 . The one or more non-transitory computer-readable storage media of claim 11 , wherein the transformation algorithm is a Fourier transform algorithm of the sine and cosine functions.
13 . The one or more non-transitory computer-readable storage media of claim 8 , wherein the at least one treatable historical health parameter comprises pneumonia.
14 . A system comprising:
one or more processors, that by executing computer readable instructions stored in memory, perform:
training a neural network, implemented using an unsupervised machine learning algorithm, to identify one or more treatable health parameters by at least:
accessing, via a network connection by a multi-agent system comprising multiple artificial intelligence (AI) agents, a database cluster that comprises an EHR system that is remote from the multi-agent system, the EHR system storing patient medical records in disparate data sources in a variety of locations that are remote from one another;
receiving, by the multi-agent system from the disparate data sources of the EHR system via the network connection, a historical patient dataset associated with a plurality of historical patients, wherein the historical patient dataset comprises one or more historical time-dependent health parameters and one or more historical time-independent health parameters, the one or more historical time-dependent health parameters comprising multiple historical blood test measurements obtained over a period of time;
converting, by the multi-agent system, the one or more historical time-dependent health parameters of the historical patient dataset into one or more converted historical time-independent health parameters using a transformation algorithm and a fuzzy logic algorithm;
generating, by the multi-agent system, multiple thousands of versions of an analytical dataset using the historical patient dataset, the analytical dataset including the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters, wherein generating the multiple thousands of versions of the analytical dataset comprises using different combinations of unsupervised learning algorithms to define different relationships between data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset;
populating, by the multi-agent system, a probabilistic graph database with the different relationships between the data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset;
generating, by the unsupervised machine learning algorithm of the neural network, a plurality of clusters from the multiple thousands of versions of the analytical dataset using the different relationships between the data elements of the analytical dataset amongst the multiple thousands of versions of the analytical dataset from the probabilistic graph database,
wherein populating the probabilistic graph database and generating the plurality of clusters comprise modifying a graph-based data structure that encodes probabilistic relationships between health parameters across the multiple thousands of versions of the analytical dataset;
identifying, using the unsupervised machine learning algorithm of the neural network, sets of treatable historical health parameters that correspond respectively to the plurality of clusters; and
ranking, using the unsupervised machine learning algorithm of the neural network, the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters relative to one another respectively within the corresponding sets of treatable historical health parameters based on each treatable historical health parameter's effect on a mortality of each of the plurality of historical patients corresponding to the respective clusters of the corresponding sets of treatable historical health parameters;
receiving, by the multi-agent system from the EHR system via the network connection, a particular patient dataset associated with a particular patient, wherein the particular patient dataset comprises one or more particular time-dependent health parameters and one or more particular time-independent health parameters, the one or more particular time-dependent health parameters comprising multiple particular blood test measurements obtained over a period of time;
identifying, using the unsupervised machine learning algorithm of the neural network, a particular cluster in the plurality of clusters that is associated with the particular patient based on a measure of similarity between the health parameters of the particular patient and the health parameters of the particular cluster; and
initiating a treatment for the particular patient based on at least one treatable historical health parameter in the particular cluster, wherein initiating the treatment comprises:
generating a recommendation to treat the at least one treatable historical health parameter in the particular cluster for the particular patient based on the ranking of the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters; and
communicating the recommendation to a computing device for display on the computing device.
15 . The system of claim 14 , further comprising identifying one or more treatable health parameters from the one or more historical time-independent health parameters and the one or more converted historical time-independent health parameters within the particular cluster.
16 . The system of claim 14 , further comprising generating a survival percentage of each one or more time-independent health parameters and the one or more converted historical time-independent health parameters, the survival percentage being generated based on an effect treatment of each of the health parameters is predicted to have on the particular patient associated with the particular patient dataset.
17 . The system of claim 14 , further comprising modeling the one or more historical time-dependent health parameters and the one or more particular time-dependent health parameters using sine and cosine functions.
18 . The system of claim 17 , wherein the transformation algorithm is a Fourier transform algorithm of the sine and cosine functions.
19 . The system of claim 14 , wherein the at least one treatable historical health parameter comprises pneumonia.
20 . The system of claim 14 , wherein converting the one or more historical time-dependent health parameters of the historical patient dataset and the particular patient dataset into the one or more converted historical time-independent health parameters further comprises:
converting the one or more historical time-dependent health parameters from time-dependent data in a time domain to time-independent data in a frequency domain using a Fourier transform algorithm;
computing one or more statistical measures of the time-independent data in the frequency domain; and
generating the one or more converted historical time-independent health parameters based at least on the one or more statistical measures of the time-independent data using the fuzzy logic algorithm.
21 . The computer-implemented method of claim 1 , wherein populating the probabilistic graph database comprises storing weighted connections representing confidences and distances between health parameters derived from the multiple thousands of versions of the analytical dataset, wherein generating the plurality of clusters comprises adjusting similarity thresholds based on the weighted connections to regulate cluster formation.
22 . The computer-implemented method of claim 1 , wherein populating the probabilistic graph database comprises encoding weighted probabilistic relationships between health parameters based on confidences derived from the multiple thousands of versions of the analytical dataset, wherein generating the plurality of clusters comprises applying a self-organizing map and a k-nearest neighbor algorithm that utilize the weighted probabilistic relationships to determine distances and cluster boundaries.