Clover Health Patent

Systems And Methods For Determining Event Probability
DOCUMENT ID DATE PUBLISHED

US 11508465 B2 2022-11-22
INVENTOR INFORMATION
NAME CITY STATE ZIP CODE COUNTRY Blumenfeld;
Ian
San Francisco CA N/A US
APPLICANT INFORMATION
NAME CITY STATE ZIP CODE COUNTRY AUTHORITY
Clover Health Jersey City NJ N/A US N/A
TYPE
assignee
ASSIGNEE INFORMATION
NAME CITY STATE ZIP CODE COUNTRY TYPE CODE
Clover Health Jersey City NJ N/A US 02
APPLICATION NO DATE FILED

16/021976 2018-06-28
US CLASS CURRENT:
1/1
CPC CURRENT
TYPE CPC DATE
CPCI G 06 N 20/00 2019-01-01
CPCI G 06 N 7/005 2013-01-01
CPCI G 16 H 10/60 2018-01-01
KWIC Hits
APPLICANT GROUP
Clover Health Jersey City NJ US
ASSIGNEE NAME (USPAT, PGPUB)

Clover Health
Abstract
Systems and methods for data unfolding are disclosed. For example, it may be desirable or necessary
to increase a data set, such as for increasing accuracy of one or more predictive models. Data set
proliferation without introducing unnecessary bias may be important for increasing such accuracy.
Described herein are system and methods that allow for data set proliferation by generating records
based on whether an event occurred with respect to an entity during multiple time intervals. A record
may be generated for each time interval and the associated data may be unfolded and disassociated,
at least partly, from other records related to the entity. Those records may then be used for data
analytics and/or predictive model generation, for example.
Background/Summary
BACKGROUND
(1) The use of data for analytics has become ubiquitous. The usefulness of data for analytics may
depend, generally, on whether a sufficient set of data has been collected. However, in some instances,
sufficient data has not been generated and/or has not been collected to provide a meaningful analysis.
Described herein are improvements in technology and solutions to technical problems that can be
used to, among other things, generate data sets of sufficient quantity and quality to increase the
accuracy of data analytics.
Description
BRIEF DESCRIPTION OF THE DRAWINGS
(1) The detailed description is set forth below with reference to the accompanying figures. In the
figures, the left-most digit(s) of a reference number identifies the figure in which the reference number
first appears. The use of the same reference numbers in different figures indicates similar or identical
items. The systems depicted in the accompanying figures are not to scale and components within the
figures may be depicted not to scale with each other.
(2) FIG. 1 illustrates a schematic diagram of an example environment for data folding and unfolding.
(3) FIG. 2 illustrates a conceptual diagram of components and/or operations associated with a record
generator for data folding and unfolding.
(4) FIG. 3 illustrates a conceptual diagram of example ranked data entries and uses of the same.
(5) FIG. 4 illustrates a conceptual diagram of example ranked data entries and further uses of the
same.
(6) FIG. 5 illustrates a flow diagram of an example process for data folding and unfolding.
(7) FIG. 6 illustrates a flow diagram of another example process for data folding and unfolding.
DETAILED DESCRIPTION
(8) Systems and methods for folding and unfolding data are described herein. Take, for example, a
system that has stored thereon and/or has access to one or more databases containing data, such as
health-related data, for example. Such data may be used by the system for one or more purposes,
such as predicting outcomes and/or events. However, without a robust data set from which to perform
data analyses, the accuracy of such predictions may be limited. Increasing available data on which to
perform analyses may improve prediction accuracy and/or allow for more robust analysis techniques
to be utilized. For example, the system may have access to data for two patients. One of the patients
may have been hospitalized since data was recorded for the patient while the other patient may not
have been hospitalized. In these examples with just two data points on hospitalization, determining a
likelihood that the patients and/or other patients will be hospitalized using the available data would
likely lead to an inaccurate result. For example, this data may indicate that patients are 50% likely to
be hospitalized during a certain period of time. This estimation may be grossly over or under predicted
for a given population.
(9) The presently-disclosed systems and methods may be utilized to increase data set size to increase
data analysis accuracy without introducing bias from historical information. For example, the system
may have access to data associated with multiple people. For each person, which may otherwise be
described herein as an entity, a determination may be made as to whether an event of interest has
occurred. By way of example, the data may indicate if each person has been hospitalized. In this
example, for each person, only one data point exists for a hospitalization event, namely, the person
either has or has not been hospitalized. To increase data points, the data for a person associated with
the system may be “folded” as described herein. For example, a time interval may be selected at
which to assess the data associated with the person. By way of example, the time interval may be 28
days. The system may then analyze the data with respect to the time intervals to determine if the
event has occurred during that time period. For example, the system may determine that for a first
time interval (i.e., the first 28-day period of data) the person was not hospitalized; for a second time
interval (i.e., the second 28-day period of data) the person was not hospitalized; for a third time
interval (i.e., the third 28-day period of data) the person was hospitalized; and for a fourth time interval
(i.e., the fourth 28-day period of data) the person was not hospitalized. In this example, the folding of
data may result in the generation of four records for the person in question. The first record may
include data associated with the person (e.g., blood pressure, age, weight, etc.) for the first period of
time and an indication that the person was not hospitalized. The second record may include data
associated with the person for the second period of time and an indication that the person was not
hospitalized. The third record may include data associated with the person for the third period of time
and an indication that the person was hospitalized. The fourth record may include data associated with
the person for the fourth period of time and an indication that the person was not hospitalized.
(10) By folding data as described above and elsewhere herein, multiple records may be generated for
each person associated with the system. In the example provided above, the original data record
associated with the person, which indicates that the person was at some point hospitalized, may be
utilized to generate four independent records with health-related data and their own indications of
hospitalization events. These records may be “unfolded” or otherwise dissociated from each other and
may be utilized to generate one or more models for, for example, determining an event probability. The
models may be trained using the unfolded data and may be utilized to prospectively determine event
probabilities for people associated with the system.
(11) For example, a system may have access to health-related data for 20,000 people, which amounts
to 20,000 records with one record for each person. Utilizing the data folding and unfolding techniques
described herein, such as using 10 time intervals during the folding processes, 200,000 records may
be generated and then utilized to generate prediction models. This more robust data set may increase
the data sampling available to the prediction models, which may allow those models to be trained
and/or fit more accurately. The models may be built utilizing, for example, one or more machine
learning techniques where features associated with the health-related data are inputs to the model.
The models may be utilized to determine a probability of an outcome occurring, such as for people
associated with the system. Those people may be ranked based at least in part on the determined
probability and the ranking may be utilized to, for example, generate a recommendation, intervention
schedule, incentive program, and/or to admit or not admit people to a given program.
(12) The prediction models generated utilizing the folding and unfolding techniques described herein
may be validated using one or more validation techniques. For example, an area-under-the-resource-
operating-characteristic-curve measurement may be utilized to gauge the accuracy of the model
generated utilizing the unfolded data. In examples, the measurement may be at least 0.84, indicating a
high degree of accuracy. Additionally, or alternatively, precision-recall characteristic measurements
may be utilized to determine precision of the prediction models. The model generation techniques
described herein may also include at least one of gradient boosting techniques and/or hyperparameter
tuning.
(13) Additionally, or alternatively, while the unfolded records may be independent of each other, an
association between records belonging to the same person may be identified and utilized when
performing model testing. In these examples, a group of records belonging to a first set of people may
be used for training purposes while another group of records belonging to a second set of people may
be used for testing purposes. By maintaining an association between records belonging to the same
person, records may be split into the two groups without introducing leakage and/or overlap to the
testing process.
(14) Additionally, or alternatively, the systems and methods described herein may identify input
features that increase the accuracy of the prediction models and other input features that decrease the
accuracy of the prediction models or have no or little effect on the prediction models. The prediction
models may be refit to utilize the features that increase accuracy while refraining from utilizing the
features that decrease accuracy or have no or little effect on accuracy.
(15) The present disclosure provides an overall understanding of the principles of the structure,
function, manufacture, and use of the systems and methods disclosed herein. One or more examples
of the present disclosure are illustrated in the accompanying drawings. Those of ordinary skill in the art
will understand that the systems and methods specifically described herein and illustrated in the
accompanying drawings are non-limiting embodiments. The features illustrated or described in
connection with one embodiment may be combined with the features of other embodiments, including
as between systems and methods. Such modifications and variations are intended to be included
within the scope of the appended claims.
(16) Additional details are described below with reference to several example embodiments.
(17) FIG. 1 illustrates a schematic diagram of an example environment 100 for data folding and
unfolding. The environment 100 may include a first data source 102, a second data source 104, and a
third data source 106, for example. As shown in FIG. 1, the first data source 102 represents a remote
system that generates and/or acquires data associated with one or more people and stores that data,
such as in a database associated with the first data source 102. By way of example, the first data
source 102 may be a hospital, clinic, and/or other type of healthcare provider. The second data source
104 represents a personal computing device associated with a given person. The computing device
may receive inputs from the person and may generate input data, which, in the examples provided
herein, may be utilized to generate and/or store health-related data associated with the person. The
third data source 106 represents a mobile device associated with another person. The mobile device
may receive inputs from the other person and may generate input data, which, in the examples
provided herein, may be utilized to generate and/or store health-related data associated with the other
person.
(18) It should be understood that while three data sources 102, 104, 106 are depicted in FIG. 1, the
number of data sources may be one, two, three, or more than three. In examples, the number of data
sources may be much more than three, such as thousands of data sources. It should also be
understood that while the types of data sources have been depicted with respect to FIG. 1 as remote
systems, personal computing devices, and mobile devices, the types of data sources may be any
device that may receive input and generate data associated with a person based on that input.
Further, while the examples provided herein are described with respect to data associated with a
person or people, the data need not be specific to a person. For example, the data may be data
associated with a device, a system, an entity that is not a person, an object, etc.
(19) The first data source 102, the second data source 104, and/or the third data source 106 may
communicate with a system 108 via one or more networks 110. The communication may include the
sending and/or receiving of data associated with a person and/or people associated with a given data
source. The communication may also include data representing recommendations and/or requests
generated by the system 108, which may be based at least in part on the data as described herein.
The system 108 may include one or more components, such as, for example, one or more processors
112, one or more network interfaces 114, and memory 116. The memory 116 may include one or more
components, such as, for example, a data storage 118, a record generator 120, a model generator
122, a training component 124, an event probability component 126, a ranking component 128, a
recommendation component 130, and/or a validation component 132. Each of the components of the
memory 116 will be described below in detail.
(20) The data storage 118 may be configured to store data received by the system 108. For example,
the data storage 118 may be configured to receive data from one or more of the first data source 102,
the second data source 104, and/or the third data source 106. The data storage 118, and/or one or
more other components of the system 108, may be configured to format the data for storage in the
data storage 118 such that the data is associated with an identifier of the person associated with the
data. For example, data associated with a first person may be received from one or more of the data
sources 102, 104, 106. Each data source 102, 104, 106 may send the data in the same or differing
formats and/or may send the data with differing identification formats. For example, data for “John
Smith” may be received from the first data source 102 with the identifier “John Smith,” while data may
be received from the second data source 104 with the identifier “J. Smith” or “Smith, John.” The data
storage 118 and/or other components of the system 108 may be configured to associate received data
such that the data is associated with the proper person in the system 108. Additionally, or alternatively,
while the data storage 118 has been described as storing data, the data storage 118 may also, or
alternatively, be configured to request and/or access data from the data sources 102, 104, 106 with or
without storing that data in the memory 116 of the system 108.
(21) The record generator 120 may be configured to perform the folding and/or unfolding processes
described herein. For example, the record generator 120 may be configured to identify, determine,
and/or receive an indication of a time interval at which to generate records. Keeping with the example
used herein, the time interval may be, for example, 28 days. It should be understood that this example
time interval is used for illustration only. Other time intervals may be utilized. The record generator 120
may also be configured to identify, determine, and/or receive an indication of a period of time during
which to generate records. For example, a particular time frame may be utilized to generate the
records and/or an amount of time from a certain date may be utilized. For example, the record
generator 120 may identify a one-year period of time during which to generate records. Alternatively,
the record generator 120 may identify a date, such as Jan. 1, 2017, from which to start record
generation either moving forward or backward in time.
(22) The record generator 120 may be further configured to identify, determine, and/or receive an
indication of the data sources from which to utilize data. For example, the record generator 120 may
identify, determine, and/or receive an indication that the data storage 118 is to be utilized for receiving
data. Additionally, or alternatively, the record generator 120 may identify, determine, and/or receive an
indication that the first data source 102, the second data source 104, and/or the third data source 106
is to be utilized in addition to or alternatively to the data storage 118.
indication of an event to analyze with respect to the data. For example, the record generator 120 may
identify, determine, and/or receive an indication that an event of interest is hospitalization. The record
generator 120 may then determine, for each time interval during the identified time period, whether the
event occurred. For example, the system 108 may determine that for a first time interval (e.g., the first
28-day period of data) a particular person was not hospitalized; for a second time interval (e.g., the
second 28-day period of data) the person was not hospitalized; for a third time interval (e.g., the third
28-day period of data) the person was hospitalized; and for a fourth time interval (e.g., the fourth 28-
day period of data) the person was not hospitalized. In this example, the time interval was identified
and/or determined to be 28 days, and the identified time period was 112 days. This process may be
performed for all or some of the people on which data is available to the system 108. In examples, the
system 108 may identify, determine, and/or receive an indication of the people, the number of people,
and/or categories of people on which to perform the processes described herein.
(24) In the example used herein, the folding of data as described above may result in the generation of
four records for each person. For the example person illustrated herein, the first record may include
data associated with the person (e.g., blood pressure, age, weight, etc.) for the first period of time and
an indication that the person was not hospitalized. The second record may include data associated
with the person for the second period of time and an indication that the person was not hospitalized.
The third record may include data associated with the person for the third period of time and an
indication that the person was hospitalized. The fourth record may include data associated with the
person for the fourth period of time and an indication that the person was not hospitalized.
each person associated with the system and/or each person selected for data folding. In the example
provided above, the original data record associated with the example person, which indicates that the
person was at some point hospitalized, may be utilized to generate four independent records with data
and their own indications of hospitalization events. The records may be unfolded, or otherwise
separated from each other as independent records for use by the system 108 as described herein.
Extrapolating from the provided example, the system 108 may have access to data for 100,000
people, which amounts to 100,000 records with one record for each person. Utilizing the data folding
and unfolding techniques described herein, such as using 10 time intervals during the folding
processes, 1,000,000 records may be generated and then utilized by the system 108.
(26) The model generator 122 may be configured to generate one or more models based at least in
part on the records generated by the record generator 120 and/or other data available to the system
108. The more robust data set generated by the folding and unfolding techniques described herein
may increase the data sampling available to the models, which may allow those models to be trained
and/or fit more accurately. The models may be generated utilizing, for example, one or more machine
learning techniques where features associated with the data are inputs to the model. The models may
be utilized to determine a probability of an outcome occurring, such as for people associated with the
system. These models may be described herein as predictive models and/or predictive layers.
(27) A predictive model may include one or more models that utilize predictive analytics to determine
one or more outcomes. Predictive analytic techniques may include, for example, predictive modelling,
machine learning, and/or data mining. Generally, predictive modelling may utilize statistics to predict
outcomes. Machine learning, while also utilizing statistical techniques, may provide the ability to
improve outcome prediction performance without being explicitly programmed to do so. A number of
machine learning techniques may be employed to generate and/or modify the layers and/or models
describes herein. Those techniques may include, for example, decision tree learning, association rule
learning, artificial neural networks (including, in examples, deep learning), inductive logic
programming, support vector machines, clustering, Bayesian networks, reinforcement learning,
representation learning, similarity and metric learning, sparse dictionary learning, and/or rules-based
machine learning.
(28) Information from stored and/or accessible data may be extracted from one or more databases,
such as for example the data storage 118, and may be utilized to predict trends and behavior patterns.
In examples, the event, otherwise described herein as an outcome, may be an event that will occur in
the future, such as whether a member will be hospitalized, a likelihood of a member being diagnosed
with diabetes, a likelihood of a member missing a medication prescription fill at a pharmacy, etc. The
predictive analytic techniques may be utilized to determine associations and/or relationships between
explanatory variables and predicted variables from past occurrences and utilizing these variables to
predict the unknown outcome. The predictive analytic techniques may include defining the outcome
and data sets used to predict the outcome. Then, data may be collected and/or accessed to be used
for analysis, such as from the data storage 118.
(29) Data analysis may include using one or more models, including for example one or more
algorithms, to inspect the data with the goal of identifying useful information and arriving at one or
more determinations that assist in predicting the outcome of interest. One or more validation
operations may be performed, such as using statistical analysis techniques, to validate accuracy of the
models, as described more fully herein. Thereafter, predictive modelling may be performed to generate
accurate predictive models for future events. Outcome prediction may be deterministic such that the
outcome is determined to occur or not occur. Additionally, or alternatively, the outcome prediction may
be probabilistic and/or stochastic, such that the outcome is determined to occur to a certain probability
and/or confidence.
(30) The model generation techniques described herein may also include at least one of gradient
boosting techniques and/or hyperparameter tuning. Gradient boosting may include, for example,
producing a prediction model in the form of an ensemble of weak prediction models, which may be
decision trees. The prediction model may be built in a stage-wise fashion and may allow optimization
of an arbitrary differential loss function. Hyperparameter tuning may include optimization of
hyperparameters during a training process. For example, a training data set may be run through the
model and the aggregate accuracy of the model may be evaluated. Hyperparameters may be tuned
based on the aggregate accuracy.
(31) The training component 124 may be configured to train the one or more models generated by the
model generator 122. Training of the models may include, for example, selecting a training population
on which to train the model(s). Selection of the training population may include selecting data
associated with a group of individuals and/or selecting data associated with a given period of time.
Features associated with the data may be identified and/or selected and may be utilized as inputs to
the model(s). Training of models may also include and/or be described herein as model fitting. Model
fitting may be based at least in part on historical data. For example, once an event is selected,
historical data indicating what data and/or data types may be relevant to predicting the event may be
utilized. The models may be modified and/or otherwise configured, based at least in part on the
historical data, to utilize the data and/or data types relevant to the event to predict the event. Model
fitting may include techniques such as linear regression and/or nonlinear regression.
(32) Additionally, or alternatively, while the unfolded records may be independent of each other, an
association between records belonging to the same person may be identified and utilized when
performing model testing. In these examples, a group of records belonging to a first set of people may
be used for training purposes while another group of records belonging to a second set of people may
be used for testing purposes. By maintaining an association between records belonging to the same
person, records may be split into the two groups without introducing leakage and/or overlap to the
testing process.
(33) Additionally, or alternatively, the training component 124 may identify input features that increase
the accuracy of the prediction models and other input features that decrease the accuracy of the
prediction models or have no or little effect on the prediction models. The prediction models may be
refit to utilize the features that increase accuracy while refraining from utilizing the features that
decrease accuracy or have no or little effect on accuracy.
(34) The event probability component 126 may be configured to determine an event probability using
the model(s) generated by the model generator 122. For example, for a given person associated with
the system 108, data associated with the person may be run through the model(s) to determine the
probability that a selected event will occur with respect to that person. Using the example provided
herein, the probability that a specific person will be hospitalized may be predicted by the system 108.
Data associated with the person may be run through the model(s) to determine a probability that the
person will be hospitalized, such as during a certain period of time. The model(s) may utilize data
available from the data storage 118, for example, to predict the event probability. This process may be
repeated for one or more other people associated with the system 108.
(35) The ranking component 128 may be configured to utilize the data generated by the event
probability component 126 to provide a ranking. For example, a group of 100 people may be analyzed
using the model(s) generated by the model generator 122. The event probability component 126 may
determine, for some or all of the people, a probability that the event (e.g., hospitalization) will occur.
Those people may be ranked by the ranking component 128 based at least in part on the determined
probability.
(36) The recommendation component 130 may be configured to generate recommendations,

intervention schedules, incentive program details, and/or to admit or not admit people to a given
program based at least in part on the ranking produced by the ranking component 128. By way of
example, the recommendation component 130 may generate recommendations associated with
information to be provided to individuals that are likely, such as based on the event probability for
those individuals being above a threshold event probability threshold, to be hospitalized. The provided
information may include a recommendation for those individuals to schedule a health screening with
their physician or other medical professional. The provided information may also indicate a healthcare
provider that may be suited to assist in such a screening. The provided information may also provide
physician-approved actions for lowering the probability of being hospitalized and, in examples, an
indication of how much the actions are likely to reduce the risk of hospitalization. The intervention
schedules may include treatment programs and/or regimens, which may be physician-approved, to
reduce the risk of hospitalization. The admission of a person to a program may include, for example,
admission to a complex care program associated with individuals having certain healthcare conditions
and/or propensities. It should be understood that while the examples of recommendations, schedules,
and admission decisions are provided herein, they are provided by way of illustration unless otherwise
noted and/or claimed. This disclosure includes the identification, determination, and/or generation of
recommendations of any and all types, particularly those relevant to and/or associated with the event
in question and/or the data being analyzed, and/or the model(s) being generated.
(37) The validation component 132 may be configured to validate the one or more models using one
or more validation techniques. For example, an area-under-the-resource-operating-characteristic-
curve measurement may be utilized to gauge the accuracy of the model generated utilizing the
unfolded data. In examples, the measurement may be at least 0.75, at least 0.76, at least 0.77, at
least 0.78, at least 0.79, at least 0.80, at least 0.81, at least 0.82, at least 0.83, at least 0.84, at least
0.85, at least 0.86, at least 0.87, at least 0.88, at least 0.89, or at least 0.90, indicating a high degree
of accuracy. Additionally, or alternatively, precision-recall characteristic measurements may be utilized
to determine precision of the prediction models.
(38) As used herein, a processor, such as processor(s) 112, may include multiple processors and/or a
processor having multiple cores. Further, the processors may comprise one or more cores of different
types. For example, the processors may include application processor units, graphic processing units,
and so forth. In one implementation, the processor may comprise a microcontroller and/or a
microprocessor. The processor(s) 112 may include a graphics processing unit (GPU), a
microprocessor, a digital signal processor or other processing units or components known in the art.
Alternatively, or in addition, the functionally described herein can be performed, at least in part, by one
or more hardware logic components. For example, and without limitation, illustrative types of hardware
logic components that can be used include field-programmable gate arrays (FPGAs), application-
specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip
systems (SOCs), complex programmable logic devices (CPLDs), etc. Additionally, each of the
processor(s) 112 may possess its own local memory, which also may store program components,
program data, and/or one or more operating systems.
(39) The memory 116 may include volatile and nonvolatile memory, removable and non-removable
media implemented in any method or technology for storage of information, such as computer-
readable instructions, data structures, program component, or other data. Such memory 116 includes,
but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM,
digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk
storage or other magnetic storage devices, RAID storage systems, or any other medium which can be
used to store the desired information and which can be accessed by a computing device. The memory
116 may be implemented as computer-readable storage media (“CRSM”), which may be any available
physical media accessible by the processor(s) 112 to execute instructions stored on the memory 116.
In one basic implementation, CRSM may include random access memory (“RAM”) and Flash memory.
In other implementations, CRSM may include, but is not limited to, read-only memory (“ROM”),
electrically erasable programmable read-only memory (“EEPROM”), or any other tangible medium
which can be used to store the desired information and which can be accessed by the processor(s).
(40) Further, functional components may be stored in the respective memories, or the same
functionality may alternatively be implemented in hardware, firmware, application specific integrated
circuits, field programmable gate arrays, or as a system on a chip (SoC). In addition, while not
illustrated, each respective memory, such as memory 116, discussed herein may include at least one
operating system (OS) component that is configured to manage hardware resource devices such as
the network interface(s), the I/O devices of the respective apparatuses, and so forth, and provide
various services to applications or components executing on the processors. Such OS component
may implement a variant of the FreeBSD® operating system as promulgated by the FreeBSD®
Project; other UNIX® or UNIX®-like variants; a variation of the Linux® operating system as
promulgated by Linus Torvalds; the FireOS® operating system from Amazon.com Inc. of Seattle,
Wash., USA; the Windows® operating system from Microsoft® Corporation of Redmond, Wash., USA;
LynxOS® as promulgated by Lynx Software Technologies, Inc. of San Jose, Calif.; Operating System
Embedded (Enea® OSE) as promulgated by ENEA AB of Sweden; and so forth.
(41) The network interface(s) 114 may enable communications between the components and/or
devices shown in environment 100 and/or with one or more other remote systems, as well as other
networked devices. Such network interface(s) 114 may include one or more network interface
controllers (NICs) or other types of transceiver devices to send and receive communications over the
network 110.
(42) For instance, each of the network interface(s) 114 may include a personal area network (PAN)
component to enable communications over one or more short-range wireless communication
channels. For instance, the PAN component may enable communications compliant with at least one
of the following standards IEEE 802.15.4 (ZigBee), IEEE 802.15.1 (Bluetooth), IEEE 802.11 (WiFi), or
any other PAN communication protocol. Furthermore, each of the network interface(s) 114 may
include a wide area network (WAN) component to enable communication over a wide area network.
(43) FIG. 2 illustrates a conceptual diagram of components and/or operations associated with a record
generator 120 for data unfolding. The record generator 120 may utilize data from one or more sources,
such as the data storage 118 to evaluate data folds and generate records, such as to increase the
quantity and/or quality of data available to the data storage 118.
(44) The record generator 120 may be configured to perform the folding and/or unfolding processes
described herein. For example, the record generator 120 may be configured to identify, determine,
and/or receive an indication of a time interval 202-210 at which to generate records. Keeping with the
example used herein, the time interval 202-210 may be, for example, 28 days. Other time intervals
may be utilized. The record generator 120 may also be configured to identify, determine, and/or
receive an indication of a period of time 212 during which to generate records. For example, a
particular time frame 212 may be utilized to generate the records and/or an amount of time from a
certain date may be utilized. For example, the record generator 120 may identify a one-year period of
time during which to generate records. Alternatively, the record generator 120 may identify a date,
such as Jan. 1, 2017, from which to start record generation either moving forward or backward in time.
The record generator 120 may be further configured to identify, determine, and/or receive an indication
of the data sources from which to utilize data. For example, the record generator 120 may identify,
determine, and/or receive an indication that the data storage 118 is to be utilized for receiving data
from.
indication of an event to analyze with respect to the data. For example, the record generator 120 may
generator 120 may then determine, for each time interval during the identified time period, whether the
event occurred. For example, the system may determine that for a first time interval 202, illustrated as
from t=0 to t=1, a particular person was not hospitalized. This is illustrated as a “0” with respect to the
first time interval 202. The system may determine that for a second time interval 204, illustrated as
from t=1 to t=2, the person was not hospitalized. The system may determine that for a third time
interval 206, illustrated as from t=2 to t=3, the person was not hospitalized. The system may determine
that for a fourth time interval 208, illustrated as from t=3 to t=4, the person was hospitalized. This is
illustrated as a “1” with respect to the fourth time interval 208. The system may also determine that for
a fifth time interval 210, illustrated as from t=4 to t=5, the person was not hospitalized. This process
may be performed for all or some of the people on which data is available to the system. In examples,
the system may identify, determine, and/or receive an indication of the people, the number of people,
five records for the person in question. For the example person illustrated herein, the first record may
include data associated with the person (e.g., blood pressure, age, weight, etc.) for the first time
interval 202 and an indication that the person was not hospitalized. The second record may include
data associated with the person for the second time interval 204 and an indication that the person was
not hospitalized. The third record may include data associated with the person for the third time
interval 206 and an indication that the person was not hospitalized. The fourth record may include data
associated with the person for the fourth time interval 208 and an indication that the person was
hospitalized. The fifth record may include data associated with the person for the fifth time interval 210
and an indication that the person was not hospitalized.
person was at some point hospitalized, may be utilized to generate five independent records with data
separated from each other as independent records for use by the system as described herein.
Extrapolating from the example provided herein, the system may have access to data for 100,000
people, which amounts to 100,000 records with one record for each person. Utilizing the data folding
and unfolding techniques described herein, such as using 5 time intervals during the folding
processes, 500,000 records may be generated and then utilized by the system.
(48) Additionally, or alternatively, a testing period 214 may be identified, determined, and/or indicated.
As illustrated with respect to FIG. 2, the testing period 214 may be from t=5 to t=present, for example.
The predictive model(s), as described more fully herein with respect to FIG. 1, may be generated
based on the data and/or records available during the first period 212, which may be described as a
development period during which the model(s) are developed. By way of example, a portion of the
records associated with the first period 212 may be utilized to train the predictive model(s). Thereafter,
the model(s) may be further trained and/or tested during a second period 214, which may be
described as a back-testing period. During the back-testing period, data not associated with data
utilized to develop the model(s) may be utilized to train and/or test the model(s). It should also be
noted that while the back-testing period 214 is depicted as utilizing data from t=5 to t=present, the
utilized data may include data from t=5 to a time that is less than t=present.
(49) FIG. 3 illustrates a conceptual diagram of example ranked data entries and uses of the same. For
example, the environment 300 may include a ranking 302. The ranking 302 may include one or more
entity entries 304-314. The entity entries 304-314 may correspond to people, entities, objects, and/or
other identifiers of entities associated with a system such as that disclosed herein. Using FIG. 3 as an
example, the entity entries 304-314 may correspond to people whose data has been acquired by a
system. The folding and unfolding techniques described herein may be performed on the data
associated with the people, and one or more prediction models may be identified and/or generated to
predict a likelihood that an event will occur. Sticking with the example provided above, the event may
be hospitalization. As shown in FIG. 3, the likelihood that the people associated with the six entity
entries 304-314 will be hospitalized in a given period of time is provided by way of percentages. It
should be understood that the likelihood of an event occurring may be expressed in terms other than,
or in addition to, percentages. For example, the likelihood may be presented as a scale, a coefficient,
and/or a narrative.
(50) As shown in FIG. 3, the first entity entry 304 corresponds to First Person, and the model(s)
described herein have been utilized to determine that First Person has a 32% likelihood of being
hospitalized within a given period of time. The second entity entry 306 corresponds to Second Person,
and the model(s) described herein have been utilized to determine that Second Person has a 29%
likelihood of being hospitalized within a given period of time. The third entity entry 308 corresponds to
Third Person, and the model(s) described herein have been utilized to determine that Third Person
has a 19% likelihood of being hospitalized within a given period of time. The fourth entity entry 310
corresponds to Fourth Person, and the model(s) described herein have been utilized to determine that
Fourth Person has a 15% likelihood of being hospitalized within a given period of time. The fifth entity
entry 312 corresponds to Fifth Person, and the model(s) described herein have been utilized to
determine that Fifth Person has a 4% likelihood of being hospitalized within a given period of time. The
sixth entity entry 314 corresponds to Sixth Person, and the model(s) described herein have been
utilized to determine that Sixth Person has a 1% likelihood of being hospitalized within a given period
of time. It should be understood that while FIG. 3 depicts six entity entries 304-314, the ranking 302
may include one, two, three, four, five, six, or more than six entity entries 304-314.
(51) Once the entities are ranked, such as via the ranking 302, the entities may be grouped. For
example, one or more thresholds may be identified, determined, and/or generated based at least in
part on information such as historical information, budgetary information, system resources, policies,
procedures, and/or government statutes and/or regulations. The thresholds may be informed by the
action to be taken in light of the event being predicted. For example, when the event being predicted is
hospitalization, the system may be configured to attempt to minimize hospitalizations, and in doing so,
may be programmed to indicate which group(s) of people are most at risk for being hospitalized so
that interventions may be put in place to mitigate such risk. For example, a first threshold may be set
at 20% probability that a person will be hospitalized. Entities that are associated with a 20% or higher
hospitalization probability may be associated with a first group 316, which may correspond to high-risk
individuals. Entities that are associated with between a 10% hospitalization threshold and a 20%
hospitalization threshold may be associated with a second group 320, which may correspond to
medium-risk individuals. Entities that are associated with less than a 10% hospitalization threshold
may be associated with a third group 324 which may correspond to low-risk individuals. It should be
understood that while three groups are depicted in FIG. 3, the groupings may include one, two, three,
or more than three groupings. It should also be understood that while three thresholds and/or
threshold ranges are provided herein, the groupings may be based on one, two, three, or more than
three thresholds and/or threshold ranges.
(52) Some or all of the groupings may be associated with a given recommendation and/or intervention.
For example, a recommendation and/or intervention may be identified, determined, and/or generated
based at least in part on a particular grouping. Using FIG. 3 as an example, Recommendation and/or
Intervention A 318 may be associated with the first group of individuals 316. Given that the first group
316 corresponds to a high-risk group, the Recommendation and/or Intervention A 318 may correspond
to a recommendation that is more severe and/or dedicates a large amount of resources, for example,
than other recommendations. In the instance where the event is hospitalization, the recommendation
may include a dietary recommendation, a physical fitness recommendation, a recommendation to
schedule a checkup with a healthcare provider (as well as, in examples, information associated with
one or more healthcare providers), etc. The intervention may include an intervention schedule such as
changes to scheduled healthcare checkups, changes to medications, changes to delivery of dietary
products and/or foods, etc. Continuing with the example above, Recommendation and/or Intervention
B 322 may be associated with the second group of individuals 320. Given that the second group 320
corresponds to a medium-risk group, the Recommendation and/or Intervention B 322 may correspond
to a recommendation that is less severe and/or dedicates less resources, for example, than
Recommendation and/or Intervention A 318. Recommendation and/or Intervention C 326 may be
associated with the third group of individuals 324. Given that the third group 324 corresponds to a low-
risk group, the Recommendation and/or Intervention C 326 may correspond to a recommendation that
is less severe and/or dedicates less resources, for example, than Recommendation and/or
Intervention B 322.
(53) FIG. 4 illustrates a conceptual diagram of example ranked data entries and further uses of the
same. For example, the environment 400 may include a ranking 402. The ranking 402 may include
one or more entity entries 404-414. The entity entries 404-414 may correspond to people, entities,
objects, and/or other identifiers of entities associated with a system, such as described with respect to
FIG. 3. As shown in FIG. 4, the likelihood that people associated with the six entity entries 404-414 will
be hospitalized in a given period of time is provided by way of percentages. It should be understood
that the likelihood of an event occurring may be expressed in terms other than, or in addition to,
percentages. For example, the likelihood may be presented as a scale, a coefficient, and/or a
narrative.
(54) As shown in FIG. 4, the first entity entry 404 corresponds to First Person, and the model(s)
described herein have been utilized to determine that First Person has a 32% likelihood of being
hospitalized within a given period of time. The second entity entry 406 corresponds to Second Person,
and the model(s) described herein have been utilized to determine that Second Person has a 29%
likelihood of being hospitalized within a given period of time. The third entity entry 408 corresponds to
Third Person, and the model(s) described herein have been utilized to determine that Third Person
has a 19% likelihood of being hospitalized within a given period of time. The fourth entity entry 410
corresponds to Fourth Person, and the model(s) described herein have been utilized to determine that
Fourth Person has a 15% likelihood of being hospitalized within a given period of time. The fifth entity
entry 412 corresponds to Fifth Person, and the model(s) described herein have been utilized to
determine that Fifth Person has a 4% likelihood of being hospitalized within a given period of time. The
sixth entity entry 414 corresponds to Sixth Person, and the model(s) described herein have been
utilized to determine that Sixth Person has a 1% likelihood of being hospitalized within a given period
of time. It should be understood that while FIG. 4 depicts six entity entries 404-414, the ranking 402
may include one, two, three, four, five, six, or more than six entity entries 404-414.
(55) Once the entities are ranked, such as via the ranking 402, the entities may be grouped. As shown
in FIG. 4, the rankings 402 may be utilized to group individuals between two or more categories, here
depicted as “admitted” or “not admitted.” As such, the ranking 402 may be utilized to identify and/or
determine which people will be admitted to a program and which people will not be admitted to the
program. An admittance threshold may be identified, determined, and/or generated and people with
event probabilities meeting or exceeding the admittance threshold may be admitted, while people with
event probabilities not meeting or exceeding the admittance threshold may not be admitted.
(56) FIGS. 5 and 6 illustrate various processes for data folding and unfolding. The processes
described herein are illustrated as collections of blocks in logical flow diagrams, which represent a
sequence of operations, some or all of which may be implemented in hardware, software or a
combination thereof. In the context of software, the blocks may represent computer-executable
instructions stored on one or more computer-readable media that, when executed by one or more
processors, program the processors to perform the recited operations. Generally, computer-
executable instructions include routines, programs, objects, components, data structures and the like
that perform particular functions or implement particular data types. The order in which the blocks are
described should not be construed as a limitation, unless specifically noted. Any number of the
described blocks may be combined in any order and/or in parallel to implement the process, or
alternative processes, and not all of the blocks need be executed. For discussion purposes, the
processes are described with reference to the environments, architectures and systems described in
the examples herein, such as, for example those described with respect to FIGS. 1-4, although the
processes may be implemented in a wide variety of other environments, architectures and systems.
(57) FIG. 5 illustrates a flow diagram of an example process 500 for data folding and unfolding. The
order in which the operations or steps are described is not intended to be construed as a limitation,
and any number of the described operations may be combined in any order and/or in parallel to
implement process 500.
(58) At block 502, the process 500 may include generating a first record corresponding to a person,
the first record including: a first portion of health-related data of the person associated with a first
period of time; and a first indication of whether the person was hospitalized during the first period of
time. For example, a record generator may be configured to perform folding and/or unfolding
processes, as described herein. For example, the record generator may be configured to identify,
determine, and/or receive an indication of a time interval at which to generate records. The record
generator may also be configured to identify, determine, and/or receive an indication of a period of
time during which to generate records. For example, a particular time frame may be utilized to
generate the records and/or an amount of time from a certain date may be utilized.
(59) The record generator may be further configured to identify, determine, and/or receive an
indication of the data sources from which to utilize data. For example, the record generator may
identify, determine, and/or receive an indication that a data storage component is to be utilized for
receiving data from. Additionally, or alternatively, the record generator may identify, determine, and/or
receive an indication that one or more other data sources are to be utilized in addition to or
alternatively to the data storage.
indication of an event to analyze with respect to the data. For example, the record generator may
generator may then determine, for each time interval during the identified time period, whether the
event occurred. For example, the system may determine that for a first time interval the person was
not hospitalized, and for a second time interval the person was hospitalized. This process may be
performed for all or some of the people on which data is available to the system. In examples, the
system may identify, determine, and/or receive an indication of the people, the number of people,
two records for each person. For the example person illustrated herein, the first record may include
with the person for the second period of time and an indication that the person was hospitalized.
person was at some point hospitalized, may be utilized to generate two independent records with data
(63) At block 504, the process 500 may include generating a second record corresponding to the
person, the second record including: a second portion of the health-related data associated with a
second period of time; and a second indication of whether the person was hospitalized during the
second period of time. Generation of the second record may be performed in the same or a similar
manner as generation of the first record, described with respect to block 502.
(64) At block 506, the process 500 may include generating a model configured to determine a
hospitalization probability, the model fit based at least in part on the first record and the second record.
For example, a model generator may be configured to generate one or more models based at least in
part on the records generated by the record generator and/or other data available to the system. The
more robust data set generated by the folding and unfolding techniques described herein may
increase the data sampling available to the models, which may allow those models to be trained
and/or fit more accurately. The models may be generated utilizing, for example, one or more machine
learning techniques where features associated with the health-related data are inputs to the model.
The models may be utilized to determine a probability of an outcome occurring, such as for people
associated with the system. These models may be described herein as predictive models and/or
predictive layers. A predictive model may include one or more models that utilize predictive analytics
to predict one or more outcomes. Predictive analytic techniques may include, for example, predictive
modelling, machine learning, and/or data mining, as described more fully above.
(65) Information from stored and/or accessible data may be extracted from one or more databases
and may be utilized to predict trends and behavior patterns. In examples, the event, otherwise
described herein as an outcome, may be an event that will occur in the future, such as whether a
member will be hospitalized, a likelihood of a member being diagnosed with diabetes, a likelihood of a
member missing a medication prescription fill at a pharmacy, etc. The predictive analytic techniques
may be utilized to determine associations and/or relationships between explanatory variables and
predicted variables from past occurrences and utilizing these variables to predict the unknown
outcome. The predictive analytic techniques may include defining the outcome and data sets used to
predict the outcome. Then, data may be collected and/or accessed to be used for analysis.
models, as described more fully herein. Thereafter predictive modelling may be performed to generate
and/or confidence.
(68) A validation component may be configured to validate the one or more models using one or more
validation techniques. For example, an area-under-the-resource-operating-characteristic-curve
measurement may be utilized to gauge the accuracy of the model generated utilizing the unfolded
data. In examples, the measurement may be at least 0.75, at least 0.76, at least 0.77, at least 0.78, at
0.86, at least 0.87, at least 0.88, at least 0.89, or at least 0.90, indicating a high degree of accuracy.
Additionally, or alternatively, precision-recall characteristic measurements may be utilized to determine
precision of the prediction models.
(69) At block 508, the process 500 may include determining, based at least in part on the model, the
hospitalization probability associated with people on which health-related data is available to the
system. For example, an event probability component may be configured to determine an event
probability using the model(s) generated by the model generator. For example, for a given person
associated with the system, data associated with the person may be run through the model(s) to
determine the probability that a selected event will occur with respect to that person. Using the
example provided herein, the probability that a specific person will be hospitalized may be predicted by
the system. Data associated with the person may be run through the model(s) to determine a
probability that the person will be hospitalized, such as during a certain period of time. The model(s)
may utilize data available from the data storage, for example, to predict the event probability. This
process may be repeated for one or more other people associated with the system.
(70) At block 510, the process 500 may include ranking the people based at least in part on the
hospitalization probability. For example, the ranking component may be configured to utilize the data
generated by the event probability component to provide a ranking. For example, a group of 100
people may be analyzed using the model(s) generated by the model generator. The event probability
component may determine, for some or all of the people, a probability that the event (e.g.,
hospitalization) will occur. Those people may be ranked by the ranking component based at least in
part on the determined probability.
(71) The process 500 may additionally, or alternatively, include identifying a hospitalization probability
threshold and identifying a portion of the people with hospitalization probabilities at or above the
hospitalization probability threshold. The process 500 may also include generating an intervention
recommendation for the portion of the people based at least in part on the hospitalization probabilities
being at or above the hospitalization probability threshold.
(72) The process 500 may additionally, or alternatively, include determining a difference between (1)
the hospitalization probability determined by the model with respect to a portion of the people and (2)
hospitalization events occurring with respect to the portion of the people. The process 500 may also
include generating a modified model based at least in part on the difference.
(73) FIG. 6 illustrates a flow diagram of an example process 600 for data folding and unfolding. The
order in which the operations or steps are described is not intended to be construed as a limitation,
and any number of the described operations may be combined in any order and/or in parallel to
implement process 600.
(74) At block 602, the process 600 may include generating a first record corresponding to a person,
the first record including: a first portion of data associated with a first period of time; and a first
indication of whether an event occurred with respect to the person during the first period of time. For
example, a record generator may be configured to perform folding and/or unfolding processes, as
described herein. For example, the record generator may be configured to identify, determine, and/or
receive an indication of a time interval at which to generate records. The record generator may also be
configured to identify, determine, and/or receive an indication of a period of time during which to
generate records. For example, a particular time frame may be utilized to generate the records and/or
an amount of time from a certain date may be utilized.
indication of the data sources from which to utilize data. For example, the record generator may
identify, determine, and/or receive an indication that a data storage component is to be utilized for
receiving data from. Additionally, or alternatively, the record generator may identify, determine, and/or
receive an indication that one or more other data sources are to be utilized in addition to or
alternatively to the data storage.
indication of an event to analyze with respect to the data. For example, the record generator may
identify, determine, and/or receive an indication that an event of interest, such as hospitalization. The
record generator may then determine, for each time interval during the identified time period, whether
the event occurred. For example, the system may determine that for a first time interval the person
was not hospitalized, and for a second time interval the person was hospitalized. This process may be
performed for all or some of the people on which data is available to the system. In examples, the
system may identify, determine, and/or receive an indication of the people, the number of people,
two records for each person. For the example person illustrated herein, the first record may include
with the person for the second period of time and an indication that the person was hospitalized.
person was at some point hospitalized, may be utilized to generate two independent records with data
(79) At block 604, the process 600 may include generating a second record corresponding to the
person, the second record including: a second portion of the data associated with a second period of
time; and a second indication of whether the event occurred with respect to the person during the
second period of time. Generation of the second record may be performed in the same or a similar
manner as generation of the first record, described with respect to block 602.
(80) At block 606, the process 600 may include generating a model configured to determine an event
probability, the model fit based at least in part on the first record and the second record. For example,
a model generator may be configured to generate one or more models based at least in part on the
records generated by the record generator and/or other data available to the system. The more robust
data set generated by the folding and unfolding techniques described herein may increase the data
sampling available to the models, which may allow those models to be trained and/or fit more
accurately. The models may be generated utilizing, for example, one or more machine learning
techniques where features associated with the health-related data are inputs to the model. The models
may be utilized to determine a probability of an outcome occurring, such as for people associated with
the system. These models may be described herein as predictive models and/or predictive layers. A
predictive model may include one or more models that utilize predictive analytics to predict one or
more outcomes. Predictive analytic techniques may include, for example, predictive modelling,
machine learning, and/or data mining, as described more fully above.
(81) Information from stored and/or accessible data may be extracted from one or more databases
and may be utilized to predict trends and behavior patterns. In examples, the event, otherwise
described herein as an outcome, may be an event that will occur in the future, such as whether a
member will be hospitalized, a likelihood of a member being diagnosed with diabetes, a likelihood of a
member missing a medication prescription fill at a pharmacy, etc. The predictive analytic techniques
may be utilized to determine associations and/or relationships between explanatory variables and
predicted variables from past occurrences and utilizing these variables to predict the unknown
outcome. The predictive analytic techniques may include defining the outcome and data sets used to
predict the outcome. Then, data may be collected and/or accessed to be used for analysis.
models, as described more fully herein. Thereafter predictive modelling may be performed to generate
and/or confidence.
(84) A validation component may be configured to validate the one or more models using one or more
validation techniques. For example, an area-under-the-resource-operating-characteristic-curve
measurement may be utilized to gauge the accuracy of the model generated utilizing the unfolded
data. In examples, the measurement may be at least 0.75, at least 0.76, at least 0.77, at least 0.78, at
0.86, at least 0.87, at least 0.88, at least 0.89, or at least 0.90, indicating a high degree of accuracy.
Additionally, or alternatively, precision-recall characteristic measurements may be utilized to determine
precision of the prediction models.
(85) At block 608, the process 600 may include determining, based at least in part on the model, the
event probability associated with a group of people. For example, an event probability component may
be configured to determine an event probability using the model(s) generated by the model generator.
For example, for a given person associated with the system, data associated with the person may be
run through the model(s) to determine the probability that a selected event will occur with respect to
that person. Using the example provided herein, the probability that a specific person will be
hospitalized may be predicted by the system. Data associated with the person may be run through the
model(s) to determine a probability that the person will be hospitalized, such as during a certain period
of time. The model(s) may utilize data available from the data storage, for example, to predict the
event probability. This process may be repeated for one or more other people associated with the
system.
(86) At block 610, the process 600 may include ranking individual people in the group of people based
at least in part on the event probability. For example, the ranking component may be configured to
utilize the data generated by the event probability component to provide a ranking. For example, a
group of 100 people may be analyzed using the model(s) generated by the model generator. The
event probability component may determine, for some or all of the people, a probability that the event
(e.g., hospitalization) will occur. Those people may be ranked by the ranking component based at least
in part on the determined probability.
(87) The process 600 may additionally, or alternatively, include identifying an event probability
threshold and identifying a portion of the group of people with event probabilities at or above the event
probability threshold. The process 600 may also include generating a recommendation for the portion
of the group of people based at least in part on the event probabilities being at or above the event
probability threshold.
(88) The process 600 may additionally, or alternatively, include determining a difference between (1)
the event probability determined by the model with respect to a portion of the group of people and (2)
event occurrences with respect to the portion of the group of people. The process 600 may also
include generating a modified model based at least in part on the difference.
(89) The process 600 may additionally, or alternatively, include generating the model based at least in
part on identifying a first group of input features that, when utilized by the model, increase an accuracy
of the model and identifying a second group of the input features that, when utilized by the model,
decrease the accuracy of the model or maintain the accuracy of the model. The process 600 may also
include refitting the model to utilize the first group of the input features and refrain from utilizing the
second group of the input features.
(90) The process 600 may additionally, or alternatively, include identifying an event probability
threshold and identifying a portion of the group of people with event probabilities at or above the event
probability threshold. The process 600 may also include causing the portion of the group of people to
be accepted into a program based at least in part on the event probabilities being at or above the
event probability threshold.
(91) While the foregoing invention is described with respect to the specific examples, it is to be
understood that the scope of the invention is not limited to these specific examples. Since other
modifications and changes varied to fit particular operating requirements and environments will be
apparent to those skilled in the art, the invention is not considered limited to the example chosen for
purposes of disclosure, and covers all changes and modifications which do not constitute departures
from the true spirit and scope of this invention.
(92) Although the application describes embodiments having specific structural features and/or
methodological acts, it is to be understood that the claims are not necessarily limited to the specific
features or acts described. Rather, the specific features and acts are merely illustrative some
embodiments that fall within the scope of the claims of the application.
Claims
1. A system comprising: one or more processors; and computer-readable media storing first computer-
executable instructions that, when executed by the one or more processors, cause the one or more
processors to perform operations comprising: receiving, from a database and at a first time, health-
related data of a person; determining a time interval for splitting the health-related data into multiple
records, wherein each of the multiple records corresponds to a second time that occurred before the
first time; generating, utilizing for a first instance of the time interval, a first record representing: a first
portion of the health-related data attributable to the first instance of the time interval; and a first
indication of whether the person was hospitalized during the first instance of the time interval;
generating, utilizing a second instance of the time interval, a second record representing: a second
portion of the health-related data attributable to the second instance of the time interval; and a second
indication of whether the person was hospitalized during the second instance of the time interval;
dissociating the first record from the second record such that the first record and the second record
are indicated as being independent of each other; storing the first record and the second record such
that the first record and the second record are indicated as being dissociated; generating a machine-
learning model configured to determine a hospitalization probability; generating a training dataset
configured to train the machine-learning model, the training dataset including the dissociation of the
first record and the second record, the health-related data, the first instance of the time interval, the
second first instance of the time interval, the first record and, the second record; training the machine-
learning model utilizing the training dataset to identify trends between the dissociation of the first
record and the second record, the health related data, the first instance of the time interval, the second
first instance of the time interval, the first record, and the second record to determine if a
hospitalization probability of a patient is improved; determining, based at least in part on the trained
machine-learning model, the hospitalization probability associated with people on which health-related
data is available to the system; and ranking the people based at least in part on the hospitalization
probability and utilizing the trained machine-learning model.
2. The system of claim 1, the operations further comprising: identifying a hospitalization probability
threshold; identifying a portion of the people with hospitalization probabilities at or above the
hospitalization probability threshold; and generating an intervention recommendation for the portion of
the people based at least in part on the hospitalization probabilities being at or above the
hospitalization probability threshold.
3. The system of claim 1, wherein generating the machine-learning model comprises generating the
machine-learning model based at least in part on features associated with the health-related data that
inform the hospitalization probability.
4. The system of claim 1, the operations further comprising: determining a difference between (1) the
hospitalization probability determined by the machine-learning model with respect to a portion of the
people and (2) hospitalization events occurring with respect to the portion of the people; and wherein
training the machine-learning model comprises training the machine-learning model based at least in
part on the difference.
5. A method, comprising: receiving, from a database and at a first time, data of a person; determining
a time interval for splitting the data into multiple records, wherein each of the multiple records
corresponds to a second time that occurred before the first time; generating, utilizing a first instance of
the time interval, a first record including: a first portion of the data attributable to the first instance of
the time interval; and a first indication of whether an event occurred with respect to the person during
the first instance of the time interval; generating, utilizing a second instance of the time interval, a
second record including: a second portion of the data attributable to the second instance of the time
interval; and a second indication of whether the event occurred with respect to the person during the
second instance of the time interval; dissociating the first record from the second record such that the
first record and the second record are indicated as being independent of each other; storing the first
record and the second record such that the first record and the second record are indicated as being
dissociated; generating a machine-learning model configured to determine an event probability;
generating a training dataset configured to train the machine-learning model, the training dataset
based at least in part on the dissociation of the first record and the second record, the health-related
data, the first instance of the time interval, the second first instance of the time interval, the first record,
and the second record; training the machine-learning model utilizing the training dataset to identify
trends between the dissociation of the first record and the second record, the health related data, the
first instance of the time interval, the second first instance of the time interval, the first record, and the
second record to determine if a hospitalization probability of a patient is improved; determining, based
at least in part on the trained machine-learning model, the event probability associated with a group of
people; and ranking, utilizing the trained machine-learning model, individual people in the group of
people based at least in part on the event probability.
6. The method of claim 5, further comprising: identifying an event probability threshold; identifying a
portion of the group of people with event probabilities at or above the event probability threshold; and
generating a recommendation for the portion of the group of people based at least in part on the event
probabilities being at or above the event probability threshold.
7. The method of claim 5, wherein generating the machine-learning model comprises generating the
machine-learning model based at least in part on features associated with the data that inform the
event probability.
8. The method of claim 5, further comprising: determining a difference between (1) the event
probability determined by the machine-learning model with respect to a portion of the group of people
and (2) event occurrences with respect to the portion of the group of people; and wherein training the
machine-learning model comprises training the machine-learning model based at least in part on the
difference.
9. The method of claim 5, wherein the machine-learning model is associated with an area-under-the-
resource-operation-curve value of at least 0.81.
10. The method of claim 5, wherein generating the machine-learning model comprises generating the
machine-learning model based at least in part on: gradient boosting techniques; and hyperparameter
tuning of nodes associated with the event.
11. The method of claim 5, wherein input features are associated with the data, and wherein
generating the machine-learning model comprises generating the machine-learning model based at
least in part on: identifying a first group of the input features that, when utilized by the machine-
learning model, increase an accuracy of the machine-learning model; identifying a second group of the
input features that, when utilized by the machine-learning model, decrease the accuracy of the
machine-learning model or maintain the accuracy of the machine-learning model; and refitting the
machine-learning model to utilize the first group of the input features and refrain from utilizing the
12. The method of claim 5, further comprising: identifying an event probability threshold; identifying a
portion of the group of people with event probabilities at or above the event probability threshold; and
causing the portion of the group of people to be accepted into a program based at least in part on the
event probabilities being at or above the event probability threshold.
13. A system, comprising: one or more processors; and computer-readable media storing first
computer-executable instructions that, when executed by the one or more processors, cause the one
or more processors to perform operations comprising: receiving, from a database and at a first time,
health-related data of a person; determining a time interval for splitting the health-related data into
multiple records, wherein each of the multiple records corresponds to a second time that occurred
before the first time; generating, utilizing a first instance of the time interval, a first record
corresponding to the person, the first record including a first indication of whether an event occurred
with respect to the person during a first instance of the time interval; generating, utilizing a second
instance of the time interval, a second record corresponding to the person, the second record
including a second indication of whether the event occurred with respect to the person during a
second instance of the time interval; dissociating the first record from the second record such that the
first record and the second record are indicated as being independent of each other; storing the first
record and the second record such that the first record and the second record are indicated as being
dissociated; generating a machine-learning model configured to determine an event probability;
generating a training dataset configured to train the machine-learning model, the training dataset
based at least in part on the dissociation of the first record and the second record, the health-related
data, the first instance of the time interval, the second first instance of the time interval, the first record,
and the second record; training the machine-learning model utilizing the training dataset to identify
trends between the dissociation of the first record and the second record, the health related data, the
first instance of the time interval, the second first instance of the time interval, the first record, and the
second record to determine if a hospitalization probability of a patient is improved; determining, based
at least in part on the trained machine-learning model, the event probability associated with a group of
people; and ranking individual people in the group of people based at least in part on the event
probability and utilizing the trained machine-learning-model.
14. The system of claim 13, the operations further comprising: identifying an event probability
threshold; identifying a portion of the group of people with event probabilities at or above the event
probability threshold; and generating a recommendation for the portion of the group of people based at
least in part on the event probabilities being at or above the event probability threshold.
machine-learning model based at least in part on one or more machine learning techniques indicating
features associated with first data associated with the person that inform the event probability.
16. The system of claim 13, the operations further comprising: determining a difference between (1)
the event probability determined by the machine-learning model with respect to a portion of the group
of people and (2) event occurrences with respect to the portion of the group of people; and wherein
training the machine-learning model comprises training the machine-learning model based at least in
part on the difference.
17. The system of claim 13, wherein the machine-learning model is associated with an area-under-the-
resource-operation-curve value of at least 0.81.
machine-learning model based at least in part on: gradient boosting techniques; and hyperparameter
tuning of nodes associated with the event.
19. The system of claim 13, wherein input features are associated with first data of the person, and
wherein generating the machine-learning model comprises generating the machine-learning model
based at least in part on: identifying a first group of the input features that, when utilized by the
machine-learning model, increase an accuracy of the machine-learning model; identifying a second
group of the input features that, when utilized by the machine-learning model, decrease the accuracy
of the machine-learning model or maintain the accuracy of the machine-learning model; and refitting
the machine-learning model to utilize the first group of the input features and refrain from utilizing the
20. The system of claim 13, the operations further comprising: identifying an event probability
threshold; identifying a portion of the group of people with event probabilities at or above the event
probability threshold; and causing the portion of the group of people to be accepted into a program
based at least in part on the event probabilities being at or above the event probability threshold.

Clover Health Patent

Uploaded by

Copyright:

Available Formats

Clover Health Patent

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Clover Health Patent

Uploaded by

Copyright:

Available Formats

Systems And Methods For Determining Event Probability

DOCUMENT ID DATE PUBLISHED

APPLICATION NO DATE FILED

ASSIGNEE NAME (USPAT, PGPUB)

BRIEF DESCRIPTION OF THE DRAWINGS

(36) The recommendation component 130 may be configured to generate recommendations,

You might also like