Invented by Praduman Jain, Josh Schilling, Dave Klein, Vignet Inc
The Vignet Inc invention works as followsMethods and systems for predicting community outcomes, including computer programs stored on computer-storage mediums. In certain implementations, monitoring information is received. This includes location tracking data indicating the locations visited by people in a particular community. Data about the community, including its characteristics and the geographic area associated with it, is provided. A predictive model is used to assess the potential of a region for disease transmission based on individual behavior patterns in the community. One or more models are trained on data that describes a variety of communities, behavior patterns, and disease outcomes for individuals over time.
Background for Using predictive models for disease onset prediction and selecting pharmaceuticals
Infectious disease such as COVID-19 is difficult to track and treat on a large-scale, particularly when it occurs in large numbers. In some cases, traditional techniques of detecting infection and administering treatment may not be able to provide an effective response that is fast enough or personalized in order to detect and treat the disease. The factors that influence the spread of disease and its impact can vary from one location to another. Different communities have different risk profiles, and therefore need different strategies for preventing, containing, or managing the spread of diseases.
In some implementations, the system can be used to monitor, predict and control the spread of COVID-19 or other diseases. The system can assess the needs and risks of each community to tailor disease-related measures, predictions and recommendations. In some cases, communities may be geographical regions. Fine-grained recommendations are made for specific areas, such as zip codes, counties, cities or neighborhoods. The system provides communities with tools that can help predict which areas are most likely to be disease hotspots based on the current behavior of the people in the community. The system can also predict future disease measures, such as hospitalizations and infections (e.g.) based on real-time monitoring information. Machine learning models can make these and other predictions using data from monitoring and outcomes for different communities.
In some cases, medical treatment devices provide one or multiple therapies to treat COVID-19 and COVID-19 symptoms. The medical device has features for data collection. These include measurement of physiological parameters and location tracking. Location tracking data can be used by the medical treatment device to determine when and if COVID-19 treatment should begin. The medical treatment device could, for example, measure the body temperature, heartbeat, respiration rate or any other parameter and determine if the user has likely contracted COVID-19 based on these parameters. The medical device can also determine, using location tracking data like GPS data, whether the device was in a region with heightened disease transmission potential of COVID-19. This makes infection by COVID-19 more likely for the user. Identification of one or more regions can be done using, for instance, location tracking data from other devices, or information about movement patterns within a community. The one or multiple regions can be generated by a medical treatment device or received from a server. The medical treatment device can automatically deliver one or several therapies based on physiological monitoring data, and a detection of an elevated disease transmission risk in one or more regions. The medical treatment device may automatically provide one or more therapies based on the physiological monitoring data and the detection of a location in an area with elevated disease transmission potential, e.g. by providing digital therapeutics interventions to promote behavior change or support specific functions (e.g. breathing, posture, exercise etc.).
The system can collect data on individuals continuously or regularly using digital platforms. It can also gather aggregated data about communities and their effects. These inputs can be used to generate predictions that are updated and to update predictive models. The system can respond to changes in conditions quickly and sometimes in real time, using data streams that show real-time behavior of individuals. The system can forecast future measures and trends related to disease, identify factors that contribute to these measures and trends and recommend actions for a more effective improvement in a community’s effort to contain and eradicate a disease.
The system’s predictions can also be used to customize care for individual patients. The system’s predictions about future trends and measures of disease in a particular community can be used for individualized disease management. If a system predicts an increase in disease prevalence or infection rate, it may, for instance, change the monitoring of the individual for disease detection (e.g. infection prediction), provide diagnostic testing, or begin or modify the treatment for that individual.
Additionally, interventions and treatment can be provided throughout a community. If a trip was planned for a large group of people, or a small family, or a business, the system could detect high disease risk and take action to manage it. The system can, for example, schedule appointments with doctors or lab tests, or check vaccination status, depending on the destination. The system can collect information on disease prevalence from different sources, like government agencies or public health organizations, in order to detect areas where COVID-19 outbreaks are present. These techniques can also be applied to other diseases such as measles or other infectious diseases. If a school field trip is being planned, the vaccine can be given out based on the risk level of a particular destination. Another example is a program for employees, where if certain employees travel, the entire office or business receives treatment or vaccinations to reduce susceptibility.
The system is able to predict hotspots for disease transmission with high precision. Rich data on geography and community behavior can be used to train machine learning models. Examples of data that can be used to train machine learning models include data about specific places, data on occupancy and traffic, map data and geographic relationships. Examples can include outcomes of disease at the individual and community level (e.g. hospitalizations, death, etc.). Monitoring data can include location tracking, which shows the locations visited by people, as well as the duration, times and activities. If contact tracing data are available, they can be used to determine when and where various interactions that lead to the transmission of disease occurred. Machine learning techniques can be used to train a model using this data. The model will incorporate relationships between travel patterns, location types and factors such as geographical factors, environmental factors, and geographic factors. The model can, for example, learn to identify relative transmission risk posed by various types of locations in combination with community characteristics or behavior. When the system evaluates a data set of community information that includes a location with these factors, the model will indicate a high-risk transmission risk.
This approach has significant advantages over other approaches. Conventional contact tracing, for example, is a tool that allows you to track the transmission chain of a particular disease between individuals who are in the same location or interact with one another. The conventional contact tracing of COVID-19 involves identifying those who have a positive diagnosis (e.g. actual disease cases), as well as their contacts. (e.g. anyone who has been within 6 feet of a positive case at least 15 minutes, starting 48 hours prior to the time the person became ill, until they are isolated).
However conventional contact tracing does have limitations.” The adoption of conventional contact tracking technology is limited and it takes time to deploy. Contract tracing, which tracks close contact between individuals, does not capture all the routes of COVID-19 transmission (e.g. inhaled drops, contaminated surfaces or aerosolized particles). Two people visiting the same place may not be detected by a typical contact trace, such as if they are outside of a threshold (e.g. 10 feet), or if they visit at different times. The disease could be transmitted through a contaminated surface, or another transmission method. The relatively high contact level required to detect two people in contact is another limitation of the traditional contact trace. A limitation of the technique is that people must be in the same location at the exact same time for contact to be registered. For example, they have to stay at the place for an overlapped period of time. These factors are useful but they may miss events which can still cause exposure to disease, for example, individuals passing through the same region, even if they do not enter close proximity or be at the same place at the exact same time. COVID-19 is spread through air and surfaces, possibly through respiratory droplets and aerosolized particles. More versatile tools are needed to detect these events.
The techniques described below allow for a variety features, including geofencing and contact tracking. These features allow for better communication, especially with those who have officially been diagnosed with COVID-19. They can identify individuals with whom they had close contacts (?close contact?) During the period of time that they were infectious, these features allow for better communication with individuals. This includes those who have been officially diagnosed with COVID-19. They can identify all individuals to whom they had close contact (?close contacts?) You can also communicate with people who have been diagnosed and are in close proximity to them. For example, family members, friends, someone living with them or random acquaintances. Communication can be given to people who visit the same place as another within a defined time period, like 16 hours or any other time frame. The exposure scores can be adjusted based on how much time has passed between visits by different people. The system can communicate with those who have observed high-risk areas. The system can provide individuals with appropriate treatment and guidance measures, such as self isolation, based on estimated exposure. The system can monitor individuals in order to keep them informed of COVID-19 risk identifiers and exposure rates. The system can provide notifications, alerts and community news about new tags that represent exposure events or significant risks.
The system allows the different influence of proximity and timing on an individual to be reflected by their risk level. The system can also incorporate other factors, such as traffic or population levels, types of business or activities for different locations. The system can then provide a more accurate assessment of the individual’s exposure level, or risk, allowing it to take more appropriate and timely disease management measures (e.g. disease prevention recommendations, tests, treatments, etc .).
In some implementations location tags that represent disease exposure events or risk can be based either on symptom reports, or infection likelihood prediction, instead or in addition to disease testing results. A location tag that represents a person’s visit can be assigned a score for disease transmission based on the symptoms reported by the user, physiological measurements tracked (e.g. with passively sensed information), detected behavioral changes, or other early indicators of infection, before a positive diagnosis is obtained. Machine learning models can be used to predict the likelihood of infection for a user. This allows a high level of confidence that a person has COVID-19 even before testing is done. These machine learning predictions, along with other data like behavior data, physiological data or self-reported symptoms, can be used to assign scores to the location tags. This allows the system to detect COVID-19 more accurately and recommends measures to slow its spread, even before widespread testing is available.
Another benefit of the system is its ability to use machine-learning to determine what types of events or contacts result in the transmission of disease, and then use that information to improve location tags over time, resulting in an increasingly accurate system.” Most conventional contact tracking systems use predefined contact measures, such as proximity distances and time overlaps, which are not always appropriate to the different types of interactions and locations that people visit. The current system is more flexible, as it allows individual tags (e.g. representing visits or events of individuals) to have geofences of different sizes and shapes, and location tags can have transmission scores that are different, and change over time. Different thresholds and contact definitions can be tailored, such as for different communities or types of locations. Due to the differences in ventilation, and other environmental factors that affect a location, it is possible for a virus to remain viable longer at a grocery shop than a departmental store, or any other type of location. The percentage of people who are vaccinated against a particular disease or who have recovered (and thus developed immunity) at a given location can also limit the risk of a transmission. The system can determine the parameters of transmission by analyzing the data on location tracking, disease outcomes, and other related data. The system can then adjust and customize the scoring of the exposure-tagged location and the areas, as well as the tag timings for these locations (e.g. including patterns of risk reduction based on elapsed times or social distance precautions) for better exposure estimations and risk measures.
In some implementations, the system can provide a technology platform which is used to select treatment and assess risk of a particular disease. The system can be tailored or personalized to each individual and used as a tool for ongoing risk assessment. The platform may include a data package that can be downloaded (e.g. software module or set configuration data) to manage the disease. This data package can then be used on a user device, such as a mobile phone. The data package may instruct the monitoring of a person, such as to obtain physiological data, sensor data or survey responses. These data can be sent to a server and processed by either the user device or the server. The system can use the data collected to estimate the level of exposure and likelihood of infection for a user. The system may also take additional measures to detect the disease. For example, it can change the monitoring procedures, (e.g. the types of data that are collected, how often data is collected, etc.). Or, the system can issue a physical test kit to detect the disease. The system can select and recommend treatments for the disease. These include medication, behavioral change, digital therapeutics etc. The system can predict many things, including a user?s personal risk for contracting COVID-19. It can also predict the severity of the disease that could be experienced by the user if he or she were to become infected. The system can take action to prepare for new risks. For example, it may encourage a user’s behavior to be changed, increase monitoring or change their monitoring methods. The system can then protect users against changing disease risks by predicting and estimating these risks, and recommending or taking actions to reduce them. These steps can reduce or prevent the severity and incidence of viral outbreaks as well as other diseases.
The system is able to detect disease earlier and with greater accuracy using new techniques for creating and updating individual baseline measurements. The system is able to generate baseline measurements for multiple aspects of the user’s behavior and physiology. Several previous systems did not use the predictive value of user behavior changes effectively to detect and manage diseases. As discussed below, in many cases behavioral changes can be an early indicator or predictor of infection, or the severity of future symptoms. However, behavioral changes may be subtle or of varying significance to different individuals. The system can update the baseline measurements over time by using repeated and ongoing data collected from mobile devices or other data sources. In some cases the baseline measures may include average values calculated from data collected over time. Baseline measures can include complex data types, including data that indicate patterns and trends in time as well as classifications for those patterns and tendencies.
The system may also be able to use data from each individual’s locality in order to improve estimates or predictions about disease exposure, infection likelihood, and many other things. The system can, for example, collect data about the location of a user’s residence, work, and general time spent, as well the type of places the user visits, and their specific travels and visits. To better predict the likelihood of exposure and infection, information about infection rates in communities where users have been is used. In this process, the system can use data from contract tracing, such as data that tracks the proximity of an individual user over time to other individuals. The system can also use data from monitoring to track the location of a user, their time in those locations, traffic or occupancy at the places, or other data.
The system can adapt the type of interaction and the level of monitoring with the user based on the predicted exposure level or likelihood for infection. As a user’s level of exposure increases, the system will automatically increase the frequency, extent, and duration of monitoring via user devices such as smartphones, smartwatches, or other wearable devices. The system can change the type of measurement made (e.g. types of sensor data) and the types or surveys, prompts to the user, or other interactions with the user that are provided by the system. The system can vary the monitoring procedures for the user to achieve a desired confidence level in the predictions.
The monitoring data and predictions can be used by the system to perform many different actions in order to manage a particular disease. For example, the system could recommend a specific testing kit, a vaccination and a regimen of administration for that vaccine, or a treatment (e.g. selecting drugs, dosages, medical devices, settings, digital therapeutics). The system will recommend these actions, as well as others, based on clinical data, monitoring data, community measures and other factors.
The monitoring of user behaviour and the use of this behavior data for disease detection and treatment selection is one of the major advantages of the current technology. In many cases, subtle behavioral changes can occur in the early stages of a disease, before a person is aware of a symptom. The present technology is able to detect early signs of infection and other risk factors by detecting subtle differences in user behavior, such as changes in frequency and intensity, or changes in how different behaviors are combined at the same time and in similar sequences. By comparing observed data to personalized baseline measurements, these types of signs can be detected more accurately. The technology also allows for earlier indicators of behavior changes to be confirmed and enhanced by follow-up interactions, such as surveys or other interactions that help pinpoint causes or factors that influence behavior. “The combination of real time behavior change detection and real time physiological monitoring, and the two being tracked and correlated together, allows for better predictive modeling and detection than just physiological monitoring.
In general behavior can be tracked in order to detect certain signs, symptoms, and effects of diseases, as well as the treatment of those diseases, even for items which are not easily detectable by physiological sensing alone. Passively sensed data can be used to infer user behavior, including location tracking, movement monitoring, device poses and usage, sounds detected by a microphone (e.g. detecting a coughing user), etc. Behavior data may include records and indications of events and activities, patterns and trends, and discrete instances. For each type of activity, a baseline measurement can be calculated, stored and periodically updated for the user. Baseline measures can be used to indicate the type of behavior, its frequency, duration, intensity and location. The baseline measures can also include whether the activity is done alone or in a group, with whom the activity takes place, and how often it occurs. Changes in sleep, diet and work habits, as well as changes in travel, physical activity, social behavior and sleep patterns, can all be indicators of disease. The system can interact with the user in real time to gather context and additional information when it detects changes in their behavior. “For example, a device could detect that a user spends more time in the bathroom than usual and send a survey to ask if they are experiencing GI distress. This is one of the many COVID-19 symptoms.
The behavior data provide the system with context for better interpreting physiological data. The behavior data, for example, can tell if a temperature measurement is taken during a high, medium or low level of activity or when resting or sleeping. It is possible to use and interpret measurements taken in different contexts or activities. The data collected is more diverse than what was used in previous systems. The present system, for example, provides new categories of measurements and evaluations that can be used to detect data patterns, disease signs, and symptoms, which would not otherwise be possible if the measurement was restricted to one context, like sleep. The system can, for example, evaluate changes in the user’s behavior at work, in their exercise routine, in their sleep patterns, etc. as indicators of possible disease. A user might have a headache, but no elevated heart rate, temperature or other physiological parameter. The user might change his behavior due to the headache. He may go to bed earlier, skip a workout, move more slowly at the office, or stay home. The system can deduce from the behavior that the user does not feel well and initiate additional monitoring and user interaction to better determine the causes of the changes.
Click here to view the patent on Google Patents.