Our SMART Technology
With the use of deep learning , Marpai achieves the dual goals of lowering the cost of claims for self-insured companies, while maintaining high quality healthcare outcomes for members. We generate and collect troves of data from all aspects of our business including healthcare claims, member engagement, internal operations and customer experience and systematically apply intelligent automation and artificial intelligence to delight our end users, improve healthcare outcomes and lower costs.
We use AI in virtually every part of our business, including the core systems that are the backbone of a third party administrator. These include:
- Claims management that enables accurate auto-adjudication of most claims, which will lower our operating costs and increase our operating efficiency. This also includes detecting and flagging suspicious claims before final adjudication.
- Member engagement and self-service options via an intuitive and easy-to-use application, which will also lower the cost to serve employers and allow members to interact with us as often as they wish. We will present the best providers in terms of quality and costs, in a manner that is centered around a member’s needs and geography.
- Care management with high-impact prediction via new, innovative AI modules, which will help us manage the overall cost of claims for our clients.
- Continuous Provider quality tracking enables us to identify the best providers and also the usual and customary practices and their related costs.
- Smart plan design including cost optimization, performance monitoring and risk prediction.
Our AI Capabilities – Deep Learning
During the past few years, we have witnessed the greatest leap in performance in the history of AI, largely due to a subfield of artificial intelligence known as “deep learning” (or “deep neural networks”). While traditional machine learning methods require extensive manual processing (e.g., for feature extraction), deep learning methods allow for “end-to-end” learning, directly operating on raw data and predicting the final outcome.
The accuracy of deep learning-based models is significantly higher than traditional machine learning models especially when more data and more data types are available. This makes deep learning very relevant for big data platforms, which involve large quantities of data and various data types. Despite the huge success of deep learning in improving the state-of-the-art computer vision, speech recognition, text understanding, and other areas, few companies have the expertise to develop and deploy deep learning-based solutions, due to the numerous technical challenges involved. We believe that is why the application of deep learning remains limited in the healthcare space and primarily confined to medical image processing today.
We are creating a robust deep learning infrastructure that would allow us to analyze nearly any data type. More importantly, this infrastructure is expected to have the capability to automatically train on multiple data types at once and obtain unified predictions. Specifically, we expect our deep learning capabilities to be used for the following data types:
- Structured data – Any form of data that results in a tabular database. Examples of such data include laboratory tests, claims data, and payment details to healthcare providers. Our deep learning models for structured data, which rely on fully connected neural networks, allow us to automatically train on any such structured databases, and incorporate the insights into the predictive models.
- Images – While traditional image processing requires cumbersome development for each specific use case, deep learning-based models can directly operate on any image data, without having to go through a preprocessing feature extraction phase. We are developing a series of deep learning models based on convolutional neural networks that could be extended for processing medical images, including radiology imaging such as MRI, CT Scan, X-Ray, Ultrasound, PET Scan, as well as processing other types of medical images, such as those in dermatology, pathology, and ophthalmology.
- Text – Traditionally, analyzing text data requires lengthy natural language processing (“NLP”). Recent breakthroughs in deep learning, especially the advent of transformer networks (e.g., BERT), allow for end-to-end training of language models which are significantly more accurate than previous NLP methods. Our infrastructure includes support for these deep learning-based language models, allowing us to automatically process textual data as well. A prime example of textual medical data is the plain text description of doctors’ notes, which includes information about the symptoms, diagnosis, and treatment. Even though some of this data, such as disease codes, appear in structured databases as well, the textual data contains significantly more information that does not appear elsewhere.
- Multi-type data – Deep learning can also be applied on multiple data sources and data types. For each data type at the first stage, the relevant deep learning models are applied (e.g., CNN for images, Transformers for text, etc.), and then, at the second stage, the processed information from various data sources is fed into a secondary deep learning model, which will provide unified predictions. This is one of the major advantages of deep learning over traditional AI, allowing it to incorporate multiple big data sources into a single unified prediction model, far exceeding the accuracy rate achieved by traditional methods.
- Data fusion – We have the ability to study different data types together. Data that are images (e.g., CT scan) with data that are in tabular form (e.g., figures from healthcare claims). Putting these different data types is data fusion.
SMART Health Plan Services System
Our SMART system is fully integrated with our user platform and can automatically deliver alerts to both internal staff, members and other stakeholders.
- Mass Scale Data Fusion – Our system fuses massive amounts of data from various sources in many formats into a single environment for analysis.
- Data – Our system includes a unified healthcare schema that can ingest and make useful any healthcare data type including claims, social determinants of health and psychographic data, blood test results and pathology and radiology images.
- Data lake – Our system includes a massive data lake with our unified schema that stores and structures healthcare data in order to enable analysis.
- AI Models – Our system automatically applies our AI models to healthcare data to enable pattern recognition, advanced querying and anomaly detection.
The FutureSight Advantage and Proactive Health
Our AI capabilities in the area of deep learning enable us to make more accurate predictions about costly healthcare events using data types that are hard or impossible to use using traditional artificial intelligence. This allows us to help our member prevent costly healthcare events and bend the cost curve for employers.
Our deep learning technology is able to process, analyze and store myriad types of data in large scale. Our AI models recognize patterns and detect anomalies, giving us greater insights on a patient’s medical trajectory. Knowing that trajectory allows us to intervene early with the best healthcare recommendations. The current data types our AI technology uses to predict healthcare needs include, but are not limited to, the following:
- Claims data
- Social determinants of health and psychographic data
- Blood test and other laboratory results
- Pathology images
- Advanced images from X-ray, magnetic resonance imaging (“MRI”) and computed tomography (“CT”)
- Electronic health records
- Health risk assessments
- Information gathered from applications used by Members
- Information gathered from wearables like Fitbit devices and the Apple watch.
Using our innovative technology, will help self-insured employers manage their employees’ healthcare needs proactively. Our AI-enabled predictions will be presented as real-time alerts that our team then uses to guide an employee towards the best preventive care management providers at the earliest possible time and to guide members to lower-cost but high-quality healthcare providers. The key components of include:
- Deep Member Profile – We understand a member’s health based on their medical history, demographic information, historical claims, and in some cases, what they shared with us as their own health risk assessments. All this data is fed through various deep learning modules and automatically processed;
- Claims Trajectory – Our AI models have studied millions of claims and have mapped out an expected trajectory for members. The predictions are based on all available data on each member. Currently, the data includes structured data, such as demographic information, historical claims, laboratory test data when available. In the future, the data will include unstructured data, such as medical images, text-based assessments and other types of health records. All these data will be fed into deep learning models for each data type, and subsequently fed into a unified deep learning model which finds patterns and insights across all data and data types available for the member; and
- Alerts – Our AI models come to life via alerts, which are triggered by an event such as a particular claim. Each alert is a new data point for each member, invoking the entire model to update its predictions. Deep learning models are especially superior for finding non-linear patterns and correlations. For example, a new data point X that apparently is completely unrelated to prediction Y, may actually affect it through complex non-linear patterns, which are very difficult for humans and traditional artificial intelligence to find, but deep learning models are very good at finding those patterns, which trigger new actionable alerts. These alerts allow the Care Guide team to start outreaching to an at-risk member.
For example, an alert may say that a Member is likely to develop Type 2 diabetes in the next twelve months, even though that member is not currently diagnosed. Our Care Guide team reaches out to the member and recommends a visit to a primary care physician for a root cause or drill down analysis. The doctor can then develop an actionable healthcare plan with the member.
Our AI-enabled predictions helps both members and their providers be proactive and get in front of costly events, so they can predict, prevent, and plan around them. Our services provide not only lower costs, but also better value for the money spent on healthcare, and therefore greatly reducing any waste related to care coordination and care delivery.
TopCare: Matching Members with High Quality Providers
Matching Members with high-quality providers is a key component of our services. Provider quality is continuously tracked and measured by feeding AI models with various provider quality databases. Monitoring providers’ outcomes, costs, and proper conduct will allow our models to score good and poor-quality providers as well as measure cost-efficiency. This information will allow us to assist members in making more informed decisions on their healthcare provider options.
Using in-depth insights from laboratory results, imaging, and electronic health records in AI models will help us detect suspicious claims and potential over-treatment. These cases can be re-examined by clinicians and may be used to educate both members and providers, or otherwise guide members to better healthcare solutions. High-performing providers will be used by our AI models for developing the “gold standard” with respect to pricing models and to understand what is usual and customary in terms of costs.
Given the high price variability for expensive medical procedures, we want to guide members to cost-effective but also high-quality providers. For instance, we may generate an alert for a member who is on a trajectory to have a knee replacement surgery over the next twelve months. In that case, we will educate the member early on his/her options in terms of highly rated orthopedic surgeons for that exact procedure within his/her geographic area. This may happen months before the actual surgery and involves the member’s choice of care. This helps reduce any waste related to not obtaining the best pricing for healthcare options. Company health plans may also allow for incentives whereby the member’s co-insurance is reduced or eliminated if he/she selects the high-quality and more cost-effective surgeon.
For our self-insured companies, we use AI to design and optimize plans and costs given the historical data of its employee population and its predicted costly healthcare events. This information forms the building-blocks of plan design, risk management, and plan optimization. Continuous monitoring of the at-risk population as well as measuring how mitigation strategies are working are also key components to lowering the overall cost of claims.
Part of the technology also tracks member outreach to see how and if the member has responded, and what communication vehicles are the most successful for a particular member (e.g. text, email, phone call). This data helps us tailor outreach that best facilitates member responses and, therefore, reduces waste related to care coordination and care delivery. The response is also fed through a feedback loop to the deep learning models to not only improve the predictions, but more importantly, the actionable alerts.