Using a combination of predictive and prescriptive information about next action, providers can close the care gap for individual patients by leveraging a combination of data sources – clinical data, patient surveys, SDOH data and datasets about consumers and behaviors – and applying artificial intelligence techniques to create those insights.
In the United States, health care is often more effective at treating disease than at preventing it. Chronic diseases are main causes illness, disability, death and rising health care costs in the country. Lifestyle-related health problems like obesity, high blood pressure and blood sugar, poor diet and smoking are linked to more than $730 billion health care spending in the United States six out of 10 American adults living with a chronic illness.
But these lifestyle diseases don’t exist in a bubble. The circumstances in which a person lives have a direct impact on health outcomes. Social determinants of health parameters (SDOH) such as income levels, food insecurity, education and housing status 30% to 55% health outcomes. In addition to being the root cause of a patient’s health issues, these SDOHs can also be a barrier to receiving the culturally and contextually appropriate care that a person needs.
Closing the care gap is in the interest of the health system. When providers can identify patients at risk of discontinuing disease treatment, the impact can be a matter of life and death. When Medicare providers identify patients with potential negative health outcomes and report them to providers, it saves both taxpayers’ money and patients’ lives.
By using a combination of predictive and prescriptive information about next action, providers can close the gap in care for individual patients. To do this, healthcare systems leverage a combination of data sources – clinical data, patient surveys, SDOH data, and consumer and behavioral datasets – and apply artificial intelligence (AI) techniques to create these informations.
Predictive and prescriptive insights defined
Predictive analytics uses historical data to create static models that predict future outcomes. In a concrete example, predictive models can be used to prepare a hospital for an influx of new diseases; something we often experienced during the peak months of the COVID-19 pandemic. These predictive models analyze past and current infection numbers to identify when a new wave of infection will begin, giving the hospital system a head start to prepare with adequate PPE, staffing plans and protocols.
Perhaps the most important aspect of using predictive analytics is knowing what to do with the information. It’s one thing to understand which patients are at higher risk of developing type 2 diabetes based on their blood tests. It’s another thing to know what the next best actions are for that specific patient and their lifestyle. Predictive information identifies patients at risk and prescriptive information recommends a set of interactions specifically for those patients.
For example, a predictive insight might tell us that Member X is at risk of non-compliance with his treatment due to a few critical factors, such as lack of reliable transportation to his doctor, his distrust of the medical system, or because English is not their first language. Based on this information, a specific set of next best actions are recommended to target these specific factors to encourage adherence to care.
Generating insights through AI and data
AI techniques such as machine learning take predictive analytics to the next level. Machine learning algorithms create predictive models using sample data. As the model receives more data, the algorithms learn and improve themselves from patterns in the data. Self-reported patient data is often combined with third-party datasets, such as publicly available data. national SDOH data or companies that collect consumer data. Each of these data sets is never 100% accurate on its own, but when combined they create a more complete picture in which to discover patterns and generate insights.
We believe it is critical to include a wide range of datasets including clinical data, SDOH, consumer data, behavioral data, and self-reported data when building these predictive models. This allows us to get a better holistic view of each patient, to understand what could potentially prevent them from having positive health outcomes (e.g. they live in a rural area, are not fluent in English, have transportation problems and are distrustful of the healthcare system) and close the care gap by prescribing a set of interactions specific to that patient’s situation (for example, to ensure that they continue to take their drugs or find a provider who speaks their mother tongue).
Use predictive and prescriptive information to close the care gap
I’ve seen how predictive models and next best action insights can make a difference in some of the most at-risk communities. One of the predictive machine learning models we use in Africa is tracking patients on antiretroviral therapy (ART) for HIV to identify patients at high risk of dropping out of the healthcare system. All predictive analytics models are built on large, multi-year, longitudinal patient datasets (e.g., 500,000 patients): data that identifies who dropped out of the healthcare system, what their CD4 count was, where they lived, etc.
Based on the model that builds correlations and identifies patients at increased risk for care dropout, providers are informed and take proactive action with high-risk patients. These interactions are fed back into the model, which uses this information to continually improve. The results are impressive. One program saw a 36% increase in retention of high-risk patients.
Looking Ahead: Using Data to Create More Personalized Next Best Actions
We live in a data-driven, data-rich world – from our smartphones in our pockets, to our Apple Watches and FitBits on our wrists, to the purchases we make online. In the next 10 to 15 years, I predict this type of personal data will be used more often to get an even more personalized picture of each patient’s health. If doctors had access to objective data on a patient’s resting heart rate, sleep patterns, number of steps they take in a day – which is currently done via subjective self-report surveys – we could build more accurate predictive models to identify at-risk patients and the next best actions needed to meet that patient’s personal needs at the time.
Like all conversations around big data, especially health data, privacy is paramount. Apple is advancing this technology with its recent update to iOS 15, which allows users to share their health data with others, such as family members or doctors through the Health app. This is a step towards closing the gap between patients and their providers. By combining this data with SDOH, we can not only create models that predict disease or disease progression, but create a personalized action plan for each patient, delivered to them in the context of their real life. Equitable healthcare is possible and data and AI will get us there.
Dr. John Sargent is the founder of BroadReach Group.