How machine learning can help improve treatment
The technology can transform analytics to be quicker and more efficient for healthcare organizations.
As the cost of healthcare has risen, health plans have seen a growing need to proactively manage members’ health. The proactive part of that is the rub; it’s far easier to react to an event than to anticipate the likelihood of an event and act to prevent it.
That’s where analytics becomes valuable. Most health plans have much of the data they need to predict health events, particularly those related to chronic disease. But getting accurate insights from that data, particularly predictive insights, has, until recently, been expensive, complex and time-consuming.
Hence, the wish list is topped by more useful analytics. It’s the one capability that can truly transform a health plan’s business model, if it could only be adopted in a way that is fast, agile and affordable.
Most CEOs are familiar with the term artificial intelligence (AI) and have at least heard the phrase machine learning. But most aren’t sure what those terms really mean. While it’s not really necessary for the average business person to know the details of how these technologies work, a basic overview is important. In this article, I’ll share a short description of the technology, then show you how it can be put to practical use to solve health plan challenges.
AI is the general term for a computing resource that can mimic human thought, incorporating new data to improve results. Machine learning is a subset of AI which puts that ability to work. Essentially, machine learning allows an algorithm to change and improve the accuracy of results as it encounters increasingly large sets of data. Imagine a school child trying to understand English grammar.
As the child encounters increasingly more complex forms of language, the child learns the difference between the present and past tenses. In that same fashion, an algorithm can discern new patterns in data as it encounters more examples. And like a human learning from a photo, a spreadsheet and a written description, the algorithm can learn from all kinds of data, structured and non-structured. This means that it can incorporate images, graphics, medical notes and other hard-to-use data sources.
This “learning” ability allows an analytic algorithm very quickly to become better at predicting the future. For example, we used this approach to rapidly develop sophisticated predictive algorithms to identify high cost and high-risk members in a large health plan’s population. We helped them identify members who were frequent users of services, particularly high cost services such as the emergency department and inpatient care. Without machine learning, developing these predictive models would have likely taken months, using expensive data scientist resources.
Using an automated machine learning solution, we were able to create these advanced algorithms and start seeing results in less than a week. That’s a huge leap forward. Moreover, this project was executed by business experts, not data scientists. Essentially, this automated machine learning approach gives “citizen data scientists” the ability to create algorithms directly. This frees up the data scientists’ time for other projects.
Our findings will enable the health plan to proactively identify their potential case management candidates, design tailored care management support and adopt early intervention strategies to keep these members healthier.
Of course, the accuracy of these results is critical. You don’t want to waste expensive resources on members who don’t need that help, and you don’t want to overlook someone who, without intervention, will develop a preventable disease. The beauty of automated machine learning lies, in part, in the ability to improve accuracy with experience by running very large data sets through multiple logics quickly. This is something no human, even expert data scientists, can accomplish. The more you use it, the more accurate the algorithm becomes. And there are built-in quality controls to validate the predictive models used and to keep the algorithm accurate as it learns.
The part of all this that a health plan executive needs to remember is that machine learning can make analytics easier, faster and less expensive to use. The actionable insight is this: ask your CIO to learn more about how analytics are being transformed by machine learning and to proactively find partners who have experience with machine learning.
That’s where analytics becomes valuable. Most health plans have much of the data they need to predict health events, particularly those related to chronic disease. But getting accurate insights from that data, particularly predictive insights, has, until recently, been expensive, complex and time-consuming.
Hence, the wish list is topped by more useful analytics. It’s the one capability that can truly transform a health plan’s business model, if it could only be adopted in a way that is fast, agile and affordable.
Most CEOs are familiar with the term artificial intelligence (AI) and have at least heard the phrase machine learning. But most aren’t sure what those terms really mean. While it’s not really necessary for the average business person to know the details of how these technologies work, a basic overview is important. In this article, I’ll share a short description of the technology, then show you how it can be put to practical use to solve health plan challenges.
AI is the general term for a computing resource that can mimic human thought, incorporating new data to improve results. Machine learning is a subset of AI which puts that ability to work. Essentially, machine learning allows an algorithm to change and improve the accuracy of results as it encounters increasingly large sets of data. Imagine a school child trying to understand English grammar.As the child encounters increasingly more complex forms of language, the child learns the difference between the present and past tenses. In that same fashion, an algorithm can discern new patterns in data as it encounters more examples. And like a human learning from a photo, a spreadsheet and a written description, the algorithm can learn from all kinds of data, structured and non-structured. This means that it can incorporate images, graphics, medical notes and other hard-to-use data sources.
This “learning” ability allows an analytic algorithm very quickly to become better at predicting the future. For example, we used this approach to rapidly develop sophisticated predictive algorithms to identify high cost and high-risk members in a large health plan’s population. We helped them identify members who were frequent users of services, particularly high cost services such as the emergency department and inpatient care. Without machine learning, developing these predictive models would have likely taken months, using expensive data scientist resources.
Using an automated machine learning solution, we were able to create these advanced algorithms and start seeing results in less than a week. That’s a huge leap forward. Moreover, this project was executed by business experts, not data scientists. Essentially, this automated machine learning approach gives “citizen data scientists” the ability to create algorithms directly. This frees up the data scientists’ time for other projects.
Our findings will enable the health plan to proactively identify their potential case management candidates, design tailored care management support and adopt early intervention strategies to keep these members healthier.
Of course, the accuracy of these results is critical. You don’t want to waste expensive resources on members who don’t need that help, and you don’t want to overlook someone who, without intervention, will develop a preventable disease. The beauty of automated machine learning lies, in part, in the ability to improve accuracy with experience by running very large data sets through multiple logics quickly. This is something no human, even expert data scientists, can accomplish. The more you use it, the more accurate the algorithm becomes. And there are built-in quality controls to validate the predictive models used and to keep the algorithm accurate as it learns.
The part of all this that a health plan executive needs to remember is that machine learning can make analytics easier, faster and less expensive to use. The actionable insight is this: ask your CIO to learn more about how analytics are being transformed by machine learning and to proactively find partners who have experience with machine learning.
More for you
Loading data for hdm_tax_topic #better-outcomes...