Descriptive Analytics: Baby steps towards Artificial Intelligence/Machine Learning
We are witnessing a rise in the use of AI/ML concepts and technologies in healthcare. Today, algorithms are already outperforming radiologists at spotting malignant tumors and guiding researchers to construct cohorts for costly clinical trials. There is also an increasing number of investments being made around AI/ML use cases and proofs-of-concept across the healthcare industry.
Clinical Decision Support (CDS) systems are great examples of artificial intelligence in healthcare. The ones that currently help identify high-risk patients who could suffer a stroke or detect diabetic retinopathy in a patient represent great strides in the right direction for preventive care. Artificial Intelligence for decision making with minimal human interactions is a work-in-progress, and there are many improvement areas.
It is very easy to be bogged down by the many mathematical & statistical concepts, neural networks, and technology demands of AI/ML. So, it might help to get back to the basics. Let’s take some baby steps towards the broader objectives using AI/ML. There is so much that can still be done with Descriptive Analytics. Put simply; Descriptive Analytics is a means to identify and describe the data present before us most effectively and efficiently possible.
Descriptive analytics has been playing a prominent role in healthcare since the time medical data has been recorded. Some of its use cases include understanding member populations or patient cohorts, identifying irregularities in insurance claims to help healthcare organizations gauge their performance.
These performance measures help in understanding the data internal to any organization, and also to compare with industry benchmarks. But sometimes, even these pre-defined measures could pose challenges if they are not supplemented with the correct algorithm/parameters or if the underlying data itself is incorrect or unorganized.
Getting the Head Start
The importance of data quality cannot be emphasized enough for descriptive analytics to give an accurate picture of the members’ health and the health of the organization. It is also necessary to understand the organization’s business goals and vision to understand the desired scope of analytics.
Each missing or irregular data element lends to compromised data quality, so identifying data gaps and fixing those would be a first step towards ensuring efficient data governance and sound descriptive analytics.
A one-size-fits-all approach will not work since each organization is different in its methods, processes, and technology usage. Every organization provides a different challenge and unique opportunity to work with disparate systems and data sources to integrate, streamline, analyze, cleanse, and build the needed structure ranging from a basic analytics dashboard to a complete control center.
Here are a few descriptive analytics use cases:
Medical Loss Ratio: The Medical Loss Ratio (MLR) describes the ratio of the amount spent vs. the healthcare payer’s premium. Here, the amount paid can be claims amount paid, administrative cost, etc. By looking at the MLR trend, the organization can keep a tab on increasing or decreasing cost, for the premium that it has earned, to take appropriate actions.
Premium Earned: It is the amount of money collected as insurance premium from all members and is the most significant revenue source for healthcare payers. Trends in Premium Earned can also show signs of opportunity.
Member Population: Payers can quickly see trends in member population by the employer group, geographical location, socio-economic factors, regulatory requirements, etc.
Per Member per Month (PMPM): PMPM can be used in different KPIs, e.g., PMPM revenue, Claim Amount PMPM OR PMPM Cost, # of Claims PMPM, # of Inbound Calls PMPM, etc.
Claim Amount: An increase or decrease in the claim amount requested by providers can indicate many factors about member populations and providers’ services. For a specific type of provider or diagnosis, this trend analysis could lead to the required course of action. Another perspective could be to check the performance of various member engagement and outreach campaigns initiated by the payer.
High Dollar Claimants: High dollar claimants can be defined differently based on every payer’s business model. Usually, payers can take high dollar claimants as members for whom the payer reimburse more than $50k through claims in a year. With the 80/20 rule, payers can track and monitor these high dollar claimants to create a course of action to reduce high dollar claims.
Auto Adjudication: Adjudication is the decision made on the claim amount requested by the provider. For a majority of payers nowadays, adjudication is an automated process. But there are many scenarios and use cases where the decision is not possible by automation. As payers, it is important to track the trend in the rate of Auto Adjudication to monitor the process, as a decrease in rate can lead to a burden on the claims processing team and manual errors or wrong decisions.
30 Days Readmission & ER Visit Trend: Amongst all member encounters, inpatient visits are usually the most expensive to payers. Additionally, in many scenarios, when the discharge plan and care coordination for a member are not adequately defined and executed, the member may get readmitted within 30 days of discharge from the care facility, for the same diagnosis. Worse yet, they may need to go for an Emergency Room visit. So, the trend analysis of readmissions and ER visits can help payers understand more about providers’ practices and member populations.
As healthcare data streams become more complex, Al/ML will be necessary to decipher interesting insights that will enrich the care process and health plans with new strategies to reduce cost and increase care quality. The journey to reach this competence level begins with a detailed understanding of descriptive analytics and ways to use them in your business process.
Learn how you can leverage the power of descriptive analytics from our Healthcare IT experts!
Reach out to us at email@example.com
Latest posts by Pankaj Kundu (see all)
- Understanding the Limitations of Inadequate Data - December 16, 2020
- The Steps to Success for Healthcare AI/ML Endeavors - December 4, 2020
- Establishing an Analytics Culture – An Overview of the Roadmap - October 19, 2020
Pankaj has vast experience ranging from claims processing engine to application of machine learning algorithms in US Healthcare. As a Healthcare Business Analyst, he is passionate about addressing healthcare data/process related challenges and ideating solutions for clients.All stories by: Pankaj Kundu