Addressing “Prior Authorization” with Machine Learning

read 2 mts

What is prior authorization?

When a patient wants to buy medicine at a store, the store owner’s system in real-time communicates with the insurance company to see if it is covered under the plan.  Some medicines are approved instantly.  

However, for some others, the insurance system suggests prior authorization (i.e. a human expert needs to check the policy and decide whether they would reimburse or not).

The store manager then informs the patient to come back in a few days to see if the medicine is authorized for reimbursement.  More than 80% of the drugs that are put under prior authorization are never bought by the patients due to inertia.  This not only has detrimental economic effects but more importantly, detrimental health effects on the patient.

The client who builds authorization software for several pharmacies wanted INSOFE’s help in building a tool to alert the doctors about the possibility of that prescription requiring prior authorization even while they are typing a medicine’s name.

Implementation

INSOFE’s team of data scientists learned that building a knowledge base from the documentations of insurance firms is extremely difficult as each one followed a different format.  The lists are also very dynamic.  Hence, the team decided to pursue an alternative approach.

Causal data of around 12 million patients is collected (this is roughly 20 days of data).  A new patient and drug are in real-time compared with this data and the closest neighbor is identified.  If the closest neighbor required prior authorization, we would alert the same for the current record.

To have INSOFE faculty and data scientists solve your business problems, prep your engineering teams to face the real world complexities, visit here

We identified false negatives as more important errors to minimize and chose recall as a metric to pursue.  We used a variety of eager and lazy learning strategies and could optimize a gradient boosting machine to get maximum performance.  The performance was then optimized to get a near real-time output across hundreds of clinics and doctors.  A simple GUI as a tooltip was designed.

Outcome

The client solved prior authorization problem efficiently and could minimize the prior authorization cases by 50% in their network.  Interestingly, they also learned many insights in this exercise that would be extremely valuable to the doctors and the insurance firms.

1+

Leave a Reply

Your email address will not be published. Required fields are marked *