Thread by @erenbali

We have spent a ridiculous amount of time trying to bring full price transparency to healthcare payments but haven’t been able to make much progress. Let me explain why this is hard 🧵

https://twitter.com/nikillinit/status/1343006389166370822

Our goal is to predict patient responsibility of a visit as soon as possible. Ideally this would be calculated real-time in our EHR so that clinicians can do cost/value trade offs with the patient while ordering things that impact the cost (procedures, diagnostic tests, etc)

The main independent variables here are
1) Complexity of the evaluation (usually 1-5)
2) Additional services like labs, procedures, medical imaging etc which are modeled as CPT codes
3) The nature of the work done (preventive, sick visit, etc) which are mapped to service types

4) Coverage of the insurance plan by service type. This is far more complex than United vs Cigna. Each plan may have arbitrary rules that impacts coverage + HMO/ACO/EPOs complicate things further
5) Contracted rates between provider and network (not the insurance or the plan).

6) Remaining deductible
7) Other details like the clinician type, modality, location etc.

The first challenge is to generate claim details (mostly CPT codes) real-time in your EHR. This is a solvable design problem. You can solve price transparency for self pay patients with it.

Next you have to get very granular details of the insurance plan. Clearing house APIs give you some of this but you hit an “edge case” 20-30% of the time.

Next challenge is digitizing your contracts which is 100x harder than it should be but Rivet Health does help here.

Here is the biggest challenge: with the exact same plan, CPT code, same literally everything you get 100s of different “total allowed amount”s from the insurance companies. On theory the allowed amount should match the contracted rates, but it only does ~40% of the time.

Two underlying problems here. First, there are a ton of things that influence “the contract” which are opaque to the provider. Secondly, insurance companies still employ an army of humans to respond to claims. You get everything from inconsistent interpretation to missing digits.

Price transparency with the incomplete data we have is really hard, particularly because an inaccurate prediction may be worse than no prediction. Companies like Oscar have automated claim processing systems, but legacy insurance infrastructures can’t support this.

I’m guessing the solution will be getting rid of the complexity altogether (case rates, full capitation, etc). Unfortunately a lot of efforts to “reduce cost” are making things even more complicated and thus indirectly increase the cost.

I should sit down and write a detailed article about this and have our team fact check it. Take this as a directionally accurate portrayal of the challenge.