Should achievability be measured at the practice-brand or practice-generic level?

[sebbacon]

Our current definition of an achievable saving is:

• Create a list of price per pill data paid in a single month. For each practice, one row for each of:
  • the generically prescribed presentation; and
  • all of its branded prescriptions
• Pick the top decile price per pill as an achievable cost
• Calculate possible savings accordingly

However, @richiecroker is concerned that it's provocative to assert this is achievable. He feels we should calculate price-per-quantity based on the per-practice mean for the generic presentation. In other words, we should assert that the average price achieved by a high-performing practice for a chemical presentation is achievable; not that a particular high-performing price should be achievable.

The current model assumes that if pill A1 is expensive and pill A2 is cheap, the ideal practice could prescribe 100% of pill A2. Rich is concerned that in reality, if the currently best-performing practices are still prescribing 20% of A1, then they are probably already representing the realistically achievable best practice.

For example: at present, if in a single month Bash Street Practice bought 10 generic Tramadol 200mg at 0.3p / pill, and 10 branded Tramulief at 0.2p / pill, these are two separate price-per-pill data points for the purposes of calculating a realistic best-case price-per-pill.

Rich's proposal is that we should consider Bash Street Practice 200mg Tramadol as a single datum of 0.25p / pill, and that the achievable target is for every practice to do as well as the practice at the top decile.

The counter argument is that just because the best-performing practices are still prescribing 20% A1, that doesn't mean there's a good reason why. By focusing only on what's already happening we might obscure some genuine possible savings across the board.

I'm not sure of the solution!

[sebbacon]

It's worth remembering that there are a bunch of other reasons 100% savings will never be achievable that we've already identified, such as (copied from the README):

• Imported drugs often vary wildly in costs due to different import routes
• The prescribing data rarely distinguishes between pack sizes. Larger pack sizes tend to be cheaper per-dose than smaller pack sizes. A prescriber has no influence over the pack size used to fulfill a prescription
• In some cases, dispensers can charge the cost of an entire pack for medicines where the prescribed quantity is smaller than an entire pack ("broken bulk")
• Inconsistencies in the data. We have tried to address these where we have found them (see below for details)
• Non-bioequivalence. Some drugs within a generic class cannot always be considered clinically equivalent.
• Licensing differences. Two drugs which are bioequivalent may not be licensed for all possible uses, so cannot be switched.

So even if we address the specific case Rich has brought up, we will still have to think about the political implications of us claiming things are "achievable". I think that means we need to do a few things:

• Decide internally what we think "achievable" means
• Write up our rationale in detail
• Unambiguously and loudly declare that no practice will ever achieve 100% of the projected savings

[sebbacon]

We've agreed that we should factor what the best practices are already achieving into the possible price-per-dose calculation.

Bearing in mind the general point about "achievability" above, we also talked about possibly adding a "slider" or "alternative scenario" to the data views for each CCG. I've referenced the two models so you can compare the data side-by-side in the README.