Future budgeting / price shock calculator

[sebbacon]

Idea

Similar to what we currently do with Price Concessions, project an organisation's spend forward based on current drug tariff data (including price concessions), known or expected future changes in the drug market (e.g. patent expirations), and that organisation's historic spending.

Discussion

• Model could incorporate extrapolating trends (by chemical? section?) forwards
• It would also have to take seasonality into account
• We can validate any model we build against historic spending
• Ancient discussion of early ideas here
• SPS publish an annual Prescribing Outlook which includes a calculator, for estimating the effect of expected future changes, we would could potentially incorporate
• Incorporating changes related to patent expiry is problematic as there are many factors, e.g. does stock arrive in the UK, drug tariff negotiations Cat M that makes it impossible to reasonably predict
• We would want to include in our data / visualisations some kind of indication of our confidence in the projection. A formal model based on regressions and current costing could include confidence intervals; but the SPS modelling involves a lot of hand-waving so including that would have to have some made-up indicators of confidence

See also #1649

[brianmackenna]

Discussion from DataLab day on 26.2

OP- forecasting methodology? Could be useful for CCGs/NHS to make a robust prescribing spend forecast. E.g. fan chart with 95% error bars. Add to list of feature development Predicted spend with range and associated possible savings (also what savings could have been if started at beginning of year). FOI NHS budget profiles Could we detect changes in people going for savings if they are coming in under/over budget?

[brianmackenna]

Things to be worked out

• General trends and project forward based on current factors
• Forward looking spend for smallest amount of things to take action
• How to filter for interestingness

1417 is related in terms of demographic changes if we include in any model

[sebbacon]

We agreed @richiecroker would start by making a model in a notebook, and we'll continue discussions from there

[brianmackenna]

87 is also relevant