This project compares the price-per-quantity between different clinically comparable products, so we can advise practices and CCGs on potential cost savings.
We work from the monthly GP Practice prescribing data. In this, each row of data describes prescribing of a presentation for one practice for that month. For example, for a given practice we might see two rows of data for tramadol: one for Tramadol Hydrochloride 300mg tablets where it was prescribed generically; and one for Tramulief 300mg tablets where it was prescribed by brand.
Within a single month, we calculate the average price-per-dose achieved by each practice for each presentation (generic and branded combined). We then approximate a realistic, best-case price-per-dose by finding the average price-per-dose which is at the cheapest decile for that presentation.
We then work out what each GP Practice or CCG could save if it prescribed that presentation at the best decile.
The current model assumes that if pill A1 is expensive and pill A2 is cheap, the perfect practice could switch to prescribing 100% of pill A2.
Most saving opportunities are possible because:
The Tariff price is usually based on the list prices of one or more supplier. Particularly in a crowded market where many brands and manufactured generics are available, the Tariff price may not reflect the cheapest prices available.
This happens most frequently in Category C, where the Tariff price is usually based on the list price of the originator brand. For example, as of Jan 2016, the Tariff price for generic Flucitasone/Salmeterol is based on the GSK list price for Seretide, which is up to 25% higher than the cheapest brand available.
The quickest way to make a cost saving in such cases is to switch to prescribing by the cheapest brand.
This is a variation of the list-price phenomenon described above. Where (for example) both tablets and capsules are available for a Category C drug, the originator brands may have abritrarily different list prices, which are then reflected in the Tariff.
For example, the Tariff price of Venlafaxine 150mg MR capsules is twice that of the equivalent tablets. There is no underlying market logic to the cost differential: the cheapest brand available is, in fact, capsules.
In this case, a large cost saving could be achieved by prescribing tablets generically, but an even larger cost saving could be achieved by prescribing a specific brand of capsulte.
For example, at the time of writing there is widespread prescribing of Nasonex, which is more than 3 times the price of the manufactured generic equivalent listed in the Tariff.
The cost saving here would be achieved by switching to prescribing by the generic presentation (for which cheap manufactured generics are available).
Expensive brands may be prescribed out of patient preference, or for historic reasons, or occasionally for clinical reasons (e.g. in the case of many antiepileptic drugs).
This is a price-gouging practice which can be applied to any presentation not listed in the Drug Tariff ("NP8").
A generic presentation which isn't in the Tariff can be invoiced by the dispenser at any available list price. There are various contractual mechanisms which allow a dispenser and a supplier to collude in sharing the profit for arbitrarily-priced drugs.
The fix is either to prescribe a specific, cheaper brand; or wait for the drug to be placed in the Tariff. Sometimes one formulation (e.g. tablets) will be in the Tariff, but another (e.g. capsules) isn't. In these cases keeping the prescription generic but switching the formulation will also realise a saving.
Two kinds of testing strips (Blood glucose, and Urine) will never appear in the tariff as products in the tariff must be listed against an official non-proprietary name, and these don't have one.
Such categories typicaly contain a very large range of functionally equivalent products with a correspondingly large range of prices.
Many drugs are only available from one supplier. In these cases, particularly where the supplier is outside the PPRS, the list price of a drug is liable to go up unexpectedly for a range of reasons, the simplest being market forces such as resource scarcity, or reducing efficiencies of scale for items with reducing demand.
Sometimes it is alleged that monopoly positions are exploited by arbitrarily inflating costs, as happened in the case of Epanutin. When the drug has a very narrow therapeutic index (such as Epanutin), the supplier will often include their name in the authorised marketing name of the product (e.g. Phenytoin Sodium Flynn). This means GPs are able to continue to prescribe the product by name (as they must for drugs with such a narrow therapeutic index), but despite being generic, no identical imported drugs can be substituted as the name as prescribed includes a trade mark.
It is very difficult to detect the difference between these two reasons for price hikes in the data, and the potential savings are similarly hard to identify from data. Sometimes there is a clinically equivalent alternative drug available, but in drugs with a very narrow therapeutic index (e.g. Epanutin) a switch is not possible.
Examples of drugs which (1) are only available from a single supplier; (2) have relatively small but significant spending; (3) have seen recent, extreme price hikes (>300%); and (4) for which there are realistically no therapeutic substitution, include:
As can be seen from the above, in most cases a saving should be achievable by switching to the cheapest brand available, rather than leaving the selection to pharmacists. In some cases a saving will also be possible by switching formulation (e.g. from capsules to tablets). Where we have included formulation swaps in our calculation, we indicate this in the data.
However, the variability in price-per-dose is sometimes outside the influence of the prescriber. As a result, it is likely that many of the projected savings will not be 100% achievable. Some of the reasons include (in rough order of their relative likely effect):
Due to these variables, an element of judgement is likely to be required to decide how much of a theorical saving is possible in practice. We flag where products may be subject to any of these conditions where it is possible to do so automatically.
We have mitigated some of this effect by assuming that the best-peforming practices are already achieving all the practically possible savings, and selecting the price-per-dose achieved by the practice at the cheapest decile. An alternative, less conservative approach, is to define a best-case price-per-dose by taking the average price-per-dose at the cheapest decile for a presentation (brands and generic) across the entire dataset. This has the effect of removing the assumption that the best-performing practices are already achieving the best real-world savings possible.
Compare the two approaches:
This is discussed further in issues #2 and #16.
The per-CCG spreadsheet has the following column names:
bnf_code: the code for the generic equivalent of that presentationbnf_presentation: the name of that presentationbnf_chemical: the chemical of the presentationcategory: the NHS Drug Tariff category of the presentationlowest_decile: the best price-per-dose we take as achievablequantity: the quantity of that presentation prescribed in the given monthprice_per_dose: the achieved price per dose in the given monthpossible_savings: the total possible savings for the given monthformulation_swap: formulations we have considered equivalent for the analysis (see Formulation Swaps, below)flag_imported: set to 1 if the product is an importflag_broken_bulk: set to 1 if the product can have broken bulk rules appliedflag_non_bioequivance: set to 1 if there are biolequivance issues between products with the generic categoryMost cost savings can be realised by prescribing by brand. This is (to some extent) problematic, as it goes against historic clinical recommendations and the way the system was originally set up to promote competitive markets.
Also, because dispensing contractors' profits are guaranteed, any savings to the NHS made by switching to cheap brands will ultimately be offset by increased costs for Category M medicines or decreased discount reductions.
However, in a situation where CCGs are the primary gatekeepers of spending it is unavoidable that they have the most incentives to become the market driver of costs. In the current situation, where some CCGs have more information and capability to act on it than others, then those CCGs will benefit disproportionately. Conversely, pharmacy contractors in such areas will have their profits reduced disproportionately to other contractors.
Therefore, any system which equalises access to cost-saving opportunities increases equity across the country in terms of CCG funding.
However, because prescriptions are "sticky" (patients and GPs are reluctant to switch them), CCGs are likely to be inefficient market actors, compared with the dispensing contractors when dealing with generically prescribed items.
Clinically, generic prescribing is recommended to avoid unecessary cognitive overhead which may lead to confusion or mistakes. For example:
In terms of market incentives, branded prescribing is not optimal because:
On the other hand, there are market efficiency problems around generic prescribing too - most significantly in "NP8" products (see above for more).
The practice of branded prescribing, when evenly applied across the country, should not unduly affect contractor profits as these are guaranteed and applied via Category M and bulk discount deductions.
However, when unevenly applied, some contractors will be significantly disadvantaged compared with others.
For this reason, the PSNC is "completely against" the practice of generic prescribing.
On the other hand, if branded prescribing cost-saving opportunities were applied evenly across the country, this would no longer be an issue.
Our algorithm assumes all products coded as equivalent in the BNF can be considered as possible substitutions.
However, there are some inconsistencies in the BNF coding which break this assumption. For example, all Gluten Free Breads are coded as equivalent (sliced bread, wholemeal bread, etc).
There are currently four such classes we know of; these are completely excluded from our analysis. The problem is fully described in issue #9.
Many different formulations (with different BNF coding) are bioequivalent, so should be considered as alternatives when considering potential cost savings.
For example, 100mg Tramadol M/R capsules and tablets have different generic BNF codes, but could be substituted; in Sept 2016, tablets cost on average 38p each in capsules 24p each.
Therefore, we consider a range of formulations as equivalent. We've created a manually-curated spreadsheet of mappings, using the following guidelines:
See issue #11 (and issue #8) for discussion.
There are over 1000 products without a generic equivalent listed in the BNF (identified by codes ending in A0; and in dm+d with a Basis of name: Other for their VMP).
Some of these can be meaningfully compared with each other, despite not having a formally coded generic equivalent. For example, Blood Glucose Test Strips are clinically equivalent but have no generic equivalency coded in the BNF (though they are grouped with the Blood glucose biosensor testing strips VMP in dm+d.
We can group them together by BNF subparagraph (Glucose Blood Testing Reagents) as every product within that paragraph is a blood glucose test strip, and the quantity field consistently refers to number of strips.
We have identified one other subparagraph that can be treated in the same way -- Urine Testing Reagents).
See issue #1 for a full discussion.
quantityTo calculate possible savings, we work out a price-per-dose by dividing the net cost for each product by its quantity.
However, in a handful of products, the definition of quantity is not consistent, giving false positives for price variation. We exclude these completely from our analysis.
See issue #12 for discussion.
Various special products are coded as being in "unspecified" categories. We exclude these. See issue #14 for discussion.
The current calculation relies on a table "tariff" for the tariff categories. This was correct as of September 2016, but changes monthly. That table should be updated before doing any analyses by tariff category.
Tests can be run with
py.test tests/test.py