Price versus clinical guidelines in primary care statin prescribing: a retrospective cohort study and cost simulation model

Objective To investigate the relative impact of generic entry and National Institute for Health and Care Excellence clinical guidelines on prescribing using statins as an exemplar. Design Retrospective analysis of statin prescribing in primary care and cost simulation model. Setting Royal College of General Practitioners Research and Surveillance Centre (RCGP R&SC) database and Prescription Cost Analysis (PCA) database. Participants New patients prescribed statins for the first time between July 2003 and September 2018. Main outcome measures Shares of new patients prescribed one of the five statins available in the British National Formulary, and cost of prescribing statins to new and existing patients in primary care in England. Results General trends of statin’ prescriptions were largely driven by a decrease in acquisition costs triggered by patent expiration, preceding NICE guidelines which themselves did not seem to affect prescription trends. Significant heterogeneity is observed in the prescription of the most cost-effective statin acrossGPs. A cost simulation shows that, between 2004 and 2018, the NHS could have saved £2.8bn (around 40% of the £6.3bn spent on statins during this time) if all GP practices had prescribed only the most cost-effective treatment. Conclusions There is potential for large savings for the NHS if new and, whenever possible, ongoing patients are promptly switched to the first medicine that becomes available as generic within a therapeutic class as long as it has similar efficacy to still-patented medicines.

compares the data over time in our two data sources: Panel (a) on the left shows the figures from the PCA dataset between 1998-2018. Panel (b) on the right shows the figures from RCGP R&SC database from 2004-2018. Top panels display the total quantity in terms of daily-defined-doses (DDDs) while the bottom panels display the shares of each of the five statins in the market. The similarity in the trends reported in panel (a) and (b) confirms that the RCGP R&SC database is a representative sample of national data of statins prescription.

Spending Savings Simulation Exercise Methodology
This appendix describes the methodology used for the spending savings simulation exercise, by which we estimate the potential savings for the NHS that could have been achieved if GPs had prescribed simvastatin or atorvastatin as active ingredients, whenever these two medicines were the prescribing standard in this market according to the observed preferences of GPs and the recommendations in national guidance. We start by describing the computations of actual and hypothetical cost for the first scenario, in which the analysis refers to the first prescription episode, i.e., the first 28 days of drug treatment, for new patients only; and then the second scenario, in which we apply the same methodology to all prescriptions issued to all existing patients being treated in every period.

First scenario
To compute actual and hypothetical cost we use information on the number of new patients treated, their initial drug treatment (i.e., a specific statin and strength), and a measure of each treatment's acquisition cost to the NHS per day of treatment. From the RCGP R&SC database, we count the number of new patients being prescribed statin treatment s in period t for the first time, denoted by .
From the Prescription Cost Analysis (PCA) series, containing data on all medicines prescribed and dispensed and their corresponding cost to the NHS at the national level, we retrieve a measure of the actual acquisition cost of each statin treatment. To compute the average acquisition cost of treatment s in period t, denoted by , we take the ratio between the Net Ingredient Cost ( ) and the corresponding Total Quantity ( ) prescribed of each different strength of statin, that is = . Cost figures are then expressed in constant 2018 GBP using the GDP deflators at market prices (see https://www.gov.uk/government/statistics/gdp-deflators-at-market-prices-and-money-gdp-march-2019-spring-statement) Since data on the total number of new patients starting treatment on each statin nationally is not publicly available, we estimate such figure by combining information from the RCGP R&SC database (which is a nationally representative sample of GP practices in England) with national aggregated data from the Prescription Cost Analysis series. Concretely, we compute the total number of new patients nationally, denoted by � , as follows: � = × , where denotes the total quantity of each statin treatment prescribed in every period in the RCGP R&SC database. Indeed, since the RCGP R&SC sample of GP practices is representative of the English general practice sector, then the ratio of new patients to total quantities prescribed in both data sources should be equivalent. Aggregating � over all treatments at the year-level, � = ∑ � , results in the figure reported in column (1) of Table 1, i.e., the estimated number of new patients treated with statins in each year.
Finally, we compute the actual cost of first prescription episodes for each statin treatment s in every period by multiplying the total number of new patients on each treatment � with the cost per day of treatment times 28, that is = � × × 28. Then we aggregate over all treatments at the year-level, = ∑ , which is the figure reported in column (2) of Table 1.
As explained above, practitioners' preferences when treating patients for CVDs risk moved towards simvastatin since its patent expiration (May 2003) until atorvastatin's patent expiration (May 2012); and from then onwards, they tended towards atorvastatin. Our cost simulation exercise extremes this observed behaviour by asking what would have been the cost savings if either simvastatin or atorvastatin had been the active ingredients originally prescribed to new patients, whenever these two medicines were the prescribing standard in specific periods. Accordingly, the hypothetical cost is constructed by substituting the originally prescribed treatment , with a therapeutically similar one, denoted by * , containing simvastatin for those first-time prescriptions issued between 2004 and May 2012, or atorvastatin for those issued after May 2012.
The therapeutic similarity criteria we use is based on the ability of each strength of each drug (e.g., 1 tablet of atorvastatin 20 mg. a day, 1 tablet of simvastatin 40 mg. a day, etc.) in reducing low-density lipoprotein (LDLP) cholesterol levels per day of treatment. The percentage reduction in low density lipoprotein cholesterol is used in NICE's CG-181 to group the five statins (and each of their corresponding strengths) according to their intensity. The relationship between the strengths of the statins and reduction in LDLP cholesterol is stated in NICE's CG-181, which in turn is based on the paper by Law et al. (2003). 16 A reproduction of this information is presented in Table B1. To make this operative, for each level of percentage reduction in LDLP cholesterol achieved by the originally prescribed treatment, i.e., a drug-strength pair, we look for the closest strength of both simvastatin and atorvastatin that achieves a similar level in LDLP reduction to the originally prescribed one. The correspondence between original treatments and the substitutes is presented in Table B2.
Columns (3) and (4) show the strength of simvastatin and atorvastatin, respectively, that achieves the closest percentage reduction in LDLP cholesterol than the original treatments listed in columns (1) and (2). For example, if a patient was prescribed atorvastatin 10 mg a day for treatment initiation before 2012, the hypothetical prescription for this patient is a treatment of simvastatin 40 mg a day, as both achieve a reduction of 37% in LDLP cholesterol. Second example, if a patient was prescribed rosuvastatin 10 mg for treatment initiation after 2012, then the hypothetical prescription for this patient is a treatment of atorvastatin 20 mg, as both achieve a reduction of 43%.  Finally, the hypothetical cost is computed by multiplying the total number of new patients times the cost of the therapeutically similar treatments * times 28, that is = � × * × 28. Then we aggregate over all treatments at the year-level, = ∑ , which is the figure reported in column (3) of Table 1.

Second scenario
The second scenario considers not only first-time prescriptions for new patients, but all prescriptions for all existing patients treated with statins. For this, we use the information on total quantity and spending from the PCA database. The actual cost is obtained by aggregating spending on all statins prescribed in each year. The hypothetical cost is computed by replacing the per unit cost of the original treatment (statins and strength) with the corresponding cost of the therapeutically similar treatment (either simvastatin or atorvastatin), as described above. Additionally, to provide an estimate of the total number of all existing patients treated with statins in every period, � (the figure reported in column (6)