Impact of EMA regulatory label changes on systemic diclofenac initiation, discontinuation, and switching to other pain medicines in Scotland, England, Denmark, and The Netherlands

Abstract Purpose In June 2013 a European Medicines Agency referral procedure concluded that diclofenac was associated with an elevated risk of acute cardiovascular events and contraindications, warnings, and changes to the product information were implemented across the European Union. This study measured the impact of the regulatory action on the prescribing of systemic diclofenac in Denmark, The Netherlands, England, and Scotland. Methods Quarterly time series analyses measuring diclofenac prescription initiation, discontinuation and switching to other systemic nonsteroidal anti‐inflammatory (NSAIDs), topical NSAIDs, paracetamol, opioids, and other chronic pain medication in those who discontinued diclofenac. Absolute effects were estimated using interrupted time series regression. Results Overall, diclofenac prescription initiations fell during the observation periods of all countries. Compared with Denmark where there appeared to be a more limited effect, the regulatory action was associated with significant immediate reductions in diclofenac initiation in The Netherlands (−0.42%, 95% CI, −0.66% to −0.18%), England (−0.09%, 95% CI, −0.11% to −0.08%), and Scotland (−0.67%, 95% CI, −0.79% to −0.55%); and falling trends in diclofenac initiation in the Netherlands (−0.03%, 95% CI, −0.06% to −0.01% per quarter) and Scotland (−0.04%, 95% CI, −0.05% to −0.02% per quarter). There was no significant impact on diclofenac discontinuation in any country. The regulatory action was associated with modest differences in switching to other pain medicines following diclofenac discontinuation. Conclusions The regulatory action was associated with significant reductions in overall diclofenac initiation which varied by country and type of exposure. There was no impact on discontinuation and variable impact on switching.

The PHARMO Database Network is a population-based network of electronic healthcare databases and combines data from different primary and secondary healthcare settings in the Netherlands. To address the objectives of the present study the Out-patient Pharmacy and the GP Database were used. The Outpatient Pharmacy Database of the PHARMO Database Network comprises GP or specialist prescribed healthcare products dispensed by the out-patient pharmacy (population 4.2 million in 2016). These data can be linked on a patient-level using probabilistic linkage to other databases. Data on indication and contraindications were obtained from the GP Database for a population of approximately 1 million. This database comprises data from electronic patient records registered by GPs. Dispensing data is recorded as ATC and diagnoses as ICPC (International Classification of Primary Care) codes or entered as free text. ICPC codes can be mapped to ICD codes. Data linked with the GP Database was available from 2007 up to 2016.

Exposures
We used prescription data to calculated non-overlapping periods of exposure and non-exposure to diclofenac over a patients entire observation period. If a prescription occurred before a previous prescription's end date, the end date of the previous prescription was moved to the day before the current prescription's start date so that the exposure periods did not overlap. The exposure file was used to create a file of exposure episodes by merging exposure periods separated by less than 92 days of non-exposure and splitting periods of non-exposure into the first 92 days (labelled recently discontinued) and the rest (unexposed). Using 92 days or longer to define discontinuation guaranteed that no patient could discontinue and re-initiate diclofenac in the same quarter. [3]

Diclofenac initiation
Diclofenac initiation was defined as a prescription for diclofenac with no exposure to diclofenac in the preceding 92 days. The denominator was the number of non-users on the first day of the time period defined as no exposure to diclofenac in the previous 92 days. The numerator was the number of these patients initiating diclofenac in the time period. This was performed for overall diclofenac initiation then stratified by indication, age category and gender. Age was classified as 0-17, 18-29, 30-39, 40-49, 50-59, 60-69, 70-79 or 80+. Read, ICD or ICPC codes were used to classify licenced indications: Crystal arthropathies; Inflammatory arthropathies including Pain and inflammation in musculoskeletal disorders, and Pain and rheumatic disease, including juvenile idiopathic arthritis which were subdivided into osteoarthritis and other inflammatory arthropathies; and osteoarthritis. The classification was based on any record dated before the end of the time point.
One-off users were defined as patients prescribed a single diclofenac prescription only. To define sporadic and chronic users we calculated a possession ratio for each patient defined by using the number of days prescribed (or supplied) assuming a standard daily dose divided by the number of days between diclofenac prescriptions. We defined sporadic users as patients with a diclofenac possession ratio of less than 1 standard day of therapy per 3 days. Patients with a diclofenac possession ratio of more than 1 standard day of therapy per 3 days were defined as chronic users. Standard daily doses were assumed to be: For estimating the duration of each diclofenac prescription, we assumed a standard diclofenac treatment regimen for each patient and prescription as if they were taking it with complete adherence. For tablets/capsules we used a total daily dose 150mg diclofenac. For example, a standard prescription consisting of 50mg strength tablets/capsules we therefore divided the quantity of tablets/capsules per prescription by this standard regimen (i.e. 3) to provide the standard duration of therapy in days. We measured trends in the average standard duration of therapy for prescriptions issued within each time period, before and after the date of the regulatory intervention.

Diclofenac discontinuation
Overall diclofenac discontinuation was defined as the number of patients with a prescription for diclofenac with no exposure to diclofenac in the 92 days following the date of that diclofenac prescription. The denominator was the number of patients prescribed diclofenac in the time period. The numerator was the number of these patients discontinuing. [4]

Switching to other alternative medicines
A switch to an alternative medicine group was defined as those patients who discontinued diclofenac (as defined above) and who then initiated a drug in the class listed in supplementary table S5. Initiation of an alternative medicine was defined as the first prescription of a drug in that class prescribed within 92 days following the date of the last diclofenac prescription.

Date of the regulatory intervention
For interrupted time series regression analysis, the date of the regulatory intervention was pre-specified as 28 June 2013.

Analytical approach
The primary analysis used quarterly time periods. For each year these were defined by the following dates:  1 st January to 31 st March = Quarter 1  1 st April to 30 th June = Quarter 2  1 st July to 30 th September = Quarter 3  1 st October to 31 st December = Quarter 4 The proposed primary analysis used interrupted time series regression to fit time trends to each series of time period data for each country. Using regression modelling we evaluated: 1. The baseline slope before the intervention time point 2. The change in slope from the baseline trend to the post-intervention trend 3. The immediate change associated with the intervention time point Before fitting all regression models, the data was visualised graphically. The effect of the intervention for each country was represented either by a step function, or by a continuous linear function representing gradual implementation (interrupted time series analysis). This choice, and whether it is necessary to model any trends prior to the intervention time point, was decided on visual inspection of the data. The analysis was done and is reported by data source.
To measure change in trends over time after the pre-specified date of the regulatory intervention, June 2013, this was achieved by fitting interrupted time series (ITS) models with a joint point at this date. 14 They were parameterised so that one parameter estimated the change in slope after vs before the join point (the coefficient of a time variable counting the number of quarters since the intervention and set to 0 before it), and another estimated any step change at the join point (the coefficient of a variable set to 1 after the intervention and 0 before it). Where there was evidence of discontinuities at other times, in either absolute rates or their slopes, the range of data was trimmed to periods immediately before and after June 2013 when trends were approximated to be linear.
We therefore assumed normal error distributions and fitted trends using weighted linear regression, the weights being the denominators in each proportion. We found no increase in the magnitude of residuals with increasing fitted values and therefore did not transform the data to log units for analysis. We performed sensitivity analyses comparing models that did and did not allow for autocorrelation and found trivial differences in the parameter estimates. The analyses presented in this report did not allow for autocorrelation. [5] SUPPLEMENTARY FIGURES Figure S1. Trends in diclofenac initiation in Denmark, the Netherlands, England and Scotland by indication. Figure S2. Trends in diclofenac initiation in Denmark, the Netherlands, England and Scotland by age.
[7] Figure S3. Trends in diclofenac initiation in Denmark, the Netherlands, England and Scotland by gender.
[8] Figure S4. Trends in diclofenac initiation in Denmark, the Netherlands, England and Scotland by exposure type.
[9] Figure S5. Trends in the mean duration of diclofenac prescriptions in Denmark, the Netherlands, England and Scotland in days.