This was a retrospective cohort study using primary care data from the Clinical Practice Research Datalink (CPRD). CPRD is a large database which contains anonymized individual patient-level medical and demographic information on approximately 8% of the UK population from more than 630 general practices. The patient population captured in the database is broadly representative of the UK population and CPRD is a reliable resource for longitudinal epidemiological research.
The total study period was between January 1, 2006 and February 25, 2011 (end of data records). The follow-up period for each patient starts after the first non-insulin antihyperglycaemic drug is prescribed (index date) and ends at the earliest of the following: date of last data collection for the practice, date of transfer out (patient leaves the practice), date of probable pregnancy, date of death of the patient, or end of study period.
Inclusion and Exclusion Criteria
We required that each patient had a prescription of antihyperglycaemic therapy (OAD or injectable) other than insulin between January 1, 2006 and December 31, 2010. Additionally, patients had to be permanently registered with a CPRD participating practice and have a valid date of birth and gender information (acceptable patients). Furthermore, all patients had to have at least 12 months of computerised data prior to the index date and the index date was later than the practice's up-to-standard (UTS) date; this is the date from which CPRD considers the practices data to be sufficiently exhaustive to be used for research purposes.
We excluded patients pregnant at index date, with a specific diagnosis of type 1 diabetes mellitus at or prior to index date, a prescription of an OAD at any time prior to index date, a prescription of insulin prior to index date, who left the practice on index date or with a diagnosis of polycystic ovary syndrome (PCOS) if metformin is the index OAD (unless the patient was suffering both T2D and PCOS).
Variables and Outcomes
The main treatment pattern outcomes were discontinuation, switching, augmentation and initiation of insulin therapy. In order to define the treatment pattern outcomes at all times, it was necessary to censor each patient's follow-up by 90 days since it is not possible to assess whether a patient has discontinued or not, or whether a switch has effectively occurred or not, if the last prescription occurs near the end of the patient's records. Treatment patterns occurring subsequently to the first change in the OAD therapy are not in the scope of this project given the myriad of possible treatment patterns.
Duration of a prescription is key to defining treatment patterns. If the pack size or daily dose were missing for a prescription then the duration was imputed from typical durations observed for that drug or drug class. Failing this, and where appropriate, a value of 30 days was used.
Discontinuation. Discontinuation was firstly defined for the index OAD and then for all OADs; the latter allows for a patient to switch between OADs until all OAD therapy is discontinued. A patient was defined to have discontinued his/her index OAD if a gap occurs of at least 60 days between the expiry of the last prescription and the start of the next prescription of that particular OAD. For discontinuation of all OAD therapy, a gap of at least 60 days was again used irrespective of the type of OAD drug and whether switching in the OAD therapy had occurred during the initial period. Time to discontinuation (i.e., persistence) used the number of days from the index date until discontinuation date.
Switching & Augmentation. A switch is defined when a new OAD replaces the previous one. For switching, diabetes therapy has to have been continuous, hence the new OAD (or insulin) has to start before the current OAD has been deemed to have discontinued. Hence, the difference between switching and augmentation is the degree of overlap between the two OAD therapies: if a new OAD is prescribed on or after the date of the last prescription of the index OAD (i.e., the overlap between the new and old drugs is less than the duration of one prescription) then we conclude that a switch has occurred, but only if the patients receives at least two prescriptions of the new drug. Augmentation occurs if the new OAD is introduced before the date of the last prescription. Again there must be at least two prescriptions of the new drug. Augmentation is defined as a prescription of a second OAD drug in addition to the index OAD without the discontinuation of the original drug.
Rates. The rates for discontinuation used as denominator time from index date until the earliest of either discontinuation or 90 days before end of follow-up whilst for switching and augmentation the follow-up time was terminated when the index OAD was discontinued. Initiation of insulin was assessed between index date and end of follow-up.
Other Variables. We assessed variables of interest at baseline, sociodemographic information, comorbidities, OAD at index date (type and monotherapy or in combination) and laboratory test results including HbA1c (mmol/mol). Comorbidities included coronary heart disease (CHD), cerebrovascular disease, congestive heart failure (CHF), nephropathy, neuropathy, retinopathy, and hypertension, defined by the presence of the corresponding diagnostic or test Read codes. In order to evaluate the trends in HbA1c, the HbA1c results were assigned to specific time periods. The baseline HbA1c was the most recent value recorded in the six months prior to index date. For each year since index date we assigned the HbA1c that was closest to the anniversary of the index date within the six months before or after this date.
We described baseline characteristics of all study patients at the time of the first administration of an OAD. These characteristics included age, gender, HbA1c and other laboratory test results, relevant comorbidities and OADs, and other co-medication use. Numbers and percentages of patients were reported for categorical variables, and the mean and median with appropriate measures of spread were given for continuous variables.
Multinomial logistic regression models were applied to estimate the association between each covariate and the likelihood of each of the treatment outcomes as compared to no change in treatment during the timeframe of interest. The timeframes for which models were applied included the baseline period (first six months after index date) and yearly intervals since index. It was not possible to execute the models for time periods beyond one year since index date due to the decreasing numbers of patients over time.
Most covariates comprise of the baseline characteristics. However, HbA1c, age and time since diagnosis were updated to reflect time period being analysed. The most parsimonious models are reported.
Data programming, management and analyses were carried out using SAS (version 9.2).
The study was approved by the Independent Scientific Advisory Committee (ISAC) that provides advice to the Medicines and Healthcare Regulatory Agency on database research (ISAC protocol 10_186). The ISAC is a non-statutory expert advisory body to provide advice on research related requests to access data provided by CPRD (and data from the Yellow Card Scheme as well).