Time trends in prescribing of type 2 diabetes drugs, glycaemic response and risk factors: A retrospective analysis of primary care data, 2010–2017

Aim To describe population‐level time trends in prescribing patterns of type 2 diabetes therapy, and in short‐term clinical outcomes (glycated haemoglobin [HbA1c], weight, blood pressure, hypoglycaemia and treatment discontinuation) after initiating new therapy. Materials and methods We studied 81 532 people with type 2 diabetes initiating a first‐ to fourth‐line drug in primary care between 2010 and 2017 inclusive in United Kingdom electronic health records (Clinical Practice Research Datalink). Trends in new prescriptions and subsequent 6‐ and 12‐month adjusted changes in glycaemic response (reduction in HbA1c), weight, blood pressure and rates of hypoglycaemia and treatment discontinuation were examined. Results Use of dipeptidyl peptidase‐4 inhibitors as second‐line therapy near doubled (41% of new prescriptions in 2017 vs. 22% in 2010), replacing sulphonylureas as the most common second‐line drug (29% in 2017 vs. 53% in 2010). Sodium‐glucose co‐transporter‐2 inhibitors, introduced in 2013, comprised 17% of new first‐ to fourth‐line prescriptions by 2017. First‐line use of metformin remained stable (91% of new prescriptions in 2017 vs. 91% in 2010). Over the study period there was little change in average glycaemic response and in the proportion of people discontinuing treatment. There was a modest reduction in weight after initiating second‐ and third‐line therapy (improvement in weight change 2017 vs. 2010 for second‐line therapy: −1.5 kg, 95% confidence interval [CI] −1.9, −1.1; P < 0.001), and a slight reduction in systolic blood pressure after initiating first‐, second‐ and third‐line therapy (improvement in systolic blood pressure change 2017 vs. 2010 range: −1.7 to −2.1 mmHg; all P < 0.001). Hypoglycaemia rates decreased over time with second‐line therapy (incidence rate ratio 0.94 per year, 95% CI 0.88, 1.00; P = 0.04), mirroring the decline in use of sulphonylureas. Conclusions Recent changes in prescribing of therapy for people with type 2 diabetes have not led to a change in glycaemic response and have resulted in modest improvements in other population‐level short‐term clinical outcomes.


| INTRODUCTION
Prescribing of glucose-lowering therapies for patients with type 2 diabetes has changed markedly in recent years. International guidelines have been updated to include a much greater choice of agents when additional therapies after metformin are required to achieve glycaemic control. [1][2][3][4] Newer drug classes including dipeptidyl peptidase-4 (DPP-4) inhibitors, sodium-glucose co-transporter-2 (SGLT2) inhibitors and glucagon-like peptide-1 (GLP-1) receptor agonists are now established alongside the longstanding options sulphonylureas, thiazolidinediones and insulin. Choice between these agents is left largely to the clinician and patient. Recent studies show that there have been marked changes in which agents are initiated after metformin, with declining use of sulphonylureas and increasing and earlier use of DPP-4 inhibitors and SGLT2 inhibitors in both the United States, Europe and the United Kingdom. [5][6][7][8] Although studies have suggested that the glucose-lowering effectiveness of agents typically added to metformin may be comparable, 1,9,10 there are well established differences between the different drug classes in weight change and side effects. GLP-1 receptor agonists and SGLT2 inhibitors are associated with weight loss, whereas DPP-4 inhibitors are weightneutral and sulphonylureas can promote weight gain. 9,10 Hypoglycaemia risk is greater with sulphonylureas and insulin relative to other agents. 9 Despite these known differences in non-glycaemic effects between agents, evidence of the impact of recent changes in prescribing on population-level patient outcomes is limited. 5,7,11,12 In the present study we aimed to describe changes in prescribing of glucose-lowering drugs for patients initiating first-to fourth-line therapy between 2010 and 2017 in the United Kingdom, a setting where prescribing does not reflect the ability of patients to pay. We further examined population-level time trends in the short-term clinical outcomes of glycaemic response, weight change, blood pressure change, hypoglycaemia and treatment discontinuation.

| Data source and data extraction
We conducted a population-based analysis of anonymized primary care data from the UK Clinical Practice Research Database (CPRD). CPRD is a population-representative database including demographic, clinical and prescription primary care records of patients. 13 Although CPRD includes full prescription records, no data on drug dispensation are available. CPRD has been extensively used to study drug prescribing and patient outcomes in type 2 diabetes. 14 We analysed data from the January 2018 release of CPRD, including all practices that were still contributing to CPRD in 2017, to ensure that changes in prescribing did not reflect changes in the practices captured in CPRD over the study period. We classified glucose-lowering drugs into drug classes according the British National Formulary sections 6

| Study population
We extracted the clinical and prescription records of all patients with type 2 diabetes who started at least one glucose-lowering drug for the first time ever between 1 January 2010 and 31 December 2017 and met CPRD quality assurance criteria. Inclusion criteria and data ascertainment were the same as those included in our previously reported CPRD cohort profile. 16 Type 2 diabetes was defined largely on the basis of prescriptions for non-insulin diabetes therapies rather than diagnostic medical codes, to minimize coding errors. 17 We excluded patients with diagnostic codes for other forms of diabetes or polycystic ovary syndrome, which can be treated with metformin. To remove patients with type 1 diabetes, whose disease may have been miscoded as type 2, we excluded patients with an age at diagnosis of <35 years or who were on insulin treatment within 12 months of diagnosis. Consequently, patients with type 2 diabetes whose firstline therapy was insulin were not included. We defined date of diabetes diagnosis as the earliest of: first prescription for a non-insulin diabetes therapy; first glycated haemoglobin (HbA1c) result ≥48 mmol/mol (6.5%); or first diabetes diagnostic code.

| Study design
The study exposure was a new first-to fourth-line drug prescription record for a patient within the study period. New drug prescriptions The primary unit of analysis was line of therapy. This meant individual patients who started more than one new therapy over the study period contributed to the analysis more than once with different lines of therapy (Flowchart S1).

| Study outcomes
For each line of therapy, we evaluated annual time trends in the drug classes initiated, and time trends in changes in HbA1c, weight, systolic and diastolic blood pressure, hypoglycaemia rates and treatment discontinuation after therapy start. To evaluate all outcomes we used a "new-user" design, which mitigated immortal time bias. 18 Patients were followed up from their drug start date until there was any change in diabetes therapy or the end of the study period specific to each outcome. A change in therapy could be the addition of a new glucose-lowering drug or the stopping of the drug of interest or any concomitant glucose-lowering drug. Patients were considered to have stopped a drug if there was a subsequent gap in prescribing of that drug for at least 6 months. 16 We defined glycaemic response (the change in HbA1c), weight change and blood pressure change as the absolute change from baseline to 6 months (6-month measure minus baseline measure). For glycaemic response, baseline HbA1c was defined as the closest HbA1c to the drug start date in the 3 months prior to the drug start date. HbA1c at 6 months was defined as the closest HbA1c value to 6 months after the drug start date (±3 months). Glycaemic response was only valid if there were no changes in glucose-lowering therapy between 2 months prior to the baseline HbA1c and the date of the 6-month HbA1c. The same approach was used for weight change and blood pressure change.
We defined hypoglycaemia as the first Read code for hypoglycaemia up to 2 years after starting a line of therapy, using a previously published Read code list for hypoglycaemia. 19 Because of the low number of hypoglycaemia events captured in primary care we We examined treatment discontinuation by estimating the proportion of patients who stopped a therapy within 3 months, 6 months and 1 year. Six months' follow-up after discontinuation was required to determine no new prescriptions were issued. All data extraction and analysis was conducted in STATA v14.0.

| Sensitivity analysis
We repeated all outcomes analysis using change in each measure from baseline to 12 months as the outcome in a distinct cohort of patients  Table S1. The mean baseline HbA1c at which second-to fourth-line therapy was initiated increased over the study period; mean baseline weight increased first-line, but there was little difference for other lines of therapy. The proportion of patients with valid measures for inclusion in the analysis of each outcome is shown in the Flowchart S1.

| Changing prescribing of glucose-lowering therapy
We found marked changes in relative prescribing of second-to fourth-line therapy (Figure 1)   Figure S2D).

| Reduction in HbA1c
Average reductions in HbA1c at 6 months were relatively constant over

| Weight change
Although there was a trend towards greater weight loss at 6 months for all lines of therapy, this was most marked with second-and third-  Figure S3A]). There was no change in diastolic blood pressure ( Figure S3B).

| Hypoglycaemia
We observed a decrease in hypoglycaemia rates for patients starting  Figure 4 and Table S2]).

| Treatment discontinuation
Treatment discontinuation at 3 months, 6 months and 1 year after initiating therapy was stable over the period 2010 to 2017 (Table S3).

| Sensitivity analysis
The baseline characteristics of the patients who were excluded as they did not have valid clinical measures were similar to those included in the analysis (Table S4). Time trends for outcomes at 12 months were similar to those at 6 months for glycaemic response ( Figure S4), weight change ( Figure S5) and blood pressure ( Figure S6).
Second-line prescribing trends and patient outcomes in the subset of patients adding a second-line drug to continued first-line metformin therapy (73% of patients included in the primary analysis) were near identical to those in the primary analysis ( Figure S7). Weight change trends were reduced when models were adjusted for drug therapy as a covariate (Table S5A) and, after adjustment for drug, there was no evidence of a difference in risk of hypoglycaemia over time (Table S5B).  ing use of sulphonylureas in this older age group. 28 The changes observed in these studies examining the overall population of patients with type 2 diabetes will lag considerably behind those observed in the present analysis of new therapy initiation, as, once initiated, a glucose-lowering therapy may be continued for decades.

| DISCUSSION
Strengths of the present study include our approach examining new prescribing, which allowed interrogation of time trends whilst accounting for the increasing prevalence of type 2 diabetes, which in the United Kingdom is attributable more recently to declining mortality rather than increasing incidence, 29,30 and means prescribing of glucose-lowering therapy is increasing in absolute terms. 6,31 Our definition of type 2 diabetes should minimize misclassification. 16 The present study provides a near-complete picture of United Kingdom prescribing because, in the United Kingdom, type 2 diabetes is largely managed in primary care. Even new therapy initiated on the advice of a specialist will usually be prescribed by the patients' primary care physician. A limitation of the present study is the weakness in the way hypoglycaemia is recorded. It is likely that many episodes of hypoglycaemia will be missing from a patients' primary care record, as mild hypoglycaemia or more severe hypoglycaemia requiring attendance in secondary care are poorly recorded; however, previous studies have provided useful insight into hypoglycaemia using similar definitions in the same dataset. 32 Although the missing records mean the absolute rates of hypoglycaemia in the present study will be an underestimate, the specificity of our key finding, a relative decrease in hypoglycaemia rates with second-line therapy, where use of sulphonylureas has markedly declined, is reassuring. Whilst the present study provides timely information on population-level trends, further observational studies, building on recent work, will be needed to establish the real-world comparative effectiveness of individual drug classes at different lines of therapy. 10,33 Our results show that prescribing of glucose-lowering therapy in Type 2 diabetes is rapidly changing towards newer, more expensive agents. Changes in prescribing appear to have pre-empted rather than reflected changes in clinical guidelines. 1 In particular, second-line prescribing of DPP-4 inhibitors increased rapidly long before treatment guidelines were updated to position them along sulphonylureas and pioglitazone as second-line options. 1 The positive trends in weight change, hypoglycaemia and blood pressure are likely to have improved the quality of life for patients, and a reduction in hypoglycaemia is also likely to have a cost benefit. 34  In conclusion, the trend towards prescribing of newer, more expensive, glucose-lowering medication in the United Kingdom has coincided, for patients initiating new therapy, with a likely reduction in hypoglycaemia rates and a modest improvement in weight and blood pressure, but little change in glycaemic response or treatment discontinuation. These results demonstrate the potential population-level impact of the rapid changes that are occurring in prescribing of glucose-lowering therapy worldwide.