Patterns of metformin monotherapy discontinuation and reinitiation in people with type 2 diabetes mellitus in New Zealand

Aim To describe the patterns of discontinuation and reinitiation in new users of metformin monotherapy in New Zealand, overall and according to person- and healthcare-related factors. Materials and methods We created a cohort (n = 85,066) of all patients in New Zealand with type 2 diabetes mellitus who initiated metformin monotherapy between 1 January 2006 and 30 September 2014 from the national data collections, and followed them until the earlier of their death or 31 December 2015. Discontinuation was defined as a gap in possession of metformin monotherapy of ≥90 days. We explored patterns of discontinuation and reinitiation using competing risks methods. Results After 1 year of follow-up, 28% of cohort members had discontinued metformin monotherapy at least once; the corresponding figures after 2 and 5 years were 37% and 46%. The proportions who reinitiated metformin monotherapy within 1, 2, and 5 years of their first discontinuation were 23%, 49%, and 73%. Discontinuation after the first reinitiation was common (48% after 1 year). Discontinuation and reinitiation varied by age, ethnicity, and other person- and healthcare-related factors. Discussion Our findings highlight the dynamic nature of metformin monotherapy use, show that substantial periods of non-use are common, and identify priority populations for interventions to facilitate adherence.


Reviewer #1
1. A flow diagram in Suppl. Figure 1 may be more informative with the number of patients affected by each excluding condition.
Thank you to the reviewer for making this helpful suggestion. The study flowchart has been updated to include the number of people in the study cohort at each point in the cohort's derivation.
2. (On a similar note as above,) How many patients were impacted by excluding "patients who permanently discontinued metformin <100 days of initiation?" A total of 3,952 (or 4.4%) of the study cohort at that point in the study cohort derivation were excluded because of the 'permanently discontinued metformin within 100 days of the first metformin dispensing' rule.
3. More information on the medication dispensing data and the insurance system in New Zealand would be helpful to understand the validity of the study. Are all dispensations captured by the database regardless of the insurance type?
Thank you for this suggestion; we always incorrectly assume everyone is familiar with New Zealand's health system! New Zealand has universal health coverage, with most health care publicly funded [1]. This includes prescription medicines. The cost of most medicines used in New Zealand is subsidised, either fully or partially, by the state Pharmaceutical Management Agency (PHARMAC). Private insurance has little involvement with prescription medicines, except in the case of a very small set of high-cost medicines not subsidised fully by the state (such as certain cancer treatments). With regard to medicines commonly dispensed outside of hospitals, prescription medicines can effectively be considered to be publicly funded for all residents. To be paid for a dispensing, community pharmacies must submit claims for payment, including details about a dispensing and the patient dispensed to. This information is collected in the Pharmaceutical Collection (Pharms), which provides the dispensing data used in this study. Because the vast majority of medicines used in the community are state subsidised, including metformin, and community pharmacies are required to claim reimbursement for subsidised dispensings, Pharms provides virtually complete capture of community pharmacy dispensings to the New Zealand population. This is a substantial advantage to using the New Zealand dispensing data collection.
To briefly clarify the nature and capture of Pharms, we have added the following to Section 2.3 (Construction of medication record):

Pharms contains details of community pharmacy dispensings of medications subsidised by the state (virtually all commonly used medications in New Zealand), providing a comprehensive information source on community medication use.
We have also expanded our brief sentence on the capture of Pharms in the Discussion to now read: We are also likely to have identified nearly all metformin dispensings to cohort members. Further information on Pharms and the other data sources cited in this manuscript are available in our previous manuscript based on the larger study [2] and cited in Section 2.3 (Construction of medication record).

Is the medication for oral diabetes treatment predominantly metformin in New
Zealand? What are the percentages of other antidiabetics used as the initial treatment?
Metformin is the predominant initial treatment for type 2 diabetes mellitus in New Zealand. A recent national study looking at initial pharmacotherapy for T2DM found that, in 2015/2016, 85% of New Zealanders with T2DM commenced pharmacotherapy with metformin monotherapy. [3] Metformin with a sulfonylurea (6%) and metformin with an insulin (5%) were the other combinations to occur more than 2% of the time in the data.
We have amended the first sentence of the second paragraph in the Introduction to now read: Metformin monotherapy is the first-line pharmacological treatment for T2DM in New Zealand [4,5] and accounts for approximately 85% of the initial pharmacological agents prescribed for T2DM [3] 5. The authors should explain the rational for including the presence of glucose test within 6 month as well as urinary ALB/CRE ratio test as covariates. What were they looking for and how are they interpreting their association with the outcomes?
We included these two covariates for two main reasons. Firstly, we were interested to see what proportion of people with T2DM initiating metformin monotherapy for the first time actually had a recent glucose test, given that the guidelines for diagnosis and management of T2DM in New Zealand explicitly refer to HbA1c level thresholds. [4] Secondly, we saw these as proxies for monitoring of T2DM by a health professional, with the hypothesis that increased monitoring by a health professional would improve adherence and reinitiation (this has been found elsewhere e.g. [6,7]). These tests are recommended as part of clinical monitoring of people with T2DM in New Zealand, and so we felt they provided a relevant proxy for clinical monitoring and engagement.
This manuscript's findings showed a decreased risk of first discontinuation, increased risk of subsequent reinitiation, and lower rates of discontinuation when these tests were performed in the six months before cohort entry or discontinuation of metformin monotherapy. These associations are consistent with what one would expect based on the studies cited above and the hypothesis that more intensive clinical monitoring improves adherence to and reinitiation of metformin monotherapy.
We have added the following to the bottom of Section 2.5 (Covariates) to clarify why the covariates were selected: These covariates were chosen because they represented important population groups, were factors identified in the literature as potentially influencing metformin adherence, or were tests recommended in New Zealand as part of clinical monitoring for people with T2DM.

Reviewer Two
1. I am only concerned that, currently, after 7 years, the prescribing patterns in New Zealand may have changed. It would therefore be of very interest to conduct a second study, using the same methodology, in a timeframe of 2014-2020.
We agree that it would be good to assess patterns of metformin monotherapy using more recent data not just to examine changes in prescribing patterns but also to monitor whether inequities between groups have changed. For the current study, we were limited in the recency of dispensing data by the lag in the availability of complete datasets (particularly hospitalisations and mortality datasets) that we relied upon for other parts of the wider project not presented in this paper. We hope that this study will serve as an important baseline for future research looking at longer time trends.
We would note that guidelines for initiating pharmacological therapy for T2DM have not changed in New Zealand since the study.
2. Since the present work focuses on the evaluation of pattern of discontinuation and reinitiation to treatment, two phases of medication adherence, it is necessary, in the introduction, to better define the concept of adherence to treatment by citing the current and recognized EMERGE guidelines on Medication Adherence (available from: https://www.espacomp.eu/project/emerge-guidelines/), explaining the 3 different phases of adherence; in addition, in the methodology, it should be subsequently indicated in detail which of the three phases should be studied.
We are familiar with the EMERGE Guidelines [8], which provide an excellent framework for reporting adherence studies, and have followed its guidance in this manuscript where applicable. However, our manuscript reports on a study of dynamic changes in adherence using the 'refill-gap method' analytical approach described by Grégoire and Moisan [9] which, we feel, does not elegantly fit within Vrijens et al.'s taxonomy of medication adherence as published (e.g. [10,11]). Since the EMERGE Guidelines mandate using Vrijens et al.'s taxonomy, we have chosen not to comply with the Guidelines fully. We hope the following discussion will explain our decision.
Vrijens et al.'s initiation, implementation and persistence framework provides a useful overarching conceptual guide and consistent vocabulary for adherence research. However, we feel it becomes unwieldy in the context of considering adherence as a dynamic process, which is where we have explicitly situated it here. If persistence is defined as the time from the first prescription to discontinuation (as per Vrijens et al.), then where does subsequent reinitiation fit in this taxonomy (it is not listed as a phase in the published papers describing the taxonomy)? Does reinitiation constitute the initiation of a new adherence 'episode'? Or, because the discontinuation was not permanent, does it not count as a true discontinuation, and therefore the person never actually leaves the implementation phase? Researchers such as Jensen et al. [12,13] have taken the former approach whilst the EMERGE Guidelines seem to imply the latter. Others (e.g. [14]) have used the EMERGE Guidelines whilst tacitly noting the limitations of Vrijens et al.'s taxonomy for adherence research which looks at medication adherence as a dynamic process. We would note that Vrijens et al.'s taxonomy is also being stretched by further developments in the field, such as the use of trajectory models to classify patterns of adherence behaviour.
As the above discussion suggests, we are not clear how our study would fit into the three phases of Vrijens et al.'s taxonomy. Instead, we have followed the 'refill-gap method' analytical approach described by Grégoire and Moisan [9]. Thus, we are not sure how describing the taxonomy nor using it to frame the reporting of our findings would add to the manuscript or help the reader understand what we did and its implications. That said, we do note that our use of a time to event analysis for calculating associations with the first discontinuation follows best practice suggested by Vrijens et al. [10,11] and Halpern et al. [15]. We have added the following (with citations) to the beginning of Section 2.6 (Statistical analyses) to make this approach explicit: We used the refill-gap method as a framework for our analyses [23], along with time-to-event analyses as recommended by Vrijens et al. [24,25].
In keeping with the EMERGE Guidelines, and the helpful conceptual clarity that Vrijens et al.'s taxonomy provides, we have substituted the term 'persistence' and its variations with 'continued', 'being dispensed', 'adherence', 'discontinuation and reinitiation', 'medication use' or 'long-term adherence' throughout the manuscript as appropriate to avoid ambiguity given that it is unclear whether what we describe would be considered persistence according with the EMERGE Guidelines and Vrijens et al.'s taxonomy.