Metformin adherence in patients with type 2 diabetes and its association with glycated haemoglobin levels

Keywords: Type 2 diabetes; medication adherence; general practice; inequality

Type 2 diabetes is a growing health issue in New Zealand (NZ) that affects twice as many Māori as non-Māori and an increasing number of younger people.^1–3 Type 2 diabetes is associated with a range of microvascular and macrovascular complications, chronic kidney disease and cardiovascular disease; the latter being the greatest cause of morbidity and mortality in this patient group.^4–6 Patients with good glycaemic control have fewer complications.^7,8 To achieve optimal outcomes and reduce the chance of diabetes-related complications, patients should aim to have their glycated haemoglobin (HbA1c) measurement <53 mmol/mol.⁹

Type 2 diabetes is mainly managed in general practice with the aim of achieving good glycaemic control and reducing cardiovascular risk factors. Management primarily involves the use of diet and exercise along with medication.¹⁰ Typically, metformin is the initial medication of choice as it has been shown to reduce insulin resistance, cardiovascular disease and mortality, while demonstrating a good safety profile at a low overall cost.¹¹ Many studies have demonstrated significant improvements in glycaemic control with metformin use, either alone or in combination therapy.^12–14 However, the efficacy of metformin depends on patients’ adherence to prescribed medication.^15,16 Many factors contribute to diabetes medication adherence, including cost, health literacy and disease awareness.^17,18 Many patients may also be non-adherent to metformin prescribing because of gastrointestinal side-effects.^19,20

Adherence to metformin and other oral hypoglycaemic agents is sub-optimal in NZ patients with type 2 diabetes.^21,22 Non-adherence is higher in patients receiving combination oral hypoglycaemic therapy and in patients who are male, of Māori ethnicity and who have high socioeconomic deprivation.²¹ In these studies, non-adherence was assessed using medication dispensing data and there was no record of whether this was because patients failed to receive a prescription or did not visit a pharmacist. This is important to know because there may be disparities between Māori and non-Māori in the medications that are prescribed rather than medications dispensed.²³ Therefore, the primary aim of this study was to characterise metformin adherence in a cohort of NZ patients with type 2 diabetes using both primary care prescription and dispensing data. This study also evaluates associations between metformin non-adherence and HbA1c levels as this has not previously been reported in NZ.

This study was part of a larger project aiming to assess the quality of diabetes care in primary care using demographic and clinical data collected from 10 general practices in the Waikato region. The full project was granted ethics approval by the New Zealand Health and Disability Ethics Committee (ref: 19/CEN/8).

In June 2019, primary care data were collected directly from the electronic patient management systems of study general practices for the period of October 2016 – March 2018. National Health Index (NHI)-identified patient information was extracted for all patients who had a confirmed diagnosis (≥12 months) of diabetes (read code C10) and were aged ≥15 years on 1 October 2016. Extracted data included age (at time of data collection in June 2019), gender, ethnicity, HbA1c values and metformin prescription data for the study period. HbA1c values were averaged to provide a mean value for each patient for the study time period. General practices were coded as rural or urban using the NZ Department of Statistics database Urban Rural Profile Categories.²⁴

Patient records were checked against the Waikato District Health Board clinical records to retrieve missing demographic and diagnosis information. Patients with type 1 diabetes were then excluded from the dataset.

Patients with type 2 diabetes who had received at least two dispensings of metformin were included for analysis. Primary care prescribing data were analysed for this cohort to determine how many metformin prescriptions were written for each patient during the 18-month study period (the ‘Prescribing Period’). These data were then linked by patient NHI to the Ministry of Health Pharmaceutical Collection database to determine how many prescriptions were dispensed. Each dispensing could be matched to only one prescription and needed to be within 3 months of the prescription date. The pharmaceutical dataset included the 21-month period between October 2016 and June 2018 (‘Dispensing Period’). The extra 3 months allowed for the dispensing of prescriptions that were written towards the end of the Prescribing Period (medications can be dispensed up to 3 months after being prescribed, while remaining valid for government subsidy).²⁵ A 90-day quantity dispensing was assumed for each prescription.

For this study, metformin adherence is characterised in two ways: (1) using medication dispensing data to define the number of days that patients had medication cover (the medication possession ratio (MPR)); and (2) by comparing the number of prescription and dispensing events to calculate the proportion of prescriptions being filled. Using the pharmaceutical dispensing data, an MPR of ≥0.8 indicated ‘good adherence’ (medication cover for at least 80% of days).²⁶ Metformin adherence was considered ‘poor’ where the MPR was <0.8.²⁶ MPR was calculated by summing the days of medication supply from the first to the last prescription (inclusive) divided by total number of days of the prescription period.²⁶

To calculate the proportion of prescriptions being dispensed, we analysed data from a subset of patients who had been prescribed metformin two to seven times during the 18-month prescribing period. These patients were classified as either having 100% prescription adherence (every metformin prescription could be linked to a dispensing event) or <100% prescription adherence (at least one prescribed medication could not be linked to a dispensing event).

Metformin adherence was described by gender, age group, ethnicity, rural or urban practice location and concurrent diabetes medications. Subgroup differences were analysed with chi-square test and Student’s t-tests, and logistic regression was used to estimate the odds ratio of a patient having 100% prescription adherence or having good adherence, adjusting for gender, age, ethnicity, rurality and diabetes treatment regimen. The average number of prescription and dispensing events is reported as mean ± standard deviation (s.d.).

Factors influencing HbA1c levels were also analysed. Student’s t-tests were used to compare the differences in means of various subgroups, and linear regression was used to determine the adjusted associations of HbA1c with metformin adherence, ethnicity, age, gender, medication regimen and rurality.

All data analyses were performed in Python 3.7 (Python Software Foundation, Beaverton, USA) using the Pandas 0.25.3, Scipy 1.3.2, and Statsmodels 0.10.2 libraries, with significance accepted at P < 0.05.

A total of 3716 patients with type 2 diabetes were initially identified in the primary care dataset, including 1595 patients (42.9%) who had at least two metformin dispensings during the prescription period. The mean number of metformin prescriptions during this period was 6.1 ± 1.7.

The demographic characteristics of this group of 1595 patients are shown in Table 1. Their median age was 65 years, with a slightly higher proportion of male patients (55.5% vs. 44.5%; P < 0.001). Nearly half of all patients (49.0%) were prescribed metformin monotherapy, with the remainder prescribed combination therapy (Table 1).

Table 1.  Demographic and clinical characteristics of patients with type 2 diabetes receiving at least two prescriptions of metformin during the 18-month prescribing period

For the 1595 patients who had received at least two dispensings of metformin, the mean overall MPR was 0.87, and three-quarters of all patients had an MPR of ≥0.8, which did not differ significantly by gender, rural or urban practice location or medication regimen (Table 2). Māori and Pasifika patients had poorer metformin adherence than patients of other ethnicities, with 63.8% and 59.6% of patients, respectively, having ≥80% medication adherence, compared to 73.8–81.2% of other ethnic groups (Table 2). This difference remained for Māori compared to NZ Europeans after adjusted analysis (OR 0.41; 95% CI 0.30–0.56; P < 0.001). Older patients were more likely to have reduced adherence (P < 0.001), although this association did not remain significant after adjustment for gender, rurality, ethnicity, and medication regimen.

Table 2.  Metformin adherence using the medication possession ratio (MPR) calculated from pharmaceutical dispensing data

Overall, 1127 (70.7%) of the 1595 patients with two or more metformin prescriptions received five to seven metformin prescriptions in the prescribing period, indicating full prescription coverage. The proportion of patients given five to seven prescriptions of metformin was lower for Māori and Pasifika patients (Table 3).

Table 3.  Dispensing adherence to metformin prescriptions for patients with type 2 diabetes who were regular metformin users (5–7 prescriptions; n = 1127).

For patients given five to seven metformin prescriptions, the majority were then dispensed, with 86.2% of patients overall having 100% prescription adherence. No significant differences in prescription adherence were seen by gender, ethnicity, rurality, or medication regimen. However, prescription adherence differed significantly with age (P < 0.001), with older patients being more likely to have 100% prescription adherence. After adjustment for other factors, patients aged 45–59 years also had a significantly lower rate of 100% prescription adherence than patients aged 60–74 years (OR 0.51; 95% CI 0.33–0.77; P = 0.002).

For patients who received less than five prescriptions of metformin during the 18-month period (n = 438), prescription adherence was still high, with more than 75% of patients having 100% prescription adherence (Figure 1).

Fig. 1.  Proportion of metformin prescriptions that were associated with a pharmaceutical dispensing event during the 18-month study period. ‘All’ = every prescription was matched to a dispensing evening; ‘n-1’, ‘n-2’ and ‘n-3’ = all except one / two / three prescriptions were matched to a dispensing event.

HbA1c values were available for 1560 of the 1595 patients with two or more metformin prescriptions (97.8%). Compared to patients who had an MPR of ≥0.8 and 100% prescription adherence, mean HbA1c was significantly higher in patients with an MPR <0.8 (67.6 vs. 59.8 mmol/mol, P < 0.001) and in patients who had <100% prescription adherence (64.1 vs. 59.7 mmol/mol, P < 0.001). Mean HbA1c was also higher in Māori (65.4 mmol/mol) than in non-Māori patients (60.5 mmol/mol; P < 0.01).

After controlling for other factors, we found that, compared with patients with an MPR of <0.8, having an MPR ≥0.8 was associated with a mean reduction in HbA1c of 4.83 overall (P < 0.001; 95% CI 3.08–6.57) and 5.00 mmol/mol for Māori (P = 0.008; 95% CI 1.30–8.70). Furthermore, having 100% prescription adherence was associated with an overall mean reduction in HbA1c of 3.20 mmol/mol (P = 0.004; 95% CI 1.01–5.38) compared with patients who had <100% prescription adherence.

The association between metformin adherence and HbA1c differed by age. In patients aged <60 years (mean age 50.6 years, mean HbA1c 67.9 mmol/mol), an MPR of ≥0.8 and 100% prescription adherence were associated with a mean reduction in HbA1c of 7.44 mmol/mol (P < 0.001; 95% CI 4.12–10.77) and 4.32 mmol/mol (P = 0.031; 95% CI 0.41–8.24), respectively. In contrast, in patients aged ≥60 years (mean age 71.7 years, mean HbA1c 58.5 mmol/mol), good metformin adherence was associated with a smaller reduction in HbA1c: MPR ≥0.8 (2.90 mmol/mol; P = 0.004; 95% CI 0.94–4.85) and 100% prescription adherence (2.66 mmol/mol; P = 0.045; 95% CI 0.06–5.26).

Our study showed disparity between Māori and New Zealand European (NZE) for metformin coverage and this is likely due to access to prescriptions rather than medication dispensing. Māori were less likely to receive the five to seven prescriptions needed to maintain 100% medication cover (and more likely to have an MPR of <0.8 and a higher mean HbA1c level); however, once prescribed, metformin dispensing was not different from other ethnic groups.

This finding of fewer metformin prescriptions for Māori patients agrees with previous reports that, in general, Māori are prescribed fewer medications than people of other ethnic groups.^21,27 This may be a result of the westernised model of health care used in NZ not being appropriate for use by many Māori,²⁸ such that Māori have been reported to be less satisfied with their access to health care in this country.²⁹ However, it is also possible that Māori may be less adherent to metformin because of other reasons such as a higher rate of side-effects.

Importantly, our study did show that if metformin was prescribed, Māori were just as likely as others to have the medication dispensed. This is a positive finding, though it disagrees with the Statistics New Zealand Family, Income and Employment survey from 2010 that reported that Māori had greater odds than European NZers of not having medication dispensed because of cost.³⁰ The reasons for this study showing a different result are unknown, but could be because of the implementation of the Māori Health Strategy in recent years³¹ or because Māori patients in this study were not from a socially deprived area and cost was not their primary concern. Alternatively, it is possible that our subset of patients is not representative of the larger regional population. In a large national review of 85,066 patients with type 2 diabetes who recently initiated metformin monotherapy, a mean MPR value of 0.81 for Waikato patients was reported.²² This is lower than the overall mean MPR of 0.88 seen in our study. However, our study includes all patients and not only patients who have recently commenced metformin therapy. We also report that older patients (who are likely to have had type 2 diabetes for longer) have higher metformin adherence. Therefore, our data are likely to be similar to that used in previous reports.

Up to one-quarter of all patients did not have continuous cover of metformin during the study period. The reasons for this will vary, but may include reduced health literacy, lack of disease awareness and acceptance,^17,18 and gastrointestinal disturbances.²⁰ Where possible, these factors need to be addressed, particularly as our study corroborates other reports that higher HbA1c levels are associated with poorer glycaemic control.³² In particular, general practitioners should take care to monitor prescription frequency for patients with less severe disease and younger patients, to minimise disease progression. Escalation or change of therapy should also be considered in patients with elevated HbA1c levels, particularly if metformin adherence is low because of adverse events.

Several diabetes medication adherence programmes have been tried with varying degrees of success.^33–35 Integrated approaches have been shown to be the most effective, though even low-cost solutions such as SMS reminders have been improved adherence rates.³³ Māori patients may not receive appropriate diabetes education as diabetes health literacy is lower in this group despite the same provision of resources.³⁶ Any diabetes intervention designed to improve adherence must be delivered in a culturally appropriate way. Further, diabetes education and intervention programmes must be age and gender appropriate. Younger patients and women, for example, have been shown to be more likely to respond to diabetes management that is associated with improvements in weight.³⁷ In contrast, older patients are often more concerned about diabetes complications.³⁸

The main strength of our study was that we were able to evaluate both primary care prescribing data and pharmaceutical dispensing data to determine where disparity in the use of metformin in primary care occurs. This has not been reported on previously and offers advantages over studies using only dispensing data as a proxy for medication use. Our study also included HbA1c levels, which allowed us to directly compare the effect of metformin non-adherence on glycaemic control.

The primary limitation of our study was that an MPR of 0.8 was used as an indicator of good adherence, suggesting that patients are taking their medication at least 80% of the time. This may not actually be the case as some studies do report that some patients choose to stockpile their medications or share them with additional family members.^39,40 Second, the assumption was made that prescriptions were all for 90 days, whereas this may not have always been the case. Third, matching of prescription to dispensing records was not an exact process, as patients who required an additional prescription because of legitimate reasons (i.e. because they lost it or if it was prescribed outside of our study practices) will appear to have less than 100% prescription adherence, and patients prescribed metformin elsewhere may have a data mismatch. Similarly, patients may also have had HbA1c tests that were not recorded in their primary care records; these would have been identified if we had also linked to laboratory data. These circumstances may skew the overall findings, although we believe that the rigour of our methodology should have reduced the incidence of these events. Finally, we suggest that it would be useful in future studies to include the number of visits that patients make in general practice to determine whether disparity in prescribing is due to differences in the number of health-care visits.

In conclusion, primary care data are invaluable for understanding medication adherence, demonstrating here that there appear to be differences in metformin use by Māori, likely as a result of them having fewer prescriptions.

The authors have no competing interests to declare.

LC, RL, RK and RP all contributed to the conception and design of the study, as well as interpretation of data. BM collected the data and CM and CL analysed it. LC, CM, RP and RL drafted the article for publication. All authors revised it for critically important intellectual content. All authors give their approval for publication.