In this study on 8515 primary care patients receiving their first prescription of an antidiabetic drug, the majority claimed their prescription. However, many of them never claimed a second prescription, and 64% remained on the initially prescribed drug 2 years later. Patients who failed to claim their prescription within 30 days were more likely to be born outside of Europe or unemployed. Factors associated with the 30% of patients not starting long-term treatment (i.e., not claiming the second prescription) included being born outside of Europe, unemployment, low income, and being either young or old. Low income, age, and birth outside Europe continued to impact the use of antidiabetic medications for persistence, and the factors associated with discontinuation also included females and having a university education.

We found that 92% of patients claimed their prescription within 30 days from the prescription date. This initiation rate aligns with previous research, which suggests that, on average, 10% of patients never claim their diabetes prescriptions [32]. However, in our study, most patients claimed their first prescription within a year. Therefore, using time windows that are too short may underestimate the initiation. Canadian researchers found one-third to discontinue within 3 months, being similar to the findings of 71% for I150 [14]. In our study, the pattern of initiation was similar across different treatment groups, but patients prescribed only insulins were more likely to never claim their prescription. Early discontinuation of insulins could be explained by some patients using them temporarily for high HbA1c levels at the time of prescription, while another antidiabetic agent is prescribed for long-term use. Polytherapy was also discontinued early, potentially for the same reason as for insulins. Although other studies have shown that insulins have lower MA compared to other antidiabetic therapies, our findings support this observation [19]. Patients prescribed monotherapies other than metformin or insulins were less likely to claim their prescriptions. This could be due to the relatively higher cost of the new antidiabetic agents compared to metformin and insulins. Additionally, insulin is fully reimbursed in Sweden without any patient co-payment. Metformin users had higher frequencies of both initiation and persistence, but still, 32.8% of patients had discontinued metformin after 24 months, and the other treatments had even higher discontinuation rates. Overall, 70% of the total population demonstrated persistence after 1 year, and 64% remained persistent after 2 years. These rates are slightly higher than previous research, which reported persistence rates for initially prescribed antidiabetic agents ranging between 33 and 61% after 6–24 months [9]. The generous definition of persistence with the anniversary method, using a time window of ± 3 months, contributed to higher rates of persistence in our study. Differences between studies may also be attributed to variations in study design or the populations included.

Demographic factors, such as age and sex, were identified as being associated with MA. The oldest age group exhibited lower treatment initiation rates, potentially due to practical challenges related to collecting the medication or the presence of comorbidities. However, both the youngest and the oldest age groups demonstrated lower persistence rates. Previous studies have shown associations between poor MA and either older age [16] or both very young and old age [13]. Patients between 50 and 79 years may be more motivated by the potential for a longer life. Women had lower persistence compared to men, which contradicts some research findings [13, 33] but aligns with other studies [16, 34]. Sex patterns were observed for both persistence to the initially prescribed medication and overall antidiabetic therapy. This could be attributed to women having higher adherence to lifestyle interventions involving diet and physical activity, thus reducing the need for medication. Discontinuing women had lower initial HbA1c levels, with a mean of 54.9 mmol/mol, compared to the total population mean of 60.2 mmol/mol. Unemployment and being born outside of Europe were the primary SE factors associated with low initiation. However, unemployment was not associated with persistence, indicating a temporary effect on adherence, potentially due to changes in routines or increased stress. This could also be influenced by the Swedish reimbursement system, due to higher patient co-payment for initial dispensations. Patients born outside Europe exhibited lower adherence across all measures, aligning with studies where non-native background was a disadvantage for adherence [15, 34, 35]. Low income was associated with lower initiation and persistence, which corresponds to the association between low SE status and poor MA reported in other studies [34]. University education was slightly associated with lower persistence, but this was not significant with a 99% confidence interval. While being married has been associated with adherence in other studies [6, 13], we only observed a slightly higher degree of claiming the first prescription among married individuals. No associations were found when comparing patients living alone to those who were married or cohabiting or when separating the married and cohabiting groups. It was not possible to stratify many SE factors in the analysis due to small samples in each group, but a pattern was seen where young women born outside of Sweden had very low persistence, which calls for further research.

This study has several notable strengths. Firstly, the separation of initiation and persistence phases and the analysis of associated factors separately provide valuable insights into targeting specific adherence phases. The findings of this study demonstrate that different factors play a significant role during different phases of adherence. Secondly, the study assessed two distinct measures for initiation, distinguishing between patients who tried the medication and those who started long-term treatment. The comparison of EHR prescriptions with pharmacy claims for initiation is an understudied aspect in MA research, offering a better understanding of patients’ therapy initiation. Additionally, the study’s 2-year follow-up period after the first dispensation date allows for tracking patients’ prescription renewal in the second year. The use of national dispensing data is also advantageous, as it encompasses all dispensed drugs in Sweden, ensuring comprehensive coverage for all patients and the ability to track patients who relocate within the country. Lastly, the availability of individual-level data on various SE factors is a rarity in the field, contributing new insights.

While the national Swedish registers provide extensive data coverage, certain limitations exist. For instance, approximately 1.3% of the study sample had missing data on education, primarily due to patients not being born in Sweden [24]. Clinical variables and diagnoses, although highly covered in EHR, possess some limitations due to their lack of structure compared to national registers. Missing data may occur in EHR variables due to the data collection time frame. The calculated eGFR was missing for 31% of patients, and 19% did not have any HbA1c measurements within 12 months before prescription. This could be attributed to HbA1c test being conducted before prescription of the antidiabetic drug and added to the EHR when analyzed at the laboratory, a few days later. Another limitation is the inability to adjust for other health-related factors such as body mass index (BMI) or smoking due to missing data, which may result in residual confounding. Income was only based on individual income and not household income, which would be more accurate when determining SE position. Another limitation for income is using the same data collection window as for the other variables, where a mean of several years might reflect SE position better. The main limitation of the regression model is the general correlation between different SE factors. Although multicollinearity testing only showed high correlations between age and occupation, in the reduced models adjusting for each SE factor separately, the effects of high income, unemployment, and country of birth were more pronounced. This suggests some overfitting of the full model and underestimation of the impact of these factors on initiation and persistence. There may also be clustering in data with patients visiting the same doctor or PHC, which could have been incorporated in a multilevel analysis. Furthermore, there are also some limitations in the calculations of persistence. The anniversary method is a rough estimation of a person’s actual usage and likely overestimates persistence. Other methods including supply gaps could, on the contrary, underestimate persistence, due to challenges in estimating stockpiling and undocumented therapy changes. Finally, when performing many statistical tests, there is always the risk of by chance gaining associations. We addressed it by adding 99% confidence intervals for the SE variables, which showed similar results.

The Uppsala region included in this study is representative of the entire country of Sweden, including both urban and rural areas and both low- and high-income neighborhoods. The generalizability of this study beyond Sweden is limited due to differences in regulations and laws for prescribing, dispensing, and reimbursement between countries.

In conclusion, the findings of this study reveal that while a majority of patients claim their prescriptions, certain SE groups exhibit lower rates of initiation or persistence. To address medication nonadherence effectively, patients from the different SE groups need support in different phases of MA. Patients born outside of Europe or with a low income might need support for both the initiation and persistence phase, while unemployed might need more support for the initiation phase and patients with young or old age for the persistence phase. By supporting patients in the phase of need, it is possible to improve MA and reduce inequalities in quality of care.