The epidemic of type 2 diabetes (T2D) is of significant concern due to increased morbidity, mortality, and health care costs. Current treatment guidelines recommend early use of sodium glucose transport protein 2 inhibitors (SGLT2i) and glucagon like peptide 1 receptor agonists (GLP1RA) for those with high cardiovascular risk, heart failure or renal disease (1). In the remainder, metformin, followed by a range of other medications are recommended for glucose lowering on a sequential, additive basis, influenced by cost, side effect profiles, and average treatment effects from clinical trials (1). Consideration of personal characteristics such as comorbidities, and circumstances such as food security, financial stability, alongside setting an appropriate target HbA1c for individuals is part of individualising treatment of T2D. Identifying subgroups or strata of patients with similar characteristics who respond better to one type of glucose-lowering therapy over another, and/or have altered risk of treatment-specific side effects, has been termed stratification of treatment response, which represents the first step towards precision medicine for T2D (2).

The glucose lowering response to non-insulin therapies that target discrete aspects of glucose metabolism is, extremely variable between individuals with T2D. Knowledge of which medications are likely to be most effective for individuals with T2D is important to minimise cost, side effects and periods of hyperglycaemia associated with initiating an ineffective medication. In the absence of knowledge about which subgroups of people with T2D respond better to a particular medication, relatively low glucose lowering response to certain treatments may be inappropriately attributed to non-adherence. There is emerging evidence for stratification of glucose lowering response to certain medications for T2D by baseline characteristics, such as ethnicity, obesity and triglyceride levels.

Ethnicity as a marker for differential medication responsiveness has been investigated as a simple tool for therapeutic stratification of T2D medications. Both SGLT2i and dipeptidyl peptidase 4 inhibitor (DPP4i) have been reported to be more effective in lowering HbA1c in people of Asian compared with European ethnicity (3). Metformin may be more effective in African Americans with younger onset T2D compared to Americans of European ethnicity (4). Alpha-glucosidase inhibitors appear to work better in South East Asians than Europeans (5).

It is unclear whether differences in treatment response by ethnicity are due to biological differences in underlying T2D aetiology, or whether ethnicity serves as a proxy for residual confounders such as level of glycaemia, concomitant diabetes therapies, dietary factors, medication adherence or other health inequities. Biological differences in underlying T2D aetiology using routinely assessed clinical markers of insulin resistance, such as obesity and/or high triglycerides (OHTG), have been shown to predict less glucose-lowering to DPP4i (6), which amplifies glucose-stimulated beta cell insulin secretion through the endogenous incretin pathway. Those with obesity have been reported to have greater glucose-lowering with thiazolidinediones (6, 7), which act to improve insulin sensitivity and lipid metabolism by stimulating the nuclear receptor peroxisome proliferator-activated receptor gamma (PPARG).

Māori are the indigenous people of Aotearoa, New Zealand and share common Polynesian ancestry with Pacific peoples. Obesity and elevated triglyceride levels are prevalent in people of Māori and Pacific ethnicities (8, 9). There are disparities in diabetes outcomes for Māori and Pacific peoples (10), hence research into biological and other factors underlying these disparities are needed. To date glucose lowering response to different medications in people of Māori or Pacific ethnicity with T2D has not been studied. In this T2D medication, Which One is Right Here (WORTH) trial, we tested the potential stratification of comparative glycaemic response to pioglitazone (thiazolidinedione) and vildagliptin (DPP4i), by Māori or Pacific ethnicity and by OHTG status.

This multicentre, two-way, randomised crossover trial showed that the glucose-lowering response to pioglitazone relative to vildagliptin was not different between people of Māori or Pacific ethnicity and other New Zealand ethnicities. However, there was a significant interaction in relative glucose-lowering by OHTG status, with 4.7mmol/mol greater reduction in HbA1c by pioglitazone relative to vildagliptin among those with OHTG. Those without OHTG had similar glucose lowering to each medication. Importantly, both treatments generated similar treatment satisfaction scores although there was greater weight gain and greater improvement in lipids and liver enzymes after pioglitazone than vildagliptin. No severe adverse events were directly related to these treatments.

While people of Māori or Pacific ethnicity had higher prevalence of OHTG, such ethnicity grouping did not predict altered glucose lowering response to pioglitazone versus vildagliptin. Within Māori and Pacific populations, there is genetic heterogeneity in risk and aetiology of T2D. A genetic variant (rs373863828, p.Arg457Gln) in the CREBRF gene, specifically found in approximately 25% people of Māori and Pacific ancestry, is associated with increased body mass index (15, 16) yet 40% lower odds of T2D (15, 16), due to enhanced insulin secretion capacity (17). Testing whether CREBRF gene variant status has utility in predicting stratified glucose lowering responses to T2D medications among people of Māori and Pacific ethnicity is currently underway (12). OHTG status has been reported to predict lower glucose response in studies with European participants (6).

In contrast to rosiglitazone, which is no longer used due to its associated increased risk of myocardial infarction and overall mortality (18), pioglitazone has demonstrable cardiovascular benefits (19, 20). This is thought to be partly due to its favourable impact on lipids (21, 22), which was also observed in this trial. In addition, our finding of improvement in liver enzymes with pioglitazone treatment is consistent with its reported benefits in non-alcoholic fatty liver disease and non-alcoholic steatohepatitis. Whilst the weight gain with pioglitazone has been highlighted as a caution for its use, our data suggests the weight gain was mild and did not reduce patient satisfaction.

The lack of interaction of reduction in liver enzymes or lipids after pioglitazone with baseline OHTG is interesting. One explanation is that the glucose lowering action of pioglitazone is mediated by ligand binding to nuclear receptor PPARG which are most abundant in adipocytes and hence greater in people with obesity (23). In contrast, the liver enzyme or lipid lowering action of pioglitazone may be mediated by other pathways that are not directly correlated with baseline OHTG status such as PPAR alpha expression (24).

Comparative medication glucose response data from this crossover study design provides evidence-based understanding of how patients would respond if they were to switch from one medication to another. While this is something that happens frequently in diabetes care, there is very limited clinical data about how patients’ glycaemic control changes when this happens or which medication they prefer. In addition, several consensus reports describe that when making treatment decisions for patients with type 2 diabetes it is important to consider not only the efficacy and safety of each medication, but also patient preference. This open-label cross-over study provides the added value of patient preference outcomes between pioglitazone and vildagliptin in the real-world setting. The protocol for a three-way crossover study including a DPP4 inhibitor, thiazolidinedione and a SGLT2 inhibitor has been described, which will provide further information on stratification of glucose response to three diabetes medications by routine clinical features ascertained at baseline (25).

A key limitation of this trial was the low follow up rate resulting in only 203 participants with valid HbA1c data after both treatments. Some of this related to difficulties encountered during the pandemic in obtaining a timely HbA1c result at the end of each study medication treatment phase and in conducting in person assessments. This is unlikely to have introduced any retention bias as each person contributed their own comparative data for both medications. Furthermore, the analysis was conducted with and without imputation on the primary analysis population, to test the robustness of the findings. Secondly, the open-label nature of the trial was selected for lower cost and complexity, which may have introduced observer and participant bias, but increased external validity. Thirdly, the crossover trial design might have led to the inappropriate estimation of the risk of weight gain with each medication, although the sensitivity analysis showed no significant carryover effect. The crossover design was selected as the best method to test differential glucose-lowering responses using HbA1c as the primary outcome measure between two diabetes medications in different patient subgroups (11). A longer treatment duration of each period could have mitigated against the slow onset and offset of each medication in absence of wash-out period. However, the 16-week consecutive treatment period was considered the most efficient study design.

It is well recognised that overall glucose lowering effects of pioglitazone are typically greater than vildagliptin. The important finding of this study is that a superior glucose-lowering response to pioglitazone versus vildagliptin is observed mainly in those with OHTG. Whilst OHTG is more prevalent in those of Māori and Pacific ethnicity, the glucose-lowering response was not significantly different by such ethnicity group in a multi-ethnic New Zealand population. This has several clinical implications. Firstly, these findings suggest that ethnicity is not a sufficient proxy for stratified medicine for pioglitazone and vildagliptin in T2D. Rather, clinical markers of insulin resistance that are routinely available in clinical practice such as BMI and triglyceride levels are able to predict whether superior glucose lowering response to pioglitazone relative to vildagliptin is expected. The recent progress in better defining novel T2D subtypes based on underpinning disease mechanisms such as age at diabetes diagnosis, GAD antibodies, C-peptide and genetic factors need to be explored for their utility in predicting other stratified glucose lowering responses (26). However, the use of non-routine biomarkers are less easily adopted in clinical practice. Secondly, the participant satisfaction data suggests people are generally not troubled by the small weight increase from pioglitazone so this medication is best selected for those with OHTG to maximise risk vs benefits. Finally, T2D treatment guidelines could include consideration of OHTG to inform treatment selection between vildagliptin and pioglitazone. Including a more prominent role for pioglitazone over vildagliptin in the presence of OHTG would be reasonable since this medication produces much greater glucose lowering in this subgroup. This is a robust finding that has now been replicated in a non-European population. Pioglitazone treatment selection will need to be considered in the context of other common indications for this medication such as fatty liver or dyslipidaemia, and relatively uncommon contraindications such as heart failure, osteoporosis and macular oedema, as part of individualised care based on patient characteristics.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by NZ Health and Disability Ethics Committee. The patients/participants provided their written informed consent to participate in this study.

RM designed the trial. YJ was responsible for the statistical analysis of the data. TRM was responsible for the genetic analysis plan. ML designed the population-specific genetic variant panel. TM assisted with genetic analysis of the data. RM and PRS obtained funding. RB, RY, RT-C, KS, GD, KAM, RD, PC, AM, NN, FK, JH, BO-W, and RP were responsible for trial implementation and data acquisition. RB and RM wrote the first draft of this manuscript and RM handled the revisions. All authors reviewed the manuscript and approved the final version for submission.