Highlights
- External validation confirms that a clinical algorithm for selecting between SGLT2 inhibitors (SGLT2i) and DPP4 inhibitors (DPP4i) is accurate across White, Black, South Asian, and Mixed/Other ethnicity groups in the UK.
- While SGLT2i predictions were robust across groups, minor recalibration was required for DPP4i predictions due to higher-than-expected observed glycemic responses.
- The predicted glycemic advantage of SGLT2i over DPP4i was most pronounced in individuals of White ethnicity (3.7 mmol/mol) compared to South Asian (2.1 mmol/mol) and Black (0.6 mmol/mol) groups.
- Results support the clinical utility of the MASTERMIND selection model as a tool for personalized diabetes management in multi-ethnic populations.
Introduction: The Shift Toward Precision Diabetology
The management of type 2 diabetes (T2D) has evolved rapidly over the last decade. With the introduction of multiple drug classes—most notably SGLT2 inhibitors and DPP4 inhibitors—clinicians are often faced with a choice of which agent to prescribe as a second- or third-line therapy. Current guidelines often provide broad recommendations, but individual response to these medications can vary significantly based on clinical phenotypes, including age, BMI, renal function, and ethnicity. Precision medicine aims to move beyond a one-size-fits-all approach by using predictive modeling to identify which patient will benefit most from a specific therapy.
However, many predictive models are developed using data from predominantly White populations, raising concerns about their generalizability and equity when applied to diverse ethnic groups. The study by Güdemann et al., published in The Lancet Regional Health – Europe, addresses this critical gap by validating a treatment selection algorithm across the major ethnicity groups represented in the United Kingdom.
The Challenge of Ethnic Diversity in Clinical Modeling
Ethnicity is a complex variable in medical research, encompassing genetic ancestry, socioeconomic factors, dietary patterns, and healthcare access. In type 2 diabetes, ethnic differences in pathophysiology—such as variations in insulin sensitivity and beta-cell function—can influence drug efficacy. For instance, South Asian populations often develop T2D at a lower BMI and younger age compared to White populations, which may alter the glycemic response to SGLT2i or DPP4i.
Before a predictive model can be safely deployed in clinical practice, it must undergo rigorous external validation. This ensures that the model’s predictions remain accurate in a real-world population that differs from the original development cohort. This study sought to determine if a model developed to predict the 6-month change in HbA1c could reliably guide therapy selection for White, Black, South Asian, and Mixed/Other patients.
Study Design and Methodology
This retrospective cohort study utilized the Clinical Practice Research Datalink (CPRD) Aurum database, a large-scale repository of UK primary care records. The researchers identified 145,556 non-insulin-treated individuals with type 2 diabetes who initiated either an SGLT2i (n = 57,749) or a DPP4i (n = 87,807) between 2013 and 2020. This cohort was entirely independent of the population used to develop the original selection model.
The population was categorized into four major self-reported ethnicity groups:
- White: 114,287 (78.5%)
- South Asian: 20,969 (14.4%)
- Black: 6,663 (4.6%)
- Mixed/Other: 3,637 (2.5%)
The primary outcome was the change in HbA1c at 6 months post-initiation. The researchers applied a closed testing procedure to evaluate if the model’s intercept or slope required recalibration for each specific ethnicity. Calibration accuracy was assessed by comparing the predicted differences in glycemic response against the observed outcomes.
Key Findings: Model Performance and Recalibration
Recalibration Requirements
The study found that the original model slightly underestimated the glycemic response to DPP4 inhibitors across all groups. Minor model adjustments (recalibration of the intercept) were necessary to account for a greater-than-predicted reduction in HbA1c for DPP4i. The required adjustments were 1.6 mmol/mol for White, 3.0 mmol/mol for Black, 2.6 mmol/mol for South Asian, and 2.6 mmol/mol for the Mixed/Other group. Interestingly, the predictions for SGLT2i response were highly accurate and did not require adjustment for the non-White ethnicity groups.
Differential Treatment Effects
After the model was updated, it successfully predicted the differential treatment effect (the difference in HbA1c reduction between SGLT2i and DPP4i) for all groups. However, the magnitude of this difference varied significantly by ethnicity:
- White: SGLT2i was predicted to reduce HbA1c by 3.7 mmol/mol more than DPP4i (95% CI 3.5–3.9).
- South Asian: SGLT2i was predicted to be 2.1 mmol/mol more effective than DPP4i (95% CI 1.6–2.6).
- Black: SGLT2i was predicted to be only 0.6 mmol/mol more effective than DPP4i (95% CI 0.5–1.7).
- Mixed/Other: SGLT2i was 2.6 mmol/mol more effective than DPP4i (95% CI 1.4–3.8).
These findings suggest that while SGLT2i generally provides a greater glycemic reduction than DPP4i across the board, the relative benefit is significantly narrower in Black and South Asian patients compared to White patients.
Expert Commentary: Clinical and Methodological Implications
The validation of this model is a significant step forward for precision medicine in diabetes. The fact that the model remained well-calibrated for all ethnicity groups after only minor adjustments is encouraging. It demonstrates that clinical features—such as baseline HbA1c, renal function, and age—capture much of the biological variance that drives drug response, regardless of a patient’s self-reported ethnicity.
However, the need for recalibration for DPP4i responses highlights a critical lesson in clinical informatics: models are not “plug-and-play.” Local factors, including changes in prescribing habits over time or population-specific characteristics not captured by the initial model, can influence performance. The researchers correctly suggest that simple recalibration should be a standard step before deploying such algorithms in new healthcare settings or diverse populations.
From a clinical perspective, the smaller differential benefit of SGLT2i in Black and South Asian patients is noteworthy. While SGLT2i remain favored for their cardiovascular and renal benefits, the glycemic “gap” between them and DPP4i is smaller in these populations. This information can help clinicians manage expectations and personalize therapy when glycemic control is the primary immediate goal.
Study Limitations
The researchers acknowledged several limitations. First, ethnicity was self-reported, which is a proxy for a complex array of biological and social factors. Second, the “Mixed/Other” category is highly heterogeneous, making it difficult to draw specific conclusions for individuals within that group. Finally, the study focused on glycemic response (HbA1c); it did not evaluate long-term cardiovascular or renal outcomes, which are also critical factors in the choice between SGLT2i and DPP4i.
Conclusion: Practical Takeaways for Clinicians
The validation of the SGLT2i-DPP4i selection model in a large, diverse UK primary care cohort provides strong evidence for its utility. The key takeaways are:
- The model accurately predicts which patients will achieve better glycemic control on SGLT2i versus DPP4i across all major UK ethnicity groups.
- Clinicians can have confidence that the clinical variables used in the model (e.g., BMI, eGFR, baseline HbA1c) are valid predictors across diverse backgrounds.
- SGLT2i generally offers superior glycemic lowering compared to DPP4i, but this advantage is less pronounced in Black and South Asian populations.
- Recalibration is a vital process to ensure that predictive tools remain accurate when applied to new or evolving patient populations.
As we move toward a future of data-driven healthcare, tools like the MASTERMIND algorithm will be essential in helping clinicians navigate the complex landscape of type 2 diabetes pharmacotherapy, ensuring the right drug reaches the right patient at the right time.
Funding and References
Funding for this study was provided by the UK Medical Research Council, the National Institute for Health and Care Research (NIHR) Exeter Biomedical Research Centre, and the EFSD/Novo Nordisk.
Reference:
Güdemann LM, Young KG, Cardoso P, et al. Validation of an algorithm for selection of SGLT2 and DPP4 inhibitor therapies in people with type 2 diabetes across major UK ethnicity groups: a retrospective cohort study. Lancet Reg Health Eur. 2025;61:101547. doi:10.1016/j.lanepe.2025.101547