Results and Conclusions

Overall, the trial emulations have replicated LEAD-5 well on the HbA1c and BMI measurements.

When the patients meet the LEAD-5 patient baseline characteristics, the trial emulations produce the same ranking between GLP-1, basal insulin and placebo as LEAD-5. All the results have suggested that GLP-1 has the best performance in terms of HbA1c, systolic blood pressure and BMI reduction if the patients meet the inclusion criteria of LEAD-5.

Experiments with real patients who do not fall into the baseline characteristics of LEAD-5 have presented different performance rankings between the three drugs. This suggests that LEAD-5 trial outcomes cannot be simply extrapolated to cover other patient populations. To this end, the virtual trial emulation models and tools are potentially very useful in terms of providing evidence to support the extrapolation of clinical trials for real-world clinical practice.

GLP-1 vs basal insulin on systolic blood pressure also well agree with LEAD-5. But there is a discrepancy on the effect of GLP-1 vs placebo on systolic blood pressure, where LEAD-5 shows no significance while our emulation still predicts significant reduction.

Although our experiments with the trial emulations on real patients have produced the right directions for the effect, we have also observed over estimation in some of the cases.

We recon the difference in the settings between the virtual and real trials potentially contributes to the discrepancy between the real trial and emulations. These factors include drugs (e.g. the virtual trial experimented with GLP-1 but LEAD-5 targeted Liraglutide, etc.), variables (i.e. several variables involved in the virtual trial emulation were not identical to the variables used in the LEAD-5 patient baseline characteristics), and the way that the clinical measurements were taken. Another limitation within the dataset is the pre-treatment history of the patients might not be complete. Overall it is not possible to replicate a completely identical cohort practically.

Due to time constraint, this study has its limitations in a number of aspects: As a proof-of-concept study, it only has replicated a single clinical trial in T2DM (i.e.LEAD-5). To further validate the concept, more evidences are needed from further studies on a wider range of clinical trials. The study has evaluated the emulation with a real trial that we already know the outcomes. For the future use of the trial emulations to address real-world clinical questions where the ground truth is unknown, we need standardised quality metrics to calibrate the emulation outcomes. The current model has made several assumptions about the causal structures. We have made several simplifications in data pre-processing.

More details can be found in Deliverable: Trial Emulation Report.