Identification of subgroups of patients with type 2 diabetes with differences in renal function preservation, comparing patients receiving sodium‐glucose co‐transporter‐2 inhibitors with those receiving dipeptidyl peptidase‐4 inhibitors, using a supervised machine‐learning algorithm (PROFILE study): A retrospective analysis of a Japanese commercial medical database

Abstract

Aims To investigate the effects of sodium‐glucose co‐transporter‐2 (SGLT2) inhibitors vs. dipeptidyl peptidase‐4 (DPP‐4) inhibitors on renal function preservation (RFP) using real‐world data of patients with type 2 diabetes in Japan, and to identify which subgroups of patients obtained greater RFP benefits with SGLT2 inhibitors vs. DPP‐4 inhibitors. Methods We retrospectively analysed claims data recorded in the Medical Data Vision database in Japan of patients with type 2 diabetes (aged ≥18 years) prescribed any SGLT2 inhibitor or any DPP‐4 inhibitor between May 2014 and September 2016 (identification period), in whom estimated glomerular filtration rate (eGFR) was measured at least twice (baseline, up to 6 months before the index date; follow‐up, 9 to 15 months after the index date) with continuous treatment until the follow‐up eGFR. The endpoint was the percentage of patients with RFP, defined as no change or an increase in eGFR from baseline to follow‐up. A proprietary supervised learning algorithm (Q‐Finder; Quinten, Paris, France) was used to identify the profiles of patients with an additional RFP benefit of SGLT2 inhibitors vs. DPP‐4 inhibitors. Results Data were available for 990 patients prescribed SGLT2 inhibitors and 4257 prescribed DPP‐4 inhibitors. The proportion of patients with RFP was significantly greater in the SGLT2 inhibitor group (odds ratio 1.27; P = 0.01). The Q‐Finder algorithm identified four clinically relevant subgroups showing superior RFP with SGLT2 inhibitors (P < 0.1): no hyperlipidaemia and eGFR ≥79 mL/min/1.73 m2; eGFR ≥79 mL/min/1.73 m2 and diabetes duration ≤1.2 years; eGFR ≥75 mL/min/1.73 m2 and use of antithrombotic agents; and haemoglobin ≤13.4 g/dL and LDL cholesterol ≥95.1 mg/dL. In each profile, glycaemic control was similar in the two groups. Conclusion SGLT2 inhibitors were associated with more favourable RFP vs. DPP‐4 inhibitors in patients with certain profiles in real‐world settings in Japan.

What have we learned about renal protection from the cardiovascular outcome trials and observational analyses with SGLT2 inhibitors?

“…Finally, in another large database study, Zhou et al studied 990 patients prescribed SGLT2 inhibitors and 4257 patients prescribed DPP‐4 inhibitors . After adjustment of baseline confounders with propensity score matching, SGLT2 inhibitor users were significantly more likely to have preservation of renal function compared to DPP‐4 inhibitor users (adjusted odds ratio 1.27 CI 1.05‐1.53, P = .0116).…”

The present and future scope of real‐world evidence research in diabetes: What questions can and cannot be answered and what might be possible in the future?

“…Beyond the intense discussion on what exactly ML means and its pros and cons compared to “classical” statistical modelling methods, it is worth noting that the use of ML algorithms in medicine has received a wider attention following the demonstration of a performance similar to human clinical decision in the field of diabetes medicine, namely the diagnosis of diabetic retinopathy . ML models have been applied in diabetes to define clusters of diabetes phenotypes; predict kidney disease, hypoglycaemia, or glucose control; identify risk factors for CVD and death in diabetes; develop prediction models for complications; or transport RCT data to a target population …”

Q-Finder: An Algorithm for Credible Subgroup Discovery in Clinical Data Analysis — An Application to the International Diabetes Management Practice Study

“…Q-Finder was applied in the field of diabetes to many other research questions, such as the detection of patient profiles that benefit the most of SGLT2i compared to DDP4i in terms of renal function preservation, using Electronic Health Record data ( Zhou et al, 2018 ; Zhou et al, 2019 ); the identification of profiles of patients who better control their blood sugar, using data from pooled observational studies ( Rollot, 2019 , “Reali project”); and the discovery of new predictors of diabetic ketoacidosis (DKA), a serious complication of type 1 diabetes, using data from a national diabetes registry ( Ibald-Mulli et al, 2019 ). Q-Finder was also successfully applied in the context of several other pathologies such as hypophosphatasia, using SNPs data ( Mornet et al, 2020 ), dry eye disease using prospective clinical trials data ( Amrane et al, 2015 ), and cancer using clinical data from RCTs ( Nabholtz, 2012 ; Dumontet et al, 2016 ; Dumontet et al, 2018 ; Alves et al, 2020 ) or transcriptomic data from a research cohort ( Adam et al, 2016 ).…”

“…For further details, an in-depth discussion of Q-Finder is also proposed in Section 14 of supplementary materials. This approach has been applied in several therapeutic areas, with published examples available ( Nabholtz, 2012 ; Eveno, 2014 ; Amrane et al, 2015 ; Adam et al, 2016 ; Dumontet et al, 2016 ; Gaston-Mathe, 2017 ; Dumontet et al, 2018 ; Rollot, 2019 ; Zhou et al, 2018 ; Ibald-Mulli, 2019 ; Zhou et al, 2019 ; Alves et al, 2020 ; Mornet, 2020 ).…”