Modeling recurrent heart failure risk in type 2 diabetes: impact of flexible HbA1c trajectories using nonhomogeneous...

Many clinical trials yielded inconsistent results regarding the effect of intensive glycated hemoglobin control on cardiovascular diseases in type 2 diabetes. We identified distinct HbA1c trajectories and their association with the recurrent hospitalization of heart failures (HHF) for patients with type 2 diabetes starting from the date of diabetes diagnosis.

In this study, we included 194,258 patients who entered the SingHealth Diabetes Registry from 2013 to 2020. Their diagnoses of type 2 diabetes spanned the years 1960-2020, encompassing HbA1c measurements, records of HHF, and other cardiovascular complications. Latent class growth models (LCGM) with splines were used to extract the subgroups with distinct HbA1c trajectories. The association between HbA1c trajectories and the recurrent risk of HHF was investigated by nonhomogeneous Poisson processes (NHPP).

Eight distinct HbA1c trajectories were identified as follows: low stable (LowS, 22.2%), moderate low ascending (ModLowA, 12.7%), moderate high ascending (ModHighA, 11.5%), moderate low descending (ModLowD, 17.2%), moderate high descending (ModHighD, 10.1%), moderate high volatility (ModHighV, 10.1%), high with a sharp decline (HighSD, 8.0%), and high volatility (HighV, 10.2%). Using the Class LowS as a reference, the hazard ratios for recurrent HHF for the other classes are as follows: 0.79 for ModLowA, 1.30 for ModHighA, 1.17 for ModLowD, 1.89 for ModHighD, 1.94 for ModHighV, 1.25 for HighSD, and 2.88 for HighV. Considering recurrent HHFs, our NHPP model demonstrated predictive capability for type 2 diabetes patients' future HHF events.

Low baseline HbA1c levels are associated with a lower risk of recurrent HHF, while poor glycemic control significantly increases this risk. Our application of LCGM with splines effectively captures flexible, long-term HbA1c trajectories, while the innovative use of the NHPP model allows for precise modeling of HHF recurrence risk. This approach provides a foundation for personalized risk predictions and future HF management by incorporating dynamically updated risk factors.