Patient clusters based on HbA1c trajectories: A step toward individualized medicine in type 2 diabetes

Abstract

AimsTo identify clinically meaningful clusters of patients with similar glycated hemoglobin (HbA1c) trajectories among patients with type 2 diabetes.MethodsA retrospective cohort study using unsupervised machine learning clustering methodologies to determine clusters of patients with similar longitudinal HbA1c trajectories. Stability of these clusters was assessed and supervised random forest analysis verified the clusters’ reproducibility. Clinical relevance of the clusters was assessed through multivariable analysis, comparing differences in risk for a composite outcome (macrovascular and microvascular outcomes, hypoglycemic events, and all-cause mortality) at HbA1c thresholds for each cluster.ResultsAmong 60,423 patients, three clusters of HbA1c trajectories were generated: stable (n = 45,679), descending (n = 6,084), and ascending (n = 8,660) trends, which were reproduced with 99.8% accuracy using a random forest model. In the clinical relevance assessment, HbA1c levels demonstrated a J-shape association with the risk for outcomes. HbA1c level thresholds for minimizing outcomes’ risk differed by cluster: 6.0–6.4% for the stable cluster, <8.0% for the descending cluster, and <9.0 for the ascending cluster.ConclusionsBy applying unsupervised machine learning to longitudinal HbA1c trajectories, we have identified clusters of patients who have distinct risk for diabetes-related complications. These clusters can be the basis for developing individualized models to personalize glycemic targets.

The Identification of Diabetes Mellitus Subtypes Applying Cluster Analysis Techniques: A Systematic Review

“…Only two studies have indicated reducing the dimensionality of the data prior to CA [12,13]. The widespread method for clustering among included publications was k-means clustering [7,8,13,[15][16][17][18]20]. Several studies performed k-means analysis only for GADA-negative individuals [7,8,18].…”

“…In addition, two publications determined the number of clusters based on principal component analysis (PCA) [12,13], one publication performed a "NbClust" algorithm that selected an optimal method for the determination of number of clusters [15], one study was based within the cluster sums of squares against the number of clusters [16], one study was based on a cosine distance metric [19].…”

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 …”

Predicting short‐ and long‐term glycated haemoglobin response after insulin initiation in patients with type 2 diabetes mellitus using machine‐learning algorithms

“…Prediction models based on such traditional regression analyses, however, have limited utility for clinical decision‐making because they do not personalize the prediction to the individual. Similarly, studies that identify clinical characteristics of patients that are associated with distinct HbA1c trajectories in patients with T2DM over the course of insulin treatment via unsupervised clustering algorithms cannot sufficiently guide clinical decision‐making . Such studies lead to inconsistent groupings of patients, suggesting that cluster analysis driven by outcome patterns is not helpful in predicting HbA1c responses using clinical variables at baseline.…”

“…Similarly, studies that identify clinical characteristics of patients that are associated with distinct HbA1c trajectories in patients with T2DM over the course of insulin treatment via unsupervised clustering algorithms cannot sufficiently guide clinical decision-making. [14][15][16][17][18] Such studies lead to inconsistent groupings of patients, suggesting that cluster analysis driven by outcome patterns is not helpful in predicting HbA1c responses using clinical variables at baseline. There is a need for a more individualized approach to support clinical decision-making and to guide personalized treatment for patients with T2DM in need of additional treatment.…”