Pre-Diabetes Isn't One Diagnosis: Six Subtypes That Predict Very Different Futures

The study, led by a team at Siriraj Hospital in Bangkok, followed nearly 5,000 adults without diabetes for a median of 8.8 years and used a machine-learning technique called k-means clustering to sort them by six routine variables: age, BMI, fasting glucose, HbA1c, HDL cholesterol and ALT (a liver enzyme). What emerged were six distinct subphenotypes with meaningfully different futures — not a smooth gradient from "a little high" to "a lot high," but discrete patterns that behave differently over time.

That distinction matters. If pre-diabetes is really a family of conditions, then a person whose HbA1c sits at 6.0% because they are lean, older and slightly insulin-resistant is being asked to follow the same advice as someone whose 6.0% comes packaged with obesity, a fatty liver and low HDL. The first person's risk curve and the second person's risk curve, this paper suggests, are not the same shape at all.

The researchers labeled their six groups descriptively. Cluster 1 was a low-risk group — younger, leaner, with relatively clean labs. Cluster 2 was older adults with mild dysglycemia. Cluster 3 combined severe dysglycemia with obesity. Cluster 4 had milder glucose elevations but obesity. Cluster 5 was the "severe dysmetabolic obese" group, where glucose, lipids and liver markers were all off at once. Cluster 6 was older adults with severe dysglycemia — and it carried the highest risk profile of all, with significantly elevated rates of macrovascular events compared with the others, according to the Siriraj cohort analysis.

What is striking is not just that the clusters differ on paper, but that they diverged in real outcomes over nearly nine years of follow-up. Some groups progressed to type 2 diabetes at substantially higher rates. Others were more likely to experience large-vessel complications — the kind that drive heart attacks and strokes. The conventional cut-points (fasting glucose 100–125 mg/dL, HbA1c 5.7–6.4%) lumped all of these people together. The clustering pulled them apart.

Subtyping is having a moment in metabolic medicine. Earlier European work famously split adult-onset diabetes into roughly five clusters with different complication profiles, and researchers have been chasing the same logic upstream into pre-diabetes ever since. The new Thai analysis is part of that wave: an attempt to replace a binary label with a map.

For readers who already track their own data — CGM curves, fasting glucose, an annual HbA1c, a lipid panel — the practical implication is subtle but real. A single number, viewed in isolation, is a weak predictor. The combination of numbers, viewed together, may be a much stronger one. A mildly elevated HbA1c in someone with a healthy BMI, normal liver enzymes and high HDL is a very different signal from the same HbA1c in someone whose ALT is creeping up and whose HDL is low.

It is worth being honest about what this study is and isn't. It is a single-center cohort from one hospital in Bangkok, with a participant pool whose average age was around 60 and which skewed slightly female. The clusters were generated using a statistical technique that finds patterns in this dataset; whether the same six groups reappear in, say, a younger European or American cohort is an open question. The authors themselves frame the work as a step toward more granular risk stratification, not a finished clinical tool.

That is also why our editors rated the evidence here as moderate rather than strong. The signal is real and the methodology is reasonable, but the clusters need to be reproduced in other populations before any clinician — or any wellness writer — should start telling people which "type" of pre-diabetes they have.

You do not need to wait for personalized pre-diabetes care to arrive to start thinking like the researchers behind it. The variables they chose are not exotic; they are the six lines that already appear on most routine bloodwork. Looking at them as a pattern — rather than scanning for a single red flag — is closer to how this research suggests risk actually clusters.

If your fasting glucose or HbA1c has nudged into the pre-diabetes range, the genuinely useful next step is not to self-sort into a cluster but to bring the full panel to a clinician who can interpret it in context. Age, body composition, family history, liver markers and lipids all carry information that a glucose number alone does not.

Personalized metabolic care is not here yet. But the direction of travel is clear, and it is away from the blunt yes/no of "pre-diabetes" and toward something more like a weather forecast: a probability, a pattern, and a set of choices calibrated to your specific sky. The six-cluster framework is a draft of that forecast — early, imperfect, and genuinely interesting.