Prediabetes: Need for changes in approach and intervention

Introduction

Prediabetes is a general term defining an intermediate stage of glycaemic dysfunction between levels considered normal and the diagnosis of diabetes. This disorder affects more than 720 million people worldwide, and it is estimated that up to 10% of individuals with prediabetes may develop type 2 diabetes mellitus (T2DM) each year.¹ However, prediabetes is not only a risk factor for T2DM but is also associated with an increased cardiometabolic risk and mortality.²

Various societies, medical associations, and governmental bodies recommend strategies for the diagnosis or prognostic assessment of individuals with prediabetes.³ However, discordance in the identification of these individuals may lead to misdiagnosis, unnecessary medication, or clinical inertia. Additionally, patients with prediabetes frequently present risk factors such as obesity and dyslipidaemia and lack effective predictive markers for T2DM and cardiovascular disease (CVD), which may limit appropriate follow-up and treatment.⁴ Promising biomarkers, such as uric acid and fructosamine, associated with glycaemic abnormalities and insulin resistance (IR), could also be involved as therapeutic targets.^5,6 However, low specificity or difficulty in clinical interpretation limits their use.

Furthermore, emerging technologies, such as artificial intelligence (AI) combined with continuous glucose monitoring (CGM), may be defining more accurate prediction of T2DM in at-risk patients, while stratification of individuals with prediabetes could provide an opportunity for more effective assessment and management.^2,7 The aim of this review is to highlight the heterogeneity of individuals with prediabetes and the need for changes in the identification and early intervention of these subjects.

Development of the concept and differences in the diagnosis of prediabetes

The glucose values established for the diagnosis of diabetes and prediabetes were initially considered arbitrary. Their cut-off values have evolved according to the presence of microvascular and macrovascular complications described in epidemiological studies. Impaired glucose tolerance (IGT), around 1979, was the first concept of prediabetes and remains to this day a diagnostic criterion for this disorder. Its assessment is performed following a 75 g anhydrous glucose load (oral glucose tolerance test [OGTT]), and it is defined as a 2-h plasma glucose value between 140 and 200 mg/dl³. In 2003, the American Diabetes Association (ADA) guidelines updated the diagnostic criterion for impaired fasting glucose (IFG), establishing the value between 100 and 126 mg/dl. However, the World Health Organization suggests maintaining the initially proposed value (110 mg/dl), as different epidemiological studies suggest that this value provides better prognostic prediction for T2DM and cardiovascular disease (CVD) compared with 100 mg/dl². In 2009, the ADA also considered that an HbA1c value ≥ 5.7%, but < 6.5%, constitutes a diagnostic criterion for prediabetes. However, the International Expert Committee (IEC) does not accept this value and proposes a cut-off between 6 and 6.5%, since up to 50% of individuals with that HbA1c would develop T2DM, compared with 25% of those with values between 5.7 and 6%³. These differences in the diagnosis of prediabetes have revealed several drawbacks. On the one hand, the established criteria (IFG, IGT, or HbA1c ≥ 5.7%) do not identify the same individuals.⁸ On the other hand, it has been shown that different diagnoses, such as IGT and IFG, differ in the mechanisms involved in glucose homeostasis. Although both individuals with IFG and IGT show reduced insulin secretion in the early phase, individuals with IGT also exhibit impaired insulin secretion in the late phase. Furthermore, individuals with IGT present marked peripheral IR with only mild hepatic IR, whereas individuals with IFG show the opposite situation⁹ (Table 1). Consequently, opinions regarding the diagnosis of prediabetes vary considerably, ranging from supporting the current definition to abandoning the concept altogether. Recently, the term intermediate hyperglycaemia has been proposed to encompass the different abnormalities within the prediabetes range and avoid a label suggesting a “pre-disease”.⁴ However, this does not solve the main problem, namely, how to identify these individuals and how to implement a more appropriate intervention.

Table 1.

Comparison of glucose-based diagnostic methods for prediabetes.

	IFG	1h-PG during OGTT	2h-PG during OGTT	HbA1c	CGM
Reference blood glucose value	Fasting ≥ 100–125 mg/dl	1h ≥ 155 mg/dl during OGTT	2h ≥ 140–199 mg/dl during OGTT (IGT)	5.7–6.4% (average glucose: approximately 117–137 mg/dl)	Adaptable, not standardised. TIR assessment has been suggested
Predominant pathophysiological abnormality (more than one defect may coexist)	β-cell dysfunction	Reduced β-cell sensitivity	Loss of β-cell function	β-cell dysfunction	Impaired early- and late-phase β-cell insulin secretion
	Impaired early-phase insulin secretion		Impaired late-phase insulin secretion
	Hepatic insulin resistance	Impaired early-phase insulin secretion	Peripheral or muscular insulin resistance	Impaired early-phase insulin secretion
Prediabetes diagnosis and diabetes prognosis	Identifies individuals with prediabetes with variable sensitivity	Suggested as an early marker of dysglycaemia	Reference standard for diagnosing prediabetes, high sensitivity	Recommended for diagnosing prediabetes with good specificity and moderate sensitivity	Not validated for diagnosing prediabetes
	RR may progressively increase with IFG, rising within the IFG range			Good predictor of T2DM at values of 5.7–6.4%
	Evidence of T2DM risk similar to IGT	Different studies have shown that individuals with NGT and elevated 1h-PG have a greater risk of developing T2DM compared with 2h-PG	As a predictor of T2DM, high specificity but variable sensitivity (depending on population and lower compared with IFG or HbA1c)	The 6.0% cut-off has shown superior prediction compared with 5.7%	Individuals with glucose levels ≥ 130 mg/dl for ≥ 10% of CGM time showed greater diabetes risk
	Low CVD risk
Strengths of the test	Most prediabetes prevalence studies are based on this test	Better predictor of prediabetes compared with 2h-PG	Largest number of T2DM and CVD prediction studies	Reflects integrated glucose levels over approximately 180 days	CGM identifies glucose variability not detected by other tests
	Recognised worldwide			Convenient and highly reproducible
	Most widely used test	Includes fasting glucose during testing	Allows assessment of glycaemic response following oral glucose challenge (assesses IFG and IGT)	Reduces daily variations caused by stress and/or illness.	Potential behavioural changes to improve lifestyle
				Universally standardised measurement
Practicality and accessibility of the test	Single blood sample	Low cost	Low cost	Single blood sample	High cost
	Low cost
	Available in areas with limited access to HbA1c	Available in areas with limited access to HbA1c testing	Availability in areas with limited access to HbA1c testing	No fasting or patient preparation required	Would require user training
Limitations of the test	Requires overnight fasting	Requires overnight fasting	Requires overnight fasting	Lower sensitivity than IFG and 2h-PG	Requires time availability and patient training for sensor use
			Need for OGTT (more than one sample)
			Sensitive to daily variation (diet or exercise)	Accuracy concerns
	Less sensitive than 2h-PG	Need for OGTT (more than one sample), at least 1 h duration	Values may vary according to time of day	Interpretation may be affected by age (children and young adults), pregnancy, renal failure, HIV infection, anaemia, haemoglobinopathies	Limited access to a few healthcare centres
			Reproducibility not as good as IFG or HbA1c
			Requires at least 2 h	Weakly associated with diabetes pathophysiology

1h-PG: 1-h plasma glucose during OGTT; 2h-PG: 2-h plasma glucose during OGTT; CGM: continuous glucose monitoring; CVD: cardiovascular disease; HbA1c: glycated haemoglobin A1c; IFG: impaired fasting glucose; IGT: impaired glucose tolerance; NGT: normal glucose tolerance; OGTT: oral glucose tolerance test; RR: relative risk; T2DM: type 2 diabetes mellitus; TIR: time in range.

Source: table created based on Lizarzaburu et al.,² ElSayed et al.,³ Tura et al.,⁹ Bergman et al.¹⁰ and Bergman et al.¹¹

Prediction of diabetes and associated complications using current criteria

The heterogeneity of prediabetes may determine different outcomes and complications according to each diagnostic criterion. It has been described that the risk of progression to T2DM depends on whether the diagnosis is based on IFG, IGT, or glycated haemoglobin A1c (HbA1c)⁹ (Table 1). Other factors such as age, low educational level, or physical inactivity have also been associated with prediabetes, and, moreover, in some cases their relationship is linked to only one criterion.³ In recent years, different reports have proposed that elevated 1-h plasma glucose (1h-PG) in individuals with normal glucose tolerance (NGT) may identify subjects in early stages of prediabetes. In this regard, the International Diabetes Federation (IDF) has proposed that 1h-PG ≥ 155 mg/dl (8.6 mmol/l) is a more sensitive and earlier marker for identifying individuals at risk of T2DM, while also showing differences in incidence compared with IGT across different populations.¹⁰ In a German cohort, 1h-PG showed greater predictive power for future T2DM than IFG, elevated 2-h plasma glucose during OGTT (2h-PG), and HbA1c ≥ 5.7%, based on comparison of the areas under the receiver operating characteristic curves (AUROC 0.84, 0.70, 0.79, and 0.73, respectively).¹¹ These findings were confirmed in different ethnic groups, such as Mexican American, Chinese, and Indian adults. Initially, glucose tolerance curve values, including 1h-PG, may be useful for stratifying individuals with prediabetes and differentiating phenotypes, also considering factors such as adiposity or impaired renal or hepatic function.^2,12

A recent meta-analysis reported that the association between prediabetes and all-cause mortality was greater in individuals with IGT compared with IFG and HbA1c.¹³ This study also found a moderate association between prediabetes and increased risk of mortality, cardiovascular events, chronic kidney disease, several types of cancer, and dementia, with relative risks up to 39% higher than in individuals with normal glucose levels. However, no associations were observed with the incidence of depressive symptoms or cognitive impairment, with low or very low certainty of evidence.

Prediabetes is not necessarily a continuum leading to diabetes

Although prediabetes constitutes a disorder with significant risk of progression to diabetes, as well as cardiometabolic diseases,³ it should not necessarily be classified as part of a “continuum”. Individuals with prediabetes may progress to different outcomes: develop T2DM, remain in the prediabetic state, or return to normal glucose levels (Fig. 1). Moreover, progression to T2DM from prediabetes may differ according to disease duration or the diagnostic criterion considered (IFG, IGT, or HbA1c ≥ 5.7%), whereas reversal from prediabetes to normoglycaemia may decrease over time and has been observed even after 11 years of follow-up according to a meta-analysis published by Richter et al.¹⁴ The same report showed that the highest cumulative incidence of T2DM in individuals with prediabetes, and the strongest association with progression to T2DM compared with normoglycaemia, was observed with the combined diagnosis of IFG and IGT (HR = 6.90; 95% CI: 4.15–11.45), whereas IGT, IFG, and HbA1c ≥ 5.7% showed lower incidence and weaker associations.

Figure 1.

Graphical representation of potential prediabetes outcomes expressed as different abnormalities in the glucose tolerance curve.

1h-PG: 1-h elevation during OGTT; CV: cardiovascular; IFG: impaired fasting glucose; IGT: impaired glucose tolerance; MASLD: metabolic dysfunction-associated steatotic liver disease; OGTT: oral glucose tolerance test.

Source: figure created based on Bergman et al.,¹¹ Wang et al.¹² and Schlesinger et al.¹³

On the other hand, in studies evaluating T2DM prevention (randomised clinical trials [RCTs]), such as the Finnish Diabetes Prevention Study, the Da Qing study, or the Diabetes Prevention Program (DPP), in individuals with IGT or IGT + IFG (only in the DPP), reduction in progression to diabetes ranged from 38% to 58% during the intervention period with lifestyle modification or pharmacological treatment (metformin in the DPP). In these studies, it was observed that patients who did not progress to T2DM either remained in the prediabetic state or returned to normal glucose levels.¹⁵ In the DPP, it was demonstrated that, after intervention and 10 years of follow-up (Diabetes Prevention Program Outcomes Study [DPPOS]), subjects who normalised their glucose levels had lower risk of T2DM and microvascular complications, particularly if they reduced their body mass index (BMI).¹⁶

These RCTs showed that not all patients with prediabetes, despite presenting risk factors such as obesity or family history of diabetes, will progress to T2DM, and some may even return to normal glucose levels. On the other hand, although progression from prediabetes to T2DM implies increased CVD risk, recent studies have shown that only those individuals with IGT may reduce this risk by reverting from prediabetes to normal glucose levels.¹⁷ Other specific clinical characteristics, such as BMI, fasting glucose, or lipid control, may influence reversal from prediabetes to normoglycaemia. A recent systematic review evaluated 19 cohort studies in Europe, America, and Asia using risk models to obtain subdistribution hazard ratios (SHR) specific to each cohort, highlighting that overweight (SHR = 0.88; 95% CI: 0.76–0.99), obesity (SHR = 0.66; 95% CI: 0.52–0.81), and reduced HDL-C concentration (SHR = 0.72; 95% CI: 0.59–0.84) were among the important factors significantly reducing reversal from prediabetes to normoglycaemia.¹⁸

Need for new markers in prediabetes

The diagnosis of prediabetes is defined according to glucose or HbA1c values. However, factors such as lipotoxicity, β-cell dysfunction, and altered incretin release are also involved in this disorder and contribute to a state of chronic inflammation. Adiponectin, a hormone with anti-inflammatory effects and a stimulatory role in fatty acid oxidation, has been suggested as a possible biomarker, as low levels are associated with peripheral IR, atherosclerosis, and metabolic dysfunction-associated steatotic liver disease (MASLD), while biomarkers such as fructosamine or glycated albumin have been shown to be superior to HbA1c, overcoming the limitation of measurement in patients with renal failure or haemolytic anaemia, although they may show low values in individuals with elevated BMI and visceral adiposity. Elevated α-hydroxybutyrate levels have also shown a significant association with IR, independently of BMI, age, or sex, and may be associated with prediabetes.¹¹ Nevertheless, the main limitation of new biomarkers is that they do not define, among those who develop IR and present risk factors, which individuals will achieve adequate compensation in insulin secretion or inevitably progress to T2DM.

Lipid markers have greater applicability in clinical practice. Elevated triglyceride levels are associated with reduced insulin secretion, whereas low HDL-C levels may be a marker of progression to T2DM. Specifically, small HDL3 subparticles, measured mainly in clinical and epidemiological studies assessing cardiovascular risk, have been associated with high triglyceride levels and are significantly elevated in individuals with prediabetes.¹⁹ It is important to mention that some Native American or Amerindian populations present low HDL-C levels as a genetic trait, particularly the ABCA1 gene variant (C230), which would not inherently constitute a disease risk factor, unlike what occurs in other ethnic groups (e.g. Caucasians). Recently, the triglyceride/glucose index (TyG) and remnant cholesterol (RC) have been described as potentially useful biochemical markers of prediabetes.²⁰ The former for assessing IR and predicting the risk of developing T2DM, comparable or even superior to traditional markers such as HOMA-IR, and the latter because of a stronger association with the diagnosis of diabetes, prediabetes, and IR compared with conventional lipids and their ratios in the general population, showing a stable relationship with these conditions even when HDL-C, LDL-C, or TG levels are within appropriate ranges.²¹ Even so, further evidence is needed to demonstrate that individuals with prediabetes and elevated RC levels are at greater risk of T2DM than those with normoglycaemia. On the other hand, it is important to mention that, in individuals with prediabetes, elevated uric acid levels and impaired renal function may independently be associated with a higher risk of developing T2DM, whereas the opposite situation may be associated with reversion to normoglycaemia.⁶

Risk scores such as FINDRISC or the ADA Risk Score, or even metabolic syndrome (MetS), may be useful, particularly in primary care, when there is adequate familiarity with and validation for their application. However, these scores were designed to assess the general population and identify individuals at risk of T2DM, not necessarily those already diagnosed with prediabetes.³ In a recent publication, we evaluated the prognostic value of 1h-PG ≥ 155 mg/dl in T2DM compared with the FINDRISC score and MetS and interestingly observed that 1h-PG was superior to both scores.²²

Risk group stratification (phenotypes) or clustering could improve detection and intervention in individuals with prediabetes. Consideration of clinical variables such as estimated glomerular filtration rate, MASLD, or even menopausal status, in addition to glucose values, may be necessary for categorisation. In this regard, Wagner et al.²³ identified six subphenotype groups in a cohort of non-diabetic Caucasian individuals based on assessment of variables such as IFG and IGT, body fat distribution (by magnetic resonance imaging), liver fat content, and genetic risk. Three of these groups had elevated glucose levels, but only two (both with significant insulin secretory deficiency) presented imminent risk of diabetes. Interestingly, one cluster showed a moderate risk of developing T2DM but a higher risk of kidney disease and all-cause mortality. This demonstrated that pathophysiological heterogeneity exists before the diagnosis of T2DM and highlights groups of individuals at greater risk of complications, even without rapid progression to T2DM. However, these clusters have shown limitations and inconsistencies in different populations, as suggested by the same author in a recent report involving samples from different communities in China, where too many metabolic variables showed substantial intraindividual variation.²⁴ Recently, a report analysing a subset of participants (n = 994) from the DPP clinical trial applied an unsupervised clustering analysis known as k-means to determine the optimal number of participant groups based solely on common clinical risk factors or on common risk factors together with more comprehensive OGTT and body composition measures. Five patient phenotype groups were identified, in which participants showed differing risk factor profiles.²⁵ The clinical model showed greater accuracy regarding time to development of T2DM, whereas the more comprehensive models better differentiated a metabolically healthy overweight phenotype.

AI tools in prediabetes

Recent advances in artificial intelligence (AI) have produced models that provide significant accuracy in predicting glucose levels in individuals with type 1 and type 2 diabetes, and even in those with prediabetes. Thus, AI integration has been associated with improvements in key clinical indicators in patients at risk of diabetes and CVD, such as the IR index, HbA1c, and continuous glucose monitoring (CGM) parameters (time above and/or below range), with positive user acceptance in the latter case, demonstrating efficiency and reinforcing the value of AI as a support tool in T2DM prediction.⁷

Recently, CGM has become a useful aid in clinical decision-making in patients with diabetes, by providing a continuous and more objective representation of glycaemic trends. In prediabetes, CGM combined with AI could improve diagnostic accuracy and management of this disorder by identifying dietary triggers and predicting postprandial responses. Some studies suggest that these approaches could complement tests such as the OGTT, facilitating a more functional assessment of glucose metabolism.²⁶ However, its use in prediabetes remains limited, mainly because of cost, CGM availability, and the absence of standardised criteria for interpretation. A study using CGM during home-based OGTT assessments (considering that a 3- or 5-point OGTT has shown utility in predicting progression to T2DM, CVD, and mortality) in individuals with prediabetes (n = 21) employed a machine learning model to classify metabolic subphenotypes (assessing muscle/hepatic IR and β-cell dysfunction). Although these findings suggest potential application in T2DM risk stratification, the small sample size and lack of external validation currently preclude extrapolation to routine clinical practice.²⁷

The most commonly used AI models include prediction networks based on time series (long short-term memory networks [LSTM]), models capable of integrating multiple data sources (multichannel transformers [MCAT]), and more advanced systems aimed at personalised interventions (hybrid systems and foundation models) (Table 2). The latter, more focused on intervention, may represent the forefront of integration into clinical settings, offering automated and individualised recommendations, although their implementation requires substantial datasets and considerable computing capacity. LSTM networks, designed for sequential data analysis, are well suited for short- and medium-term glucose prediction (e.g. 15, 30, or 60 min) through learning temporal patterns in CGM data. In clinical practice, their implementation may be more feasible in the near term, as they are simpler and more specific models focused on short-term glucose prediction.

Table 2.

Recent AI models evaluated for prediabetes identification and diabetes prognosis.

Applied model (reference)	Objective	Study population	Methodology	AI model or system compared	Outcome
LSTM	Short-/medium-term glucose prediction	Healthy individuals with no associated comorbidities, short-term follow-up	Validation with CGM and temporal models; n < 100	Linear models (simple regression)	Useful for point-in-time prediction, without long-term stratification capability
				With CGM data at short/specific time intervals
Multi-channel transformers	Analysis of CGM curves combined with contextual data (physical activity, sleep, food intake, medication administration, emotional stress, etc.)	- Prediabetes subcohorts from AI-READI	Validation on stratified subsets.	• LSTM • GRU (Gated Recurrent Unit) • MLP (Multilayer Perceptron)	- Detection of metabolic subphenotypes with differential risk of progression to DM
		• Stratification by metabolic sub-phenotypes (hepatic IR, muscular IR, β-cell dysfunction). • Classification according to circadian rhythm, lifestyle, glycaemic variability, psychosocial factors	n > 300	With other CGM series + contextual data (analytical metabolic profile, sociodemographic data, lifestyle, chronic stress)	• Better adaptation to individual patterns vs simple LSTM • ↓RMSE 10% • ↓MAE 15% • ↑ accuracy in hyperglycaemia prediction
Hybrid systems	Improved accuracy in time series considering key events (attention mechanisms	Prediabetes and individuals with risk/metabolic syndrome, stratified by intake, estimated insulin release, physical activity	Supervised training with cross-validation; comparisons with pure LSTM and MLP.	LSTM and basic neural networks (DNN-XAI)	• ↑Sensitivity to postprandial variations • Better adaptation to individual patterns vs simple LSTM
			n = 200	Multivariate with CGM data, carbohydrate ratios, insulin and exercise
Foundation models	Risk stratification for progression to DM and cardiovascular events	Healthy individuals without associated comorbidities, long-term follow-up	Pretraining with > 10 million CGM readings; longitudinal study; validation across multiple cohorts; risk analysis and T2DM progression; n > 1 million	LSTM and traditional transformers	- Detection of metabolic subphenotypes with differential risk of progression to DM (higher AUC 0.8–0.9 at 12 months)
				HbA1c, GMI, OGTT; integrated CGM data from different CGM devices	• Precision and personalised medicine

AUC: area under the curve; CGM: continuous glucose monitoring; CV: cardiovascular; DM: diabetes mellitus; DNN-XAI: deep neural network–explainable artificial intelligence; GMI: glucose management indicator; GRU: Gated Recurrent Unit; HbA1c: glycated haemoglobin A1c; LSTM: long short-term memory; MAE: mean absolute error (CGM metric); MLP: Multilayer Perceptron; RMSE: root mean square error (CGM metric).

Basic neural networks: automated learning or machine learning models.

Source: table created based on Stroebel et al.,²⁵ Ji et al.²⁶ and Metwally et al.²⁷

Other more complex models, such as MCAT, integrate physical activity, diet, or sleep data and generate automated recommendations. Among these, GluFormer and AttenGluco, trained with CGM data from healthy, prediabetic, and diabetic individuals, have shown high efficacy in T2DM prediction. The former enabled highly accurate and scalable predictions of diabetes progression and cardiovascular risk, demonstrating superiority compared with HbA1c in various cohorts,²⁸ whereas the latter integrated CGM and physical activity data (from the AI-READI database), showing superior performance compared with LSTM models, achieving a significant reduction in glucose excursion error measures (10% and 15% lower RMSE and MAE, respectively), with favourable results in patients with prediabetes²⁹ (Table 2). These models represent a promising but still experimental avenue because of their high computational cost, the need for large data volumes, and the lack of technical and clinical validation.

Role of medical intervention in prediabetes

During the prediabetes stage, there is an opportunity to prevent or delay progression to T2DM, in addition to reducing the risk of developing cardiometabolic complications. On the other hand, follow-up of at-risk patients provides an opportunity to avoid delayed diagnosis or clinical inertia by enabling early diagnosis of T2DM. However, how should we decide, or which characteristics should we consider, when initiating pharmacological or non-pharmacological treatment in these individuals? Individualising people with prediabetes would probably be the most appropriate approach. Grouping by risk categories or clusters, as described previously, could allow more efficient and cost-effective strategies to reduce the disease burden associated with prediabetes and, despite representing a “non-dynamic” assessment of risk, could be useful in differentiating intervention strategies, with more intensive management in individuals with obesity and/or impaired hepatic or renal function.² In contrast, those who are younger, with lower postprandial glucose or HbA1c values, without associated features of MASLD or subcutaneous adiposity, would have lower risk, suggesting a less intensive intervention, with observation and regular monitoring only.²⁴ An important point to consider is that it has been demonstrated that, following intervention, patients with prediabetes who return to normal glucose values have a lower risk of developing T2DM.¹⁶ Phenotyping of these individuals could therefore regard return to normoglycaemia as a favourable prognostic state for the development of T2DM.

There is sufficient evidence that lifestyle modification interventions have been successful and highly effective in reducing the risk of progression to T2DM, as demonstrated by the DPP, Finnish DPP, and Da Qing studies. However, the conditions under which these studies were conducted are not necessarily replicable in the general population. In the DPP trial, the number needed to treat (NNT) to prevent one case of diabetes over 3 years of follow-up was 7, despite participants undergoing an intensive intervention programme. Included subjects had mildly elevated IFG, IGT, and overweight, with a diabetes conversion rate of 29% in the control group⁴. Despite differences in at-risk populations, lifestyle modification should remain a fundamental component of medical care recommendations.

Regarding pharmacological treatment, benefit has been demonstrated in reducing the risk of developing T2DM with metformin and pioglitazone. The DPP and ACT NOW studies reduced the incidence of T2DM by up to 60% in treated subjects.¹⁵ However, these medications do not have permanent effects on IR or β-cell dysfunction, which probably explains the lack of a sustained effect. On the other hand, treatment of other complications associated with prediabetes has not produced the expected results. Although a post hoc analysis of the STOP-NIDDM study reported beneficial effects of acarbose on cardiovascular events in individuals with prediabetes (cumulative event incidence 4.7% in the placebo group versus 2.2% in the acarbose group),¹ the recent publication from the DPPOS group reported that neither metformin nor lifestyle intervention reduced major cardiovascular events in patients with prediabetes over 21 years, despite long-term prevention of diabetes.³⁰ The use of drugs in prediabetes remains controversial, and “overtreatment” should be avoided. However, treatment of associated complications in the prediabetic patient may be beneficial. Obesity, dyslipidaemia, MASLD, and elevated uric acid may be present in individuals with prediabetes and MetS,^2,24 so treating these conditions may benefit the patient, depending on the intervention goal and the risk–benefit balance of pharmacological treatment.

Conclusions

Prediabetes is a risk state, but not necessarily a continuum towards T2DM. Certain factors may determine progression to disease, but also the possibility of returning to normoglycaemia in individuals with prediabetes.

Stratification or phenotypic clustering of patients with prediabetes appears to be key for the assessment and treatment of these individuals. Glycaemic values, clinical variables, and the use of technological tools such as CGM combined with AI may be necessary in this process. However, integration of AI into clinical practice requires larger studies, standardisation of methods, and critical evaluation of its true cost-effectiveness.

Personalised intervention is necessary. Lifestyle changes remain the cornerstone, preventing in many cases the development of cardiometabolic complications and progression to T2DM, as has been demonstrated. Pharmacological treatment should be guided by the results of risk stratification, with the aim of achieving specific clinical control objectives.

Key messages

• •

  Initial assessment of individuals with prediabetes should focus on identifying the predominant glycaemic abnormality (IFG, IGT, elevated 1h-PG, or HbA1c).

• •

  It is essential to assess other clinical or biochemical risk characteristics, such as obesity, TyG index, MASLD, or uric acid levels, for appropriate patient risk stratification (phenotyping).

• •

  Artificial intelligence (AI) combined with continuous glucose monitoring (CGM) is emerging as a key tool for the early diagnosis and prognostic assessment of prediabetes.

• •

  The management and treatment of patients with prediabetes represent a continuum and are based on appropriate initial assessment and risk stratification.

Funding

This research received no specific grant from any public, commercial, or not-for-profit funding agency.

Acknowledgements

The authors thank Dr Alonso Garro-Mendiola for his collaboration in the development of Fig. 1 and the Endocrinology Department of Hospital Universitario Fundación Jiménez Díaz for the logistical and academic support provided in the preparation of this review.