Various classification methods for diabetes mellitus in the management of blood glucose control

Abstract

In the era of precision medicine, the classification of diabetes mellitus has evolved beyond the traditional categories. Various classification methods now account for a multitude of factors, including variations in specific genes, type of β-cell impairment, degree of insulin resistance, and clinical characteristics of metabolic profiles. Improved classification methods enable healthcare providers to formulate blood glucose management strategies more precisely. Applying these updated classification systems, will assist clinicians in further optimising treatment plans, including targeted drug therapies, personalized dietary advice, and specific exercise plans. Ultimately, this will facilitate stricter blood glucose control, minimize the risks of hypoglycaemia and hyperglycaemia, and reduce long-term complications associated with diabetes.

Key Words: Diabetes classification; Glycaemic control; Personalised treatment; Soft clustering; Precision medicine

Core Tip: The prevalence of diabetes is increasing, and blood glucose control is of utmost importance. This article explores various diabetes mellitus classification methods, for example, incorporating phenotypic subtypes, genetic subtypes, and soft clustering, and highlights the limitations of traditional classification methods. Updated classification methods can help develop personalised treatment plans, which are important for optimal diabetes management.

INTRODUCTION

The prevalence of diabetes mellitus (DM) has been steadily increasing, with a global prevalence of 10.5% among people aged 20-79 years in 2021, which is projected to rise to 12.2% by 2045[1,2] Advances in medical technology have shown that stem cells can differentiate into pancreatic β-cells, but clinical application of stem cells for restoring insulin secretion is still experimental[3]. The development of novel drugs, such as oral insulin[4], long-acting insulin analogs[5], and glucagon-like peptide-1 receptor agonists[6], is advancing, but these are not yet widely available or approved. Despite significant advances in the treatment of DM in recent decades, only 50% of patients with DM establish a glycosylated haemoglobin level of < 7%[7]. Effective glycaemic control remains a crucial means of preventing organ dysfunction and DM-related complications.

Different classification methods of DM, a complex chronic disease, directly influence the formulation of treatment plans and the efficacy of blood sugar control. For example, latent autoimmune diabetes in adults (LADA) is classified into LADA type 1 [high-titer LADA with a glutamic acid decarboxylase antibody (GADA) titer ≥ 180 U/mL] and LADA type 2 (low-titer LADA with a GADA titer < 180 U/mL). In patients with LADA type 1, the relatively rapid decline of islet function resembles classic type 1 DM (T1DM). These patients usually need to monitor their blood sugar closely and adopt an intensive treatment regimen combining basal insulin with prandial insulin. In contrast, patients with LADA type 2 have better islet function and can often manage with lifestyle interventions combined with oral hypoglycemic medications[8]. This study aimed to explore the role of DM classification methods in glycaemic control, with the ultimate goal of optimising DM management strategies.

LIMITATIONS OF TRADITIONAL CLASSIFICATION METHODS

Traditional classification methods of DM mainly rely on simple distinctions based on pathogenesis and clinical manifestations, and misdiagnosis is not uncommon. T1DM is caused by the autoimmune destruction of pancreatic islet β-cells, leading to complete insulin deficiency. However, certain types of DM, such as LADA, or some monogenic forms of DM, e.g., maturity-onset DM of the young (MODY)[9], that manifests during adolescence, can present symptoms similar to those of T1DM, resulting in misdiagnoses. Type 2 DM (T2DM) is characterized by insulin resistance accompanied by progressive insulin secretion deficiency, or insulin secretion deficiency with or without insulin resistance. However, some patients with T2DM develop ketoacidosis[10] under stress conditions (severe infections, surgeries, etc.), which is prone to being misdiagnosed as T1DM. Additionally, some subtypes of T2DM, e.g., in obese patients with abnormal fat distribution and metabolic disorders[11], are not easily distinguished using traditional classification methods, leading to inconsistencies in the selection of treatment regimens. Traditional classification methods can also fail to comprehensively cover and accurately diagnose various types of DM with clear etiology: For example, DM caused by pancreatic exocrine disease (such as pancreatitis, post-pancreatectomy, etc.) may show only a slight increase in blood glucose levels initially, leading to misdiagnosis as T2DM, rather than identifying the underlying cause as pancreatic disease; gestational DM (GDM) refers to abnormal glucose metabolism first detected or developed during pregnancy. However, recent studies show that some women with GDM had undiagnosed abnormal glucose metabolism before pregnancy[12]. Thus, the traditional classification method has limitations in clinical practice, which can lead to misdiagnosis of various types of DM. With continued medical research and developments in precision medicine, better classification and diagnostic methods are needed to improve DM diagnosis and treatment[13].

CLASSIFYING DIABETES BY PHENOTYPIC SUBTYPE

In 2018, Ahlqvist et al[14] proposed a reproducible algorithmic classification method, the Ahlqvist classification. Based on six clinical variables: Age, body mass index, glycated haemoglobin, homeostasis model assessment 2 (HOMA2)-insulin resistance index, HOMA2-β-cell function index, and GADA, newly diagnosed DM patients are classified into five subtypes: Severe autoimmune diabetes (SAID), severe insulin-deficient diabetes (SIDD), Severe Insulin-Resistant Diabetes (SIRD), mild obesity-related diabetes (MOD), or mild age-related diabetes (MARD)[15]. SAID consists of T1DM or LADA, in which patients are usually younger at onset, are of normal or low body weight, have a defective immune system, and tend to have higher and more fluctuations in blood glucose. It is important that this group of patients undergoes enhanced eye examination and care because of the higher risk of blindness, and early initiation of insulin therapy and optimisation of immunosuppressants. Patients with SIDD are relatively young, have high blood glucose levels, and a greater risk of renal failure. They must pay close attention to renal function indicators, such as blood creatinine and urea nitrogen, to control their blood pressure, and reduce their protein intake[16]. Patients with SIRD are overweight or obese, have metabolic abnormalities and severe insulin resistance, but relatively good pancreatic function. Patients with SIRD can improve insulin sensitivity with medications such as metformin and by combining dietary control, reducing the intake of high-calorie, high-fat foods, and exercise therapy to increase physical activity[17,18]. Patients with MOD and MARD suffer from mild insulin resistance and insufficient insulin secretion, resulting in mildly elevated blood glucose levels, yet overall, the situation is relatively stable. Usually, blood glucose can be effectively controlled by three months of dietary adjustment, moderate exercise, and regular blood glucose monitoring. Otherwise, medications such as metformin, insulin secretagogues, dipeptidyl peptidase-4 inhibitors, and acarbose are required for treatment[19]. Pedersen et al[20] categorised T2DM into three subtypes: Classic, hyperinsulinaemic, and insulinopenic T2DM based on β-cell function, insulin sensitivity, and HOMA2 segmentation. These authors discovered that patients with hyperinsulinaemic T2DM had a reduced risk of diabetic retinopathy and disease progression compared with patients with classic T2DM. Serum cystatin C levels may also refine risk stratification for mortality and vascular outcomes in patients with prediabetes or DM[21].

These clustering approaches are superior to traditional DM classifications because they identify patients at a high risk of DM complications at the time of diagnosis and provide information on the underlying disease mechanisms to guide the choice of treatment. Based on this information, a multidisciplinary team composed of endocrinologists, dietitians, exercise therapists, etc. can be established. Dietitians formulate personalized dietary plans according to the phenotypic characteristics of patients. For example, for patients with obvious abdominal obesity and insulin resistance, should increase the intake of dietary fibre and limit carbohydrate intake. Exercise therapists design exercise plans according to the patients' physical condition and phenotypic characteristics to help them improve their metabolic status, thereby achieving good control of blood glucose. However, classification based on phenotypic subtype is based mainly on biochemical indicators and simple demographic characteristics, and fails to fully consider the impact of environmental, socioeconomic, cultural and other factors on the onset, development and treatment response of DM. Moreover, it does not adequately reflect the diversity of the patient population, e.g., ethnicity, region, and lifestyle, which can affect the applicability of the classification results. Some patients on the "fringe" of a subtype may not fully meet the criteria of that subtype, and conversion between subtypes can occur. This new phenotypic classification method requires further research to establish clear guidelines for the definition and management of these patients.

CLASSIFYING DIABETES BY GENETIC SUBTYPE

Beyond phenotypic information, genetic data can be harnessed to subclassify DM. For monogenic genetic disorders, genetic testing can provide patients with a definitive diagnosis and can better characterise different genetic disease subtypes. For example, patients with MODY caused by glucokinase mutations are characterised by mild, non-progressive hyperglycaemia with a low risk of complications and usually do not require treatment[22]. Individuals with HFN1A- and HNF4A-MODY can achieve good control with sulfonylureas or glucagon-like peptide 1 receptor agonists without insulin[23]. Advanced genotyping techniques can be used to predict the risk of childhood onset[24] and the propensity for insulin therapy in patients with LADA[25]. A more comprehensive understanding of the disease can be attained by assembling genetic components into a novel risk score. The Singapore Genome Institute[26] constructed the T1DM macro gene Z-scoring tool using the single-cell data of 46 individuals with overt T1DM, and classified the patients into different molecular subtypes. In subtypes with severe immune cell dysregulation, the immune system may have launched a more intense attack on the pancreatic islet β-cells: More severe β-cell damage equates to lower insulin secretion, and thus these patients are more likely to experience blood glucose fluctuations and difficulty in maintaining blood glucose levels within the normal range. Suzuki et al[27] identified eight clusters of exponential single nucleotide variants with distinct profiles associated with T2DM and 37 cardiometabolic phenotypes. Genome-wide association study (GWAS) analysis has identified more than 1000 gene loci that are significantly associated with T2DM[28], and mutations or aberrant expression of some of the key genes (e.g., TCF7 L2) can lead to abnormal insulin secretion, increased insulin resistance, and glucose metabolism disorders, which contribute to the development of T2DM[29]. Therefore, DM-related genetic testing programs should be promoted in medical institutions with appropriate facilities. For patients with a family history of DM, early onset disease, or atypical clinical manifestations, genetic testing is recommended to enable more accurate decision-making for subsequent treatment. Genotyping uses germline genetic markers is not affected by disease progression or treatment. However, the current research in this field is not without limitations: genetic data has mainly focused on European populations, severely restricting applicability to other ethnic groups; and detail on how genetic subtypes influence disease progression and treatment responses is lacking. It is evident that the newly proposed T1DM macro gene Z risk score requires further validation before it can be widely applied in clinical practice.

CLASSIFYING DIABETES THROUGH “SOFT” CLUSTERING

“Soft” clustering methods represent a data processing technique to dissect the heterogeneity within groups of patients with DM. People with DM belong to multiple clusters of varying degrees rather than being explicitly categorised into a single cluster[30]. In the IMI DIRECT study, researchers identified four quantitative prototypes based on soft clustering of 32 phenotypes, reflecting different patterns of dysfunction in the etiologic process of T2DM[31]. Udler et al[32] applied Bayesian non-negative matrix factorisation clustering to GWAS results for 94 independent T2DM genetic variants and 47 DM-related traits and identified five robust clusters of T2DM loci and traits, each with distinct tissue-specific enhancer enrichment based on analysis of epigenomic data from 28 cell types. DM presents a continuum of dysfunction in newly diagnosed individuals, and soft classification methods (such as prototype analysis) can detect prototypes that reflect a continuum of dysfunction combinations over the course of an individual's underlying aetiology. A big data analysis platform is needed for integration of multiple sources of patient data, including clinical, genetic, and lifestyle data. By using soft clustering algorithms to analyse this data, doctors can quickly understand the clustering characteristics of patients and formulate personalized treatment plans. However, the clinical application of soft clustering still faces some challenges: It is difficult to determine which phenotypes are the main drivers of clustering, and there is a lack of clarity in prioritizing and interpreting these phenotypes. Therefore, we encourage medical institutions to conduct clinical studies based on the soft clustering classification method to verify the effectiveness and practicality of this method in the treatment of DM. Through continuous research, the soft clustering model and treatment plans can be optimised, and the findings can be gradually translated into clinical practice to improve blood glucose control.

SUBCLASSIFYING THE HIGH-RISK DIABETES POPULATION

Previous studies[33] integrating phenotypic values (such as oral glucose challenge, magnetic resonance imaging-measured body fat distribution, and hepatic lipid content) and polygenic risk for T2DM identified six clusters with varying propensities for DM and comorbidities in a population without DM but at elevated risk. Some of these clusters exhibited characteristics analogous to those of the African American Study of Diabetes Remission clusters, albeit with inconsistent membership in incidence clusters. More aggressive interventions (e.g., pharmacotherapy) combined with lifestyle interventions are needed to reduce the risk of developing T2DM in individuals with high-risk DM subphenotypes (e.g., subphenotypes 5 and 6)[34]. For different subgroups of high-risk populations, more targeted preventive strategies can be developed according to their specific risk factors and characteristics. However, there are numerous influencing factors for high-risk DM populations, and it is not easy to formulate a comprehensive, accurate, and simple-to-implement subclassification standard. Different classification methods may lead to the same population being classified into different categories, affecting the accuracy and consistency of the diagnosis.

CONCLUSION

The heterogeneity of DM is reflected in various aspects, including genetics, phenotype, and environmental factors[35]. Integrating multiple types of data and conducting in-depth research on DM subtypes among different populations can greatly enhance our understanding of the complex pathogenesis of DM, and is helpful for predicting disease progression in individual patients. Moving forward, a large number of clinical trials with rigorous designs, sufficient sample size, and covering different populations are needed to verify the clinical significance of the various classification methods[36]. Additionally, effective decision support tools based on the integration of multi-source data and advanced algorithms should be developed to analyse patient data in real time. These additional studies and tools will reshape medical standards for DM. By accurately classifying DM subtypes and providing customized medical services according to the characteristics of individual patients, we can effectively control blood glucose, reduce the occurrence of complications, lower medical costs, and provide patients with higher-quality, more efficient, and personalised medical services[37].

ACKNOWLEDGEMENTS

We would like to express our gratitude to the team members who contributed ideas and thoughts to this article.