Type 2 diabetes mellitus (T2DM) is a highly prevalent endocrine and metabolic disease worldwide (1), which causes a variety of chronic complications such as cardiovascular disease (CVD), non-alcoholic fatty liver disease (NAFLD), renal disease, diabetic neuropathies, amputation, blindness, mortality, some brain lesions and so on (2–5). T2DM and its chronic complications bring a huge burden to the global medical and economy (6, 7). Meanwhile, chronic complications of T2DM are the leading causes of CVD, blindness, end-stage renal disease, and high mortality in China (7–9). Currently, experts divide the chronic complications of T2DM into microvascular complications, atherosclerotic cardiovascular disease (ASCVD), nervous system complications (NSC), diabetic foot (DF), and others (eye, oral cavity, skin lesions, etc.) (10), to facilitate clinical diagnosis and classification of complications. Unfortunately, the current classification measure mainly based on the location of target organs of lesion damage cannot effectively reduce the occurrence of chronic complications of T2DM or slow down their progress, because this classification method fails to take into account the pathology, mechanisms, and associated risk factors of various chronic complications. Another not-so-optimistic situation is that most patients with T2DM often develop multiple chronic complications at the same time, which greatly increases the difficulty of treatment for clinicians. Therefore, we need a more accurate division way to differentiate the T2DM patients with chronic complications, it could help doctors to clearly stratify these patients based on their clinical characteristics, the most important of all, to give effective intervention targeting the distinguished risk factors of different hierarchical patients.

Recently, cluster analysis based on clinical features has played an important role in clinical treatment strategies and disease management in many diseases. Ahlqvist E et al. performed a cluster analysis of 8980 patients with adult-onset diabetes based on 6 variables (glutamate decarboxylase antibodies, age at diagnosis, BMI, HbA1c, homoeostatic model assessment 2 estimates of β-cell function, and insulin resistance), and identified 5 replicable clusters (11). They found that different clusters of diabetic patients had significantly different characteristics and risks of diabetes complications, which helps the implementation of precision medicine for diabetes and its complications (11). Meanwhile, they confirmed their results in 3 independent cohorts. Subsequently, more studies reproduced the results of cluster analysis in multiple regions, including China (12, 13). Furthermore, based on this method, some researchers conducted cluster analysis on patients with T2DM, and compared the differences in clinical characteristics, disease progression, treatment response, and complication status of different subgroups, providing a basis for the new strategy for treatment of T2DM (14, 15). Similarly, this method of cluster analysis has also obtained meaningful results in the subgroup classification of diseases such as metabolic-associated fatty liver disease (MAFLD) (16), idiopathic pulmonary fibrosis (IPF) (17), ASCVD (18), bronchiectasis (19), and so on. Furthermore, these results have innovative implications for the early diagnosis of diseases, precise grouping, determining the severity of diseases, and providing optimization of clinical care.

In addition, several studies have explored the potential impact of serum metabolite levels on the development of T2DM and its complications. In 2022, an Updated Systematic Review and Meta-analysis of Prospective Cohort Studies investigated the relationship between metabolomics and the risk of T2DM, the findings revealed significant associations between certain metabolites and the risk of developing type 2 diabetes (20). Recently, according to the study conducted by Wang S et al., there is a positive correlation between a range of amino acids and serum carnitine with T2DM (21). And these studies have identified the related biomarkers, such as branched-chain amino acids, metabolites of phenylalanine, metabolites involved in energy metabolism and lipid metabolism, thereby improved the prevention and treatment of T2DM and its complications (22).

This observational study included the real data of hospitalized patients with chronic complications of T2DM. We aimed to identify characteristic clusters of inpatients with chronic complications of T2DM by K-means clustering based on common clinical variables. We also sought to further analyze the risk of associated chronic complications in each cluster and to compare the differences in metabolite levels among the clusters.

The K-means method is a traditional and relatively simple machine learning method (28). It had been used for iterative subspace projection and clustering, consensus clustering (29), disease phenotype recognition (11, 16), and the results hold significant value in the clinical progression study of the disease. In this retrospective real data study, 4 clusters of patients with chronic complications of T2DM were divided using K-means cluster analysis based on 11 common clinical variables. Each cluster showed substantially different phenotypes in clinical patterns and the risks of microangiopathy, ASCVD, and NSC were significantly differential enriched in each cluster. In the current analysis, we also found distinct metabolic characteristics and specific metabolites in each cluster.

Cluster analysis based on clinical features is of great value in disease subtype decision-making. Several published studies in recent years have tried to phenotype T2DM patients. These studies proved novel hierarchical clusters of T2DM patients could express significant differences including clinical characteristics, disease progression, complications, and treatment responses (14, 15, 30). To the best of our knowledge, limited studies focused on T2DM patients with chronic complications. In this current study, we stratified T2DM patients with chronic complications into 4 distinct clusters based on clinical data: the Low-Risk cluster, the Medium-Low-Risk cluster, the Medium-Risk cluster, and the High-Risk cluster. The four clusters showed different risks of the accumulation of chronic complications for T2DM patients, the High-Risk cluster had 55.55% of the patients with comorbidity of three chronic complications. Moreover, patients in the High-Risk cluster had more risk to have microangiopathy (OR:9.644, p<0.001, after adjusting for 20 covariates), especially for DR (OR:15.602, p<0.001, after adjusting for 20 covariates). That means patients in the High-Risk cluster need to pay more attention to preventing the occurrence of multiple complications of T2DM.

What is special is that, compared with the Low-Risk cluster, the ASCVD risk of the Middle-Low-Risk cluster was higher (OR:1.440, p=0.003, no variables were adjusted), and it was also relatively high in the High-Risk cluster (OR >1, although p<0.05, regardless of adjustment for covariates), but the Medium-Risk cluster shows a lower ASCVD risk (OR:0.526, p=0.006, after adjusting for 20 covariates). Tracing back to the variable characteristics of each cluster in Table 2 , it was found that the corresponding levels of SBP and DBP in the Middle-Low-Risk cluster and High-Risk cluster were higher, and also showed relatively higher ASCVD risk values; while the blood pressure control level of patients in the Middle-Risk cluster was slightly better than the former two clusters, meanwhile, the middle-risk group showed a lower risk of ASCVD, regardless of model 1 or model 2, with a p-value of less than 0.05. This result reflects that the risk of ASCVD in patients may be closely related to blood pressure levels. Previous studies have also suggested that blood pressure is one of the risk factors for ASCVD in patients with T2DM (31), and a meta-analysis also found that antihypertensive treatment reduces the risk of CVD events in patients with T2DM (32). Stefano Ciardullo et al. also found that the seasonal change of SBP is the main factor leading to the seasonal variation of ASCVD risk score in patients with T2DM (33).

It was also found that patients in the Medium-Low-Risk cluster had a lower risk of developing NSC (OR=0.805, p=0.043), but this trend lost statistical significance after adjusting for 20 covariates. This may be related to the joint effect of multiple NSC risk factors in the pathogenic process. Xiuxiu Liu et al. pointed out that the risk factors for diabetic peripheral neuropathy (DPN) are: the duration of diabetes, age, glycosylated hemoglobin A1c(HbA1c), and DR, but BMI, smoking, total triglyceride (TG), and total cholesterol (TC) did not increase the risk of DPN (34). The study on the risk factors of DPN in young patients with diabetes found that risk factors for DPN in youth with type 1 diabetes were older age, longer diabetes duration, smoking, increased DBP, obesity, increased LDL-C and triglycerides (TG), and lower HDL-C. While in youth with T2DM, risk factors were older age, male sex, longer diabetes duration, smoking, and lower HDL-C (35). In addition, an observational study of childhood and adolescents with diabetes found that people with T2DM were more likely to develop DPN than those with type 1 diabetes (36). Overall, the pathogenesis and core pathogenic factors of neurological complications of type 2 diabetes still need to be further studied.

In the metabolite analysis of the 4 clusters, we found that 24 metabolites had statistically significant differences among the clusters, and screened out metabolites with typical characteristics in some clusters. Glutaryl carnitine and Butylcarnitine increased while Tyr decreased in the High-Risk cluster, and Thr increased in the Low-Risk cluster. These findings will provide new possibilities for risk intervention targeting metabolite levels in patients with chronic complications of T2DM. These characteristic metabolites can be used as biomarkers for the screening, diagnosis, and prediction of chronic complications of T2DM, which will help to understand the metabolic information and metabolic pathways of chronic complications of T2DM and explore the possible mechanisms of complications, thus providing possible targets for its treatment (22). For example, J Ricardo Lucio-Gutiérrez et al. used partial least squares analysis (PLS-DA) to obtain the PLS-DA model for distinguishing type 2 diabetic nephropathy based on urinary metabolites, and it was well validated (37). This provides a new diagnostic option for the diagnosis of asymptomatic DN in patients with T2DM.

In the correlation analysis of characteristic metabolites and chronic complications, we found that Thr was negatively correlated with microangiopathy(especially DN), but positively correlated with ASCVD. And Tyr was negatively correlated with microangiopathy (regardless of DN or DR) and NSC. Both Butylcarnitine and Glutaryl carnitine were positively associated with ASCVD risk. Cornelia G Bala et al. found that the disruption of the tyrosine biosynthesis pathway might be disrupted related to the high oxidative stress response in patients with type 2 diabetes (38). Oxidative stress is a key pathogenic factor in diabetic complications and can lead to the development of microangiopathy and ASCVD (39, 40). Changes in metabolite levels may reflect the occurrence of multiple pathogenic mechanisms. Therefore, further research on the underlying mechanisms involved in changes in other metabolites is particularly important for the study of disease progression. And thereby reducing the huge disease burden of diabetes (41).

Our study is based on real data of hospitalized patients without any missing values, which can better reflect the practical situation of patients with chronic complications of T2DM. With a relatively ideal amount of data and a balanced male-to-female ratio, the cluster analysis obtained the cluster characteristics of patients with a high risk of chronic complications of T2DM. The relationship between metabolic changes and chronic complications was analyzed from the level of metabolites, which provides new ideas and targets for the prevention and treatment of chronic complications of T2DM and is helpful for the precise treatment of chronic complications of T2DM. However, our study also has certain limitations. It is a single-center study, which only includes patients from Dalian, Northeast China, and has not yet included subjects from other countries and regions. At the same time, this study was also retrospective and failed to observe the progression of chronic complications in each subcluster. In order to extend our research and comprehensively promote the guiding significance of our research results to clinical decision-making, it is very necessary to carry out corresponding prospective follow-up research and observe the treatment response. In the future, we will continue to improve this part of the research to enhance its value of the research.

The complication status of patients with chronic complications of T2DM has obvious cluster characteristics, and the performance of different target organ damage risks is not completely consistent. However, the current classification of complications has not considered the impact of these pathogenic risk factors on the occurrence of complications. The clustering results of this study reflect the clustering characteristics of different risk levels of target organ damage and the clustering characteristics of multiple complications. Different clusters also had significant differences in the levels of most metabolites, which helps to explore the potential role of the pathogenic mechanism of metabolic pathway alterations in the progression of complications. This will provide novel decision support for the prevention and treatment of chronic complications of T2DM.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by the Ethics Committee of the Second Affiliated Hospital of Dalian Medical University. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and institutional requirements.

CW, YL, KD and CL performed data analysis, interpretation and manuscript writing. CW, HZ and JW supervised the data collection and research collaboration. CW, CL, GW, HZ and XL participated in data collection and literature search. HZ, CW and XL designed the experiment and supervised the overall progress. All authors contributed to the article and approved the submitted version.