Subgrouping patients with type 2 diabetes using behavioural and clinical factors: a cross-sectional study in a hospital-based setting in Thailand

Abstract

Objectives

This study aimed to identify distinct patient subgroups based on glycaemic control (glycosylated haemoglobin (HbA1c)), self-efficacy and self-management in patients with type 2 diabetes mellitus (T2DM), and to examine differences in outcomes and identify key predictors associated with cluster characteristics.

Design

Cross-sectional study.

Setting

Chronic disease clinic at Thasala Hospital, Thailand.

Participants

Participants with T2DM were recruited using a consecutive sampling approach during their scheduled clinic visits on predefined days and times. A total of 440 participants were included in the final analysis.

Outcomes measures

The three variables used for K-means cluster analysis were HbA1c, self-efficacy scores and self-management scores. HbA1c values were obtained from medical records, while self-efficacy and self-management were assessed using the Thai versions of the Diabetes Management Self-Efficacy Scale and the Diabetes Self-Management Scale. Demographic and clinical characteristics were included as predictor variables in multiple linear regression analyses.

Results

Four clusters were identified. Cluster 1 (moderate profile, n=124) had fair glycaemic control (HbA1c=7.9%) and moderate self-efficacy (mean=70) and self-management (mean=47). Cluster 2 (underperforming, n=136) exhibited poor glycaemic control (HbA1c=8.7%), regardless of high self-efficacy (mean=79) and low self-management (mean=40). Cluster 3 (high performers, n=135) demonstrated fair glycaemic control (HbA1c=7.5%) with the highest levels of self-efficacy (mean=84) and self-management (mean=51). Cluster 4 (high risk, n=45) had very poor glycaemic control (HbA1c=9.4%) and the lowest scores for both self-efficacy (mean=56) and self-management (mean=34). Regression analysis confirmed the heterogeneity across clusters, with varying predictors and explained variance (adjusted R² ranging from 0.014 to 0.182 across significant models).

Conclusions

The findings highlight the distinct behavioural and clinical profiles among patients with T2DM. Cluster 4 patients with the poorest glycaemic and behavioural outcomes may benefit from intensive behavioural support and closer clinical monitoring, whereas Cluster 2 patients, showing high self-efficacy but poor self-management, indicate the need for structured, skills-based interventions. Clusters 1 and 3 showed balanced profiles, suggesting less urgent need for intervention and potential to maintain current management.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study employed a multidimensional approach by clustering individuals with type 2 diabetes mellitus based on clinical (glycosylated haemoglobin), psychological (self-efficacy) and behavioural (self-management) dimensions, providing a comprehensive view of patient heterogeneity.

• Subgroup-specific regression analyses were conducted using theory-based predictors to explore outcome-specific associations within each cluster, offering insight into differential needs and supporting more tailored intervention strategies.

• The fixed-cluster approach used in regression analysis assumes that cluster membership is known with certainty; this may underestimate SEs and inflate significance levels, and results should be interpreted in light of this limitation.

• Self-efficacy and self-management were measured using self-reported questionnaires, which may introduce response or recall bias and limit the objectivity of behavioural assessments.

Introduction

Type 2 diabetes mellitus (T2DM) is a major global public health concern, affecting millions of individuals and placing a significant burden on healthcare systems worldwide. As of 2021, approximately 536 million adults were living with diabetes, and it is projected to increase to over 783 million by the year 2024.^{1 2} The number of diabetics in Thailand increased significantly from 4.1 million in 2011 to 6.1 million in 2021, reflecting this worldwide trend.¹ This increase has been attributed to a combination of factors, including increased intake of added free sugar (up to 20% of total energy), reduced levels of physical activity and population ageing.³ Furthermore, data from 2011 to 2018 indicate that only 33.0–35.6% of Thai patients with T2DM were able to reach appropriate glycaemic control (glycosylated haemoglobin (HbA1c)<7.0%).⁴ This challenge contributes to the high incidence of diabetes-related complications and healthcare costs.^{5 6}

Effective glycaemic control, commonly assessed by HbA1c, remains a primary goal in the treatment of T2DM because it is directly related to better patient outcomes and the avoidance of long-term problems.⁷ Pharmacological interventions are primarily evaluated based on their ability to achieve optimal glycaemic targets.⁸ Despite advances in medical treatment, long-term glycaemic control remains suboptimal for a substantial proportion of patients. This has led to increasing recognition of the importance of behavioural and psychological factors in influencing diabetes outcomes. Among these, self-management, which is the individual’s ability to manage medication, diet, physical activity and monitoring, is widely recognised as a critical component of successful diabetes care.^{9 10} In parallel, self-efficacy, defined by Bandura as an individual’s belief in their capacity to perform specific behaviours, plays a pivotal role in motivating and sustaining these self-care activities.^{11 12} Individuals with higher self-efficacy are more likely to initiate and maintain diabetes-related self-care than those with low self-efficacy.¹³ Importantly, self-efficacy supports effective self-management, and both have been shown to be associated with improved glycaemic control (HbA1c).^{14 15} Therefore, these three dimensions; HbA1c, self-management and self-efficacy are inter-related and collectively reflect the clinical, behavioural and psychological domains that shape diabetes outcomes. In this study, these variables were selected as key indicators for clustering patients, enabling identification of meaningful subgroups based on their comprehensive care profiles.

Although the clinical, behavioural and psychological aspects of T2DM have been extensively studied, few studies have examined how these domains interact within distinct patient subgroups. Cluster analysis has increasingly been applied in diabetes research to identify subgroups of patients with shared characteristics. However, most of these studies have relied primarily on metabolic or biomedical markers while overlooking behavioural and psychological factors such as self-management and self-efficacy.^{16 17} The majority of these studies have been conducted in Western populations, limiting their generalisability to Southeast Asian contexts, including Thailand. The heterogeneity in clinical outcomes, health behaviours and psychosocial factors among patients with T2DM suggests that a one-size-fits-all approach may be insufficient. There is a need for research that incorporates a more holistic set of variables, integrating both biomedical and psychosocial dimensions, to better inform personalised diabetes care strategies.

To address this gap, this study aims to identify distinct subgroups of Thai patients with T2DM by applying cluster analysis based on three key variables: glycaemic control (HbA1c), self-management and self-efficacy. These variables reflect the clinical, behavioural and psychological dimensions of diabetes care, respectively. Following cluster identification, multiple regression analyses were then conducted within each cluster to explore demographic and clinical predictors and to confirm variations across clusters. This two-step approach illustrates the heterogeneity of T2DM and supports the need for more personalised care.

Methods

Research design and participants

This cross-sectional study was conducted among patients with T2DM attending the chronic disease clinic at Thasala Hospital, a medium-sized public hospital (220 inpatient beds) located in a suburban area of Nakhon Si Thammarat Province, Thailand. Data were collected during a single visit between January and May 2022 through a combination of self-administered questionnaires, structured interviews and review of electronic medical records.

Eligible participants were adults aged 20–80 years who had been diagnosed with T2DM for at least 2 years, based on WHO criteria (fasting plasma glucose ≥126 mg/dL or HbA1c ≥6.5%). Participants were also required to have a scheduled HbA1c test appointment on the day of data collection. Patients with a history of type 1 diabetes or anaemia were excluded. Recruitment was carried out using a consecutive sampling approach. On predefined clinic days and time slots, all patients who met the inclusion criteria and expressed interest were invited to participate. Because the research team did not have access to patient lists in advance, recruitment was conducted entirely on-site during clinic hours. No individual was chosen for inclusion based on clinical traits or results. All participants gave their written informed permission before the data collection began.

The required sample size was calculated using a single population proportion formula, with an expected prevalence (P) of 50%, a 95% confidence level (Z=1.96) and a 5% margin of error (d). A prevalence of 50% was chosen as a conservative estimate to ensure an adequate sample size, as no prior data were available regarding the prevalence of the combined clustering variables (HbA1c, self-efficacy and self-management) in this population. Using a 50% prevalence is a conventional approach in descriptive research when prior data on outcome prevalence are not available.¹⁸ The minimum sample size was estimated to be 384 participants. To account for potential incomplete data, 20% was added, resulting in a target of 460 participants for data collection.

Study variables

Three continuous variables including glycaemic control (HbA1c), self-efficacy and self-management were used as clustering variables to identify patient subgroups. In addition, these variables were continuous outcome variables in regression analyses. HbA1c values were extracted from the electronic medical records and reflected laboratory results from the date of data collection. Self-efficacy and self-management were measured using validated Thai-language questionnaires as follows.

The Thai version of the Diabetes Management Self-Efficacy Scale was used to measure self-efficacy. This tool consists of 20 items with four domains including diet (9 items), monitor (4 items), physical (4 items) and regimen (3 items). Each item is scored on a 5-point Likert scale, with 1 being ‘strongly disagree’ and 5 being ‘strongly agree’. The total possible score for the questionnaire was 100 points, with higher scores indicating greater levels of self-efficacy. This tool has been found to have good psychometric qualities.¹⁹

The Thai version of the Diabetes Self-Management Scale was used to assess self-management. This questionnaire contains 13 items with responses ranging from 0 to 7 in four domains including diet (5 items), exercise (2 items), blood (2 items) and foot care (4 items). The self-management questionnaire had a maximum total score of 91 points, with higher scores reflecting better self-management practices. This tool was reported to have good criterion and content validity.²⁰

Demographic and clinical variables were used as predictor (exposure) variables in regression analyses to examine associations with three outcomes: glycaemic control (HbA1c), self-efficacy and self-management. The selection of these predictors was informed by their theoretical relevance and previous literature reporting associations with each of the three outcomes of interest. Demographic characteristics included gender (male/female), age (continuous), age at T2DM diagnosis (continuous), education level (categorised as primary school or lower, high school and undergraduate or higher) and body mass index (BMI, continuous). Clinical variables included treatment regimen (oral antidiabetic drugs (OAD) only vs OAD plus insulin), comorbid conditions (hypertension and dyslipidaemia, each coded as present/absent) and presence of microvascular complications (yes/no). All categorical predictors were appropriately coded, and no transformations were applied to the outcome variables.

Data collection

Data collection was conducted in the clinic waiting area while participants were waiting to see their physician, shortly after they had received their queue number at the outpatient registration desk and completed laboratory tests as part of their follow-up orders. After confirming eligibility and obtaining informed consent, the first investigator recorded demographic data through a brief interview. This investigator was also responsible for later reviewing the participant’s electronic medical record. The second and third investigators then administered the questionnaires on self-efficacy and self-management, respectively. If participants had visual impairments or difficulties with reading, the questionnaire was read aloud to them by the investigators to ensure accurate understanding and response. All investigators were licensed health professionals working with patients with diabetes in collaboration with physicians and the multidisciplinary care team and had been trained in standardised data collection procedures to minimise bias and ensure consistency and accuracy.

All questionnaire-based data were collected in the morning, prior to each participant’s scheduled medical consultation. Later in the afternoon of the same day, the first investigator reviewed the electronic medical records to extract relevant clinical data for analysis.

Patient and public involvement

Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Statistical analysis

Statistical analysis was performed using SPSS for Windows V.23 (IBM, Armonk, New York, USA). A p value of 0.05 was used to determine statistical significance. Descriptive statistics, including mean, SD and range, were used to summarise continuous variables, while frequencies and percentages were used for categorical variables.

K-means clustering was performed using raw scores of three variables: HbA1c, self-efficacy and self-management. Standardisation was not applied in order to retain the clinical interpretability of the cluster centres (eg, actual HbA1c percentages). The number of clusters (K) was determined by testing three, four and five clusters’ solutions and comparing their statistical separation and clinical interpretability. Because SPSS’s K-means procedure does not provide silhouette scores, a two-step cluster analysis was additionally performed to confirm model quality among 3–5 clusters. The silhouette scores were comparable across the tested solutions. Consequently, the 4-cluster solution was selected as it offered the best balance of model fit and clinical relevance. The interdistance between each cluster was calculated to reflect multivariate differences across the three clustering variables.

One-way analysis of variance (ANOVA) was used to validate the differences between the identified clusters on the three clustering variables: HbA1c, self-efficacy and self-management. The analysis tested whether the means of each variable differed significantly across the four clusters. F-statistics and corresponding p values were reported to indicate the significance of between-cluster variation.

Multiple linear regression analyses were conducted to examine the associations between demographic and clinical characteristics and each of the three outcome variables: HbA1c, self-efficacy and self-management scores. Based on theoretical relevance, the predictors for glycaemic control included age, age at diagnosis, gender, education level, BMI, treatment regimen and comorbidities (hypertension and dyslipidaemia).²¹,²³ Predictors associated with self-efficacy were gender, education level, BMI, microvascular complications and comorbidities.²⁴,²⁷ For self-management, relevant predictors included age, age at diagnosis, gender, education level, BMI, treatment regimen and comorbidities.²⁸,³⁴

Although all predictors were selected based on previous literature, findings from past studies varied depending on the population and research focus. As such, it was anticipated that not all theoretically supported variables would significantly predict outcomes in this study population. Therefore, univariate linear regression was used as a screening step to identify candidate variables with p values less than 0.10. These variables were then included in the final multiple regression models using the enter method. Separate models were conducted for the overall sample and for each of the four clusters, resulting in a total of 15 models (five groups and three outcomes). Results are presented with adjusted R² and p values to indicate model fit and statistical significance. This approach allowed for the identification of both common and cluster-specific predictors, supporting the overall aim of characterising heterogeneity among patients with T2DM.

All statistical analyses were conducted using data from participants with complete information on all study variables. Missing data were handled by listwise deletion, resulting in a final analytical sample of 440 participants. No specific statistical adjustments were made to account for the sampling strategy, as all eligible participants during the study period were included consecutively. Sensitivity analyses were not performed.

Results

A total of 648 participants were assessed for eligibility. Of these, 188 participants did not meet the inclusion and exclusion criteria. A total of 460 participants with T2DM were enrolled in this study. Due to incomplete data from 20 participants, 440 participants were included in the final data analysis (figure 1). Demographic and clinical characteristics are presented in table 1.

Figure 1. Flow diagram. DM, diabetes mellitus.

Table 1. Demographic and clinical characteristics (N=440).

Variables	Value, mean (SD) or n (%)	Range
Age (year), mean (SD)	58.98 (9.53)	25–80
Age at diagnosis (year), mean (SD)	50.31 (8.69)	21–75
Body mass index (BMI) (Kg/m2), mean (SD)	28.10 (4.96)	17.75–46.11
Gender, n (%)
Male	88 (20)	–
Female	352 (80)	–
Education level, n (%)
Primary school or lower	331 (75.2)	–
High school	65 (14.8)	–
Undergraduate and higher	44 (10.0)	–
Treatment, n (%)
OAD	348 (79.1)	–
OAD and insulin	92 (20.9)	–
Comorbidity, n (%)
Dyslipidaemia	353 (80.2)	–
Hypertension	284 (64.5)	–
Microvascular complication, n (%)	158 (35.9)	–

OAD, oral antidiabetic drugs.

Three key variables, including HbA1c, self-efficacy score and self-management score, were used for cluster analysis. Based on K-means clustering, participants were grouped into four subgroups, as shown in table 2 and figure 2. The ANOVA results confirmed the validity of the cluster variables among subgroups with a p value of less than 0.001. Cluster 1 (moderate profile, n=124) was characterised by fair glycaemic control, moderate self-efficacy and moderate self-management. Cluster 2 (underperforming, n=136) showed poor glycaemic control, high self-efficacy and low self-management. Cluster 3 (high performers, n=135) exhibited fair glycaemic control, the highest self-efficacy and the highest self-management. Cluster 4 (high risk, n=45) demonstrated very poor glycaemic control, the lowest self-efficacy and the lowest self-management.

Table 2. Final cluster centres.

Cluster variables	All (N=440) Mean (SD)	Cluster				F	P value
		1 (n=124)	2 (n=136)	3 (n=135)	4 (n=45)
HbA1c	8.20 (1.58)	7.9	8.7	7.5	9.4	28.26	<0.001*
Self-efficacy (0–100)	75.56 (9.87)	70	79	84	56	454.64	<0.001*
Self-management (0–91)	44.81 (7.10)	47	40	51	34	209.46	<0.001*

^*The mean difference is significant at the 0.05 level.

HbA1c, glycosylated haemoglobin.

Figure 2. Cluster profiles based on HbA1c, self-efficacy and self-management scores. HbA1c, glycosylated haemoglobin.

The K-means cluster analysis produced distances between final cluster centroids, reflecting multivariate differences in the three clustering variables (HbA1c, self-efficacy and self-management). These values were interpreted comparatively, with larger distances indicating greater behavioural and clinical dissimilarity; however, no standardised cut-off criterion was used to define meaningful separation. Among all cluster pairs, Clusters 3 and 4 exhibited the greatest multivariate difference, suggesting that patients in Cluster 4 had substantially poorer glycaemic control and lower self-care behaviours than those in Cluster 3. Intercluster distances are shown in table 3.

Table 3. Distances between final cluster centres.

Cluster	1	2	3	4
1		11.789	14.849	18.233
2	11.789		11.560	23.777
3	14.849	11.560		32.140
4	18.233	23.777	32.140

The results from the multiple linear regression confirmed the heterogeneity among Thai patients with T2DM. The table 4 presents the initial predictors in univariate analysis and their p value. The final predictors retained in each model and the corresponding adjusted R² values, reflecting the variance explained in HbA1c, self-efficacy and self-management. Subgroup analyses revealed that the proportion of variance explained differed across clusters, suggesting the influence of individual-level factors in predicting outcomes. For HbA1c, the overall model explained 18.2% of the variance (adjusted R²=0.182, p<0.001), with treatment regimen, dyslipidaemia and education level retained as significant predictors. Among subgroups, Cluster 2 had the highest explained variance (adjusted R²=0.177, p<0.001), with treatment regimen emerging as a significant predictor. For self-efficacy, the overall model accounted for 1.4% of the variance (adjusted R²=0.014, p=0.018), with education level showing as a significant predictor. Among subgroups, Cluster 3 had the highest explained variance (adjusted R²=0.045, p=0.04), with education level retained as a key explanatory variable. For self-management, the overall model explained 5.8% of the variance (Adjusted R²=0.058, p<0.001), with BMI, treatment regimen and dyslipidaemia emerging as significant predictors. Among subgroups, Cluster 3 had the highest explained variance (Adjusted R²=0.097, p<0.001), with gender showing statistically significant associations (detailed multiple linear regression results are shown in online supplemental table 1).

Table 4. Regression models predicting HbA1c, self-efficacy and self-management from demographic and clinical characteristics in the overall sample and cluster subgroups.

	Univariate linear regression (p values for individual predictors)	Multiple linear regression (p values for predictors retained in model)	Adjusted R2	Model p value
HbA1c
All subject (N=440)	Age (p=0.001), age at diagnosis (p<0.001), gender (p=0.052), education level (p=0.05), BMI (p=0.45), treatment regimen (p<0.001), hypertension (p=0.65), dyslipidaemia (p=0.008)	Age (p=0.25), age at diagnosis (p=0.23), gender (p=0.16), education level (p=0.03)*, treatment regimen (p<0.001)*, dyslipidaemia (p=0.02)*	0.182	<0.001
Cluster 1 (n=124)	Age (p=0.09), age at diagnosis (p=0.05), gender (p=0.93), education level (p=0.42), BMI (p=0.60), treatment regimen (p<0.001), hypertension (p=0.62), dyslipidaemia (p=0.07)	Age (p=0.36), age at diagnosis (p=0.97), treatment regimen (p<0.001)*, dyslipidaemia (p=0.09)	0.153	<0.001
Cluster 2 (n=136)	Age (p=0.099), age at diagnosis (p=0.003), gender (p=0.46), education level (p=0.36), BMI (p=0.82), treatment regimen (p<0.001), hypertension (p=0.38), dyslipidaemia (p=0.69)	Age (p=0.95), age at diagnosis (p=0.33), treatment regimen (p<0.001)*	0.177	<0.001
Cluster 3 (n=135)	Age (p=0.08), age at diagnosis (p=0.04), gender (p=0.05), education level (p=0.20), BMI (p=0.35), treatment regimen (p<0.001), hypertension (p=0.40), dyslipidaemia (p=0.004)	Age (p=0.39), age at diagnosis (p=0.93), gender (p=0.06), treatment regimen (p<0.001)*, dyslipidaemia (p=0.008)*	0.172	<0.001
Cluster 4 (n=45)	Age (p=0.49), age at diagnosis (p=0.20), gender (p=0.47), education level (p=0.44), BMI (p=0.66), treatment regimen (p=0.16), hypertension (p=0.58), dyslipidaemia (p=0.43)	–	–	–
Self-efficacy
All subjects (N=440)	Gender (p=0.17), education level (p=0.03), BMI (p=0.17), hypertension (p=0.08), dyslipidaemia (p=0.48), microvascular complication (p=0.23)	Education level (p=0.02)*, hypertension (p=0.06)	0.014	0.018
Cluster 1 (n=124)	Gender (p=0.62), education level (p=0.62), BMI (p=0.63), hypertension (p=0.67), dyslipidaemia (p=0.66), microvascular complication (p=0.56)	–	–	–
Cluster 2 (n=136)	Gender (p=0.38), education level (p=0.83), BMI (p=0.67), hypertension (p=0.02), dyslipidaemia (p=0.36), microvascular complication (p=0.42)	Hypertension (p=0.02)*	0.036	0.015
Cluster 3 (n=135)	Gender (p=0.08), education level (p=0.03), BMI (p=0.10), hypertension (p=0.09), dyslipidaemia (p=0.74), microvascular complication (p=0.65)	Gender (p=0.46), education level (p=0.049)*, BMI (p=0.30), hypertension (p=0.17)	0.045	0.040
Cluster 4 (n=45)	Gender (p=0.37), education level (p=0.85), BMI (p=0.79), hypertension (p=0.27), dyslipidaemia (p=0.82), microvascular complication (p=0.19)	–	–	–
Self-management
All subject (N=440)	Age (p=0.03), gender (p=0.099), education level (p=0.79), BMI (p<0.001), age at diagnosis (p=0.03), treatment regimen (p<0.001), hypertension (p=0.83), dyslipidaemia (p=0.03)	Age (p=0.21), gender (p=0.34), BMI (p=0.01), age at diagnosis (p=0.59), treatment regimen (p=0.001), dyslipidaemia (p=0.04)	0.058	<0.001
Cluster1 (n=124)	Age (p=0.49), gender (p=0.37), education level (p=0.42), BMI (p=0.21), age at diagnosis (p=0.80), treatment regimen (p=0.27), hypertension (p=0.51), dyslipidaemia (p=0.21)	–	–	–
Cluster2 (n=136)	Age (p=0.27), gender (p=0.28), education level (p=0.03), BMI (p=0.095), age at diagnosis (p=0.59), treatment regimen (p=0.24), hypertension (p=0.43), dyslipidaemia (p=0.97)	Education level (p=0.06), BMI (p=0.22),	0.032	0.043
Cluster3(n=135)	Age (p=0.56), gender (p<0.001), education level (p=0.62), BMI (p=0.04), age at diagnosis (p=0.14), treatment regimen (p=0.099), hypertension (p=0.13), dyslipidaemia (p=0.49)	Gender (p=0.001), BMI (p=0.47), treatment regimen (p=0.14)	0.097	0.001
Cluster4 (n=45)	Age (p=0.68), gender (p=0.18), education level (p=0.99), BMI (p=0.94), age at diagnosis (p=0.68), treatment regimen (p=0.75), hypertension (p=0.24), dyslipidaemia (p=0.33)	–	–	–

^*The mean difference is significant at the 0.05 level.

BMI, body mass index; HbA1c, glycosylated haemoglobin.

Discussion

This study identified four distinct subgroups of Thai patients with T2DM, based on their levels of glycaemic control, self-efficacy and self-management. Together, these variables offer a comprehensive perspective on diabetes status by capturing clinical, psychological and behavioural dimensions. HbA1c serves as a standard indicator of long-term glycaemic control, while self-efficacy and self-management reflect the patient’s capacity and actions in managing their condition. The four clusters, including moderate profile, underperforming, high performers and high risk, demonstrated meaningful differences in both clinical outcomes and self-care behaviours, underscoring the heterogeneity of T2DM within this population. The findings are consistent with previous studies that have identified diverse phenotypes in people with T2DM, although many of those studies focused more on metabolic, demographic or comorbidity-based factors.³⁵,⁴⁰ In contrast, this study incorporated behavioural elements into the clustering process, providing additional insight for designing more personalised care strategies.

Among the identified subgroups, the most contrasting profiles were found in Cluster 3 and Cluster 4. Cluster 3 (high performers), defined by fair glycaemic control along with the highest levels of self-efficacy and self-management, likely represents patients who are actively engaged in their care and capable of sustaining lifestyle changes. In contrast, Cluster 4 (high risk) displayed very poor glycaemic control in combination with the lowest behavioural scores, suggesting a group that may require intensive, multidisciplinary interventions. These results are consistent with previous research emphasising the influence of behavioural factors, particularly self-efficacy and self-management, on diabetes outcomes.^{39 41} The noticeable differences in HbA1c across the groups highlight the need for more personalised care plans. Patients with poor self-care capacity, as seen in Cluster 4, may benefit from structured education, frequent follow-ups and psychosocial support. Meanwhile, individuals in Cluster 3 are likely to maintain good control with minimal pharmacotherapy and targeted lifestyle interventions. This aligns with findings by Zhang et al, who emphasised the importance of tailoring intervention strategies to match the diverse needs of subgroups within the T2DM population.⁴²

Interestingly, Cluster 2 (underperforming) presented a contrasting pattern, with high self-efficacy but poor glycaemic control and low self-management. This suggests that perceived confidence does not necessarily lead to effective glycaemic control when self-care practices are lacking. Peter et al reported that although individuals with T2DM had knowledge of lifestyle modifications, their actual practice remains poor.⁴³ Conversely, previous studies have shown that strong self-efficacy is generally associated with better self-management behaviours.^{44 45} This inconsistency may reflect the underlying heterogeneity of T2DM. The current findings suggest a complex interaction between psychological perceptions and real-world disease management, consistent with results from other cluster-based analyses in T2DM populations.^{46 47} Contributing factors may include external barriers such as limited social support, environmental constraints or psychological distress.^{48 49} Mizokami-Stout et al found that lower diabetes-related distress is linked to improved glycaemic control.⁵⁰ These results indicate that effective diabetes self-care involves more than just confidence. Without practical strategies and adequate support, patients may struggle to manage their condition. Supportive approaches such as coaching or digital tools could help improve adherence and encourage sustainable behaviour change.

The subgroup-specific regression analyses for HbA1c, self-efficacy and self-management further support the relevance of the identified clusters. Notably, the proportion of variance explained by the predictive models differed across clusters, reaffirming the heterogeneity among individuals with T2DM. Cluster 2 showed the highest explained variance for HbA1c, whereas Cluster 3 exhibited the highest explained variance for both self-efficacy and self-management. Regarding predictors, although the variables were selected based on theoretical relevance and prior evidence, only a subset demonstrated statistically significant associations with the outcomes in this population. These findings suggest that, although previous studies have shown oral medications and insulin therapy to be commonly used strategies for controlling blood glucose levels,⁵¹ pharmacological treatment alone may not fully explain the variations in HbA1c observed in this population. Factors such as dyslipidaemia and education level also emerged as significant predictors, highlighting the multifactorial nature of glycaemic control. Both treatment regimens and dyslipidaemia also played key roles in predicting self-management behaviours. These results support the importance of addressing both pharmacological and non-pharmacological strategies, including appropriate management of diabetic dyslipidaemia, to reduce the risk of coronary heart disease.⁵² Education level appeared as a significant predictor, highlighting its central role in effective diabetes management. Previous studies have also confirmed the benefits of patient education in improving outcomes among individuals with T2DM.⁵³ In addition, hypertension was a significant predictor of self-efficacy in Cluster 2. This finding aligns with previous studies indicating that hypertension, which is one of the most prevalent comorbidities among individuals with T2DM,⁵⁴ can influence self-efficacy and overall disease management. BMI was also included in the predictive model for self-management, aligning with a previous study that identified an obesity-related cluster among patients with T2DM.⁵⁵

The findings from this cluster analysis offer important clinical implications for managing individuals with T2DM. Identifying subgroups based on HbA1c levels, self-efficacy and self-management behaviours provides a foundation for delivering more personalised care. Although HbA1c remains a central indicator of glycaemic control, it should not be the sole focus in diabetes management. Psychological and behavioural factors such as self-efficacy and self-management play a critical role in shaping outcomes. Among Thai patients, those with low confidence and poor self-care behaviours may require more intensive support, while those who are more engaged may benefit from reinforcement and goal-oriented interventions. While previous studies have used cluster analysis in T2DM populations, the selection of variables has varied considerably, reflecting the complex and diverse nature of the disease. This variability poses challenges for standardising care and emphasises the need for context-specific strategies. Moreover, demographic and clinical factors such as education level, BMI, comorbidities including hypertension and dyslipidaemia and treatment regimen all contribute to the unique characteristics of each subgroup. These elements should be considered when designing interventions to ensure they align with individual patient needs and can effectively improve outcomes. To address the diverse needs identified across clusters, interdisciplinary collaboration is essential. Healthcare teams involving physicians, nurses, dietitians, physiotherapists, diabetes educators and mental health professionals can work together to provide holistic and tailored care that supports both medical and behavioural aspects of diabetes management within the Thai context.

Despite its strengths, this study has some limitations. Self-efficacy and self-management were assessed using self-reported questionnaires, which may lead to bias. The generalisability of findings may be limited to populations with similar demographic and clinical characteristics. Although raw scores were used in the clustering process to preserve clinical interpretability, we acknowledge that differences in variable scales may have introduced bias. Furthermore, regression analyses were performed using fixed cluster assignments, which do not account for uncertainty in cluster membership and may underestimate SEs. This limitation should be considered when interpreting the regression results. Future research should explore the longitudinal stability of these clusters and evaluate the effectiveness of cluster-specific interventions and consider incorporating theory-driven approaches such as Directed Acyclic Graphs to improve model specification and causal interpretation.

Conclusion

This study highlights the heterogeneity in self-care behaviours and clinical outcomes among individuals with T2DM. Four distinct clusters were identified: moderate profile, underperforming, high performers and high risk. The identification of these distinct patient subgroups provides valuable insights for tailoring interventions and optimising diabetes management. Incorporating personalised strategies based on behavioural and clinical profiles may lead to improved health outcomes and more efficient use of healthcare resources. These findings support the need for stratified care approaches that move beyond one-size-fits-all models. Based on theoretical utility, future research should explore the translation of these subgroup-based strategies into real-world clinical practice through policy integration and interdisciplinary care models, particularly in resource-limited or culturally specific settings.

Data availability statement

Data are available upon reasonable request.