Cluster analysis of patient characteristics, treatment modalities, renal impairments, and inflammatory markers in diabetes mellitus

Type 2 diabetes mellitus (T2DM) is caused by an interplay of various factors where chronic hyperglycemia and inflammation have central role in its onset and progression. Identifying patient groups with increased inflammation in order to provide more personalized approach has become crucial. We hypothesized that grouping patients into clusters according to their clinical characteristics could identify distinct unique profiles that were previously invisible to the clinical eye. A cross-sectional record-based study was performed at the Primary Health Care Center Podgorica, Montenegro, on 424 T2DM patients aged between 30 and 85. Using hierarchical clustering patients were grouped into four distinct clusters based on 12 clinical variables, including glycemic and other relevant metabolic indicators. Inflammation was assessed through neutrophil-to-lymphocyte (NLR) and platelet to lymphocyte ratio (PLR). Cluster 3 which featured the oldest patients with the longest T2DM duration, highest hypertension rate, poor glycemic control and significant GFR impairment had the highest levels of inflammatory markers. Cluster 4 which featured the youngest patients, with the best glycemic control, the highest GFR had the lowest prevalence of coronary disease, but not the lowest levels of inflammatory markers. Identifying these clusters offers physicians opportunity for more personalized T2DM management, potentially mitigating its associated complications.

www.nature.com/scientificreports/T2DM and its complications to some extent, but comprehensive care for individuals with T2DM often necessitates the management of other risk factors such as hypertension and dyslipidemia, as well as lifestyle modifications 9 .
New classes of drugs that have been developed act on different systems and they not only reduce hyperglycemia, but have beneficial effects on other cardiovascular risk factors like reducing blood pressure, reducing progression of renal impairments and promoting weight loss 10 .They have also shown anti-inflammatory effects 11 .Therefore, the choice of a pharmacological agent should be based on a holistic understanding of the patient's clinical profile, including the potential role of inflammation.
Concerning the management of this multifactorial chronic disease, a special attention should be paid on the older adult population given the fact that they face with frequent occurrence of comorbidities 12 .This imposes the need for an individual approach in the adults with T2DM.Glycemic control targets also differ between the different population groups.The less stringent target HbA1c has been recommended for older T2DM patients with multiple comorbidities and long duration of diabetes 13 .Therefore, the need to divide T2DM patients into phenotypes, in order to bring closer the concept of complexity of T2DM is of utmost importance.
Within this context, grouping patients into clusters according to their clinical characteristics appears as a very useful tool in identifying patterns and nuances that were previously invisible to the clinical eye, which was also the aim of the present study.This could help clinicians to tailor interventions ensuring that each patient receives care in accordance with their unique profile, potentially minimizing the risk of complications and comorbidities.

Results
The study included 424 patients, 208 males and 216 females.The average age of the population was 66.19 ± 11.14 years, ranging from 26 to 91 years.Participants were predominately non-smokers with average duration of the disease less than 10 years (8.67 ± 4.93 years).About 90% had hypertension, with an average SBP of 134 mmHg and DBP of 82.11 mmHg.
Table 1 presents the basic demographic and clinical features of the study population.
Coronary disease was the most prevalent manifestation of CVD.Hyperlipidemia was spotted in 82.3% of patients, but only 60% of them were using hypolipidemic drugs (i.e.statins).Most patients were on non-insulin medications, and over a third (34.5%) were receiving insulin treatment.
The average serum glucose level was 7.91 ± 2.95 mmol/L and HbA1c 7.28 ± 1.66% suggesting that many patients had elevated blood sugar levels.Most of the patients had decreased levels of GFR and increased levels of ESR (Table 2).
The Cluster 3 included the oldest patients (p < 0.001) with the longest duration of T2DM (p < 0.001) and the highest percentage of hypertension (p < 0.001).Patients from cluster 3 had the highest mean level of urea (p < 0.001), creatinine (p < 0.001), and the lowest GFR value (p < 0.001).This cluster had the highest percentage of patients diagnosed with significant GFR impairment (G3-G5) (p < 0.001) and the highest percentage of patients using DDP4i (p < 0.001) and insulin (p < 0.001).The Cluster 4 included the youngest patients, with the shortest disease duration, the lowest levels of fasting glycaemia (p = 0.040) and HbA1c (p < 0,001).Patients in this cluster also had the lowest mean levels of urea and creatinine, and the highest GFR level (Table 3).The prevalence of CVD was significantly different among clusters (p < 0.001).CVD were significantly more prevalent in Cluster 2 and Cluster 3 compared to Cluser 4 (p < 0.001 for both).

Discussion
When treating patients with T2DM, one of the main challenges physicians face is understanding the complexity of individual patient profiles.This complexity arises not only from the multidimensional nature of T2DM itself, but also from the multitude of associated comorbidities and underlying pathophysiological processes 14,15 .It implies that course of the disease can be highly variable.While some patients with T2DM face rapid deterioration, others maintain stable for extended period of time.This makes long-term planning challenging 16 .
Recognizing the important role of inflammation in the T2DM pathophysiology and its associated comorbidities has highlighted the need to identify groups of patients prone to increased inflammation as well as to identify www.nature.com/scientificreports/risk factors associated with increased inflammatory responses 17 .In this way clinicians can more effectively adjust therapeutic approaches and provide better control of the disease that extends beyond glucose control.
According to the authors' best knowledge, the studies that applied clustering method to group the patients with T2DM targeting inflammation, comorbidities and therapy regimens are scarce.We found only one study that used clustering to pair inflammatory and clinical parameters in patients with T2DM.However, it was conducted on a considerably smaller sample size than our study 14 .
We identified four distinct profiles of patients with T2DM based on their clinical and demographic characteristics.Each cluster had its unique characteristics and differed in terms of age, disease duration, associated conditions, and biochemical profiles.The Cluster 3 featured the oldest patients with the longest duration of T2DM, who also had the lowest levels of GFR and exhibited poor glycemic control.Patients from this cluster had the highest level of NLR and PLR which means that the Cluster 3 had the most pronounced subclinical inflammation since the correlation between increased NLR and PLR values and inflammation in T2DM is well established in the literature 18 .These findings are in alignment with other studies suggesting that inflammation might be associated with a more advanced or prolonged stage of T2DM, poor glycemic control and low GFR [19][20][21][22] .
T2DM is considered to be age-related disease.It is characterized by chronic activation of the innate immune system which can be increased by over-nutrition and aging process 23 .Over-nutrition in addition to genetic predisposition and lack of physical activity leads to obesity.Particularly in cases of central adiposity, this can trigger adipose tissue dysfunction, prompting macrophage infiltration and a subsequent surge in inflammatory cytokine release 24 .Chronically elevated inflammatory biomarkers promote insulin resistance and hyperglycemia.Furthermore, chronic hyperglycemia sustains persistent inflammation creating a cycle where inflammation exacerbates glucose metabolic disturbances, further aggravating the body's metabolic equilibrium 24 .This can explain why patients with higher levels of HbA1c like in the Cluster 3 exhibit the higher level of inflammation, as determined by higher NLR, PLR and neutrophil count.
Chronic hyperglycemia and inflammation have detrimental effects on various organs including kidneys 25 .These effects manifest as changes in the microvasculature, particularly in the thickening of the capillary basement membrane impacting arterioles in the glomeruli, retina, myocardium, skin, and muscle.Such alterations in the glomeruli play a crucial role in the onset and progression of diabetic nephropathy 6 .In a recent study, it was found that an increased NLR and PLR were not only significantly correlated with diabetic nephropathy but were also proposed as predictors and prognostic risk markers of diabetic nephropathy 26 .Our findings align with this, highlighting the interrelationship between kidney function and inflammatory responses.Specifically, once kidneys are damaged, they can further exacerbate inflammatory responses in the body 27 .This interplay is reflected in the Cluster 4, where good kidney function corresponds well with moderate inflammation markers, potentially suggesting a protective mechanism against intense inflammation.Furthermore, the Cluster 3 had the highest percentage of patients diagnosed with diabetic nephropathy and coronary disease.Previous studies also showed that higher NLR level was associated with an increased prevalence of CVD and diabetic nephropathy pointing out the important role that inflammation plays in development of such complications 28 .The coexistence within a single cluster highlights their interconnected nature 6,7 .
Patients in Cluster 4, who are characterized with the best clinical performances, have higher mean level of inflammation (as indicated by NLR and PLT markers) than patients in Cluster 1 (who have lower renal function and more CVD and worse metabolic indicators) but this difference did not reach statistical significance.Another feature of the clusters is the fact that patients in Cluster 4 have significantly lower level of inflammation (as indicated by NLR and PLT markers) than patients in Cluster 3, who are the worst with respect to the presence of CV comorbidities, and are also the oldest ones.One of the possible reasons for such discrepancies includes the wide range of age of studied diabetic patients that could have influenced the characteristics of clusters, in addition to differences in medications use.Furthermore, there are complex relationships between age, gender, postmenopausal status, T2DM duration, body shape, BMI categories, HbA1c, and inflammatory marker values as observed in previous studies 29,30 .www.nature.com/scientificreports/Although the Cluster 3 had the highest levels of inflammatory markers, we observed paradoxically low levels of total cholesterol and LDL.Considering the high prevalence of coronary artery disease in the Cluster 3, it might be plausible that these patients have been treated aggressively with lipid-lowering therapies in the past or might still be under such treatment.
Notably, the Cluster 3 also demonstrated a pronounced percentage of retinopathy cases, although this association did not reach statistical significance.These results are in line with a study conducted by Ciray et al. 31 that found no independent association between NLR and diabetic retinopathy.While some research has suggested NLR as a potential diagnostic biomarker for diabetic retinopathy, the association remains debated 28 .The highest percentage of patients with diabetic neuropathy was in the Cluster 2 which also had the pronounced levels of NLR and PLR but significantly lower than in the Cluster 3.This could be explained by multifactorial nature of the retinopathy and neuropathy where inflammation is just one aspect of a broader pathophysiological picture 32 .
Furthermore, it's worth noting that Cluster 4 which included the youngest patients with the lowest levels of fasting glycaemia and HbA1c, along with the highest GFR and relatively short disease duration presented with surprisingly higher inflammation markers compared to Cluster 1. Patients from the Cluster 1 also showed some unfavorable characteristics like patients from Cluster 2 and 3 including older patients with high percentage of hypertension and decreased GFR who had the worst glycemic and lipid control.Yet, despite these seemingly adverse factors, this cluster surprisingly exhibited the lowest levels of inflammatory markers.Medication regimen could be a contributing factor to these observed levels of inflammation in the Cluster 1. Namely, these group of patients had the highest percentage of patients on oral therapy, with Metformin being the most commonly used.Even though there was not a statistically significant difference between the clusters regarding Metformin use, we believe its presence played a pivotal role in reducing inflammation levels as suggested in different studies which showed that Metformin has potent anti-inflammatory effect through inhibiting secretion of pro-inflammatory Table 3. Clinical characteristics of clusters.BMI, body Mass Index; T2DM, type 2 diabetes mellitus; HbA1c, hemoglobin A1c; GFR, glomerular filtration rate; TC, total cholesterol; HDL, high-density lipoprotein; LDL, low-density lipoprotein; TRG, triglycerides; GLP-1r, glucagon-likepeptide-1 receptor agonists; SGLT2i, sodium-glucose cotransporter-2 inhibitors; DPP4i, dipeptidyl peptidase-4 inhibitors. 1 ANOVA, 2 Kruskal-Wallis's test, 3 Chi-squared test, a vs Cluster 1 p < 0.05, b vs Cluster 2 p < 0.05, c vs Cluster 3 p < 0.05.www.nature.com/scientificreports/cytokines from activated macrophages 33,34 .In accordance with this, study by Mohammed et al. 35 revealed a dose-dependent effect of Metformin on the reduction of NLR in T2DM patients.Furthermore, Cluster 1 had the highest percentage of patients using sulfonylureas, which also appear to have some anti-inflammatory effect www.nature.com/scientificreports/but less potent than metformin 36 .Despite the documented anti-inflammatory properties of insulin evidenced by both in vitro and animal studies-such as modulation of molecular pathways, reduction of pro-inflammatory cytokine expression, and augmentation of anti-inflammatory mediators-this cluster had the lowest percentage of patients using insulin 37 .

Cluster
The Cluster 1 had the highest percentage of patients using Angiotensin-Converting Enzyme Inhibitors.These drugs, while primarily recognized for their antihypertensive effects, also exhibit anti-inflammatory, antiproliferative, and antioxidant properties through their action on angiotensin II receptors 38 .This could have further contributed to the reduced inflammation levels observed in this cluster.Although with the lowest levels of inflammatory markers, patients from the Cluster 1 still had the higher percent of patients with diabetic complications, especially coronary artery disease compared to Cluster 4. It is possible that current snapshot of inflammatory markers might not provide a comprehensive history and inflammation may have decreased over time, perhaps due to medication or lifestyle modifications still resulting in coronary artery disease from previously elevated inflammation.However, it is crucial to emphasize that the Cluster 3 showed the highest level of inflammation and had the most pronounced incidence of coronary artery disease.This correlates with findings from prior research indicating that an elevated NLR is closely associated with the progression of coronary atherosclerosis.Increased ratios typically align with a deteriorating cardiovascular risk profile and increased complexity and severity of coronary artery disease confirming the established relationship between inflammation and cardiovascular complications in T2DM patients 39 .
We expected a higher percentage of patients to be using medications with proven cardiovascular and renal benefits (SGLT2i, GLP-1r) in Cluster 3, as it had the highest percentage of patients with renal impairment and coronary heart disease 40 .These drug classes have shown superiority in terms of cardiovascular and renal outcomes compared to DPP4i in patients with T2DM, as demonstrated in a meta-analysis that included 23 cardiovascular outcome trials 41 .However, in addition to Metformin, patients from this cluster more commonly used DPP4i.Other studies have yielded similar results, indicating that despite the proven benefits of SGLT2i and GLP-1r, physicians predominantly continue to prescribe DPP4i.This trend can be explained by clinical inertia 42 .
Our study had some limitations.The first limitation is a cross-sectional design of the study since it allows us to observe association between variables, but it limits us when making casual conclusions.Also, the wide range of age of diabetic patients included in the study could have influenced the characteristics of clusters.Another limitation derived from record based data, because there might be inaccuracies or missing information from medical records.For instance, there could be potential underreporting or misclassification of some clinical conditions based on the ICD-10 codes.While use of prescribed medication was recorded we did not provide data about dietary habits and consumption of over the counter drugs which could both influence inflammation levels.
In conclusion, it is worth to note that inflammation is one of the key contributors to disease T2DM pathophysiology and it is associated with variables like age, disease duration, glycemic control, kidney function and medication regimens.Still, it is important to emphasize that inflammation is not the only factor contributing to the development and progression of T2DM and its complications.Other factors like genetic predisposition, comorbidities, lifestyle choices, changes in metabolic control over the time all play significant role in disease progression.This also emphasizes the need to personalize approach in managing T2DM.In that sense, the identification of these distinct clusters provides invaluable insights.Beyond glycemic control, an integrated approach
To further evaluate cardiovascular risk factors, we considered the most recently reported body mass index (BMI) as well as mean values of systolic (SBP) and diastolic blood pressure (DBP) over the previous 12 months.

Statistical analysis
Data are presented as mean ± standard deviation, frequencies, and percentages.We analyzed data in R using two-step clustering method similar to Ahlqvist and colleagues 46 .
In the first step, the optimal number of clusters was determined to be 4 by using silhouette analysis (using the pam function) on a series ranging from 2 to 8 clusters.In the second step, hierarchical clustering with Gower distances (accommodate continuous, categorical, and binary variables) was performed to determine different profiles of diabetes patients.The dendrogram (Fig. 1) visualizes the results of patient clustering based on the following variables: age, BMI, T2DM duration, smoking status, hypertension, metformin, sulfonylurea, glucagon-likepeptide-1 receptor agonists (GLP-1r), sodium-glucose cotransporter-2 inhibitors (SGLT2i), dipeptidyl peptidase-4 inhibitor (DPP4i), fasting glycaemia, Hba1c, urea, creatinine, GFR, GFR category, TC, HDL, LDL,

Figure 1 .
Figure 1.Hierarchical clustering of the diabetes patients.

Table 1 .
Demographic and clinical characteristics data of the study population.BMI, body mass index; SBP,

Table 4 .
Clinical characteristics and inflammation markers in relation to the cluster analysis.