This prospective observational study included 220 participants recruited between January 2022 and January 2023 from COVID-19 hospitalization centers operated by the Iraqi Ministry of Health. The study cohort comprised 160 patients diagnosed with T1DM and confirmed positive for SARS-CoV-2 by reverse transcription-polymerase chain reaction (RT-PCR) and were hospitalized in endocrinology departments, as well as 60 age- and sex-matched healthy individuals who served as controls.

Controls subjects were recruited during routine clinical screening or blood donation programs conducted within the same institutions. All controls tested negative for SARS-CoV-2 by RT-PCR at enrolment and had no history of COVID-19 infection. Individuals with metabolic disorders, inflammatory diseases, or acute infections were excluded from the healthy control group.

According to the World Health Organization (WHO) interim guidance for the clinical management of COVID-19, T1DM patients with COVID-19, were stratified into two subgroups: a mild/moderate group (n100) and a severe group (n=60)). Mild disease was defined by the presence of symptoms without evidence of pneumonia or hypoxia. Moderate disease was characterized by clinical signs of pneumonia (e.g., fever, cough, dyspnea, or fast breathing) without features of severe pneumonia. Severe disease was defined by the presence of at least one of the following criteria: oxygen saturation (SpO₂) < 90% on room air, respiratory rate > 30 breaths/min, or signs of severe respiratory distress (17).

Written informed consent was obtained from all participants, and the study protocol was approved by the Research and Ethical Considerations Committee (RECC) of the Ministry of Health of Iraq.

Eligible participants were adults aged 18 to 75 years. Patients in the T1DM group had a confirmed diagnosis of T1DM for at least one year in addition to laboratory-confirmed COVID-19 infection by RT-PCR. Individuals were excluded if they presented with autoimmune diseases other than T1DM, a history of malignancy or immunosuppressive therapy within the past six months, pregnancy or lactation, or incomplete medical records.

At admission, demographic and clinical information was recorded, including age, sex, body mass index (BMI), comorbidities (such as hypertension, cardiovascular disease, and obesity), and smoking status. Clinical assessment involved measurement of vital signs, including systolic and diastolic blood pressure, heart rate, and oxygen saturation. Hospitalization data, including length of stay, requirement for mechanical ventilation, and survival outcomes, were also documented.

Systemic inflammatory markers were measured using standard laboratory procedures. CRP and D-dimer concentrations were quantified using the ichroma™ II analyzer (Boditech Med, South Korea) in accordance with the manufacturer’s instructions. The neutrophil-to-lymphocyte ratio (NLR) was calculated by dividing the absolute neutrophil count by the absolute lymphocyte count obtained from the complete blood count (CBC). All measurements were performed under standardized laboratory conditions to ensure analytical accuracy and reproducibility.

Serum samples were collected from all participants within 24 hours of hospital admission and stored at -80 °C. Serum concentrations of cytokines were quantified using commercially available enzyme-linked immunosorbent assay (ELISA) kits according to the manufacturers’ protocols.

Interleukin-6 (IL-6) and interleukin-10 (IL-10) levels were measured using ELISA kits from BT Laboratory (Shanghai, China; catalog numbers E0090Hu and E0102Hu, respectively). The analytical sensitivity of the IL-6 assay was 1.03 pg/mL, while that of the IL-10 assay was 2.59 pg/mL. Interleukin-12 (IL-12) concentrations were determined using an ELISA kit from myBioscience (San Diego, CA, USA; catalog number MBS8801265), with a sensitivity of 3.3 pg/mL. Interleukin-18 (IL-18) levels were measured using a human IL-18 ELISA kit from Invitrogen™ (Thermo Fisher Scientific; product code 10079322), with a sensitivity of 6.25 pg/mL.

Optical density was measured using a microplate reader. Cytokine concentrations were calculated from standard curves generated using recombinant cytokine standards and were expressed in picograms per milliliter (pg/mL). All assays were performed in triplicate.

Genomic DNA was extracted from whole blood collected in EDTA tubes using the QIAamp DNA Blood Mini Kit (Qiagen, Chatsworth, CA, USA) following the manufacturer’s instructions. DNA concentration and purity were assessed using a NanoDrop spectrophotometer (Thermo Fisher Scientific, USA).

Polymerase chain reaction (PCR) amplification was carried out in a final reaction volume of 25 µL containing 100 ng of genomic DNA, 12.5 µL of 2× Green Master Mix (Promega, USA), 0.75 µL of outer forward primer (0.4µM), 0.75 µL of outer reverse primer (0.4 µM), 1.5 µL of inner forward primer (0.2 µM), and 1.5 µL of inner reverse primer (0.2 µM) (Table 1).

Briefly, The thermal cycling conditions used for amplification of IL-6 (rs1800795), IL-10 (rs1800872), IL-12 (rs568408), and IL-18 (rs1946518) polymorphisms consisted of an initial denaturation at 95 °C for 2 minutes, followed by 35 cycles of denaturation at 95 °C for 20 seconds, annealing at 57 °C for 20 seconds, and extension at 72 °C for 1 minute, with a final extension step at 72 °C for 5 minutes.

PCR products were separated by electrophoresis on a 2% agarose gel stained with ethidium bromide, and visualized under ultraviolet illumination using a Gel Doc XR Plus system (Bio-Rad, USA) (Figure 1).

All statistical analyses were performed using SPSS version 21.0 (SPSS Inc., Chicago, IL, USA) and GraphPad Prism 9 (GraphPad Software Inc., La Jolla, CA, USA). Quantitative data were summarized using descriptive statistics, with continuous variables expressed as mean ± standard deviation (SD), whereas non-normally distributed variables were expressed as median with interquartile range (IQR). Data normality was assessed using the Shapiro-Wilk test for all clinical and biochemical parameters, including interleukins (IL-6, IL-10, IL-12, IL-18), CRP, D-dimer, SpO₂, and NLR. Correlation analyses were performed using Spearman’s rank correlation coefficient, as the data showed a non-normal distribution. Comparisons between patient and control groups were performed using the independent samples t-test for normally distributed variables and the Mann-Whitney U test for non-parametric data. To account for multiple comparisons across cytokines and inflammatory markers, p-values were adjusted using the Holm-Bonferroni correction. All statistical test were two-tailed, and p-value <0.05 was considered statistically significant.

Genotype and allele frequencies of cytokine gene polymorphisms were compared using the Chi-square test or Fisher’s exact test, as appropriate. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated to estimate relative risk. A p-value < 0.05 was considered statistically significant. A Bonferroni correction was applied for genotype comparisons.

Supervised machine-learning analyses were performed exclusively within the cohort of T1DM patients (n = 160) to identify inflammatory biomarkers predictive of disease severity. Model inputs included seven clinically relevant inflammatory markers: IL-6, IL-10, IL-12, IL-18, CRP, D-dimer, and NLR. No a priori feature elimination was applied; instead, feature selection was implicitly handled through Gini impurity reduction in the Decision Tree model and mean decrease in Gini impurity in the Random Forest model.

The dataset was randomly split into training (70%) and testing (30%) subsets using stratified sampling to preserve class proportions. Class imbalance (mild/moderate n = 100 vs. severe n = 60) was addressed by applying inverse class-frequency weighting during model training. A Decision Tree classifier was constructed using the Gini impurity criterion with constrained maximum depth to limit overfitting and enhance interpretability. In parallel, a Random Forest classifier comprising 100 trees was implemented, with square-root feature sampling at each split, to improve predictive performance and model stability via bootstrap aggregation.

Model performance was evaluated on the independent test set using confusion-matrix–derived metrics, including sensitivity, specificity, and classification accuracy were derived from the confusion matrix, based on the numbers of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). These terms were defined as follows:

These metrics were used to evaluate model performance on the independent test set.

The study involved 220 participants divided into two groups: 160 patients with T1DM and COVID-19, and 60 healthy controls. The mean age was 55.66 ± 8.70 years in the patient group and 54.82 ± 8.55 years in the control group, showing no statistically significant difference (p = 0.523). Body Mass Index (BMI) analysis revealed that a higher proportion of patients fell into the overweight and obese categories compared to the control group (Table 2).

As shown in Figure 2, inflammatory and immunological biomarkers are presented as median with IQR. All measured biomarkers showed significant differences between patients and controls (Mann-Whitney U test, all p < 0.0001), and these associations remained statistically significant after Holm–Bonferroni correction for multiple comparisons (α=0.0071).

Serum levels of IL-6 (Figure 2A), IL-10 (Figure 2B), IL-12A (Figure 2C), and IL-18 (Figure 2D) were markedly increased in patients compared with controls (IL-6: 67.03 [57.03-73.91] vs. 2.11 [1.73-2.45] pg/mL; IL-10: 14.77 [10.92-17.95] vs. 4.11 [3.9-4.32] pg/mL; IL-12: 8.1 [6.22-10.15] vs. 4.31 [3.69-4.95] pg/mL; IL-18: 193.3 [175.5-212.3] vs. 8.09 [7.6-8.5] pg/mL), indicating a pronounced pro-inflammatory and immunoregulatory activation in the patient group.

Similarly, levels of CRP (Figure 2E) and D-dimer (Figure 2F) were substantially higher in patients than in controls (CRP: 39.6 [34.18-44.70] vs. 1.8 [1.5-2.27] mg/L; D-Dimer: 4.45 [4.03-4.97] vs. 0.20 [0.18-0.24] mg/L), reflecting systemic inflammation and activation of coagulation pathway.

Furthermore, the NLR (Figure 2G) was significantly increased in patients compared with controls (8.2 [0-10.2] vs. 1.82 [1.49-2.16]; p < 0.001), underscoring its potential role as an indicator of immune imbalance dysregulation and disease severity. Collectively, these findings reveal a distinct inflammatory and immunological signature in affected individuals relative to healthy subjects.

Figure 3 illustrates the Spearman correlation coefficients (r) among the studied biomarkers in comorbid T1DM and COVID-19.

A moderate positive correlation was observed between IL-12 and IL-18 (r = 0.41, p<0.001), suggesting a coordinated pro-inflammatory immune response. In contrast, IL-12 exhibited a weak negative correlation with IL-6 (r = -0.22, p=0.005), indicating a potential regulatory divergence within cytokine signaling pathways.

Additionally, CRP showed a weak positive association with NLR (r = 0.27, p=0.001), supporting a possible link between systemic inflammation and hematological alterations. Overall, the correlation matrix underscores the complex interplay between cytokine levels and systemic inflammatory responses in patients.

The genotypic and allelic distributions of IL-6 (rs1800795), IL-10 (rs1800872), IL-12 (rs568408), and IL-18 (rs1946518) polymorphisms were compared between COVID-19 patients with T1DM and healthy controls (Table 3). No significant differences were observed in either allele or genotype frequencies for IL-6 rs1800795 and IL-12 rs568408 polymorphisms (p > 0.05), suggesting no direct association with disease susceptibility.

In contrast, a significant association was identified for IL-10 (rs1800872), with both allelic and genotypic distributions differing markedly between patients and controls (p < 0.001, which remained significant after correction for multiple comparisons, α = 0.0125). The G allele was significantly more frequent in patients (76%) than in controls (52%) and was associated with increased susceptibility to COVID-19 in T1DM patients (OR = 2.85; 95% CI = 1.83 - 4.43). Furthermore, the GG genotype predominated among patients (65%) compared with controls (28%) and showed a strong association with disease susceptibility (OR = 4.07; 95% CI = 1.74-9.52). In contrast, the TT and TG genotypes were significantly more prevalent in healthy individuals but did not reach a statistical significance.

For IL-18 (rs1946518), the GG genotype was more prevalent among patients (17%) compared with controls (7%). However, this distribution did not translate into a statistically significant association with disease susceptibility.

To further investigate the predictive power of inflammatory and genetic biomarkers for classifying COVID-19 severity in patients with T1DM, we employed supervised machine learning models (Figure 4). A Decision Tree classifier, using key immunological variables, selected IL-6 as the primary splitting variable at the root node. This selection reflects the optimal reduction in Gini impurity at the initial decision step rather than overall predictor dominance. The Random Forest model, which evaluates variable importance across multiple trees, consistently ranked IL-18 as the most significant biomarker at the root node, followed by CRP and IL-6. This model achieved an overall accuracy of 85.6%, with a sensitivity of 88.3% and a specificity of 81.7%. To enhance performance and mitigate overfitting, a Random Forest classifier, an ensemble of 100 trees, was subsequently implemented. This model demonstrated superior predictive power, yielding an accuracy of 91.2%, a sensitivity of 92.5%, a specificity of 89.1%, and a mean AUC of 0.975. Feature importance analysis from the Random Forest confirmed and reinforced the initial findings, ranking IL-18 (0.276), CRP (0.202), and IL-6 (0.175) as the top three predictors (Figure 5). These results underscore the critical role of hyperinflammatory markers, particularly IL-18 and CRP, in stratifying COVID-19 risk in T1DM patients and highlight the potential of ensemble methods like Random Forest as robust tools for clinical decision-support systems.

Comparative analysis of key inflammatory biomarkers revealed significantly elevated serum levels in T1DM patients with severe COVID-19 compared to those with presenting mild/moderate disease. As shown in Table 4, IL-18 concentrations were markedly higher in the severe group (median [IQR]: (218 [208.4-230] pg/mL vs. 181.9 [165.8-192.3] pg/mL; p<0.0001), as were CRP (46.7 [44.2-48.7] mg/L vs. 35.7 [32.33-39] mg/L; p<0.0001) and IL-6 (76.23 [68.28-78.5] pg/mL vs. 60.76 [42.85-59.48] pg/mL; p<0.0001). All associations remained statistically significant after adjustment for multiple comparisons using the Holm–Bonferroni correction.

These univariate findings were further validated using supervised machine-learning approaches: a Decision Tree classifier identified IL-18 as the primary node for disease severity stratification. Consistently a Random Forest model ranked IL-18, CRP, and IL-6 as the three most influential predictors of severe disease. The concordance between elevated serum biomarker levels and their high predictive importance underscores the critical role of this hyperinflammatory triad in determining COVID-19 severity among T1DM patients.