3.1 Descriptive statistics
We employed listwise deletion to manage missing data (46), which ranged from 0% for demographics (age, gender, ethnicity) to > 20% for BMI, and > 40% for diastolic/systolic blood pressure (40.1% each), and waist/hip ratio (40.6%), to as high as 60% for education level (61%), HbA1c (60.3%), and HDL-C (60.1%) (see Figure 2). Despite the limitations of listwise deletion, this approach was preferred to inputting (replacing) missing data using estimated parameters (e.g., expectation maximisation). The latter methods require assumptions of multivariate normality, which is problematic with categorical variables (e.g., SRH, MetS) (47). Regardless, we performed sensitivity analysis to compare the effects of listwise deletion versus expectation maximisation on regression results.
Of 584 patients diagnosed with T2DM, 353 patients (60.3%) met the criteria for MetS. The percentage of patients meeting each individual diagnostic criterion are as follows: HDL-C < = 0.9 mmol/L (35 mg/dl) (n = 391 (67%)), waist/hip ratio = > 0.85cm (n = 316 (54.1%)); BMI > 30 kg/m2 (n = 229 (39.2%)); diagnosed with hypertension by a doctor or nurse; (n = 370 (63.4%)): systolic blood pressure > 140 mmHg (n = 82 (14%)) and diastolic blood pressure > 90 mmHg (n = 14 (2.4%)). Just over a quarter of patients had a HbA1c > 48 mmol/mol (n = 167 (28.6%)). The percentage of participants per SRH category were ‘very good’ (9.8%), ‘good’ (32.9%), ‘fair’ (34.8%), ‘bad’ (16.1%), and ‘very bad’ (6.5%). Thus, just over 40% of patients reported ‘good’/’very good’ health.
Table 1 shows means, SDs, and frequencies for the overall sample and by MetS status (cases versus non-cases). MetS cases were significantly less likely to report ‘very good’/‘good’ SRH (χ2 (1, N = 583) = 13.344, p < 0.001). There were no group differences in demographic factors or systolic/diastolic blood pressure (all p’s > 0.01). However, MetS cases were significantly more likely than non-cases to be HDL-C deficient (HDL-C < = 0.9 mmol/L (35 mg/dl)) (χ2 (1, N = 583) = 92.768, p < 0.001), and generally overweight (BMI > 30 kg/m2), (χ2 (1, N = 583) =159.041, p < 0.001), but less likely to be centrally obese (waist/hip ratio = > 0.85cm), (χ2 (1, N = 583) =12.960, p < 0.001). MetS cases were also more likely to be hypertensive (χ2 (1, N = 583) = 231.923, p < 0.001), but show better glycaemic control (HbA1c > 48 mmol/mol), (χ2 (1, N = 583) = 45.034, p < 0.001).
Respondents smoked an average of 2.28 cigarettes a day, and consumed alcohol 5.6 times in the past 12 months. The sample met WHO thresholds for obesity (BMI (kg/m2) > 0.30 (M = 31.22)), high central adiposity (waist/hip ratio (cm) > 0.9 (men) (M = 1.00), > 0.85 (women) (M = 0.91)), and poor glycaemic control (HbA1c > 48 mmol/mol) (M = 57.50). HDL-C levels were normal (i.e., above minimum thresholds of < 0.9 mmol/L in men (M = 1.19) and < 1.0 mmol/L in women (M = 1.31)). Systolic/diastolic blood pressure values were also below the critical thresholds of >140/90 mmHg (M = 129/69.72).
Independent samples t-tests comparing cases and non-cases showed the former group had significantly higher BMI (kg/m2), exceeding the threshold for obesity (M = 33.41 versus 27.54), t(459.82) = -12.74, p < 0.001, greater waist/hip ratio (M = 0.98 versus 0.94), t(343.70) = -4.22, p < 0.001, and lower serum HDL-C (M = 1.18 versus 1.30), t(183.65) = 2.69, p < 0.01. There were no group differences in blood pressure, HbA1c, or lifestyle factors (all p’s > 0.01).
……………………………………………..
[Insert Table 1 about here]
Sample characteristics by metabolic syndrome status
………………………………………………
……………………………………………..
[Insert Table 2 about here]
Final regression models predicting metabolic factors from self-rated health and metabolic covariates in the whole sample
………………………………………………
3.2 Hypothesis 1: Does SRH predict metabolic abnormalities in T2DM patients?
Table 2 shows results of bootstrapped hierarchical multiple regression predicting metabolic abnormalities. SRH significantly predicted HDL-C (mmol/L) (Model 2) (β = -0.17, p = 0.015), increasing the explained variance, ∆R2 = 0.029, F (1, 176) = 6.035, p = 0.015. However, adjusting for metabolic factors (Model 3) negated this association, accounting for an additional 6.7% of the variance in HDL-C (∆R2 = 0.067, F (5, 171) = 2.976, p = 0.013).
SRH failed to predict systolic blood pressure (mmHg) (Model 2). Adding metabolic covariates (Model 3) significantly improved the model (∆R2 = 0.254, F (5, 171) = 13.269, p < 0.001), primarily due to diastolic covariance (β = 0.53, p < 0.001). Similarly, SRH failed to predict diastolic blood pressure (mmHg), whereas adding metabolic factors significantly improved model fit (∆R2 = 0.271, F (5, 171) = 15.660, p < 0.001), mainly due to systolic effects (β = 0.47, p < 0.001) and HbA1c (mmol/mol) (β = 0.18, p = 0.003).
The association between SRH and HbA1c (mmol/mol) was significant (β = -0.20, p = 0.008) prior to adjusting for metabolic covariates (Model 2) (∆R2 = 0.082, F (1, 176) = 7.241, p = 0.008). Adding metabolic variables (Model 3) significantly improved the model (∆R2 = 0.084, F (5, 171) = 3.454, p = 0.005), negating the SRH-HbA1c relationship (p = 0.04). Finally, SRH failed to predict anthropometric criteria (BMI, (kg/m2), waist/hip ratio (cm)) (Model 2). Including metabolic factors explained additional variance for both BMI (∆R2 = 0.090, F (5, 171) = 3.835, p = 0.003) and waist/hip ratio (∆R2 = 0.069 F (5, 171) = 4.027, p = 0.002).
……………………………………………..
[Insert Table 3 about here]
Final regression models predicting metabolic factors from self-rated health and metabolic covariates in T2DM patients with MetS
……………………………………………..
3.3 Hypothesis 2: Does SRH predict metabolic abnormalities in T2DM patients by MetS status?
Table 3 shows the results for T2DM patients who met MetS diagnostic criteria. Crucially, SRH failed to predict any metabolic variable (Model 2) prior to adjusting for metabolic covariates (Model 3) (all p’s > 0.01).
BMI was predicted by both age (β = -0.44, p = 0.001) and gender (β = -0.42, p = 0.009). Gender also predicted waist/hip ratio (p < 0.001), while age predicted diastolic blood pressure (p = 0.001). Adding metabolic predictors (Model 3) significantly improved the predicted variance for systolic blood pressure (∆R2 = 0.286, F (5, 63) = 5.517, p < 0.001) and diastolic blood pressure (∆R2 = 0.229, F (5, 63) = 5.395, p < 0.001).
Table 4 shows coefficients for patients who did not meet MetS criteria (i.e., T2DM-only patients). Again, SRH failed to predict any metabolic factor (Model 2), prior to accounting for metabolic covariates (all p’s > 0.01). Adjusting for metabolic variables (Model 3) explained significant additional variance for both systolic (∆R2 = 0.211, F (5, 96) = 7.069, p < 0.001) and diastolic (∆R2 = 0.286, F (5, 96) = 9.236, p < 0.001) blood pressure.
……………………………………………..
[Insert Table 4 about here]
Final regression models predicting metabolic factors from self-rated health and metabolic covariates in T2DM patients without MetS
………………………………………………
3.4 Exploratory analysis by gender
Research suggests men are more sensitive to physical health (48). Given the strong associations between gender and anthropometric markers observed here (see above), we decided to rerun regression analysis stratified by gender. The results are shown in Table 5. SRH significantly predicted HDL-C (mmol/L) in male patients (Model 2) (β = 0.25, p = 0.01), accounting for a significant 6.1% increase in the explained variance, after accounting for demographic and lifestyle factors, ∆R2 = 0.061, F (1, 93) = 6.712, p = 0.011.
Adjusting for metabolic factors (Model 3) did not negate the association between SRH and HDC-C (β = 0.25, p = 0.01) in males and failed to improve the model (∆R2 = 0.095, F (5, 88) = 2.253, p = 0.056). SRH also predicted HbA1c (mmol/mol) in female patients (Model 2) (β = -0.31, p = 0.007), explaining 8.4% variance (∆R2 = 0.084, F (1, 77) = 7.696, p = 0.007). Adjusting for metabolic abnormalities (Model 3) significantly improved the model, predicting another 15% of the variance (∆R2 = 0.156, F (5, 72) = 3.287, p = 0.01), but did not nullify the SRH-HbA1c association (β = -0.27, p = 0.01). SRH failed to predict the other metabolic variables, irrespective of metabolic adjustment (all p’s > 0.01).
……………………………………………..
[Insert Table 5 about here]
Final regression models predicting HDL-C and HbA1c from self-rated health and metabolic covariates in males and females
………………………………………………
3.5 Sensitivity analysis
We reanalysed the data with expectation maximisation applied to missing values, to compare the effects of different methods for resolving incomplete data (list wise deletion versus EM). As observed in previous analysis, SRH failed to predict HDL-C (mmol/L), waist/hip ratio (cm), and systolic/diastolic blood pressure (mmHg) after adjusting for metabolic covariates (all p’s > 0.01). However, contrary to expectations, SRH significantly predicted BMI (kg/m2) after metabolic adjustment (Model 3) (β = -0.12, p = 0.002). Furthermore, the previously significant SRH - HbA1c association was no longer reliable (β = -0.06, p = 0.10). Collapsing the data by MetS status (cases versus non-cases) did not change the results: SRH failed to predict any metabolic variable after adjusting for metabolic covariates (Model 3) (all p’s > 0.01). When the data was collapsed by gender the previously significant association between SRH and HDL-C in males was negated (β = 0.10, p = 0.04). Furthermore, SRH now failed to predict HbA1c (mmol/mol) in females (β = -0.07, p = 0.20). Overall, sensitivity analysis indicated some findings may be affected by the management of missing data using expectation maximisation algorithms.
3.6 Structural equation modelling
We used SEM to explore direct and indirect associations between SRH and metabolic abnormities. We were curious to see whether relations between SRH and metabolic factors are indirect, mediated by lifestyle factors (e.g., SRH negates health-protective behaviours, which in turn precipitate metabolic dysfunction) (8). Model fit was based on standard criteria: chi-square χ2 (CMIN) (p > 0.05), χ2 (CMIN)/df < 5.00, root mean square error of approximation (RMSEA) < 0.07, comparative fit index (CFI) ≥ 0.95, Tucker and Lewis Index (TLI) ≥ 0.95, and normed fit index (NFI) ≥ 0.95 (49). Metabolic factors were allowed to affect SRH, that in turn was allowed to predict lifestyle factors, which then affected metabolic variables (representing a vicious cycle in which lifestyle was a mediating factor). SEM analysis using IBM SPSS AMOSTM (version 26), with specification search, generated 192 candidate models, none of which provided a satisfactory fit. The ‘best’ model (BIC (Bayesian Information Criterion) = 0, χ2 (CMIN)/df < 5.00) suggested a cyclical relationship between HDL-C, SRH, and alcohol intake. However, this model did not satisfy most other fit criteria: CMIN (p < 0.05), RMSEA (> 0.07), CFI (< 0.95), and TLI (< 0.95)) and was therefore discarded.