Statistical analyses

Analyses were performed using Stata version 14.2 (StataCorp LP, College Station, TX, USA). Within each disease-positive sample, we estimated the crude prevalence of unawareness of that disease, overall and stratified by 10-year age groups and ethnicity. To evaluate the determinants of unawareness of the three diseases, we also compared background, sociodemographic, and clinical and biochemical characteristics between patients who were aware and unaware of having each disease using the chi-squared test or Fisher’s exact test for categorical variables as appropriate, and the student’s t test for continuous variables. Variables that were significantly associated with the outcomes in univariable analyses (i.e. P < 0.05; see Table ​Table1)1) or those that were found to be associated with unawareness of the condition(s) in previous studies were then assessed for an independent association with unawareness of each disease by simultaneous inclusion as covariables in a multivariable logistic regression model. These variables include age, gender, race, BMI (continuously and categorically), income, occupation, marital status, housing, number of non-CVD-related comorbidity, and mutually for each other.

Baseline characteristics of patients by lack of awareness of diabetes, hypertension and hypercholesterolemia

HDB Housing Development Board

*Includes ocular (cataract, myopia, age-related macular degeneration, glaucoma, diabetic retinopathy, eye trauma) and systemic non-CVD (e.g. thyroid disease) conditions

Values in italics: P <0.05

To estimate the independent relationship between disease unawareness and clinical control, we constructed a multivariable logistic regression model with unawareness as the exposure and poor clinical control of the disease as the outcome (i.e. SBP ≥ 140 mmHg or DBP ≥ 90 mmHg for hypertension, HbA1c > 7% for diabetes and total cholesterol ≥ 6.2 mmol/L for hypercholesterolemia), adjusted for variables that were associated with the outcomes in univariable analyses (see Additional file 1: Table S1) or were found to be associated with the respective poor disease control in previous studies. These variables include age, gender, ethnicity, body mass index, self-reported non-cardiovascular comorbidity, income, education, smoking, occupation, marital status, housing, duration of diabetes (for lack of awareness of diabetes only), and mutually for each other. All estimates were presented with 95% confidence intervals.