Proc surveymeans

I intake can vary between population groups. The observed I intake from natural sources
(excluding iodised salt and supplements) was estimated for several population
characteristics (age, sex, BMI, income of the head of the household, education (for
children, highest education of parents), season, region) using Proc SurveyMeans (SAS 9.2).
With linear regression (Proc SurveyReg (SAS 9.2); P<0·05
statistically significant) it was studied how the intake of I naturally present in foods
and population characteristics were associated, similar to Huybrechts et al.⁽²⁶⁾. These analyses were performed separately for children (7–18 years) and
adults (19–69 years). The variables I intake and I intake/418 kJ (100 kcal) were natural
log transformed to improve normality of the data and residuals. Potential outliers were
detected and analyses were performed including and excluding these outliers to study the
impact on the results. Linear regression was also performed to study the association
between the intake of two food groups (milk and cheese) contributing most to the intake of
I naturally present in foods (excluding iodised salt and supplements) and to the intake of
bread (a main source of iodised salt), among consumers only. These intake variables were
square root transformed to improve normality of the data and residuals, except for cheese
intake among adults that was transformed with a natural log.

We selected a 15^th percentile threshold to define persistent high cost utilization since this threshold has been commonly used in prior studies.[40 (link)–42 (link)] However, results could be sensitive to how persistent high cost utilization is defined, and therefore impact the interpretation of our findings. Therefore, we tested robustness of the final, fully-adjusted model results using sensitivity analyses that alternatively defined group classification using 90^th/10^th and 80^th/20^th (HIGH/LOW) percentile thresholds.[6 (link),42 (link),45 (link),46 ]
All regression and cost estimation analyses were conducted using survey procedures (PROC SURVEYLOGISITIC and PROC SURVEYMEANS; SAS v.9.3, SAS Institute Inc., Cary, NC) to account for the complex sampling design of the survey and employed person-level SAQ sampling weights to adjust for questionnaire non-response.[47 ] Taylor series linearization was used for variance estimation. Alpha was set at p = 0.05 for all analyses.

All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). Changes in MetS prevalence were stratified by sex and age because the effects of the COVID-19 pandemic have been reported to vary according to sex and age. We also performed pre-specified subgroup analyses based on marital status, education level, household income, and occupation. The values for categorical variables were calculated using SAS (PROC SURVEYFREQ; SAS Institute Inc.), and the values for continuous variables were similarly calculated using SAS (PROC SURVEYMEANS; SAS Institute Inc.). If p-value <0.05, the difference was considered to be statistically significant.