*3.3. Mortality*

The National Death Index, a civil registry for all Spanish residents, was used to ascertain the vital status and date of death for all participants from 25 July 2011 to March 2018. Vital status was also updated during household visits in the follow-up assessment by asking respondents' relatives. A final update was conducted on the 31 October 2018 by once again consulting the National Death Index. Twelve participants that appeared as deceased had no information about their date of death. Thus, we estimated their date of death as occurring at the mid-point between the date of interview at baseline and 31 October 2018.

## *3.4. Statistical Analysis*

Unweighted frequencies, weighted proportions, and means were used for descriptive analyses. Distinct levels of fruit and vegetable consumption were compared using the Rao-Scott chi-squared test statistic (which adjusts for complex sample design) [35] for categorical variables and one-way ANOVA test for continuous variables.

Mortality was the outcome for the analyses. Kaplan–Meier survival curves and log-rank test statistics were used to estimate the time to death (from the first interview) stratified by levels of fruit and vegetable consumption. Participants who were alive at the end of the observational period (31 October 2018) were censored.

We conducted unadjusted and adjusted Cox proportional hazards regression models to explore the association between fruit and vegetable consumption and risk of all-cause mortality. The adjusted model included levels of fruit and vegetable consumption (with "low" level as the reference category) plus other potential confounders at baseline (gender, age, educational level, cohabiting, smoking status, level of physical activity, obesity, and number of chronic conditions). Interactions between levels of fruit and vegetable consumption and covariates were explored in the adjusted models. Hazard ratios (HRs) with their 95% confidence intervals (CI) were calculated.

The assumption of proportionality was explored by calculating plots of cumulative hazard functions across the independent variables. Violation in the assumption was not found. All analyses were performed using Stata version 13 for Windows (SE version 13, StataCorp: College Station, TX, USA) taking into account complex sampling design. Weights were used to adjust for di fferential probabilities of selection within households, and post-stratification corrections to the weights were made to match the samples to the socio-demographic distributions of the Spanish population. Statistical significance was set at *p* < 0.05.
