*2.5. Data Analysis*

Firstly, analyses described the sample and checked the internal consistency of the CORE-OM instrument. Then, inferential analyses were performed to ascertain the effect on the mental distress indicator over time. For this analysis, the direct effect of time and, separately, the interaction of the effects with predictors were tested.

To evaluate the effect of time, repeated measures were considered, accounting for the variation between subjects through the Multilevel Modeling (MLM) analysis [28–30]. Each of the study stages was categorized as a level, regardless of the time interval between them (i.e., stages one, two, and three). MLM handles non-participation across the three stages by estimating a linear path of any two scores completed. Slope against time was first treated as fixed (stable among participants) then a random effect of time, i.e., different score slopes per participant, were also allowed.

After the identification of changes over time in the NR score, analyses were performed to find out predictors of this change. Stable (gender and age) and time-varying predictors (exercise practice, support in domestic activities, people available to talk, psychological and psychiatric follow-up) were tested separately. Though the latter are time-varying predictors, moderate and high levels of correlation between repeated measures were found (τb coefficient varying from 0.371 to 0.694) and only levels at the first participation were entered. Separate MLM analyses were performed for each of the potential predictors of change. Finally, score changes for participants who answered at least twice were evaluated using the Reliable Change Index (RCI), which indicates, for each participant, whether changes were larger than would be expected 5% of the time based on the reliability of the measure [31,32]. For each of the two time periods, the RCI allowed participants to be categorized as deteriorating, not reliably changing, or improving.. These proportions were compared across the two time periods, and we used this to find participants with deterioration and to inspect qualitative responses provided in the open question about concerns during the pandemic. For this simple analysis, we used the categories previously reported [23] that were identified using the Consensual Qualitative Research for simple qualitative data method—CQR m [33]—in the first stage of study.
