How anxiety and depression mediate the link between sleep quality and health perception during crisis periods

Table of Contents

Study design

We made use of information from the Federal University of Ouro Preto’s (UFOP) population-based serological survey, “COVID-Inconfidentes: Epidemiological surveillance of COVID-19 in the region of Inconfidentes/MG.” The analyzed data was collected in two medium-sized towns in the south-central region of Minas Gerais, Brazil, which is one of the country’s major iron ore-producing regions, between October and December 2021.

Conglomerate carried out the sample design in three phases: household, resident, and census sector. The National Household Sample Survey (PNAD)¹⁶, the Family Budget Survey (FBS)¹⁷, the “Saúde em Beagá” survey¹⁸, and National Health Survey (NHS)¹⁹ served as the foundation for this design. As a result, the census sectors were used as the main sampling units in the study design. They were chosen with a probability proportionate to the number of households, and the number of households was taken as a measure of size from the summary of the 2010 population census by census sectors. In order to minimize the possibility of selecting a sample from non-representative sectors (primary units), prior stratification was carried out before selecting the primary units, taking into account the average income, based on data from the 2010 demographic census of the Brazilian Institute of Geography and Statistics (IBGE). Consequently, the final sample was guaranteed to be representative of the three socioeconomic strata (wages less than one, wages between one and three, and wages greater than four).

The updated list of current household units in the primary sampling units (selected census sectors) was used to select the households that made up the secondary sampling units in a methodical manner. Both private homes and housing units in communal residences with occupants make up the household units. Following the census sectors’ selection, the household selection interval (k) for the systematic sampling was computed using the following formula: k = Ni/(xi/ni), where Ni is the total number of households in the census sector, xi is the sample size, and ni is the number of households to be selected. This method covered the entire geographic area and produced a proportionate number of homes per sector. Following the selection of the first home in the census sector based on IBGE indications, the next household was systematically sampled in accordance with the household selection interval (k).

The individuals who were chosen through a basic random sampling process constituted the tertiary sampling units. A list of every adult resident in the chosen home was created, and one resident was chosen at random to take part in the study. A previous publication provides further details regarding data collection²⁰.

In-person interviews were conducted with the adult residents (≥ 18 years) in their homes using an electronic form. Living habits, general health status, and sleep quality were characterized by thematic components of the questionnaire. Information assessed was collected using scales built and validated in Brazil. An instrument designed for this survey was used to ask specific questions about societal and financial traits, among other things. The previously published scales are explained in the method below, and we’ve also included the complete survey instrument with all the questions as additional information.

Exposure: poor sleep quality

To assess the quality of sleep, the Pittsburgh Sleep Quality Index was used. With an overall reliability coefficient of 0.82, the Brazilian version of the PSQI demonstrated high internal consistency. This test consists of 19 questions divided into 7 components, each with a score ranging from 0 to 3: subjective quality of sleep (C1), sleep latency (C2), duration of sleep (C3), habitual efficiency of sleep (C4), and sleep disturbances (C5) The overall score is calculated by adding up the scores, with the highest score indicating the lowest quality of sleep²¹. Sleep was the two categories used in this study: good quality (PSQI score ≤ 5) and poor quality (PSQI score > 5) sleep.

Outcome: self-rated health

The outcome variable, self-rated health, was assessed by asking, “In general, how do you rate your health?”¹³. Very poor, poor, regular, good, and very good were the options for the response. Following that, this variable was split into two categories: good (which included the very good and good categories) and poor (which included the very poor, poor, and regular categories).

Mediators: mental health

The presence of depression or anxiety was evaluated as one of the mediation factors. We used medication use, self-report of medical diagnosis, and symptoms to assess these factors. We classified participants with anxiety or depression if they met at least one of these requirements.

The generalized anxiety disorder scale (GAD-7) and the patient health questionnaire (PHQ-9) scales, both translated and validated in Portuguese, were used for the symptom criterion^22,23. The GAD-7 is a seven-item, four-point Likert scale with a maximum score of 21 points²⁴. Nine questions make up the PHQ-9, a four-point Likert scale with a score range of 0 to 27 points²⁵. Scores of 10 or higher on both scales were regarded as suggestive of symptoms of anxiety and depression, respectively^24,25.

The self-report criterion was asked if the participants had ever been given a medical diagnosis of anxiety or depression.

We inquired about every medication that each participant took and categorized them using the Anatomical Therapeutic Chemical (ATC) code of the World Health Organization. We took into account drugs with codes N03 (antiepileptics), N05 (psycholéptics), N06 (psychoanaleptics) and N07 (other medications in the nervous system)²⁶.

Covariates

In the study, the assessment of multiple covariates was included. There were categories for men and women. 18–34 years, 35–59 years, and 60 years and older were the three age groups established. Black, brown, indigenous and yellow skin color were classified in same category. Marital status was different between married and single people. The educational attainment of participants was categorized based on the number of years of formal education they completed: 0–8 years, 9–11 years, and 12 or more years. The three categories of family income were 2 to 4 MW, 4 MW, and 2 minimum wage. Two categories for employment status were yes and no. There were two categories of smoking behavior. Drinking alcohol was categorized as either yes or no, according to the survey.

Sedentary and physical activity behaviors were among the movement behaviors evaluated for movement behaviors. The World Health Organization established guidelines for evaluating physical activity, which stated that a person was deemed physically inactive if they performed less than 150 min of moderate-intensity physical activity or less than 75 min of vigorous-intensity physical activity per week²⁷. The total amount of time spent sitting was used to classify sedentary behavior; values ≥ 9 h were considered elevated²⁸.

A self-reported medical diagnosis assessment was included in the study to evaluate comorbidities. The chronic diseases included cancer, heart disease, diabetes, asthma, lung disease, thyroid issues, hypertension, and chronic kidney disease. Participants were categorized as either having no comorbidities (no) or at least one comorbidity (yes) based on their self-reported medical history.

Statistical analysis

To adjust the natural weights of the design and address issues brought on by the lack of response or refusal to provide information, the sample weight of each chosen unit (census sector, household, and individual) was first determined independently for each city. To calibrate the natural expansion factors, new weights for every sample member must be estimated. The ratio of the number of census sectors in the sample that were chosen from the city “i” to the number of census sectors in the hole city “i” is denoted by “Ni” and represents the probability of selecting a census sector in each city in the sample. The formula dij/Dij, where “dij” is the number of sampled households in sector “j” of the city “i” and “Dij” is the number of households in sector “j” of the city “i,” was used to determine the probability that the household in census sector “j” of municipality “i” would be selected. The likelihood of each person living in the chosen household is calculated as 1/(the number of household members who are 18 years of age or older)²⁹.

Stata software version 15.1 (Stata Corporation, College Station, Texas) was used for the analyses, and the command “svy”—which takes into account a complex sample design—was employed. To evaluate the effects of the complex sampling design on our results, we calculated the design effect (DEFF) and coefficient of variation (CV) for key estimates. The DEFF, which compares the variance under the complex design to that expected under simple random sampling (SRS), remained consistently low, with values near or below 3. This indicates a minimal inflation of variance due to the design. Additionally, the CV, which measures relative variability, was generally below 10–15%, reflecting the robustness and precision of the estimates obtained. These metrics affirm the reliability of our findings while underscoring the representativeness and statistical rigor of the data analysis.

Percentages and confidence intervals (CI95%) are used to display the data. A significance level of 0.05 was chosen. Utilizing logistic regression analysis, the relationship between sleep quality and self-rated health was estimated. The multivariate model included additional variables as covariates, with sleep quality being the primary independent variable. We used the directed acyclic graph (DAG), a graphical tool that shows the causal relationships between the variables of interest, to choose the covariates (Fig. 1). Without confounding, collision, or collinearity biases, the DAG can be used to determine which variables should be changed to estimate the association of sleep quality with self-rated health. Any variables that were linked to self-rated health and sleep quality but weren’t mediators, colliders, or offspring of the exposure or result were considered potential confounding variables. The minimum set of confounding variables was chosen using the backdoor criterion to prevent erroneous associations and adjustments³⁰. The minimal and sufficient set of variables—age, sex, family income, use of alcohol and tobacco, and chronic illnesses—were added to the model. These factors may have varying effects on self-rated health and the quality of sleep. The theoretical model created for this study is shown visually in Supplementary Table 1, “Health-related pathways overview illustrated in the directed acyclic graph (DAG)”.

Furthermore, a mediation analysis was carried out using the Karlson Holm Breen (KHB) method. This approach, which is mediated by anxiety and depression, calculates the overall, direct, and indirect effects of sleep quality on self-rated health. It assesses the statistical significance of the mediation and breaks down the logistic regression coefficients into total, direct, and indirect effects³¹. We examined a few prerequisites, including the theoretical causal structure, the path analysis, and the interaction, before performing the mediation analysis. Based on theory and research, we established the causal structure, which shows that anxiety and depression have a direct and indirect impact on self-rated health due to poor sleep quality. To confirm whether there was a significant relationship between the variables involved in the mediation, we carried out a path analysis. To determine the influence of the mediators on the variation in the effect of sleep quality on self-rated health, we conducted an interaction analysis. In the logistic regression model, we accounted for interactions between anxiety and depressive symptoms and sleep quality. We predicted that the interaction would not be significant, which would mean that for all mediator levels, the effect of sleep quality on self-rated health would remain constant. This is exactly what we discovered (Supplementary Table 2).

link