Análise do consumo alimentar e a situação de insegurança alimentar de mulheres adultas usuárias do SUS de João Pessoa, Paraíba

Abstract

Understanding eating behavior is essential to direct effective nutritional interventions, since it reflects the population's eating choices and practices, influenced by changes over the years. These changes contribute to the nutritional transition that affects eating patterns. However, socioeconomic factors directly impact these patterns, among low-income women, who face greater social vulnerability and difficulties in accessing quality food. Food insecurity, marked by concern, uncertainty, and deprivation of adequate food, affects millions of Brazilians, being more intense among families headed by women. In 2022, approximately 125.2 million people faced some degree of this problem, highlighting the need for public policies aimed at reducing inequalities and ensuring food security. The objective of this study was to analyze food consumption and the food insecurity situation of adult women users of the SUS in João Pessoa. This was a cross-sectional study nested in a prospective cohort, with 268 women aged 18 to 59 years recruited from 10 Basic Family Health Units for the analysis of food consumption, and 262 women for the assessment of food insecurity. Face-to-face interviews were conducted to characterize the sample, assess nutritional status, assess food consumption from 24-hour dietary recalls, assess food insecurity, assess state-trait anxiety inventory and the WHO quality of life questionnaire. Dietary patterns were identified by exploratory factor analysis, using principal components extraction, considering factors with eigenvalue greater than 1 and varimax rotation. Adherence to these patterns was assessed by means of comparison tests (T-Student and ANOVA), while the association with independent variables was analyzed by multiple logistic regression, with each factor dichotomized as a dependent variable. In addition, 14 machine learning models were applied to classify and identify associated factors. The best set of hyperparameters was selected based on the area under the ROC curve, in addition to the accuracy metrics, F-measure, precision, recall, specificity and Kappa coefficient. Four dietary patterns were identified: “Unhealthy”, “Mixed 1”, “Mixed 2” and “Traditional”, which explain 36% of the total variance of the sample. Being pregnant was associated with greater adherence to both the “Unhealthy” pattern (OR: 7.62; 95% CI: 3.59–16.19) and the “Traditional” pattern (OR: 6.00; 95% CI: 2.82–12.78). Being over 35 years old was associated with the “Traditional” pattern (OR: 2.10; 95%CI: 1.06–4.18) and low education level with the “Unhealthy” pattern (OR: 3.10; 95%CI: 1.18–8.12). The RDA model demonstrated good performance in classifying food insecurity, reaching 81% accuracy, that is, correctly classifying eight out of every 10 women. Its accuracy was 73%, indicating that, for every 10 women predicted to be food insecure, seven were correctly classified. Although income did not stand out directly among the most important variables, receipt of government benefits showed a direct relationship in the importance of the variables. Even so, the tendency to choose unhealthy foods among women of lower socioeconomic status should be considered in health strategies that promote nutritional education and access to healthy foods. In addition, greater adherence to the traditional pattern among older women suggests that the worsening of diet quality may be a recent phenomenon. Machine learning models demonstrated the ability to identify vulnerable populations, contributing to the promotion and prevention of food insecurity in an agile and effective manner. Among them, RDA presented the best performance when including sociodemographic and mental health factors.