Data-driven decision making for the screening of cognitive impairment in primary care: a machine learning approach using data from the ELSA-Brasil study.

The systematic assessment of cognitive performance of older people without cognitive complaints is controversial and unfeasible. Identifying individuals at higher risk of cognitive impairment could optimize resource allocation. We aimed to develop and test machine learning models to predict cognitive impairment using variables obtainable in primary care settings. In this cross-sectional study, we included 8,291 participants of the baseline assessment of the ELSA-Brasil study, who were aged between 50 and 74 years and were free of dementia. Cognitive performance was assessed with a neuropsychological battery and cognitive impairment was defined as global cognitive z-score below 2 standard deviations. Variables used as input to the prediction models included demographics, social determinants, clinical conditions, family history, lifestyle, and laboratory tests. We developed machine learning models using logistic regression, neural networks, and gradient boosted trees. Participants' mean age was 58.3±6.2 years, 55% were female. Cognitive impairment was present in 328 individuals (4%). Machine learning algorithms presented fair to good discrimination (areas under the ROC curve between 0.801 and 0.873). Extreme Gradient Boosting presented the highest discrimination, high specificity (97%), and negative predictive value (97%). Seventy-six percent of the individuals with cognitive impairment were included among the highest ranked individuals by this algorithm. In conclusion, we developed and tested a machine learning model to predict cognitive impairment based on primary care data that presented good discrimination and high specificity. These characteristics could support the detection of patients who would not benefit from cognitive assessment, facilitating the allocation of human and economic resources.

Data-driven decision making for the screening of cognitive impairment in primary care: a machine learning approach using data from the ELSA-Brasil study

The systematic assessment of cognitive performance of older people without cognitive complaints is controversial and unfeasible. Identifying individuals at higher risk of cognitive impairment could optimize resource allocation. We aimed to develop and test machine learning models to predict cognitive impairment using variables obtainable in primary care settings. In this cross-sectional study, we included 8,291 participants of the baseline assessment of the ELSA-Brasil study, who were aged between 50 and 74 years and were free of dementia. Cognitive performance was assessed with a neuropsychological battery and cognitive impairment was defined as global cognitive z-score below 2 standard deviations. Variables used as input to the prediction models included demographics, social determinants, clinical conditions, family history, lifestyle, and laboratory tests. We developed machine learning models using logistic regression, neural networks, and gradient boosted trees. Participants' mean age was 58.3±6.2 years, 55% were female. Cognitive impairment was present in 328 individuals (4%). Machine learning algorithms presented fair to good discrimination (areas under the ROC curve between 0.801 and 0.873). Extreme Gradient Boosting presented the highest discrimination, high specificity (97%), and negative predictive value (97%). Seventy-six percent of the individuals with cognitive impairment were included among the highest ranked individuals by this algorithm. In conclusion, we developed and tested a machine learning model to predict cognitive impairment based on primary care data that presented good discrimination and high specificity. These characteristics could support the detection of patients who would not benefit from cognitive assessment, facilitating the allocation of human and economic resources.

Machine learning for predicting chronic diseases: a systematic review.

OBJECTIVES: We aimed to review the literature regarding the use of machine learning to predict chronic diseases. STUDY DESIGN: This was a systematic review. METHODS: The searches included five databases. We included studies that evaluated the prediction of chronic diseases using machine learning models and reported the area under the receiver operating characteristic curve values. The Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis scale was used to assess the quality of studies. RESULTS: In total, 42 studies were selected. The best reported area under the receiver operating characteristic curve value was 1, whereas the worst was 0.74. K-nearest neighbors, Naive Bayes, deep neural networks, and random forest were the machine learning models most frequently used for achieving the best performance. CONCLUSION: We found that machine learning can predict the occurrence of individual chronic diseases, progression, and their determinants and in many contexts. The findings are original and relevant to improve clinical decisions and the organization of health care facilities.

Income inequality and self-reported health in a representative sample of 27 017 residents of state capitals of Brazil.

Background: The association between income inequality and health has been analyzed predominantly in developed countries with modest levels of inequality. The study aimed to analyze the association between income inequality and self-reported health (SRH) in the adult population of the 27 Brazilian capitals. Methods: Individuals aged 18 years or older from the National Health survey residing in Brazilian capitals in 2013 were analyzed (n = 27 017). Bayesian multilevel models were applied after controlling for individual factors and area-level socioeconomic characteristics. Results: We found a significant association between income inequality and SRH, even after controlling for individual and contextual factors. The results indicate greater odds of poor SRH among those living in areas with medium (OR = 1.31, 95% CI: 1.17-1.47) and high income inequality level (OR = 1.39, 95% CI: 1.24-1.56). Income inequality remained significantly associated with SRH, even after controlling for other contextual socioeconomic characteristics, such as local illiteracy rate, violence and per capita income. Conclusions: The study highlights the importance of the individual and contextual characteristics associated with SRH. Our findings suggest that city-level income inequality can have a detrimental effect on individual health, over and above other contextual socioeconomic characteristics and individual factors.

Patterns and predictors of health service use among people with mental disorders in São Paulo metropolitan area, Brazil.

AIMS: Important transformations in psychiatric healthcare (HC) delivery have been implemented in Latin America during the beginning of 21st century. However, information on current service uses patterns is scant, obstructing the estimates and proper planning of service needs for general population. The current investigation aims to describe patterns and estimates predictors of 12-month HC use by individuals with mental disorders in São Paulo metropolitan area, Brazil. METHOD: Data are from São Paulo Mental Health Survey, a cross-sectional multistage representative study. Participants were face-to-face interviewed in their household, using a structured diagnostic interview, the World Mental Health Survey Initiative version of the Composite International Diagnostic Interview. A total of 5037 respondents, non-institutionalised, aged 18 years and older were interviewed. The response rate was 81.3%. We determined the percentages of individuals with 12-month DSM-IV anxiety, mood and substance disorders that received treatment in the 12 months prior to assessment in main service sectors (specialty mental health, general medicine, human services (HS), and complementary and alternative medicine). The number of visits and percentage of individuals who received treatment at minimally adequacy also was estimated. Multilevel regression controlled contextual variables that influenced the use of service and treatment adequacy. RESULTS: Only 10.1% of respondents used some HC service in the 12 months prior to assessment for their psychiatric problems, including 3.9% of them being treated either by a psychiatrist, 3.5% by a non-psychiatrist mental health specialist, 3.3% by a general medical (GM) provider, 1.5% by a HS provider and 1.4% by a complementary and alternative medical provider. In general, those participants who received service in the mental health specialty sector reported more visits than those in the GM sector (median 3.9 v. 1.5 visits). The cases seen in specialty sector outnumber those visiting GM treatment in terms of minimally adequate treatment (54.6 v. 23.2%). The likelihood of receiving treatment was significantly greater among individuals diagnosed with any anxiety and mood disorder, presenting more severe disorders, and with possession of HC insurance. CONCLUSIONS: The great majority of individuals with an active mental disorder in São Paulo were either untreated or insufficiently treated. Awareness and training programmes to GM professionals are advocated to improve recognition, care take and referral to specialty care when needed. Proper integration among HC sectors is recommended.

Latin American Study of Nutrition and Health (ELANS): rationale and study design.

BACKGROUND: Obesity is growing at an alarming rate in Latin America. Lifestyle behaviours such as physical activity and dietary intake have been largely associated with obesity in many countries; however studies that combine nutrition and physical activity assessment in representative samples of Latin American countries are lacking. The aim of this study is to present the design rationale of the Latin American Study of Nutrition and Health/Estudio Latinoamericano de Nutrición y Salud (ELANS) with a particular focus on its quality control procedures and recruitment processes. METHODS/DESIGN: The ELANS is a multicenter cross-sectional nutrition and health surveillance study of a nationally representative sample of urban populations from eight Latin American countries (Argentina, Brazil, Chile, Colombia, Costa Rica, Ecuador, Perú and Venezuela). A standard study protocol was designed to evaluate the nutritional intakes, physical activity levels, and anthropometric measurements of 9000 enrolled participants. The study was based on a complex, multistage sample design and the sample was stratified by gender, age (15 to 65 years old) and socioeconomic level. A small-scale pilot study was performed in each country to test the procedures and tools. DISCUSSION: This study will provide valuable information and a unique dataset regarding Latin America that will enable cross-country comparisons of nutritional statuses that focus on energy and macro- and micronutrient intakes, food patterns, and energy expenditure. TRIAL REGISTRATION: Clinical Trials NCT02226627.