Uso do escore de propensão na avaliação da efetividade de dois programas escolares de prevenção ao uso de drogas
Data
2023-05-18
Autores
Souza, Felipe Granado de [UNIFESP]
Orientadores
Andreoni, Solange [UNIFESP]
Tipo
Tese de doutorado
Título da Revista
ISSN da Revista
Título de Volume
Resumo
Introdução: Como a unidade de randomização em um ensaio controlado randomizado por conglomerado não é o indivíduo, mas sim a unidade social à qual ele pertence, pode haver um desequilíbrio nas características dos níveis individuais entre os grupos de tratamento na avaliação basal. Este trabalho buscou realizar duas análises secundárias para verificar a efetividade de dois programas de prevenção ao uso de drogas, o PROERD/“Caindo na Real” usando o propensity score matching em dois ensaios controlados randomizados por conglomerados (5º e 7º ano) e o programa #Tamojunto2.0 utilizando o propensity score weighting no contexto de amostras complexas. Métodos: Os desfechos de interesse no programa PREORD foram o Primeiro uso de drogas na vida e Uso recente de drogas, registrados na avaliação pós-intervenção. Já no programa #Tamojunto2.0 foram a Incidência de iniciação ao uso de drogas e a Prevalência do uso no último mês, registrado na avaliação pós-intervenção. A imputação múltipla foi usada para lidar com dados ausentes. O propensity score matching foi empregado para reduzir o desequilíbrio da linha de base entre os grupos e, em seguida, equações de estimativa generalizada foram usadas para avaliar o efeito do programa PROERD para diferentes drogas no banco de dados imputado e nos casos completos. Já no programa #Tamojunto2.0 foi usado o método propensity score weighting para reduzir o desequilíbrio da linha de base entre os grupos e, para analisar seu efeito nos diferentes tipos de drogas, foi feita a regressão logística considerando a amostra complexa. Resultados: Para o programa PROERD o propensity score matching tornou os grupos equilibrados em suas características basais, uma vez que foi detectada diferenças entre os grupos para algumas covariáveis. Na análise de efeito, não foram encontradas evidências da efetividade do programa PROERD para o 5º e 7º anos. No programa #Tamojunto2.0 os resultados mostraram que os grupos ficaram balanceados em suas características de linha de base após a aplicação do propensity score weighting e, na análise do efeito do programa, os resultados indicaram menores chances para a iniciação ao uso de álcool (OR=0,75, IC95%[0,63; 0,90]) e tabaco (OR=0,81, IC95%[0,66, 0,98]), em relação ao grupo controle. Conclusão: A utilização do propensity score matching nas análises dos ensaios controlados randomizados por conglomerados apresentou resultado satisfatório no equilíbrio das características basais da amostra do estudo. A aplicabilidade do PROERD a alunos de uma cultura diferente daquelas para as quais o programa foi criado ainda é incerta, pois os resultados não suportam sua divulgação independentemente do paradigma de análise utilizado. Já o programa #Tamojunto2.0 se apresentou como um possível programa de prevenção ao uso de drogas para se tornar uma política pública nacional. Adicionalmente, o uso do propensity score weighting mostrou ser um método satisfatório para equilibrar as características de linha de base neste ensaio controlado randomizado por conglomerado, no contexto de amostras complexas.
Introduction: As the unit of randomization in a cluster randomized controlled trial is not the individual, but rather the social unit to which they belong, there may be an imbalance in individual-level characteristics between treatment groups at baseline. This research sought to perform two secondary analyzes to verify the effectiveness of two drug use prevention programs, PROERD/“Caindo na Real” using propensity score matching in two cluster-randomized controlled trials (5th and 7th grades) and the #Tamojunto2.0 program using propensity score weighting in the context of complex samples. Methods: The outcomes of interest in the PREORD program were First drug use and Recent drug use, recorded in the post-intervention assessment. In the #Tamojunto2.0 program, they were the Incidence of drug use initiation and the Prevalence of use in the last month, recorded in the post-intervention evaluation. Multiple imputation was used to deal with missing data. Propensity score matching was employed to reduce baseline imbalance between groups, and then generalized estimation equations were used to assess the effect of the PROERD program for different drugs on the imputed database and in the complete cases. In the #Tamojunto2.0 program, the propensity score weighting method was used to reduce the baseline imbalance between groups, and to analyze its effect on different types of drugs, logistic regression was performed considering the complex sample. Results: For the PROERD program, propensity score matching made the groups balanced in their baseline characteristics, since differences between groups were detected for some covariates. In the effect analysis, a lack of evidence of the effectiveness of the program was found for both years. In the #Tamojunto2.0 program, the results showed that the groups were balanced in their baseline characteristics after the application of the propensity score weighting, and in the analysis of the effect of the program, the results were positive for the initiation of alcohol use (OR = 0.75, 95% CI [0.63; 0.90]) and tobacco (OR = 0.81, 95% CI [0.66, 0.98]). Conclusion: The use of propensity score matching in the analysis of cluster randomized controlled trials showed a satisfactory result in balancing the baseline characteristics of the study sample. The applicability of PROERD to students from a different culture than those for which the program was created is still uncertain, as the results do not support its dissemination regardless of the analysis paradigm used. The #Tamojunto2.0 program was presented as a possible drug use prevention program to become a national public policy. Additionally, the use of propensity score weighting proved to be a satisfactory method for balancing baseline characteristics in this cluster-randomized controlled trial in the context of complex samples.
Introduction: As the unit of randomization in a cluster randomized controlled trial is not the individual, but rather the social unit to which they belong, there may be an imbalance in individual-level characteristics between treatment groups at baseline. This research sought to perform two secondary analyzes to verify the effectiveness of two drug use prevention programs, PROERD/“Caindo na Real” using propensity score matching in two cluster-randomized controlled trials (5th and 7th grades) and the #Tamojunto2.0 program using propensity score weighting in the context of complex samples. Methods: The outcomes of interest in the PREORD program were First drug use and Recent drug use, recorded in the post-intervention assessment. In the #Tamojunto2.0 program, they were the Incidence of drug use initiation and the Prevalence of use in the last month, recorded in the post-intervention evaluation. Multiple imputation was used to deal with missing data. Propensity score matching was employed to reduce baseline imbalance between groups, and then generalized estimation equations were used to assess the effect of the PROERD program for different drugs on the imputed database and in the complete cases. In the #Tamojunto2.0 program, the propensity score weighting method was used to reduce the baseline imbalance between groups, and to analyze its effect on different types of drugs, logistic regression was performed considering the complex sample. Results: For the PROERD program, propensity score matching made the groups balanced in their baseline characteristics, since differences between groups were detected for some covariates. In the effect analysis, a lack of evidence of the effectiveness of the program was found for both years. In the #Tamojunto2.0 program, the results showed that the groups were balanced in their baseline characteristics after the application of the propensity score weighting, and in the analysis of the effect of the program, the results were positive for the initiation of alcohol use (OR = 0.75, 95% CI [0.63; 0.90]) and tobacco (OR = 0.81, 95% CI [0.66, 0.98]). Conclusion: The use of propensity score matching in the analysis of cluster randomized controlled trials showed a satisfactory result in balancing the baseline characteristics of the study sample. The applicability of PROERD to students from a different culture than those for which the program was created is still uncertain, as the results do not support its dissemination regardless of the analysis paradigm used. The #Tamojunto2.0 program was presented as a possible drug use prevention program to become a national public policy. Additionally, the use of propensity score weighting proved to be a satisfactory method for balancing baseline characteristics in this cluster-randomized controlled trial in the context of complex samples.