Extrapolação de dados com curvas de sobrevida: uma proposta metodológica para dados agregados
Extrapolação de dados com curvas de sobrevida: uma proposta metodológica para dados agregados
Data
2021
Autores
Cordeiro, Solange Borges da Silva
Journal Title
Journal ISSN
Volume Title
Publisher
Instituto Nacional de Cardiologia
Resumo
Introdução: Muitas pesquisas clínicas possuem curto tempo de seguimento dos participantes, tornando-se necessária a adoção de métodos de extrapolação de dados nos modelos de avaliação econômica. Apesar disso, as propostas metodológicas disponíveis trabalham com os dados em sua forma individualizada, cenário nem sempre factível ao cotidiano da Avaliação de Tecnologias em Saúde (ATS). Assim, o objetivo deste estudo foi reproduzir um método para extrapolação com curvas de sobrevida aplicável a dados agregados, que são dados de sobrevida que não estão disponíveis a nível de indivíduos (Individual Patiente data – IPD). Métodos: Inicialmente, revisou-se as diretrizes e recomendações sobre extrapolação de dados e em seguida, as aplicaram em um estudo de caso. A revisão rápida consistiu em busca no Medline (via Pubmed), nas 10 principais agências de ATS e comunicações enviadas a rede internacional de agências de ATS, The International Network of Agencies for Health Technology Assessment (INAHTA). Foram incluídas diretrizes e demais documentos com recomendações explícitas relacionadas ao tema da extrapolação de dados de sobrevida na avaliação econômica em saúde, sem restrição de período, nos idiomas inglês, espanhol e português. Os dados foram extraídos em formulário padrão e realizada síntese qualitativa. O estudo de caso consistiu nas aplicações de modelos não paramétricos e paramétricos de análise de sobrevivência em duas terapias (anastrozol e fulvestranto) no contexto do câncer de mama metastático com dados de 10 anos de seguimento nos desfechos de sobrevida livre de progressão (SLP) e sobrevida global (SG) para dois braços de tratamento. Para o propósito do trabalho, buscou-se a aplicação do ajuste de curvas de sobrevida com o método de mínimos quadrados não lineares aos dados agregados. Os autores do estudo primário foram contactados sobre a possibilidade de compartilhamento de dados individualizados para a eventual comparação e validação do método. Resultados: Foram encontrados 306 documentos, três foram incluídos no processo de seleção. As principais distribuições recomendadas e elegidas para o estudo de caso foram: Exponencial, Weibull, Log-normal, Log-logística, Gama generalizada e Gompertz. Na verificação da aplicação no estudo de caso nos dados dos dois braços de tratamentos, as distribuições Weibull e gama generalizada foram as que melhor se ajustaram aos dados de SG, seja na abordagem individualizada quanto na agregada, segundo critérios estatísticos e de inspeção visual. Para a SLP foram as curvas log-logística e log-normal para o Anastrozol. No caso do tratamento com fulvestranto, foram as curvas log-normal e gama generalizada para os dados individualizados e Gompertz e gama generalizada nos dados agregados. Conclusão: Foi possível mostrar a factibilidade do ajuste de dados com curvas de sobrevida a partir de dados agregados com o objetivo de extrapolação de dados em modelos econômicos. Apesar das divergências, ao se compararem os modelos sugeridos por ambos os métodos, observam-se que seria pouco provável a existência de impacto relevantes para a tomada de decisão. Tal método pode ser útil nas situações de impossibilidade de acesso aos dados individualizados.
Introduction: Many clinical researches have a short follow-up time of the participants, making it necessary to adopt methods for extrapolating data in the economic evaluation models. Despite this, the methodological approaches available only handle with individualized patient data, a scenario not always feasible in Health Technology Assessment (HTA). Thus, the objective of this study was to reproduce a method for extrapolation with survival curves applicable to aggregated data, which are survival data that are not available at the individual level (Individual Patiente data - IPD). Methods: Initially, the guidelines and recommendations on data extrapolation were reviewed and then these recommendations were applied in a case study. The rapid review consisted of searching Medline (via Pubmed), the main HTA agencies and the international HTA network members (INAHTA). Methodological guidelines and other documents with explicit recommendations related to the topic of extrapolation of survival data in health economic evaluation were included, without publication date restriction, in English, Spanish and Portuguese languages. The data were extracted in a standard form and a qualitative synthesis was carried out. The case study consisted of the application of non-parametric and parametric models of survival analysis in two therapies (anastrozole and fulvestrant) in the context of metastatic breast cancer with 10-year follow-up data on progression-free survival (SLP) and overall survival (SG) outcomes. For this purpose, we have applied the adjustment of survival curves to the aggregated data with nonlinear least squares methods. The authors of the primary study were contacted about the possibility of sharing individual data for comparison and validation of the method. Results: 306 documents were retrieved, of which only three were included at the end of the selection process. The main distributions recommended and chosen for the case study were: Exponential, Weibull, Log-normal, Log-logistic, Generalized gamma and Gompertz. In verifying the application in the case study in the data of the two treatment arms, the Weibull and generalized gamma distributions were the ones that best fit the SG data, both in the individualized and in the aggregate approach, according to statistical and visual inspection criteria. For SLP were the log-logistic and log-normal curves for Anastrozole. In the case of treatment with fulvestrant, it was the log-normal and generalized gamma curves for the individualized data and Gompertz and generalized gamma for the aggregated data. Conclusion: It was possible to show the feasibility of adjusting data with survival curves from aggregated data in order to extrapolate data in economic models. Despite some divergences, when comparing the models suggested by both methods, it was observed that the existence of relevant impacts for decision-making would be unlikely. Such a method can be useful in situations where it is impossible to access individualized data.
Introduction: Many clinical researches have a short follow-up time of the participants, making it necessary to adopt methods for extrapolating data in the economic evaluation models. Despite this, the methodological approaches available only handle with individualized patient data, a scenario not always feasible in Health Technology Assessment (HTA). Thus, the objective of this study was to reproduce a method for extrapolation with survival curves applicable to aggregated data, which are survival data that are not available at the individual level (Individual Patiente data - IPD). Methods: Initially, the guidelines and recommendations on data extrapolation were reviewed and then these recommendations were applied in a case study. The rapid review consisted of searching Medline (via Pubmed), the main HTA agencies and the international HTA network members (INAHTA). Methodological guidelines and other documents with explicit recommendations related to the topic of extrapolation of survival data in health economic evaluation were included, without publication date restriction, in English, Spanish and Portuguese languages. The data were extracted in a standard form and a qualitative synthesis was carried out. The case study consisted of the application of non-parametric and parametric models of survival analysis in two therapies (anastrozole and fulvestrant) in the context of metastatic breast cancer with 10-year follow-up data on progression-free survival (SLP) and overall survival (SG) outcomes. For this purpose, we have applied the adjustment of survival curves to the aggregated data with nonlinear least squares methods. The authors of the primary study were contacted about the possibility of sharing individual data for comparison and validation of the method. Results: 306 documents were retrieved, of which only three were included at the end of the selection process. The main distributions recommended and chosen for the case study were: Exponential, Weibull, Log-normal, Log-logistic, Generalized gamma and Gompertz. In verifying the application in the case study in the data of the two treatment arms, the Weibull and generalized gamma distributions were the ones that best fit the SG data, both in the individualized and in the aggregate approach, according to statistical and visual inspection criteria. For SLP were the log-logistic and log-normal curves for Anastrozole. In the case of treatment with fulvestrant, it was the log-normal and generalized gamma curves for the individualized data and Gompertz and generalized gamma for the aggregated data. Conclusion: It was possible to show the feasibility of adjusting data with survival curves from aggregated data in order to extrapolate data in economic models. Despite some divergences, when comparing the models suggested by both methods, it was observed that the existence of relevant impacts for decision-making would be unlikely. Such a method can be useful in situations where it is impossible to access individualized data.
Description
Palavras-chave
Medical Oncology, Guidelines as Topic, Technology Assessment, Biomedical, Cost-Effectiveness Evaluation, Survival Analysis, Oncologia, Guias como Assunto, Avaliação da Tecnologia Biomédica, Avaliação de Custo-Efetividade, Análise de Sobrevida
Citação
Cordeiro SBS. Extrapolação de dados com curvas de sobrevida: uma proposta metodológica para dados agregados. Dissertação [Mestrado Profissional em Avaliação de Tecnologias em Saúde]. Instituto Nacional de Cardiologia; 2021.