“Network Meta-Analysis using Bayesian Methododoly : an Application for the Treatment of Multiple Myeloma” e “Alguns Tópicos sobre Métodos de Monte Carlo via Cadeias de Markov com Saltos Reversíveis”

 

  • Drª Susana Esteves – IPO Lisboa e Drª Raquel Nicolette – Universidade de Aveiro
  • FCUL (DEIO) – Campo Grande – Bloco C6 Piso 4 – Sala 6.4.30 – 15h
  • Quarta-feira, 26 de Maio de 2010
 Abstract

1)

The recent introduction of the novel drugs thalidomid, bortezomib and lenalidomide changed the treatment paradigm for multiple myeloma (MM).The combination regimen melphalan-prednisone-thalidomid (MPT) is the current standard treatment in newly diagnosed elderly patients who are ineligible for hematopoietic stem cell transplantation. In recent studies, other combination regimens have demonstrated substantial activity in this clinical setting such as:melphalan-prednisone-bortezomib (MPV), melphalan-prednisone-bortezomib-thalidomide (MPV-T), thalidomide-dexamethasone (TD) and lenalidomide-dexamethasone (RD). However no data is currently available from randomized trials that directly compare these treatment regimens with MPT. To overcome this problem we used statistical methods that combine direct and indirect evidence in a single analysis while fully respecting randomization. We aimed to perform a systematic review and a network meta-analysis to compare new treatment regimens using objective response as an efficacy endpoint.

Methods:

Relevant randomized controlled trials (RCTs) were identified using PubMed and reviewing conference abstracts. Studies were subsequently selected according to pre-established criteria. Formal synthesis of evidence was made on a log-odds scale using a Bayesian hierarchical random-effects model. The analysis was done in WinBUGS using the Gibbs sampling method.

Results:

Fifteen RCTs were included in the analysis, involving a total of 4857 MM patients. There was a trend to improved odds of response to treatment with MPV (OR 1.41, 95%HPD 0.36-2.58), MPV-T (OR 2.28, 95%HPD 0.27-5.01) and RD (OR 1.64, 95%HPD 0.17-3.91) in comparison with MPT, although not statistically significant. The odds of response to TD was lower than MPT (OR 0.57, 95% HPD, 0.2-1.0). The regimens with higher absolute probability of response were MPV-T (0.81, 95%HPD 0.63-0.95), MPV (0.75, 95%HPD 0.60-0.89) and RD (0.74, 95%HPD 0.51-0.94).

These results suggest that RD, but not TD, can be a good alternative to MPT particularly in cases where it might be desirable to have an orally administered treatment regimen. MPV also showed promising results making it a good alternative to MPT in particular in patients with high-risk of thromboembolism. Nevertheless a more comprehensive analysis considering both efficacy and safety endpoints is required and is currently underway. In addition, these results should be interpreted with caution given the limited data available and the subsequent uncertainty evidenced by the wide range of some HPD intervals and by the variability of the probability of response found in the sensitivity analysis.

Conclusion:

Healthcare decisions often involve choosing from a selection of treatment options. The methodology used in this analysis allowed simultaneous comparison of multiple treatment options through assessment of all direct and indirect evidence considered relevant in this clinical setting. By allowing for comparisons that have never been (and will probably never be) tested in RCTs, this approach reveals a great potential for use in clinical decision making and in the design of future studies.

Abstract

2)

Nos últimos anos as mais diversas áreas de modelação estatística obtiveram um considerável desenvolvimento devido à possibilidade do uso exaustivo de métodos computacionais. Um dos métodos que merece destaque é o Markov Chain Monte Carlo (MCMC), contudo apresentam a limitação de não permitir saltos de dimensão entre as amostras dentro de uma cadeia de Markov. Os métodos Reversible jump Markov chain Monte Carlo (RJMCMC) fornecem uma ferramenta para simulação MCMC em situações onde a dimensão do espaço do parâmetro pode variar entre as iterações da cadeia. Para a construção deste amostrador, Peter Green (1995) definiu um conjunto de movimentos, conhecidos como saltos reversíveis, criando um esquema onde qualquer movimento que pode alterar a dimensão da cadeia deve ser reversível, ou seja, deve ser possível voltar ao estado anterior, em um movimento posterior. Uma das possíveis aplicações dos métodos MJMCMC é na área de selecção de modelos. Neste seminário tem-se por objectivo apresentar uma revisão sobre este amostrador e alguns exemplos da sua aplicação.