Article type
Year
Abstract
Background: Factorial design randomized controlled trials (FDRCTs) serve well the purpose of answering several clinical questions through the conduct of a single trial. In fact, FDRCTs can be considered as a special case of multi-arm comparison trial. A little has been published regarding data extraction from these trials to be used in a meta-analysis (MA).
Objectives: To illustrate how the results of MA can change depending on the method of data extraction from FDRCTs.
Methods: We identified 66 FDRCTs in cancer after a MEDLINE search. Four studies met inclusion criteria as having survival as the primary outcome and availability of data for all comparisons including, importantly, reporting statistical interaction between treatments. Data were extracted using three different approaches: 1) Standard treatment was compared against each experimental treatment, which is the current common approach in data extraction from multi-arm trials (e.g. standard arm A vs experimental arm B, C or D). 2) All experimental treatments were pooled and compared against the standard arm (e.g. A vs B+C+D) and 3) pooling and comparison were performed according to the authors' hypothesis (e.g. pooled standard arm AC versus pooled experimental arm BD etc.).
Results: We compared the overall effects of experimental versus standard treatment according to the method of data extraction. When each experimental arm was compared individually against the standard arm for survival (approach 1), or pooled according to approach 2, the overall results favoured the experimental treatments (OR 0.71,95% CI [0.60,0.84], p=0.00009, OR 0.70,95% CI [0.54,0.90], p=0.005, respectively (Figure approach 1 and 2). However, the results did not favour either the experimental or standard treatment when the data were pooled according to the authors' hypothesis (OR 0.85, 95% CI [0.71, 1.02], p=0.07, Figure approach 3).
Conclusions: The results of a meta-analysis based on FDRCTs may significantly differ depending on the method of data extraction. Such a difference can be attributed to not taking into account the interaction between the regimens. The investigators should be vigilant and not pool the data when there is an interaction between the treatment arms, as incorrectly extracted data in MA can produce flawed results and lead to incorrect conclusions.
Objectives: To illustrate how the results of MA can change depending on the method of data extraction from FDRCTs.
Methods: We identified 66 FDRCTs in cancer after a MEDLINE search. Four studies met inclusion criteria as having survival as the primary outcome and availability of data for all comparisons including, importantly, reporting statistical interaction between treatments. Data were extracted using three different approaches: 1) Standard treatment was compared against each experimental treatment, which is the current common approach in data extraction from multi-arm trials (e.g. standard arm A vs experimental arm B, C or D). 2) All experimental treatments were pooled and compared against the standard arm (e.g. A vs B+C+D) and 3) pooling and comparison were performed according to the authors' hypothesis (e.g. pooled standard arm AC versus pooled experimental arm BD etc.).
Results: We compared the overall effects of experimental versus standard treatment according to the method of data extraction. When each experimental arm was compared individually against the standard arm for survival (approach 1), or pooled according to approach 2, the overall results favoured the experimental treatments (OR 0.71,95% CI [0.60,0.84], p=0.00009, OR 0.70,95% CI [0.54,0.90], p=0.005, respectively (Figure approach 1 and 2). However, the results did not favour either the experimental or standard treatment when the data were pooled according to the authors' hypothesis (OR 0.85, 95% CI [0.71, 1.02], p=0.07, Figure approach 3).
Conclusions: The results of a meta-analysis based on FDRCTs may significantly differ depending on the method of data extraction. Such a difference can be attributed to not taking into account the interaction between the regimens. The investigators should be vigilant and not pool the data when there is an interaction between the treatment arms, as incorrectly extracted data in MA can produce flawed results and lead to incorrect conclusions.