Article type
Year
Abstract
Background: Systematic review (SR) of randomized controlled trials (RCT) is reckoned the gold standard for informing treatment choice. Decision analyses (DA) inform health care policy decisions in absence of RCTs or SR of RCTs. However, oversimplification of real world scenarios in DA can be problematic. We have previously shown that DA disagrees with RCTs in 50% of cases. However, it is not known how often the results of DA and SR of RCTs either agree or disagree.
Objective: To compare the conclusions of DA and matching SR of RCTs.
Methods: We searched PubMed up to 2008 for DAs comparing at least two interventions followed by SRs that matched the DAs based on patient population, intervention, control, and outcome criteria (PICO). From each DA and SR, we extracted data on PICO, conclusion, and impact of sensitivity analyses on the conclusion. Agreement between DA and SR was based on matching of respective conclusions.
Results: From 42,704 retrieved DA citations, we found matching SR for 38 comparisons (Figure1). We found 74% (28/38) agreement between the conclusions of the DAs and the SRs. The sensitivity analyses conducted in either DA or SR did not impact the agreement. Two DA design characteristics were significantly associated with agreement: use of single versus multiple data source (p = 0.048) and use of meta-analysis data (p = 0.040).
Conclusions: This first study quantifying the correlation between the results of DA and SR of RCTs suggests a high level of agreement. Use of meta-analysis data and use of multiple sources of data appear to impact the agreement of conclusions between DA and SR of RCTs.
Objective: To compare the conclusions of DA and matching SR of RCTs.
Methods: We searched PubMed up to 2008 for DAs comparing at least two interventions followed by SRs that matched the DAs based on patient population, intervention, control, and outcome criteria (PICO). From each DA and SR, we extracted data on PICO, conclusion, and impact of sensitivity analyses on the conclusion. Agreement between DA and SR was based on matching of respective conclusions.
Results: From 42,704 retrieved DA citations, we found matching SR for 38 comparisons (Figure1). We found 74% (28/38) agreement between the conclusions of the DAs and the SRs. The sensitivity analyses conducted in either DA or SR did not impact the agreement. Two DA design characteristics were significantly associated with agreement: use of single versus multiple data source (p = 0.048) and use of meta-analysis data (p = 0.040).
Conclusions: This first study quantifying the correlation between the results of DA and SR of RCTs suggests a high level of agreement. Use of meta-analysis data and use of multiple sources of data appear to impact the agreement of conclusions between DA and SR of RCTs.
Images
