Understanding differences in results from literature-based and individual patient meta-analyses: an example from...

Meta-analyses are increasingly used to summarise observational data however a literature meta-analysis (LMA) may give different results to the corresponding individual patient meta-analysis (IPMA). This study compares the published results of equivalent LMAs and IPMAs, highlighting factors that can affect the results and therefore impact on clinical interpretation of meta-analyses.

Univariate results from published meta-analyses of prospective observational outcome data were compared, as were the number of studies, patients and length of follow-up. The absolute difference in survival was calculated. The association between severe diastolic dysfunction (RFP) and death post acute myocardial infarction (AMI) and in chronic heart failure (HF) were used as clinical examples.

The IPMA hazard ratio was lower that the LMA odds ratio: AMI hazard ratio 2.67 (95% confidence interval 2.23 to 3.20), odds ratio 4.10 (3.38 to 4.99); HF hazard ratio 2.42 (2.06 to 2.83), odds ratio 4.36 (3.60 to 5.04). The IPMAs contained most of the studies from the LMAs as well as additional unpublished data, and a longer length of follow-up was available in the IPMAs (AMI 3.7 vs 2.6 yr, HF 4.0 vs 1.5 yr). Restricting analysis to the same studies in both the LMA and IPMA resulted in a similar difference in effect sizes between methods to those found in the published analyses.

The result of a meta-analysis is affected by whether study level or individual patient data have been used, and the variant of analysis that is required. Awareness and consideration of these factors is important for clinical interpretation of meta-analyses.