Title: An improved Henmi-Copas confidence interval for random effects meta-analysis with small number of studies
Authors: Masayuki Henmi - The Institute of Statistical Mathematics (Japan) [presenting]
Satoshi Hattori - Osaka University (Japan)
Tim Friede - University Medical Center Goettingen (Germany)
Abstract: The DerSimonian Laird condence interval for the average treatment effect in meta-analysis is widely used in practice when there is heterogeneity between studies. However, it is well known that its coverage probability can be substantially below the target level of 95 per cent. It can also be very sensitive to publication bias. For solving this problem, a confidence interval (HC interval) has been proposed by using the fixed effects estimate as the center of the interval in the random effects setting. Although the HC interval has better coverage probability than the DerSimonian-Laird confidence interval and is less sensitive to publication bias, its coverage probability is still below the target level especially when the number of studies included in meta-analysis is small. This is because the HC interval uses the DerSimonian-Laird estimate of the between-study variance in its construction and this estimate is less accurate as the number of studies gets smaller. We propose an improved version of the HC interval by replacing the DerSimonian-Laird estimate with a more accurate estimate. Its performance is examined by simulation studies and we apply it to a real-data example for illustration.