P343: Meta-analysis of Observational Studies in Rare Diseases
Poster Presenter
Li Huang
Principal Biostatistician
Phastar United States
Objectives
This study aims to apply random-effects and fixed-effects meta-analysis to estimate incidence rates, exposure-adjusted incidence rates (EAER), and their confidence intervals, depending on different levels of heterogeneity among observational studies in rare diseases.
Method
We compared fixed-effects and random-effects models meta-analysis for estimating incidence rates, EAER, and confidence intervals using the HKSJ adjustment. Real-world data and simulations assessed heterogeneity and cohort size impact in only two-study scenarios.
Results
Our findings highlight the critical role of heterogeneity assessment in selecting an appropriate meta-analysis model. The combined incidence rates were higher, and confidence intervals were narrower under the fixed-effects model compared to the random-effects model, as the fixed-effects model accounts only for within-study variance and assumes low heterogeneity (I² < 50%). In contrast, the random-effects model incorporates both within-study and between-study variance, making it more appropriate for high heterogeneity (I² > 50%).
For recurrent adverse events with a constant rate over time, EAER should be used instead of incidence rate. In the special setting of only two studies which could be presented in real-world studies, a simulation study was conducted. The study sample sizes A and B were set 1000:1000, 25:25, 1000:200, 125:25 and I2 were set to 95%, 75%, 50%, 25% 10% indicated from high to low heterogeneity among two studies, which leads to 20 different combinations in 4 scenarios. In scenario 1 (similar and bigger sample size) and scenario 3(bigger and unequal sample size), CI are quite different between Fixed effect model and random effected model regardless the I^2values and HKSJ Pooled CI is narrower compare to DL pooled; the combined EAER are similar between fixed and random effect model as the I^2 values <50% but different as the I^2 values >50%
In scenario 2 (similar and small sample size) and scenario 4 (small and unequal sample size), CI are quite different between Fixed effect model and random effected model regardless the I^2values and HKSJ Pooled CI is longer compare to DL pooled; the combined EAER showed quite different between fixed and random effect model as the I^2 values >50% but little different as the I^2 values <50%
Conclusion
Meta-analysis is a crucial tool for systematically reviewing observational studies, particularly in rare diseases. Selecting an appropriate model is essential, with I² commonly used to assess heterogeneity. When heterogeneity is low (I² = 50%), a fixed-effects model is preferred, whereas a random-effects model is more appropriate when heterogeneity is moderate to high (I² > 50%). For I² values greater than 75%, indicating high heterogeneity, random-effects meta-analysis provides a better pooled estimate. The application of the HKSJ method improves confidence interval coverage, especially in meta-analyses with few studies and small sample sizes, compared to the DL method. This approach enables the effective application of meta-analysis in observational studies, even in cases with extremely small study sizes.
