Monte Carlo Simulation for Area Under the Curve Estimation from Digitized Meta-Analysis Data

Meta-analyses frequently require estimating area under the curve (AUC) from digitized figures when raw participant data is unavailable. We systematically compare two approaches: standard trapezoidal integration with error propagation versus a Monte Carlo method that samples synthetic response curves and integrates over posterior distributions. Using 3,920 synthetic datasets from seven biological response functions (skewed Gaussian, biexponential, gamma, log-normal, Weibull, Bateman, and inverse Gaussian), we varied participant numbers (5-40) and timepoints (4-10) while normalizing all curves to AUC=1.0. The standard method consistently underestimated true AUC, especially for skewed or long-tailed functions, while Monte Carlo produced near-unbiased estimates with superior accuracy across all conditions. Performance improvements were most pronounced under sparse data conditions typical of small studies. Validation on real glycemic response data showed Monte Carlo yielding narrower uncertainty bounds and more statistically informative results. These findings demonstrate the first large-scale benchmarking of AUC estimation from graphical data and recommend Monte Carlo approaches for meta-analytic applications requiring integration of digitized response curves.