Evaluation of Down's syndrome screening population data sets by simulation: analyser‐specific parameters may be superior to meta‐analysis‐derived parameters

Summary Objective:  Choice of parameter sets used to calculate Down's syndrome risks is complicated. Published population statistics were compared with assay-specific parameters to optimise screening efficiency. Design:  Weight-corrected Gaussian population statistics for α-fetoprotein (AFP), human chorionic gonadotropin (HCG) and unconjugated oestriol (uE3), expressed as log10 multiples of median (MoM) were established for a Belgian cohort of 748 unaffected pregnancies. Using Cuckle's method and Access®-specific data, Down's syndrome parameters were tailored to the Belgian cohort. Correlated marker triplets for affected and unaffected pregnancies were modelled and combined with maternal age to calculate term risks for Trisomy 21. Receiver-Operator-Curve (ROC) analysis was performed to identify the optimally-performing population set. Results:  Log-normal distributions for the Access® markers had geometric mean MoM values close to zero and standard deviation values equal to 0.1460 (AFP), 0.2185 (HCG) and 0.1317 (uE3). Correlation between AFP and other markers was significant (p < 0.001). Correlation between HCG and uE3 was not significant (p = 0.4818). The median ratio between the lowest and highest risk outcomes for the test MoM set was 4.3. Areas under ROC curves differed significantly (p < 0.001) between the models and the analyser-assay specific parameters resulted in the largest area. At a 1 in 250 threshold, sensitivity and specificity were 69% and 96%. At false-positive rates (1-specificity) = 5%, sensitivity was 72.5%. Conclusion:  Population parameters significantly affect risk outcome and hence screening performance. Highest efficiency may be obtained with parameters tailored to an assay-specific population model. Consequently models from literature, without knowledge of the assay/analyser combination may lead to suboptimal performance.