Precision of risk in multidimensional likelihood ratio based screening with first trimester screening for Down's syndrome as a special case

Abstract

Objectives - The development of computer programs for the calculation of risks and Monte Carlo estimation of the precision of such risks in likelihood ratio based screening with multiple gaussian distributed risk markers and a priori risks. A quantitative study of the variation of risk estimates in first trimester screening for Down's syndrome as a function of the variation of markers, and comparison of the results with published information on the variation of risk estimates in quality control schemes. Methods - Algorithms for calculations in the multidimensional normal distribution and procedures for Monte Carlo simulation of risk distributions were implemented in the S-PLUS programming language and used to construct programs producing risk estimates and risk distributions. Parameters of risk marker distributions and correlations were obtained from the scientific literature. Results - In screening for Down's syndrome during the first trimester the variation in risk estimates increased with increasing variation of biochemical and biometric markers, and the a posteriori risk may vary with at least a factor of 2-4. Conclusions - Risk estimates are not reasonable parameters in quality control systems. Instead, screening programmes should be controlled through careful monitoring of the distribution of risk estimates, in particular the screen positive rate, and control of the quality of the bio-chemical and biometric data. Furthermore, the correct classification of samples submitted for proficiency testing into screen positive and screen negative cases should be checked.