Multiple-marker screening for Down's syndrome: a method of assessing the statistical robustness of proposed tests.

OBJECTIVES: Antenatal screening for Down's syndrome relies on the use of multiple markers in combination. Markers that are highly correlated can cause statistical instability. We used the maximum variance inflation factor (VIF(max)) to determine whether a screening test using multiple markers was robust to imprecision in the estimation of the marker distribution parameters. METHODS: The VIF(max) for a specified screening test was calculated from the correlations between markers in Down's syndrome pregnancies for six tests: integrated and serum integrated tests without repeat measurements, both tests with repeat measurements across trimesters analysed in the standard way, and both tests with repeat measurements analysed as cross-trimester (CT) marker ratios. The screening performance of each test using published parameter values, in terms of the false-negative rates for a 3% false-positive rate (FN(3)), were calculated for simulated populations with medians 0.2 standard deviations (SD) higher or lower than the published values (to reflect imprecision in parameter estimation) for pregnancy-associated plasma protein A and unconjugated oestriol in affected pregnancies. For each test, the VIF(max) value was compared with the coefficient of variation of the FN(3) (FN(3) CV). An independent set of 27 Down's syndrome pregnancies was used to determine how many had meaningless low risks (<1 in 10,000) with each test. RESULTS: Tests with VIF(max) values greater than 5 had FN(3)CV values over 50%, but those with VIF(max) values less than 5 had FN(3) CV values less than 21%. The numbers of Down's syndrome pregnancies with meaningless low risk estimates in the independent set were 18 (64%) in tests with VIF(max) values > or =5 and none for those with values <5. CONCLUSION: VIF(max) values of 5 or more suggest instability. The tests using CT marker ratios were stable (VIF(max) < 3), but the tests using repeat measurements in the standard manner were not (VIF(max) > 5).