All markers were expressed as multiples of the gestation-specific median (MoMs) for unaffected pregnan­cies and log₁₀ transformed. Meta-analysis was used to derive the DS mean and the difference in variance and covariance between affected and unaffected pregnancies, taking the weighted average of the individual values. If there was insufficient data to calculate differences, the parameters were estimated directly by taking the weighted average between studies. Except for the ultrasound mark­ers only non intervention studies were included in the analysis. In intervention studies the results are acted on clinically and so are subject to ‘viability’ bias that will skew the results. This bias arises because a proportion of those with abnormal marker levels and termination of pregnancy would have been destined to miscarry, whereas non viable DS pregnancies with normal levels will not be known to the investigators.

Serum Meta-Analysis. Fourteen second trimester inhibin studies were included [3-16]. Five studies were used to estimate the correlation between first trimester serum PAPP-A and second trimester serum markers [16-20].

Ultrasound Meta-Analysis. There are 11 first trimes­ter NT studies where MoMs were either reported or could be derived from a figure or table in the publication [16,21-30]. The meta-analysis included five intervention studies. The results of one prospective intervention study, the mas­sive multi-center Fetal Medicine Foundation study [31], have been used to allow for this statistically and estimated that the skew increases the mean by 11% [23]. This factor was applied to the five intervention studies before taking the weighted average of all 11.

There were four studies in which second trimester NF values in MoMs were either reported or could be derived from a figure in the publication [32-35]. Three were inter­vention studies and so subject to viability bias, although this was much less of a problem than for NT: viability is higher in the second trimester than in the first, multiple ultrasound markers were used and most were already at high DS risk.

Other Parameters. The existing meta-analyses were used except that instead of the a single mean for free-β hCG throughout the first trimester, gestation-specific val­ues were used, derived from a regression curve [36].

The unaffected variance and covariance in unaffected pregnancies was obtained from 29,516 women screened in Leeds: 11,019 having serum markers measured at 10-13 weeks, 18,497 at 14-18 weeks and 5,177 having NT and center-specific MoMs. The standard deviation was esti­mated by the 10^th to 90^th centile range on a log scale di­vided by 2.563, and the covariance was estimated by ex­cluding outliers exceeding three SDs from the mean. Sepa­rate variance and covariance for the serum markers were derived for each week of gestation from 10 to 13.

Modeling Performance. Detection rates (DRs) and false-positive rates (FPRs) were estimated by numerical integration [37]. The maternal age specific risk curve was from Cuckle et al. [38], and the maternal age distribution was taken to be Gaussian with a mean age of 27 and SD of 5.5 years [39]. Performance was expressed in three ways: the DR for a fixed 5% or 1% FPR; FPR for a fixed 75% or 85% DR; DR and FPR for a fixed 1 in 200, 250 or 300 term risk cut-off.