Combining Multiple Markers Into a Risk Based Screen

As a marker for risk of Down syndrome, MSAFP by itself measured solely for the purpose of estimating the risk of Down syndrome would not be powerful enough to be justified. However, because MSAFP is widely used for ONTD screening, use of its modest additional information about the risk of Down syndrome is easily justified. Screening for Down syndrome also benefited from screening for ONTD in that the concept of the MoM had already been developed for ONTD screening (see "Using the MoM to Adjust for Gestational Age" above). Like MSAFP, the concentrations of all the markers for fetal trisomy screening are expressed as MoMs. Being a ratio (the patient's concentration divided by the median concentration at that gestational age), the MoM has no units. The MoM compensates for the concentration changes that occur with gestational age and enables combinations of multiple markers on a common scale.

Initially, MSAFP was used in Down syndrome screening in a binary fashion: any value below a fixed cutoff (0.50 MoM) indicated increased risk. However, the fact that the MSAFP distribution in Down syndrome pregnancies was log Gaussian (55) soon led to the use of Down syndrome LRs for a range of MSAFP concentrations (Fig. 8), in a manner similar to that used in ONTD screening. With the risk associated with a woman's age as the background prevalence of Down syndrome, patient-specific risks could be generated from the combination of the patient's maternal age and her midtrimester MSAFP result (113,114).

In 1988, the maternal serum concentrations of uE3 and hCG were, like AFP, shown to be independent of maternal age, only weakly correlated to one another, and conforming to Gaussian distributions (115) (see Fig. 8). Using the published means and SDs for the unaffected and Down syndrome populations, these three biochemical risk variables were

Fig. 8. Second-trimester maternal serum marker frequency distributions in unaffected and Down syndrome pregnancies for AFP, uE3, total hCG, and DIA. For any marker concentration within the limits of the distributions, an LR can be calculated. Two such examples are shown in the figure for MSAFP, with an increased risk (4:1) and a decreased risk (1:4) compared to background.

Fig. 8. Second-trimester maternal serum marker frequency distributions in unaffected and Down syndrome pregnancies for AFP, uE3, total hCG, and DIA. For any marker concentration within the limits of the distributions, an LR can be calculated. Two such examples are shown in the figure for MSAFP, with an increased risk (4:1) and a decreased risk (1:4) compared to background.

combined in a trivariate Gaussian LR algorithm: "triple-marker screening" (116). (In 1992, the distributions of the uE3 marker were refined (55), replacing the previous linear MoM values with their log equivalents, which better fit the Gaussian distribution.) In the triple marker screening algorithm, the three LRs are multiplied together, with allowance for the published slight correlations that exist between the various marker pairs. The composite multiple-marker likelihood ratio is then multiplied by the maternal age-related a priori risk to yield a patient-specific risk of Down syndrome pregnancy.

In the manner just described, each patient in a screened population will have a calculated risk of the targeted disorder Down syndrome. The population distribution of these risks can be modeled from the underlying marker distributions. Overlapping risk distributions exist for the unaffected and Down syndrome populations, with the Down syndrome population risks being very much higher. The degree of separation of the two populations is sufficient to support screening based on the triple marker Down syndrome risk as the screening variable (see Fig. 9). Modeling the risk distributions has enabled rational decisions for selecting a risk cutoff in Down syndrome screening programs.

When triple marker screening first started, the cutoff from maternal age screening was retained; namely, the term risk of 1:385 (1:290 at midtrimester) was selected because this is the Down syndrome risk at age 35. This choice avoided legal challenges because offering amniocentesis to patients over the age of 35 was deeply embedded in medical

Risk of Down syndrome pregnancy at term

Fig. 9. Down syndrome screening using the combined risk of Down syndrome as the screening variable. The computed risk is based on maternal age and the results of four maternal serum biochemistry tests: AFP, uE3, total hCG, and DIA.

practice and in the minds of the public. Fortuitously, this cutoff achieved the desired increase in Down syndrome detection without increasing the FPR. The new information about Down syndrome risk obtained from the biochemical markers was used to improve detection.

With detection rates at an acceptable level (>70%), attention turned to reducing the FPR from the current 7-10%. DIA was becoming analytically feasible in the late 1990s. The log MoM concentration of DIA conforms to a Gaussian distribution in both the unaffected and Down syndrome populations (Fig. 8), and DIA is essentially independent of AFP and uE3, and only moderately correlated with the concentration of hCG. DIA could therefore be added to the triple marker algorithm by simple mathematical extension of the modeling (Fig. 9) to include the fourth marker (117-119), thereby enabling a four-marker prenatal maternal serum screen for Down syndrome based on AFP, uE3, hCG, and DIA.

Was this article helpful?

0 0

Post a comment