OBJECTIVES: In 2006, the American College of Medical Genetics was tasked with recommending conditions to include in state NBS programs. It developed an 18 question survey and assigned scores to various answers, summing them to arrive at a total. Mean score determined one of three entry points to an algorithm, EPA (>1200, 1000-1200, and <1000), that determined a condition’s final recommendation: Core, Secondary target or not recommended. The objective of this research is to explore the influence of uncertainty on the GAII recommendation and identify if missing survey responses could have changed the final score from 1224, the EPA and the recommendation.

METHODS

: Data for 2 variables were available out of 18 questions, each having 3 missing responses. We developed boundary estimates (Manski,1989) whereby bounds on missing response answer scores are combined with probabilities of being missing to develop bounds for the mean score. This avoids reliance on the (implicit and questionable) ACMG assumption that the means of missing and observed data are equal.

RESULTS

: The total score change for the two questions combined implied a maximum loss of 4 points to a score of 1220, insufficient to alter the EPA or, therefore, the final recommendation.

CONCLUSIONS

: The analysis we have done implies no changes to the ACMG recommendation for GAII, but it clearly understates the uncertainty regarding final scores due to examining missing variables in only 2 of the 18 variables in the survey. There were 31 additional missing responses in the survey, a complete accounting for which may imply uncertainty of the original EPA. ACMG also ignored conventional issues of sampling uncertainty – that the sample mean is an estimate with uncertainty. Incorporating these additional elements may question the EPA and recommendation for GAII.