Statistical sensitivity analysis for the estimation of fetal alcohol spectrum disorders prevalence

Abstract

Fetal alcohol spectrum disorders (FASD) are thought to be the leading cause of developmental disabilities worldwide. Accurate estimates of the prevalence of FASD have been lacking. An improved estimate of FASD prevalence in the U.S. was recently reported in a study where multistage methods of active case ascertainment of first-grade children with FASD were utilized in four regions in the U.S. Each method relied on parental consent and therefore had potential nonresponse bias. We consider weighted approaches, where the weights were formed using the distribution of observed variables in the population from which the samples were drawn. However, there are likely other unobserved variables that affect both non-response and FASD outcome. We describe sensitivity analyses using methodology developed for causal inference. The results show feasible regions of FASD prevalence under certain assumptions, and provide a framework for explaining the non-response bias due to the unobserved variables.

Fetal Alcohol Spectrum Disorders in a Pacific Southwest City: Maternal and Child Characteristics

“…Nevertheless, we did have cases of FAS and pFAS where maternal exposure to risky drinking could not be documented. Across the sample, however, we found that 26.4% of women reported any drinking in pregnancy (Table ), which is consistent with California and national survey data (Denny et al, ; Tan et al, ; Xu et al, ).…”

“…In the Pacific Southwest site, they were 90.0 per 1,000 (95% CI 65.9 to 118.6) in the first year and 84.4 per 1,000 children (95% CI 61.2 to 112.3) in the second year. Furthermore, in subsequent sensitivity analyses based on population‐based survey data on any alcohol use in pregnancy and race/ethnic group from women in the Pacific Southwest region, the prevalence estimates for FASD ranged from 87.3 per 1,000 (95% CI 63.3 to 115.9) to 82.3 per 1,000 (95% CI 56.9 to 115.9; Xu et al, ).…”

Sensitivity analysis of treatment effect to unmeasured confounding in observational studies with survival and competing risks outcomes

“…For the distribution of the unobserved confounder we used binary 0, 1 with probability 0.5 each, which were recommended and used throughout the book by Rosenbaum 1 . It is also possible to incorporate normally distributed U , such as in Shen et al 5 and Xu et al, 21 in which case the probit link in model () allows closed‐form marginal propensity scores given X after integrating out U . The part of the EM algorithm would be similar to that under the proportional hazards mixed‐effects model and Monte Carlo approximation would be needed in the E‐steps 16 …”

“…However, such bounds can be very difficult to derive for complex outcomes like what we consider here in the presence of right censoring, which is unlike in Shen et al 5 where it is possible to derive these bounds for binary or continuous outcomes without censoring. Also evident in Shen et al 5 is that parametric settings are often needed in order to aid in the interpretation of the sensitivity parameters in the corresponding nonparametric settings, and extensive simulations have to be conducted in order to determine sensible ranges for these sensitivity parameters 21 …”

“…Also evident in Shen et al 5 is that parametric settings are often needed in order to aid in the interpretation of the sensitivity parameters in the corresponding nonparametric settings, and extensive simulations have to be conducted in order to determine sensible ranges for these sensitivity parameters. 21 ORCID Ronghui Xu https://orcid.org/0000-0002-2822-0561…”