TY - JOUR
T1 - Selection bias induced by reproductive stoppage in estimates of recurrence risk for autism spectrum disorders
AU - Beenstock, Michael
N1 - Publisher Copyright:
© 2021
PY - 2021/11
Y1 - 2021/11
N2 - Background: In observational studies it is implicitly assumed that the data are representative of the general population. Self-selection by individuals may undermine this assumption, inducing sample selection bias. In studies of recurrence risk of autism spectrum disorder (ASD) it is implicitly assumed that parents who had no further children face the same recurrence risk as parents who had further children. If this assumption is false, estimates of recurrence risk may be too high or too low. Method: Two canonical statistical methods for detecting and correcting for sample selection bias, including incidental truncation and inverse probability weighting, are applied to population cohort data for Israel. The data comprise 8205 younger siblings of 9117 children diagnosed with ASD, among which there were 371 recurrences. Reproductive stoppage occurred in 4216 out of 9117 families. Participants are distinguished by demographics (gender, age), ethnicity (Jews and Arabs), religiosity, birth cohort and socioeconomic status. Results: Correcting for selection bias using the incidental truncation method, the average risk of recurrence is 3.83 (95 %CI: 3.41 %–4.24 %) instead of 4.53 % (95 %CI: 4.08 %–4.98 %). The bias has p-value 0.082. Since the incidental truncation method may be sensitive to parametric assumptions regarding the joint distribution of unobserved heterogeneity in stoppage and recurrence risk, robustness tests are carried out using alternative parametric assumptions, including copulas. Results using the method of inverse probability weighting were unsatisfactory because they were sensitive to hidden confounders. Conclusions: Studies of recurrent risk of ASD should take account of potential selection bias induced by reproductive stoppage.
AB - Background: In observational studies it is implicitly assumed that the data are representative of the general population. Self-selection by individuals may undermine this assumption, inducing sample selection bias. In studies of recurrence risk of autism spectrum disorder (ASD) it is implicitly assumed that parents who had no further children face the same recurrence risk as parents who had further children. If this assumption is false, estimates of recurrence risk may be too high or too low. Method: Two canonical statistical methods for detecting and correcting for sample selection bias, including incidental truncation and inverse probability weighting, are applied to population cohort data for Israel. The data comprise 8205 younger siblings of 9117 children diagnosed with ASD, among which there were 371 recurrences. Reproductive stoppage occurred in 4216 out of 9117 families. Participants are distinguished by demographics (gender, age), ethnicity (Jews and Arabs), religiosity, birth cohort and socioeconomic status. Results: Correcting for selection bias using the incidental truncation method, the average risk of recurrence is 3.83 (95 %CI: 3.41 %–4.24 %) instead of 4.53 % (95 %CI: 4.08 %–4.98 %). The bias has p-value 0.082. Since the incidental truncation method may be sensitive to parametric assumptions regarding the joint distribution of unobserved heterogeneity in stoppage and recurrence risk, robustness tests are carried out using alternative parametric assumptions, including copulas. Results using the method of inverse probability weighting were unsatisfactory because they were sensitive to hidden confounders. Conclusions: Studies of recurrent risk of ASD should take account of potential selection bias induced by reproductive stoppage.
KW - Incidental truncation method
KW - Inverse probability weighting
KW - Recurrence risk of ASD
KW - Reproductive stoppage
KW - Selection bias
UR - http://www.scopus.com/inward/record.url?scp=85114824143&partnerID=8YFLogxK
U2 - 10.1016/j.rasd.2021.101863
DO - 10.1016/j.rasd.2021.101863
M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???
AN - SCOPUS:85114824143
SN - 1750-9467
VL - 89
JO - Research in Autism Spectrum Disorders
JF - Research in Autism Spectrum Disorders
M1 - 101863
ER -