Testing Family Confounding with Linked Claims and Sibling Comparisons

Source published October 8, 2026 · Resource reviewed October 8, 2026

The research question

Can shared familial factors explain an association seen in claims?

Data files and cohort construction

South Korean NHIS and Japanese JMDC records linked mothers to children born in 2010–2017, followed through 2023. Primary analyses required maternal age of at least 20; exclusions included missing socioeconomic information, specified childhood conditions, and death within the first year.

Measures and analytic design

Maternal influenza codes J09–J11 defined prenatal exposure; children’s first recorded neuropsychiatric diagnoses defined outcomes. Stabilized propensity weights balanced measured characteristics before Cox survival models. Exposure-discordant siblings were compared within mother-specific strata, accounting for shared family factors.

Robustness checks and interpretation

Checks assessed measured balance, younger mothers, exposure timing, and Japanese replication. Population associations attenuated in sibling comparisons. Disagreement between designs is an important confounding signal, not an inconvenience to discard.

Practical application: use family linkage to challenge an association

The following are FastHSR implementation considerations, not additional procedures claimed for the study. A research team with authorized family identifiers could compare a population estimate with a within-family estimate. This can reveal whether a finding changes when stable family characteristics are held more nearly constant. Family relationships cannot be inferred reliably from a shared address alone.

Decisions to settle before reuse

Prespecify how maternal records, births, children, and siblings are linked. Audit one-to-many relationships and retain a linkage-quality flag. Align pregnancy exposure windows and child follow-up before constructing outcomes. Separate baseline confounders from variables that arise after exposure; adjusting for a downstream event can create a new bias.

Limitations and local application

Claims miss unrecorded infection and diagnoses. Sibling designs cannot remove all pregnancy-specific confounding; survivor and disease exclusions limit generalizability. For a local application, report which families contribute discordant siblings and how they differ from the full cohort. Use enrollment histories to distinguish no diagnosis from no observation. Examine whether care-seeking and diagnostic access differ across exposure groups. A Medicaid/CHIP adaptation would require permitted family linkage, validated local coding, and coverage-continuity rules; the overseas definitions should not be copied unchanged.

Source article and supplement

Kong J, Hong S, Jo H, et al. Maternal Influenza Infection During Pregnancy and Neuropsychiatric Disorders in Offspring. JAMA Network Open. 2026;9(10):e2637902. doi:10.1001/jamanetworkopen.2026.37902.

• Read the primary journal article
• Article’s Supplemental Content: Supplement 1 methods, diagnosis definitions, and sensitivity tables

The full article and supplement listing were reviewed on October 8, 2026. The separate supplement download could not be retrieved. Confirm linkage procedures, outcome codes, and observation windows in Supplement 1 before replication.

Frequently asked question

Does a sibling comparison eliminate all confounding?

No. It helps account for shared, stable family factors but cannot remove every difference between pregnancies or every measurement error.

Explore more UpEvidence