>>24873987 (pb)
The main biases and methodological limitations of the Henry Ford Health System draft study (Lamerato et al., ~2020) and its later reanalysis are well-documented by the institution itself, independent biostatisticians, and epidemiologists. These issues prevented the original draft from advancing past internal review and make causal claims about vaccines unreliable.
Here are the primary biases:Baseline imbalances / confounding (groups not comparable): Vaccinated and unvaccinated children differed significantly on nearly every measured baseline characteristic. The unvaccinated group had more males, more White children, lower rates of prematurity, low birth weight, respiratory distress at birth, and birth trauma. These factors are independently linked to later chronic conditions. Unmeasured confounders (socioeconomic status, parental education, environmental exposures, neighborhood factors in the Detroit area, parental health-seeking behavior) almost certainly remain. Standard statistical adjustment for the measured variables does not eliminate this problem.
Detection / ascertainment / surveillance bias: Vaccinated children had substantially higher healthcare utilization (roughly 7 encounters per year on average versus about 2 for unvaccinated children). Conditions diagnosed via clinical visits (asthma, speech disorders, ADHD, developmental delays, learning disabilities, etc.) are far more likely to be recorded in the group that sees doctors more often. Sensitivity analyses that excluded children with zero visits still left large utilization differences and did not resolve the bias.
Differential follow-up time (surveillance/age bias): Median follow-up was much shorter for the unvaccinated group (approximately 461 days versus 970 days for vaccinated children). About 25% of unvaccinated children were observed only to around 6 months of age and 75% only up to age 3—before many of the studied conditions (neurodevelopmental, behavioral, mental-health diagnoses) are typically identified. Vaccinated children were followed longer and into the ages when diagnoses peak, inflating apparent rates even if true incidence were identical.
Small unvaccinated sample and zero-cell problems: Only ~1,957 completely unvaccinated children versus ~16,511 vaccinated. Many specific outcomes had zero or near-zero events in the unvaccinated arm, making rate ratios unstable or undefined under conventional modeling. The reanalysis switched to simple proportional contrasts to highlight differences across 22 categories, but this does not correct the underlying selection, detection, or follow-up biases.
Henry Ford Health explicitly cited these issues (incomparable groups, small unvaccinated sample, and markedly shorter observation periods for unvaccinated children) as the reason the draft “did not meet the rigorous scientific standards” and was never submitted for publication.
Independent reviewers (including biostatisticians and infectious-disease specialists) have described the combination of these biases as “fatal” to causal inference. Larger, better-controlled studies using national registries have not found similar associations between childhood vaccination and the chronic conditions examined.