Validation of Autism Spectrum Disorder Diagnoses in Large Healthcare Systems with Electronic Medical Records.

The team asked: are kids labeled "ASD" in big hospital computer files really on the spectrum?

They pulled 1,272 electronic records that had at least one autism code. Two experts then rescored each file using standard checklists. They counted how many kept the label.

Eight out of every ten records still looked like ASD after expert review. Having two separate autism codes in the file, not just one, was the strongest clue the label was correct.

Sawyer et al. (2014) ran the same two-code rule on insurance claims one year earlier. They hit 87% accuracy, almost matching the 81% here. The small gap comes from using claims versus full hospital charts.

Hedley et al. (2016) show clinician gut feelings also work: when a doctor says "I'm 90-100% sure," they're almost always right. Together these papers say both data rules and human judgment can flag true cases.

Wieckowski et al. (2025) flip the coin: they found one in four kids who screened positive never got any code at all. So requiring two codes catches real cases but may miss kids who never reach the system.

If you pull caseloads from electronic files for research or billing, demand at least two ASD diagnosis codes. This simple filter keeps roughly four out of five records accurate without extra chart review. Pair it with quick clinician confidence ratings when time allows, and track kids who screen positive but still lack codes so they don't fall through the cracks.