Table 3.

Comparison of automated method for identification of implausible measurements with physician judgment for a stratified random selection of measurements

Primary analysis
Type of measurement	Inlier/outlier	Automated method results	Concordant (%)	n/Total
Weight	Inlier	Plausible	98	49/50
		Implausible	96	48/50
	Outlier	Plausible	100	50/50
		Implausible	84	42/50
Height	Inlier	Plausible	98	49/50
		Implausible	82	41/50
	Outlier	Plausible	94	47/50
		Implausible	94	48/50
Overall			93	373/400
Special error types
Duplicates			100	40/40
Corrected unit errors			95	38/40
Corrected switchesa			89	34/38

^aOnly 38 of the 40 corrected switches were selected for validation because of sex-stratified selection.