Table 1.

Sensitivity and positive predictive value of computerized diagnoses for identifying ankylosing spondylitis in 2603 adults at first recorded diagnosis with 1 or more diagnoses^a

Operational definition	Number in population	Number in stratified sampleb	Number of true-positives in stratified samplec	Number of false-positives in stratified samplec	Sensitivity,d % (95% CI)	Percentage of diagnoses for AS (ICD-9 code 720.0)	PPV of algorithm to find AS, % (95% CI)e
≥ 2 diagnoses in primary care or ≥ 1 diagnosis in rheumatology	2603	129	80	49	100c	80	62 (60–64)
≥ 2 diagnoses, any department	2353	102	67	35	96 (95–97)	80	66 (64–68)
≥ 1 diagnosis, rheumatology	1575	83	61	22	72 (70–74)	94	73 (71–76)
≥ 2 diagnoses, rheumatologyf	1325	56	48	8	67 (64–69)	96	81 (79–83)

^aPatients aged 18 years or older, from inpatient or outpatient data from Kaiser Permanente Autoimmune Disease Registry, Northern California, 1996 to 2009.

^bSubjects were sampled for chart review on the basis of the number of visits, the department in which the diagnosis was made, use of disease-modifying antirheumatic drugs, and presence of comorbid autoimmune conditions. The latter 2 variables did not improve the sensitivity or specificity of the algorithm.

^cClassified on the basis of medical record review.

^dBy definition, given that our search strategy required at least 1 physician diagnosis of ICD-9 code 720.X, and recognizing the likelihood of underascertainment.

^eCalculated by dividing the number of true-positives by the number in the stratified sample. For example, on the first row, 80 (true-positives)/129 (number in the stratified sample) = 62%.

^fBoldface indicates most specific algorithm (maximized positive predictive value).

AS = ankylosing spondylitis; CI = confidence interval; ICD-9 = International Classification of Diseases, Ninth Revision; PPV = positive predictive value.