A new likelihood model for analyses of pharmacoepidemiologic case–control studies which avoids decision rules for determining latent exposure status

Table 3 Estimated association between NSAID use and severe upper gastrointestinal bleeding

Exposure definition	n	Crude OR (95% confidence interval)	Adjusted OR (95% confidence interval)	Upper/lower confidence limit ratio for adjusted OR
Fixed window, 30 days	9,453
Conditional		7.06 (6.17—8.06)	5.17 (2.40—11.11)	4.62
Ordinary		6.91 (6.19—7.71)	3.85 (2.09—7.07)	3.38
Fixed window, 90 days	12,662
Conditional		4.96 (4.46—5.51)	4.73 (2.72—8.23)	3.02
Ordinary		4.87 (4.43—5.36)	3.52 (2.19—5.65)	2.58
WTD treatment probability	39,119
Conditional		6.99 (6.35—7.69)	4.37 (3.62—5.28)	1.46
Ordinary		6.90 (6.28—7.58)	3.94 (3.29—4.72)	1.43
Joint likelihood model	39,119	5.57 (5.08 ‐ 6.05)	2.52 (1.59 – 3.45)	2.18

Case–control study of 3568 cases and 35,552 controls. See text for technical description of exposure definitions and covariate adjustment. Conditional results are based on conditional logistic regression which accounts for matching on age and sex. Ordinary results are based on ordinary logistic regression (unconditional), but where sex and age are included as covariates, both in the crude and adjusted analyses