Table. Time-Dependent Associations Between Postdiagnosis Aspirin Use and Overall Survival for Each Biliary Tract Cancer Site^a.

	Gallbladder		Cholangiocarcinoma		Ampulla of Vater		Overlapping Lesions
	No. of Events/No. at Riskb	HR (95% CI)	No. of Events/No. at Riskb	HR (95% CI)	No. of Events/No. at Riskb	HR (95% CI)	No. of Events/No. at Riskb	HR (95% CI)
Overall
Nonusers	499/600	1 [Reference]	1198/1419	1 [Reference]	116/186	1 [Reference]	360/437	1 [Reference]
Users	54/67	0.63 (0.48-0.83)	123/140	0.71 (0.60-0.85)	26/38	0.44 (0.26-0.76)	39/47	0.68 (0.50-0.92)
Prediagnosis Aspirin Use
Nonusers	145/145	1 [Reference]	383/383	1 [Reference]	33/33	1 [Reference]	119/119	1 [Reference]
Prevalent usersc	49/50	0.69 (0.50-0.94)	114/114	0.78 (0.65-0.95)	26/26	0.41 (0.22-0.79)	37/38	0.76 (0.54-1.07)
No Prediagnosis Aspirin Use
Nonusers	354/455	1 [Reference]	815/1036	1 [Reference]	83/153	1 [Reference]	241/318	1 [Reference]
Incident usersc	5/17	0.57 (0.28-1.17)	9/26	0.37 (0.21-0.64)	0/12	0.21 (0.03-1.56)	2/9	0.34 (0.12-0.94)
P value for interactiond	.03		<.001		.005		.005

Abbreviations: BTC, biliary tract cancer; HR, hazard ratio; NA, not applicable.

^aAdjusted for sex, history of heart disease, statin use (current, former, never), presence of comorbidities, age at diagnosis, and year of diagnosis. Aspirin use was modeled as time dependent and the baseline hazard was stratified by prediagnosis aspirin use.

^bThe results presented used Cox regression where aspirin was modeled as time dependent (eg, individuals could switch between use and nonuse status). The numbers represent aspirin use at the time of BTC diagnosis.

^cPrevalent users were defined as patients with 2 or more aspirin prescriptions before BTC diagnosis. Incident users were defined as patients who only initiated aspirin use on or after the BTC diagnosis date.

^dP values for interaction were estimated by putting a cross-product term in the models for postdiagnosis use and prediagnosis use.