Table 2.

Adjusted (left) Cox model-estimated and (right) marginal structural model-estimated HRs (95% CI) for the association between falls and fractures and anticholinergic burden in the OAB cohort (n=154 432), including the subgroup aged >65 years (middle); Truven MarketScan databases 2007–2015

	Cox model*				Marginal structural model*
	Overall population		Subgroup aged >65 years		Overall population
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
By anticholinergic burden level versus no burden†
Low (1–89)	1.2 (1.2 to 1.3)	<0.001	1.1 (1.0 to 1.2)	0.006	1.2 (1.1 to 1.2)	<0.001
Medium (90–499)	1.3 (1.2 to 1.4)	<0.001	1.2 (1.1 to 1.3)	<0.001	1.2 (1.1 to 1.3)	<0.001
High (500+)	1.4 (1.3 to 1.4)	<0.001	1.2 (1.1 to 1.3)	<0.001	1.3 (1.3 to 1.4)	<0.001
By age category versus ≤45
46–55	1.3 (1.2 to 1.3)	<0.001	1.7 (1.6 to 1.7)‡	<0.001	1.2 (1.2 to 1.3)	<0.001
56–65	1.5 (1.4 to 1.6)	<0.001			1.5 (1.4 to 1.6)	<0.001
66–75	2.3 (2.2 to 2.4)	<0.001			2.3 (2.1 to 2.5)	<0.001
76–85	3.4 (3.2 to 3.6)	<0.001			3.5 (3.3 to 3.9)	<0.001
86+	5.0 (4.6 to 5.4)	<0.001			5.6 (5.0 to 6.3)	<0.001
Sex
Women versus men	1.5 (1.5 to 1.6)	<0.001	1.6 (1.5 to 1.7)	<0.001	1.5 (1.5 to 1.6)	<0.001
Comorbidity categories at baseline
Cardiovascular diseases§	1.1 (1.1 to 1.1)	0.018	1.2 (1.1 to 1.2)	<0.001	1.1 (1.0 to 1.1)	0.043
Neurological impairments	1.5 (1.4 to 1.6)	<0.001	1.7 (1.5 to 1.8)	<0.001	1.5 (1.4 to 1.6)	<0.001
Endocrine, nutritional and metabolic diseases	1.1 (1.1 to 1.2)	<0.001	1.2 (1.1 to 1.4)	<0.001	1.2 (1.1 to 1.3)	<0.001
Cardiovascular disease×neurological impairments	1.1 (1.0 to 1.2)	0.042	1.0 (0.9 to 1.1)	0.945	1.1 (1.0 to 1.2)	0.048
Cardiovascular disease×endocrine, nutritional and metabolic diseases	1.0 (1.0 to 1.1)	0.750	0.9 (0.8 to 1.0)	0.118	0.9 (0.8 to 1.0)	0.219
Neurological impairments×endocrine, nutritional and metabolic diseases	1.1 (1.0 to 1.2)	0.092	1.0 (0.9 to 1.1)	0.786	1.0 (0.9 to 1.2)	0.558

*The Cox models were implemented using function coxph from the R package survival V.2.41–3. The marginal structural model was implemented using function coxph from R package survival V.2.41–3, using the weight argument to apply time-varying weights and setting a cluster term for enrolment ID for robust variance estimation. Time-varying weights were calculated using function ipwtm from R package ipw V.1.0–11 and based on a multinomial logistic regression model (using a generalised logit link) with categorical time-varying anticholinergic burden as the outcome, where age, sex and time-varying comorbidity categories, as well as all two-way interactions between them, were included as predictor variables.

†Level of anticholinergic burden assessed using the closest 6-month measure prior to the fall or fracture.

‡Cardiovascular disease=cerebrovascular disease+stroke.

§For the subgroup analysis among those aged >65 years, age categories for comparison were 65 to <74 years, vs >75 vs <75 years.

OAB, overactive bladder.