Table 4

Validity assessment and results of observational studies

Study	Ascertainment of outcome	Ascertainment of exposure	Adjustments	Limitations and bias	ICS exposure (BDP equivalents where available)	OR (95% CI)
Gonnelli, 2010^w27	Physicians assessed the lateral CXR using MorphoExpress software	Direct interview of patients and checking of medical records	Age, gender, BMI, and COPD severity	Cross-sectional design limits ability to adjust for confounders, or to establish temporal association. Exposure data from patient interviews are subject to recall bias	≤750 μg	1.26 (0.98 to 1.89)
					750–1500 μg	1.36 (0.93 to 1.72)
					>1500 μg	1.40 (1.04 to 1.89)
Johannes, 2005^w28	ICD-9 codes for non-verterbral fractures, and claims for physician or hospital care	Pharmacy claims for ICS use in the past year prior to fracture	Demographics, medical conditions, medications including oral corticosteroid use, and health utilisation for underlying respiratory disease	No spirometric definition of COPD. Reliance on ICD codes and insurance claims. Relatively few subjects aged ≥65 years. Lack of adjustment for important confounding variables. Funded by manufacturer of ICS	30 days prior vs no current use	0.86 (0.59 to 1.25)
					90 days prior vs no recent use	1.02 (0.77 to 1.36)
					1–167 μg	0.88 (0.64 to 1.19)
					168–504 μg	0.82 (0.54 to 1.26)
					505–840 μg	1.22 (0.67 to 2.25)
					> 840 μg	1.05 (0.53 to 2.07)
Lee, 2004^w29	ICD-9 codes for non-vertebral fractures. No specific validation for this study	Based on outpatient pharmacy claims database	Co-morbidities, medications, annual hospitalisation and oral corticosteroid use	No spirometric definition for COPD. Reliance on ICD coding, lack of adjustment for potentially important confounders, no lung function data. Funded by manufacturer of ICS	Current user (last 30 days) vs non current	1.20 (0.94 to 1.54)
					Recent user (last 90 days) vs non-recent user	1.14 (0.95 to 1.37)
					ICS <300 μg vs no ICS	0.83 (0.66 to 1.04)
					ICS 300–699 μg vs no ICS	0.96 (0.78 to 1.17)
					ICS >700 μg vs no ICS	1.20 (0.95 to 1.52)
McEvoy, 1998^w30	Lateral lumbar and thoracic x-rays independently reviewed by blinded radiologists.	Computerised pharmacy records, ICS use ≥4 puffs a day ≥6 months of past year, and no more than 2 brief oral steroid courses.	Smoking history, FEV₁, Baseline Dyspnoea Index, Activity Limitation Score and General Health Status Index	Analysis restricted to male patients >50 years. Cross-sectional design limits the ability to adjust for all confounders	ICS use ≥4 puffs a day for at least 6 months of past year	1.38 (0.71 to 2.69)
Pujades-Rodríguez, 2007^w31	Any fracture recorded in electronic medical records (13% were hip, and 9% were wrist)	ICS exposure based on electronic prescribing records	Age, predicted FEV₁ and oral corticosteroid use, and matched for sex and general practice	No spirometric definition of COPD. Reliance on general physician record for diagnosis of COPD and outcome/exposure ascertainment	Any ICS use	1.12 (0.97 to 1.29)
					≤100 μg	1.06 (0.88 to 1.28)
					101–200 μg	0.99 (0.78 to 1.27)
					201–400 μg	1.17 (0.95–1.44)
					401–800 μg	1.21 (0.97–1.51)
					801–1600 μg	1.13 (0.87 to 1.46)
					≥1600 μg	1.74 (1.00 to 3.01)
WEUSRTP1127 Miller, 2010^{w32 w34}	OxMIS and Read codes for non-vertebral fracture	Based on ICS use in the electronic medical record year prior to index date	COPD hospitalisation, BMI, smoking status, concomitant medication vertebral fractures, co-morbidities	Reliance on electronic medical record coding, lack of adjustment for potentially important confounders, no lung function data. Funded by manufacturer of ICS	Use in past 12 months vs non-use past year	1.25 (1.07 to 1.47)
					Current use (13–25 days) vs non-use in past year	1.10 (0.84 to 1.46)
					Recent use (26–52 days) vs non-use in past year	1.36 (1.04 to 1.77)
					Medium (750 μg) vs none	0.90 (0.67 to 1.20)
					High (1500 μg) vs none	0.92 (0.69 to 1.24)
					Very high (2000 μg) vs none	1.06 (0.68 to 1.66)
WWE113669, 2008^w33	OxMIS and Read codes for first non-vertebral fracture	Based on ICS use in the electronic medical record 1 year prior to index date	COPD hospitalisation, BMI, smoking status, concomitant medication vertebral fractures, co-morbidities	No spirometric definition of COPD. Reliance on electronic records, no lung function data. Funding source: manufacturer of ICS	Current user (last 30 days) vs non-use past year	1.42 (1.23 to 1.65)
					Recent user (last 31–90 days) vs non-use past year	1.35 (1.16 to 1.58)
					Low dose vs none	1.39 (1.16 to 1.66)
					Medium dose vs none	1.51 (1.24 to 1.83)
					High dose vs none	1.32 (1.13 to 1.55)

BDP, beclomethasone; BMI, body mass index; COPD, chronic obstructive pulmonary disease; CXR, chest x-ray; FEV₁, forced expiratory volume in the first second of expiration; ICD, International Classification of Diseases; OxMIS, Oxford Medical Information Systems.