Table 4—

Characteristics of observational studies of inhaled corticosteroids(ICS) and cardiovascular (CV) events or mortality

Study [Ref.]	Participants	Design	Data source	Duration weeks	Participants included	ICS exposure	Ascertainment of ICS exposure	Outcomes	Outcome ascertainment	Risk estimates ICS (versus controls) with outcomes	Bias and limitations
Fan 38	2654 ICS users, 5398 non-users	PC	US VA DB	78	1. Outpatient visit or hospitalisation with ICD-9 diagnosis of COPD.2. Using at least one pulmonary prescription during 90 days before index visit.3. Age >45 yrs.4. Participant in the ACQUIP RCT for ≥1 yr.	Becl (44%), triam (45%), fluni (6%), flu (5%)	Prescription for inhalers, subdivided by 90-day exposure periods.“User” if prescription filled for >80% of each 90 days.	Death	Records date of death for veterans whose families file for the benefit.	HR of death ICS versus controls.Low dose (HR 0.96; 95% CI 0.69–1.33); medium/high doses (HR 0.86; 95% CI 0.67–1.10)Adjusted for distance from hospital, pulmonary prescriptions, prior hospitalisations, comorbidity.	Significant baseline differences in age, marital status, race, comorbidity, COPD clinic visits and hospitalisations, respiratory prescription use.Assumes VA system provides records on all prescriptions and patients compliant.No ascertainment of death by checking, DB known to identify 98.8% of deaths among Medicare-eligible patients.
Huiart 39	371 cases, 1864 controls	Nested CC	Canadian Health Insurance DB in Saskatchewan	130	New patients with COPD.1. At least 3 inhaler prescriptions.2. Age >55 yrs with no previous use or ICS in 5 yrs.3. Register in health plan for 5 yrs.4. Entry time 3rd prescription between Jan 1990–Dec 1997.Matched for age, duration and frequency of exacerbations.Excluded lengthy hospital admissions.	Becl or bud	Exposed if recently received ICS prescription in 12 months prior to index date.	MI	Discharge diagnosis from the hospital DB, and death from vital statistics.Compared to medical charts, diagnostic agreement in hospital DB as high as 97%.	RR for AMI with ICS:crude RR 0.86 (95% CI 0.68–1.09); adjusted RR 0.83 (95% CI, 0.63–1.08),adjusted for AMI risk factors and severity of COPD: sex, HTN, DM, HLD, CV disease, number of COPD exacerbations and respiratory prescription.	Important baseline differences: sex, risk factors and history of CVD, hospitalisations in past 3 months.No dose information available; duration of exposure extrapolated from quantity dispensed, rather than compliance. Matching for COPD severity inaccurate as relied on co-prescription exposure.Limited data on smoking.
Lee 4	32130 cases, 320501 controls for mortality (3159 cases, 31534 controls for CV death)	Nested CC	US VA, CMS and NDI	260	1. Age >45 yrs of US VA care for >1 yr. 2. ICD-9 diagnosis of COPD between Oct 1, 1999 and Sept. 2003 at two or more outpatient visits within 12 months or primary diagnosis of COPD.	NA	Any exposure to respiratory prescription in the 180 days before index date. No details on accuracy of pharmacy data.	Death, CVD	DB captures approximately 98% of veteran deaths. Random sample of 40% with CV death: IHD disease, cardiomyopathy, cardiac arrest, or arrhythmias.	Adjusted OR for ICS: all-cause death 0.80 (95% CI 0.78– 0.83); CVD 0.80 (95% CI 0.72–0.88), adjusted for respiratory and cardiac prescriptions, exacerbations, and presence of CV disease.	Imbalances in comorbid cardiac conditions and cardiac prescription use, and COPD exacerbations. Confounding by indication. Did not ascertain COPD severity or smoking status. Unlikely to find significant differences due to low rates of exposure.
					Matched on sex, age, region, and year of diagnosis.						Immortal time bias. Uncertain accuracy of diagnosis of CVD.
Macie 5	1629 ICS users, 2393 non-users	Cohort, nested CC	Manitoba Population Health Research Repository (Canada)	39	1. Discharged from hospital with primary ICD code of COPD between Apr 1996–Mar 2000.2. Age >35 yrs.3. Resident of the province ≥1 yr prior and 1 yr post-discharge until death.Patient who died within 90 days of hospital discharge excluded.	NA	Provincial retail pharmacies capture all dispensing of prescriptions except for in-patient drugs.DB has anonymous recipient IDs and information about the prescription dispensed.	Death, CVD	Death from any cause in the 275 days after hospital discharge. Analysis of death certificates for: COPD, asthma; CV and other causes.	Adjusted HR for death with ICS within 90 days of discharge.Aged >65 yrs, 0.75 (0.61–0.91).Note: absolute mortality among age >65 yrs: 11.7% in ICS users; 13.1% in non-users.ICSs reduced the risk of death by 23% (95% CI 6–37%) and CV deaths by 38% (95% CI 11–57%) compared to bronchodilators.	Significant baseline differences: age, Charlson score, respiratory prescriptions. Confounding exists.No prescription data while hospitalised so patients admitted in first 90 days of study received less ICS and likely to be high mortality group.Difficulty in judging actual drug usage, and cause of death.
Mapel 40	786 ICS users, 397 SABA users	RC	Two regional MCOs in the USA	96	1. Aged >40 yrs.2. Had >90-day use of an ICS or SABA.3. At least two outpatient visits or one hospital admission with relevant ICD-9 code between 1995–2000.Follow-up 1st day following 90-day period to death or disenrolment.	NA	Pharmacy claims for ≥90-day cumulative of: 1) ICS alone, 2) LABA alone, or 3) ICS + LABA.	Death	Death obtained from state vital statistics records and matched on name, SSN and DOB.	HR for mortality: ICS versus SABA 0.59 (95% CI 0.46–0.78).	Significant baseline differences: age, sex, presence of asthma, disease severity. Confounding by indication.No ascertainment of exposure, compliance or use.Misclassification bias in COPD diagnosis (asthma).Uncertain risk of immortal time bias.
Mapel 41	742 ICS users, 1832 SABA users	RC, nested CC	Four US integrated HMOs	156	At least one in-patient or two outpatient visits for COPD based on ICD-9 codes from Sep 2000 to Sep 2001.Matched on sex, age, data survival follow-up.	Flu/sal, any ICS	Pharmacy records of prescriptions using National Drug Codes. Duration extrapolated from units dispensed. Prescriptions reviewed for missing data.	Death	Death during study period Sept 2000–Sept 2003 ascertained through local vital statistic registers, matched by name, sex DOB and SSN.	HR for mortality ICS versus SABA 0.76 (0.61–0.95).HR for mortality in propensity matched cohort: ICS versus SABA 0.75 (0.58–0.97).RR for death with ICS in nested CC RR 0.86 (95% CI 0.46–0.96).	Significant baseline differences: age, hospitalisation for COPD, comorbidities. Discrepancy between prescription and actual use.Duration extrapolated.Industry funded.
Sin 43	11481 ICS users, 11139 non-users	RC	CIHI hospital discharge DB in Ontario (Canada)	52	1. ICD-9 code for at least one hospitalisation with main discharge diagnosis of COPD between April 1, 1992 and March 31, 1997.2. Aged >65 yrs.3. Survive ≥30 days after discharge.	Becl, bud, triam, fluni	Ontario Drug Benefit database.	Death	Captures all decedents of Ontario, including date of death.	Adjusted RR for death with ICS use (within 90 day post-discharge RR 0.71 (95% CI, 0.65 to 0.78), adjusted for age, sex, Charlson score exacerbations and co-medication. Stratified analysis on Charlson score, sex, and age did not change association.	Significant baselines differences: age, Charlson index, prescriptions and emergency visits.Confounding by indication.No verification of actual medication usage. Risk of immortal time bias.
Sin 42	3343 ICS users, 3397 non-users	RC	CIHI hospital discharge in Alberta (Canada)	128	1. ICD-9 code for at least one hospitalisation with main discharge diagnosis of COPD between April 1, 1994 and March 31, 1998.2. Aged >65 yrs.3. Excluded those with primary diagnosis of asthma.	ICS	Crosschecked with Alberta Blue Cross DB with record of quantity and date dispensed.Doses imputed.	Death	Vital statistics with electronic file of all deceased persons in Alberta, including dates of death.	All cause death for ICS users versus non-users.Crude RR 0.61 (95% CI 0.56–0.66). Adjusted RR 0.75 (95% CI 0.68–0.82), adjustments for age, sex, comorbid conditions, ICU stay and use of other pulmonary prescription.	Significant baseline differences in age, Charlson comorbidity score and use of prescriptions.Unclear, which ICS studied, and what doses.No verification of actual use.Unclear risk of immortal time bias.
Soriano 45	431 ICS users, 3620 non-users	RC	UK GPRD	156	1. Newly diagnosed COPD between 1990–1999.2. Aged >50 yrs.3. Received ≥3 prescriptions over 6 months.4. Alive for 6 months after entry.	Flu	Prescriptions in 90 days post-hospital discharge.	Death	Date of death identified with Specific OMIS code.	HR for fluticasone (3 yr death).HR 0.62 (95% CI 0.45–0.85), Adjusted for age, sex, year of entry, smoking, comorbid conditions and oral steroid use.	Significant baseline differences: age, sex, prescription use, comorbidity.No verification of actual use.Risk of immortal time bias.Industry funded.
Soriano 44	3049 ICS users, 627 SABA users	RC	UK GPRD 1990–1999	52	1. Newly diagnosed COPD.2. Aged >50 yrs.3. 1st hospitalisation for COPD related condition.4. At least 1 prescription for respiratory condition in first 90 days after discharge.	Beclo, bud, flu	Based on prescription by GP in 90 days post-discharge.	Death	NA	Death rate in 1 yr 17.1% in ICS users versus 24.3% in control group.	Significant baseline differences in age, smoking prescription, and hospitalisation.Confounding by indication.Ascertainment of mortality.Unclear risk of immortal time bias.Drug dosage unclear.Actual use may differ.Industry funded.
					5. Excluded if died or re-admitted within 30 days.
Tkacova 46	55 ICS users, 90 non-users	RC	Slovakian university hospital	52	1. Documented outpatient clinic visit or an in-patient hospitalisation with primary diagnosis of COPD.2. At least one bronchodilator before the index visit or hospitalisation date.3. LTOT prescription between 1996–2002.4. Stable condition at the time of LTOT initiation.Prescribing physician recorded details of prescription at time of initiation of O₂.	Becl (n = 45), flu (n = 6), bud (n = 4)	Whether patients were or were not receiving ICS at time of LTOT initiation.	Death	Survival data from records of home O₂ vendors captured death.	HR for death with ICS use.HR 0.38 (95% CI 0.18–0.79). Adjusted HR 0.46 (95% CI 0.21–0.98), adjusted for age, sex, co-medication, P_a,O₂ and P_a,CO₂.ICS users had significantly greater 1-yr survival (46/55 or 84%) than non-users (54/90 or 60%).(Log-rank statistics p<0.05).	Significant baselines differences in use of SABA, arterial hypertension and severity of COPD.Misclassification as exposure status based on single time-period, and patients may have started on ICS after that.Risk of confounding as no adjustment for risk factors.
Vollmer 47	863 regular ICS users, 996 irregular ICS users, and 1043 non-users	RC	US Kaiser Permanente	208	1. Aged >50 yrs as of Jan 2000.2. Enroled between 1998–1999.3. Outpatient or ED visit with ICD-9 COPD code.4. ≥6 prescriptions of respiratory drugs, with ≥1 prescription dispensed Sept–Dec 1999.	NA	EMR.One or more dispensing event during each of the three preceding 2 months.	Death	Vital statistics registers of Oregon and Washington. 29.5% had incomplete follow-up.	RR for death (no asthma): never smokers: 0.47 (0.19–1.16, p = 0.10); ex-smokers: 0.76 (0.56–1.02, p = 0.07); current smokers 1.09 (0.65–1.83, p = 0.74); adjusted for age, sex, comorbidity, use of home O₂, pulmonologist, COPD hospitalisation or ED care, and propensity to use ICS.	Significant baselines differences: smoking, co-existing asthma, pulmonologist care, ED care.No verification of compliance.Incomplete follow-up.Industry funded.

PC: prospective cohort; VA: Veterans Affairs; DB: database; ICD: International Statistical Classification of Diseases and Related Health Problems; COPD: chronic obstructive pulmonary disease; ACQUIP: Ambulatory Care Quality Improvement Project; RCT: randomised controlled trial; Becl: beclometasone; triam: triamcinolone; fluni: flunisolide; flu: fluticasone; HR: hazard ratio; CC: case–control; bud: budesonide; MI: myocardial infarction; RR: relative risk; AMI: acute myocardial infarction; HTN: hypertension; DM: diabetes mellitus; HLD: hyperlipidaemia; CVD: cardiovascular disease; CMS: Center for Medicaid Services; NDI: National Death Index; NA: not available; IHD: ischaemic heart disease; ID: identification; SABA: short-acting β-agonist; MCO: managed care organisation; LABA: long-acting β-agonist; RC: retrospective cohort; SSN: social security number; DOB: date of birth; HMO: health maintenance organisation; Sal: salmeterol; CIHI: Canadian Institute of Health Information; ICU: intensive care unit; GPRD: General Practitioner Research Database; OMIS: Oxford Medical Information Systems; GP: general practitioner; LTOT: long-term oxygen therapy; P_a,O₂: arterial oxygen tension; P_a,CO₂: arterial carbon dioxide tension; ED: emergency department; EMR: Electronic Medical Record.