Aliment Pharmacol Ther 2012; 35: 165–174
Background Cytochrome P450 inhibition by proton pump inhibitors (PPIs) may attenuate the effectiveness of clopidogrel.
Aim To examine whether PPI use modifies the association between clopidogrel use and major adverse cardiovascular events (MACE) after percutaneous coronary intervention (PCI) with stent implantation, using time-varying drug exposure ascertainment.
Methods We conducted this population-based cohort study in Western Denmark (population 3 million) using medical databases. We identified all 13 001 patients with coronary stent implantation between 2002 and 2005 and ascertained their reported comorbidities. During the recommended 12-month postintervention treatment period, we tracked use of clopidogrel and PPI and the rate of MACE. We used Cox regression to compute hazard ratios (HRs), controlling for potential confounders.
Results During follow-up, one or more prescriptions were redeemed by 91% of patients for clopidogrel and by 21% of patients for PPIs. Of the patients, 15% experienced a MACE. The adjusted HR for MACE comparing clopidogrel use with non-use was 0.57 [95% confidence interval (CI): 0.44–0.74] among PPI users and 0.47 (95% CI: 0.42–0.53) among PPI non-users, yielding an interaction effect (i.e. relative rate increase) of 1.20 (95% CI: 0.91–1.58). PPI users treated from before PCI had a 25% increased rate of MACE compared to PPI non-users, independent of clopidogrel use [adjusted HR = 1.24 (95% CI: 0.97–1.58) for clopidogrel users and 1.26 (95% CI: 0.97–1.63) for clopidogrel non-users].
Conclusions The use of PPIs as a class did not modify the protective effect of clopidogrel, but its use was associated with major adverse cardiovascular events itself, particularly among patients having used PPIs before percutaneous coronary intervention.
The thienopyridine clopidogrel is currently a mainstay in tertiary prevention of vascular events in patients with coronary artery disease or ischaemic stroke.1 Clopidogrel is a pro-drug that is metabolized by hepatic cytochrome P450 (CYP) enzymes (primarily the 2C19 and 3A4 isoforms) to an active thiol metabolite, which irreversibly inhibits the binding of adenosine-5-diphosphate (ADP) to the platelet P2Y12-receptor. Patients with high residual ADP-inducible platelet reactivity are at increased risk of major adverse cardiovascular events (MACE) after percutaneous coronary intervention (PCI).2
Several proton pump inhibitors (PPIs) are metabolized by CYP2C19 and thus may interact with clopidogrel metabolism.3 Intense debate is ongoing about whether the diminished ex vivo antiplatelet effect of clopidogrel through concomitant PPI use translates into adverse clinical outcomes.3, 4 The importance of this interaction arises from the large number of PCIs performed annually, the increasing use of drug-eluting stents with the associated necessity for long-term clopidogrel treatment,1 and the possibility of preventing an adverse interaction by avoiding co-administration of PPIs.3
No population-based study has examined the clinical outcome of the clopidogrel-PPI interaction in patients receiving coronary stents, with assessment of clopidogrel and PPI use in a manner that accounts for discontinuation and restart of therapy, allowing clinical quantification of the interaction effect.5 To clarify these issues, we examined in detail whether PPI use modified the association between clopidogrel use and MACE after coronary stent implantation, and whether clopidogrel users were at increased risk of MACE when concomitantly administered a PPI.
We conducted this population-based cohort study using medical databases in Western Denmark, which has 3 million inhabitants (55% of the Danish population). The Danish National Health Service provides universal tax-supported health care, guaranteeing unfettered access to general practitioners and hospitals, and partial reimbursement for prescribed medications, including clopidogrel and PPIs. Accurate and unambiguous linkage of all registries at the individual level is possible in Denmark using the unique central personal registry number assigned to each Danish citizen at birth and to residents upon immigration.6
Patients and procedures
We used the Western Denmark Heart Registry (WDHR) to identify all PCIs performed between 1 January 2002 and 30 June 2005.7 Since 1999, this registry has collected patient and procedure data from all cardiac intervention centres in Western Denmark.7 We defined the first PCI during the inclusion period as the ‘index PCI’ and the date of the procedure as the ‘index date’. We did not include patients treated by balloon angioplasty without stenting.
Participating centres are high-volume facilities, each performing more than 1000 PCIs per year. The interventions were performed according to current standards, with the interventional strategy (including balloon angioplasty, pre- or postdilatation, choice of stent, use of direct stenting, and administration of periprocedural glycoprotein IIb/IIIa inhibitor) left to the operator’s discretion.7
We used the Danish Nationwide Prescription Database (DNPD)8 to identify all redeemed prescriptions for clopidogrel and PPIs.9 Thienopyridines and PPIs were available by prescription only during the study period. As no prescriptions were filled for ticlopidine, no alternative ADP receptor inhibitor was included in the study. Relevant Anatomical Therapeutic Chemical codes are provided in the Appendix S1.
The recommended daily maintenance dose of clopidogrel for secondary prevention of ischaemic vascular events in Denmark is 75 mg (one tablet) daily for up to 12 months.9 Thus, for study purposes the number of days supplied from a dispensed clopidogrel prescription corresponded to the number of tablets per package. Packages available in the Danish market contained 28 or 84 tablets.9 We computed the number of days exposed by adding 7 days to the number of days supplied. This buffer allowed for a 7-day gap to occur between prescription redemptions before a patient was considered to have discontinued the medication.
Use of the following PPIs was recorded in the DNPD: esomeprazole, lansoprazole, omeprazole, pantoprazole and rabeprazole.10 As in the case of clopidogrel, we computed the number of days exposed for PPIs. We defined current users of clopidogrel and PPI, individually, at a given point in time as patients exposed by the most recent prescription redeemed. In a time-varying manner, patients thus contributed time-at-risk as a current user or as a non-user of each drug.
Major adverse cardiovascular events
In line with the recommended duration of clopidogrel treatment, we identified MACE occurrences within 12 months after the index date.9 We defined MACE as a first occurrence of myocardial infarction (MI), ischaemic stroke, stent thrombosis, target lesion revascularization, or cardiac death. A committee of cardiac specialists, blinded to the history of medication use, reviewed relevant records to determine the occurrence of stent thrombosis and cardiac death, diagnoses that originally were not included in Danish medical registries.7
Myocardial infarction and ischaemic stroke
We used the Danish National Registry of Patients (DNRP), covering all nonpsychiatric hospitals since 1977 and emergency room and outpatient clinic visits since 1995, to identify admissions for MI and ischaemic stroke.11 Associated International Classification of Diseases (ICD) codes are provided in the Appendix S1.
Stent thrombosis and target lesion revascularization
Based on review of original medical records and catheterisation angiograms, the cardiac specialist committee adjudicated the occurrence of definite stent thrombosis as defined by the Academic Research Consortium.7 We defined target lesion revascularization as a repeat PCI or coronary artery bypass grafting of the index lesion, identified from the WDHR.7
We obtained data on all-cause mortality from the Danish Civil Registration System.6 This registry has recorded vital statistics – including date of birth, change of address, date of emigration, and exact date of death – for the Danish population since 1968.6 The cardiac specialist committee then reviewed original paper death certiﬁcates obtained from the National Registry of Causes of Death, which has collected data on dates and causes of death in Denmark since 1943.12 Deaths were classiﬁed as either cardiac or noncardiac, based on the underlying cause recorded on the death certificates. Cardiac death was defined as an evident cardiac death, unwitnessed death, or death from unknown causes.13
We obtained information on potential confounders (diabetes, hypertension and obesity) from diagnoses recorded in the DNRP between 1977 and the index date. To ensure complete identification of patients with diabetes, we also searched the DNPD for any use of antidiabetic drugs from 1995 to the index date. From the WDHR, we retrieved procedure-specific data, including the year of index PCI, PCI indication (ST-segment elevation MI, non-ST-segment elevation MI or unstable angina pectoris, or stable angina pectoris), number of treated arteries (1, 2, or 3 or more), number of implanted stents (1, 2, or 3 or more), lesion type (A, B1, B2, or C),14 and stent type (drug-eluting or bare-metal stent).7 We used the DNPD to obtain information on use of the following medications: aspirin, calcium channel blockers, statins, vitamin K antagonists, nonselective nonsteroidal anti-inflammatory drugs, cyclooxygenase-2 selective inhibitors, and systemic glucocorticoids. Associated ICD and ATC codes are provided in the Appendix S1.
We characterised the patients on the basis of medical, procedural and demographic variables. We followed all patients from the index date until the date of MACE, noncardiac death, emigration, or 12 months of follow-up, whichever came first. Among clopidogrel and PPI users, we examined the proportion taking medication at end of follow-up, based on the number of days exposed since the last prescription redemption. We stratified the analyses according to whether patients had initiated therapy before or after index PCI.
Time-varying exposure assessment allowed patients to be considered exposed to different medications over time: that is, clopidogrel plus a PPI, clopidogrel without a PPI, a PPI without clopidogrel, or no use of clopidogrel or a PPI. This approach permitted comparison of MACE frequency per cumulative time-at-risk associated with each of the four exposure categories. We illustrated graphically how the event rates associated with these four categories progressed relative to each other over time.
We examined whether PPIs as a class modified the association between clopidogrel and MACE, by comparing current use of clopidogrel with non-use, in subgroups of patients with or without concomitant PPI use. We used Cox proportional hazards regression to compute hazard ratios (HRs) with 95% confidence intervals (CIs). The ‘interaction effect’ is the exponentiated coefficient for the interaction term in the model, that is, the ratio of the stratum-specific HRs.5 The interaction effect estimates the relative hazard rate increase (or decrease) in patients with concomitant use of clopidogrel and a PPI, beyond that expected from the independent effects of each drug alone.5 An interaction effect other than 1.0 suggests that concomitant PPI use modifies any protective effect of clopidogrel. We used the Wald χ2 test to assess the null hypothesis of no interaction.
In regression analyses, we adjusted for the following potential confounders: age, gender, diabetes, hypertension, obesity and time-varying use (calculated from the number of days exposed) of aspirin and of calcium channel blockers and lipophilic statins (drugs with potential for interaction with clopidogrel).15 To examine the confounding impact of our measures of diabetes, hypertension, and obesity, we fit a minimally adjusted model omitting these variables. As the results from the minimally adjusted model were similar to those from the fully adjusted model, they are not further reported. We repeated the analyses stratifying by age, gender, PCI indication, and presence/absence of diabetes. We also repeated the analyses for the most commonly used PPIs (esomeprazole, lansoprazole, omeprazole and pantoprazole) for the individual outcomes included in MACE, and for presence of upper gastrointestinal bleeding. Because there were few instances of stent thrombosis, ischaemic stroke, and upper gastrointestinal bleeding, we do not report further on these individual outcomes.
Finally, we examined whether clopidogrel modified the association between PPI use and MACE, by comparing current PPI use with non-use, in subgroups of patients with or without concomitant clopidogrel use. To examine the impact of new use (starting after PCI) and longer term PPI use (starting before PCI),16 we repeated the analysis in subgroups of patients with or without one or more filled PPI prescriptions before index PCI.
We identified 13 001 patients who had undergone coronary stent implantation (Table 1). The median age on the index date was 64 years and 28% were women. The indications for PCI were STEMI for 3790 (29%) patients, non-STEMI or unstable angina pectoris for 3987 (31%) patients, and stable angina pectoris for 4876 (37.5%) patients. During follow-up, 11 859 (91%) patients filled at least one prescription for clopidogrel and 2742 (21%) filled at least one prescription for a PPI. Among patients using clopidogrel after PCI, only 45% continued treatment until end of follow-up (Table 2). This proportion was 54% among PPI users, highest among longer-term users (64%) and lowest among new users (39%).
|All patients (n = 13 001)||MACE patients* (n = 1890)|
|Female||3599 (27.7)||533 (28.2)|
|<60 years||4763 (36.6)||585 (31.0)|
|60–69 years||3949 (30.4)||825 (27.9)|
|≥70 years||4289 (33.0)||777 (41.1)|
|Clopidogrel||11 859 (91.2)||1097 (58.0)|
|Proton pump inhibitors||2742 (21.1)||271 (14.3)|
|Esomeprazole||1260 (9.7)||119 (6.3)|
|Lansoprazole||719 (5.5)||60 (3.2)|
|Omeprazole||421 (3.2)||45 (2.4)|
|Pantoprazole||765 (5.9)||65 (3.4)|
|Rabeprazole||18 (0.1)||1 (0.1)|
|Aspirin||11 231 (86.4)||906 (47.9)|
|Vitamin K antagonists||889 (6.8)||93 (4.9)|
|Nonselective NSAIDs||1409 (10.8)||75 (4.0)|
|COX-2 inhibitors||1322 (10.2)||78 (4.1)|
|Oral glucocorticoids||943 (7.3)||70 (3.7)|
|Calcium channel blockers||3016 (23.2)||244 (12.9)|
|Statins||9720 (74.8)||763 (40.4)|
|Diabetes||1390 (10.7)||267 (14.1)|
|Hypertension||389 (3.0)||76 (4.0)|
|Obesity||82 (0.6)||12 (0.6)|
|Year of study entry|
|2002||3112 (23.9)||496 (26.2)|
|2003||3722 (28.6)||561 (29.7)|
|2004||3986 (30.7)||542 (28.7)|
|2005||2181 (16.8)||291 (15.4)|
|STEMI||3790 (29.2)||862 (45.6)|
|Non-STEMI or unstable angina pectoris||3987 (30.7)||508 (26.9)|
|Stable angina pectoris||4876 (37.5)||461 (24.4)|
|Other||348 (2.7)||59 (3.1)|
|Number of treated arteries§|
|1||10 184 (78.3)||1472 (77.9)|
|2||2366 (18.2)||356 (18.8)|
|≥3||339 (2.6)||47 (2.5)|
|Number of stents§|
|1||10 761 (82.8)||1483 (78.5)|
|2||1720 (13.2)||308 (16.3)|
|≥3||458 (3.5)||93 (4.9)|
|A||2684 (20.6)||310 (16.4)|
|B||7884 (60.6)||1200 (63.5)|
|C||2427 (18.7)||380 (20.1)|
|BMS||8847 (68.0)||1428 (75.6)|
|DES||3548 (27.3)||378 (20.0)|
|BMS and DES||606 (4.7)||84 (4.4)|
|N||Drug use at end of follow-up*|
|No (%)||Yes (%)|
|Ever use of clopidogrel after PCI†||11 859||6557 (55.3)||5302 (44.7)|
|Ever use of clopidogrel before PCI‡||1698||811 (47.8)||887 (52.2)|
|No clopidogrel prescription before PCI||10 161||5746 (56.5)||4415 (43.5)|
|No clopidogrel prescription after PCI||1142||1102 (96.5)||40 (3.5)§|
|Ever use of clopidogrel before PCI‡||169||129 (76.3)||40 (23.7)§|
|No clopidogrel prescription before PCI||973||973 (100)||–|
|Ever use of a PPI after PCI†||2742||1263 (46.1)||1479 (53.9)|
|Ever use of a PPI before PCI‡||1622||578 (35.6)||1044 (64.4)|
|No PPI prescription before PCI||1120||685 (61.2)||435 (38.8)|
|No PPI prescription after PCI||10 259||10 183 (99.3)||76 (0.7)§|
|Ever use of a PPI before PCI‡||1432||1356 (94.7)||76 (5.3)§|
|No PPI prescription before PCI||8827||8827 (100)||–|
Overall, 1890 (15%) patients experienced a MACE during the 12-month follow-up. The rates of MACE per 1000 person years were 154 for concomitant clopidogrel and PPI use, 104 for clopidogrel without PPI use, 267 for PPI without clopidogrel use, and 263 for no use of either drug (Table 3 and Figure 1).
|Clopidogrel use||Unadjusted hazard ratio (95% CI)||Interaction effect‡ (95% CI)||P§||Adjusted hazard ratio¶ (95% CI)||Interaction effect‡ (95% CI)||P§|
|Any PPI**||−||973||677||263||104||0.38 (0.34–0.42)||1.28 (0.97–1.69)||0.08||0.47 (0.42–0.53)||1.20 (0.91–1.58)||0.19|
|+||102||138||267||154||0.48 (0.37–0.63)||0.57 (0.44–0.74)|
|Esomeprazole||−||1039||759||264||108||0.39 (0.35–0.43)||1.30 (0.85–2.00)||0.23||0.39 (0.35–0.43)||1.32 (0.86–2.03)||0.20|
|+||36||56||238||153||0.50 (0.33–0.77)||0.51 (0.34–0.78)|
|Lansoprazole||−||1050||787||263||109||0.39 (0.35–0.44)||1.06 (0.61–1.83)||0.85||0.39 (0.35–0.44)||1.09 (0.63–1.89)||0.75|
|+||25||28||289||138||0.41 (0.24–0.71)||0.43 (0.25–0.74)|
|Omeprazole||−||1053||796||263||110||0.39 (0.35–0.44)||1.01 (0.54–1.87)||0.99||0.39 (0.36–0.44)||1.00 (0.54–1.87)||0.99|
|+||22||19||288||145||0.40 (0.21–0.73)||0.40 (0.21–0.73)|
|Pantoprazole||−||1056||782||264||109||0.39 (0.35–0.43)||1.45 (0.82–2.56)||0.21||0.39 (0.35–0.43)||1.47 (0.83–2.60)||0.19|
|+||19||33||254||154||0.56 (0.32–0.99)||0.57 (0.32–1.01)|
The adjusted HR for MACE comparing clopidogrel use with non-use was 0.57 (95% CI: 0.44–0.74) among PPI users and 0.47 (95% CI: 0.42–0.53) among PPI non-users (Table 3), yielding an interaction effect of 1.20 (95% CI: 0.91–1.58). The results were consistent for MI, TLR, and cardiac death, as separate outcomes (Table S1). We observed no substantial difference from the overall results in subgroups based on age, gender, PCI indication, or presence/absence of diabetes (data not shown).
The adjusted HR for MACE, comparing use of PPIs as a class with non-use, was 1.40 (95% CI: 1.17–1.68) among clopidogrel users and 1.16 (95% CI: 0.95–1.43) among clopidogrel non-users (Table 4). The results were similar for individual PPIs among clopidogrel users (Table S2). The adjusted HR comparing new PPI use with non-use was 1.61 (95% CI: 1.13–2.29) among clopidogrel users and 0.98 (95% CI: 0.58–1.67) among clopidogrel non-users (interaction effect 1.64, 95% CI: 0.87–3.11). Patients with a longer-term PPI use had a 25% increased rate of MACE compared to PPI non-users, independent of whether or not they were using clopidogrel (adjusted HR 1.24, 95% CI: 0.97–1.58 for clopidogrel users and 1.26, 95% CI: 0.97–1.63 for clopidogrel non-users).
|Unadjusted hazard ratio (95% CI)||Interaction effect‡ (95% CI)||p§||Adjusted hazard ratio¶ (95% CI)||Interaction effect‡ (95% CI)||p§|
|PPI use overall|
|− Clopidogrel||1.18 (0.96–1.44)||1.28 (0.97–1.69)||0.08||1.16 (0.95–1.43)||1.20 (0.91–1.58)||0.19|
|+ Clopidogrel||1.51 (1.26–1.81)||1.40 (1.17–1.68)|
|− Clopidogrel||1.25 (0.97–1.62)||1.02 (0.71–1.45)||0.92||1.26 (0.97–1.63)||0.98 (0.69–1.40)||0.93|
|+ Clopidogrel||1.28 (1.00–1.63)||1.24 (0.97–1.58)|
|− Clopidogrel||1.01 (0.60–1.72)||1.69 (0.89–3.20)||0.11||0.98 (0.58–1.67)||1.64 (0.87–3.11)||0.13|
|+ Clopidogrel||1.71 (1.20–2.44)||1.61 (1.13–2.29)|
In this population-based cohort study of 13 001 patients undergoing PCI, clopidogrel use was associated with a markedly reduced rate of MACE within 12 months after coronary stent implantation, independent of PPI use. Use of PPIs individually or as a class did not modify the protective effect of clopidogrel substantially. However, PPIs use was associated with an increased rate of MACE itself, particularly among longer-term users.
Several studies have reported on the effect of coadministering PPIs to clopidogrel.3, 4, 17, 18 Gilard et al19 first drew attention to this potential drug-drug interaction in a placebo-controlled trial of 124 patients receiving coronary stents. They found a reduced ex vivo antiplatelet effect of clopidogrel when combined with omeprazole.19 Several other ex vivo studies supported such an effect for concomitant use of omeprazole,20–22 but not lansoprazole, pantoprazole, or esomeprazole.21–25 Recently, however, a post hoc analysis of the PRINCIPLE-TIMI 44 trial demonstrated a decreased inhibition of platelet aggregation of a 600 mg clopidogrel loading dose associated with concomitant use of all these four PPIs individually.26
Observational studies have showed similar inconsistency regarding whether concomitant clopidogrel and PPI use is27–31 or is not 26, 32–37 associated with adverse clinical outcomes. The only study that randomly allocated PPI treatment to clopidogrel users was the Clopidogrel and the Optimization of Gastrointestinal Events Trial (COGENT).38 This study concluded that there was no apparent increased cardiovascular risk by using clopidogrel and omeprazole concomitantly compared with clopidogrel alone,38 although the confidence limits allowed for a 44% relative increase in cardiovascular risk.3 Similarly, a post hoc analysis of the TRITON-TIMI 38 trial found no clinically relevant cardiovascular risk associated with concomitant clopidogrel and PPI use.26
Our finding that PPI use was associated with an increased rate of MACE among clopidogrel users is in line with three meta-analyses estimating that concomitant clopidogrel and PPI use is associated with a 30–40% increased risk for MACE compared with clopidogrel use alone.4, 17, 18 Importantly, the cardiovascular risk associated with PPI use in our study seemed independent of clopidogrel use. This finding is consistent with previous reports.32, 35, 39 The apparent increase in risk associated with PPI use may be due to the characteristics of the patients who use them.32, 35, 39 In Denmark PPIs are prescribed mainly for clear indication such as peptic ulcer disease or gastroesophageal reflux disease, and not routinely in combination with dual antiplatelet therapy. Our nonrandomized design is vulnerable to confounding by unmeasured variables (e.g. cardiovascular risk factors not routinely recorded in registry data such as smoking, alcohol use, lipid levels, and body mass index) and residual confounding from imperfect measured variables (in our study, e.g. diabetes, hypertension, or obesity) might also play a role.32, 39 Thus, it is important to note that the cardiovascular risks associated with PPI use may not necessarily reflect a direct drug effect.
One major limitation of most previous studies was an inability to quantify the isolated interaction effect on clinical endpoints.26, 27, 29–31, 33–35, 37 Initiating clopidogrel therapy may not be a valid proxy for exposure status throughout follow-up, because patients may stop and restart treatment or may discontinue completely before the end of the recommended treatment period, for example due to intolerable side effects, as it was seen for more than half of our patients. To quantify the actual interaction effect, it is therefore necessary to examine and compare the rate of MACE associated with use of clopidogrel alone, clopidogrel plus a PPI, a PPI alone, or neither. Through our time-varying clopidogrel and PPI ascertainment, we avoided the assumption that once a patient initiated a medication it was continued for the remainder of the recommended treatment period.
Strength and limitations
A number of issues should be considered when interpreting our study’s results. Its population-based design within the setting of a tax-supported universal healthcare system largely eliminated selection biases. Data in the prescription database are virtually complete.8 Although we had to use prescription data as a proxy for drug use, we based drug exposure information on actual dispensing at pharmacies.8 Copayment requirements increased the likelihood of compliance. We were able to calculate the number of days exposed from the number of days of medication supplied, increasing the accuracy of exposure information. We also accounted for patient adherence behaviour by allowing up to 7-day gaps between prescription refills.40 These advanced methods of defining exposure reduced the likelihood of nondifferential misclassification,41 despite reliance on assumptions regarding consumption of dispensed medications. Discontinuation of PPIs could be associated with poor health.42 If so, the high rate of PPI discontinuation throughout follow-up would draw the effect estimates towards unity and thus cannot explain the findings of an increased risk associated with PPI use. The results for individual PPIs and individual outcomes were limited by wide confidence intervals, making it difficult to rule out small, but potentially clinically relevant, risks. Also, we cannot not rule out that the cardiovascular risk associated with PPI use is independent of clopidogrel use32 only among longer-term PPI users and not among patients initiating PPI treatment after PCI. Use of WDHR and DNRP data to ascertain study outcomes has previously been validated,7, 43 and the DNPD has been shown to be accurate and complete.8
Information on drug use and hospitalizations were collected independently from medical databases, avoiding reliance upon self-report and thus reducing the potential for differential misclassification.44
In conclusion, use of PPIs as a class did not modify the protective effect of clopidogrel after coronary stent implantation. PPI use was, however, associated with MACE itself, particularly among patients having used PPIs before PCI.
Declaration of personal interests: JAB has served as a consultant to Bayer. Declaration of funding interests: This study was supported by the Clinical Epidemiology Research Foundation, Denmark. DJR was supported by a VA HSR&D Career Development Award. The funding sources had no role in the design, conduct, analysis or reporting of this study. The contents of this work do not necessarily represent the views of affiliated medical departments or the Danish and US governments.