Author + information
- Received May 2, 2008
- Revision received July 22, 2008
- Accepted August 6, 2008
- Published online November 11, 2008.
- ↵⁎Reprint requests and correspondence:
Dr. Luis A. García Rodríguez, CEIFE, Almirante 28 (2°), Madrid 28004, Spain
Objectives We studied the association between the frequency, dose, and duration of different nonsteroidal anti-inflammatory drugs (NSAIDs) and the risk of myocardial infarction (MI) in the general population. We verified whether the degree of inhibition of whole blood cyclooxygenase (COX)-2 by average circulating drug levels can be a surrogate biochemical predictor of the risk of MI by NSAIDs.
Background There is evidence that both traditional NSAIDs and selective inhibitors of COX-2 may increase the risk of MI.
Methods From the THIN (The Health Improvement Network) database, we identified 8,852 cases of nonfatal MI in patients 50 to 84 years old between 2000 and 2005 and conducted a nested case-control analysis. We correlated the risk of MI with the degree of inhibition of platelet COX-1 and monocyte COX-2 in vitro by average therapeutic concentrations of individual NSAIDs.
Results The risk of MI was increased with current use of NSAIDs (relative risk [RR]: 1.35; 95% confidence interval [CI]: 1.23 to 1.48). The risk increased with treatment duration and daily dose. We found a significant correlation between the degree of inhibition in vitro of whole blood COX-2 (r2 = 0.7458, p = 0.0027), but not whole blood COX-1 (r2 = 0.0007, p = 0.947), and the risk of MI associated with individual NSAIDs that lacked complete suppression (≥95%) of platelet COX-1 activity. Individual NSAIDs with a degree of COX-2 inhibition <90% at therapeutic concentrations presented an RR of 1.18 (95% CI: 1.02 to 1.38), whereas those with a greater COX-2 inhibition had an RR of 1.60 (95% CI: 1.41 to 1.81).
Conclusions Our findings suggest that the variable risk of MI among NSAIDs that do not inhibit platelet COX-1 completely and persistently is largely related to their extent of COX-2 inhibition.
Nonsteroidal anti-inflammatory drugs (NSAIDs), a chemically heterogeneous group of agents that comprises traditional (t)NSAIDs and NSAIDs selective for cyclooxygenase (COX)-2 (coxibs), are commonly used in the general population for treating pain and inflammatory conditions. They act mostly through the inhibition of COX-2–dependent prostanoids (1).
NSAIDs are distinguished on the basis of their COX-isozyme selectivity in vitro, described as the ratio of the concentrations required to inhibit the activity of the isozymes by 50% (IC50 for COX-1/IC50 for COX-2). This is assessed using the human whole blood assays that evaluate the effects of drugs on platelet COX-1 and monocyte COX-2 (2,3). They are capacity indexes of COX-isozyme activities to generate prostanoids from endogenous sources of arachidonic acid, and their pharmacological inhibition is not influenced by different pathological conditions. Thus, the dose of aspirin for cardioprotection selected by the assessment of thromboxane (TX) B2 levels in the whole blood assay in healthy and young subjects (4) was appropriate also for elderly patients with cardiovascular (CV) disease (5).
The assessment of COX-1/COX-2 ratios in vitro describes an experimental COX-isozyme selectivity that mirrors the chemical features of the different NSAIDs. It showed that COX-2 selectivity of NSAIDs is a continuous variable preventing separation of tNSAIDs from coxibs (6). The whole blood assays also permit estimation of achieved COX-isozyme selectivity in humans, which is the ratio of isozyme inhibition at a given plasma concentration. Importantly, achieved selectivity of NSAIDs varies as a consequence of the dose administered.
The introduction of NSAIDs selective for COX-2, which were developed to reduce the risk of serious gastrointestinal complications—dependent, at least in part, on the inhibition of COX-1—while achieving comparable efficacy, has raised new concerns now centered on CV safety (7). Originally, these concerns were reported solely for coxibs. Subsequently, we have learned that some tNSAIDs may share a CV risk similar to selective COX-2 inhibitors. The body of evidence consists of data from clinical trials (8) and from a growing number of observational studies, most of which have been performed using large automated databases (9).
Inhibition of TXA2-dependent platelet function in vivo occurs when platelet COX-1–dependent capacity to synthesize TXA2 (as assessed by measuring serum TXB2 levels) is reduced ≥95% (10). In fact, recent findings suggest that local release of tiny concentrations of TXA2 from activated platelets may play an important role in platelet thrombus formation. They activate the tyrosine-kinase–based signaling pathway (11), which may translate into full platelet activation in the presence of weak platelet agonists or subthreshold concentrations of stronger agonists. This leads to the concept of functional COX-2 selectivity by NSAIDs, namely, inhibition of COX-2 in the presence of an insufficient reduction of platelet COX-1 activity to translate into inhibition of platelet function.
In the present study, we used data from the THIN (The Health Improvement Network) database to evaluate the association between prospectively collected information on the frequency, dose, and duration of different types of NSAIDs and the risk of nonfatal myocardial infarction (MI) in the general population. Additionally, we verified the functional COX-2 selectivity by average circulating concentrations of the doses of tNSAIDs and coxibs, mostly taken from the population of the THIN database, and found that most tNSAIDs were as COX-2 selective as coxibs with respect to platelet function. Then, we aimed to address the hypothesis that the degree of inhibition of whole blood COX-2 in vitro by plasma concentrations corresponding to the average NSAID therapeutic dose in patients (an index of drug potency/exposure) (12–14) predicts the relative risk (RR) of MI for each individual NSAID functionally selective for COX-2 observed in the general population.
A detailed description of the methods used in the THIN analysis can be found in the Online Appendix.
THIN nested case-control study
We conducted a population-based, retrospective cohort study with nested case-control analysis using data from the THIN database in the United Kingdom (15). The study cohort included patients ages 50 to 84 years between January 2000 and October 2005 with at least 2 years of enrollment with the general practitioner and at least 1 health contact in the previous 3 years before their start date. Patients with cancer were excluded from the cohort. All members of this cohort were followed up from the start date until the earliest occurrence of one of the following end points: MI detection, cancer, 85th birthday, death, or end of study period. The final cohort consisted of 716,395 persons followed up for an average of 4.1 years.
We applied similar methods of case ascertainment and validation as performed in 2 recent studies (16,17). We identified patients with a recorded diagnosis of MI during the study period (n = 12,499), resulting in 10,653 patients considered incident cases of MI after review of their computerized patient profile. We then validated a random sample of 500 patients with the primary care physicians. The collaboration of the primary care physicians was excellent (response rate of 91%), and we achieved a similar high confirmation rate (95%) compared with previous studies (16,17). We retained for the analysis nonfatal incident cases of MI, defined as patients alive within 1 month after the occurrence of MI; the final number of cases was 8,852. A group of 20,000 control subjects were randomly selected from the list of eligible person-days (index date) and frequency matched to the cases on sex, age within 1 year, and calendar year.
We categorized exposure to NSAIDs into mutually exclusive time windows: current, when the supply of the most recent prescription lasted until index date or ended in the 7 days before the index date; recent, when it ended between 8 and 90 days before the index date; past, when it ended between 91 and 365 days before the index date; and nonuse, when there was no recorded use in the year before the index date. The category of current use was further subdivided: single, when there was use of only 1 individual NSAID within the month before the index date; multiple, when the patient received prescriptions for 2 or more individual NSAIDs within the week before the index date; and switcher, when there was use of only 1 NSAID in the week before the index date but there was use of at least 1 other NSAID in the window 8 to 30 days before the index date.
Among current single users, we evaluated the effect of NSAIDs according to duration, dose, and plasma half-life/formulation. We analyzed the risk among new users, defined as patients who were free of any NSAID use in the year preceding the first NSAID prescription after start date (18). We also performed a sensitivity analysis using a 0 to 30 days time window to define current use.
In vitro whole blood assays
We selected the specific cutoff values in milligrams between low-medium daily dose and high daily dose of NSAIDs used in the THIN population. The inhibitory effects toward platelet COX-1 and monocyte COX-2 by therapeutic concentrations of different NSAIDs (12–14) were determined in vitro using the human whole blood assays (3,4). The doses (Cmax) were as follows: naproxen 750 mg (253 μM), ibuprofen 1,200 mg (39 μM), meloxicam 7.5 mg (3 μM), celecoxib 200 mg (1.8 μM), rofecoxib 25 mg (1 μM), indomethacin 75 mg (3 μM), etoricoxib 90 mg (3.6 μM), diclofenac 100 mg (0.8 μM), piroxicam 20 mg (16.6 μM), and etodolac SR 400 mg (26 μM). Peripheral venous blood samples were withdrawn from 10 healthy subjects (ages 22 to 35 years), who gave their written informed consent. The in vitro study was approved by the local ethics committee, and it did not need Institutional Review Board review.
We estimated the odds ratio and 95% confidence intervals (CIs) for MI associated with NSAID use compared with nonuse using unconditional logistic regression with Statview 5.0.1 (SAS Institute, Cary, North Carolina). The mechanism of control sampling used in our study (known as incidence density sampling) means that the likelihood of being selected as a control is proportional to the person-time at risk (19). Under this design, the odds ratio is an unbiased estimator of the incidence rate ratio, also commonly referred as RR. Conditional logistic regression models stratified for matching factors (age, sex, and calendar year) produced similar results (data not shown). We used a p value <0.05 as the threshold of statistical significance. We adjusted for matching variables, use of other medications (aspirin, antihypertensive drugs, statins, other lipid-lowering drugs, oral steroids, and warfarin), comorbidity (diabetes mellitus, hypertension, hyperlipidemia, rheumatoid arthritis, osteoarthritis, anemia, coronary artery disease [CAD], and cerebrovascular disease), smoking, body mass index, and health care utilization indicators. Variables that changed the estimates of NSAID use by 10% or more were included in the multivariable model. Specific stratified analyses were performed by sex, age, and history of CAD. Linear regression analysis was performed using InStat (GraphPad, San Diego, California) to estimate correlations between relative risk obtained from the THIN study and indexes of COX inhibition of individual NSAIDs. Concentration-response curves were fitted and IC50 (drug concentration required for obtaining 50% of inhibition) values were analyzed with PRISM (GraphPad).
The overall incidence rate of nonfatal MI in our population, ages 50 to 85 years, was 4.1/1,000 person-years. The incidence was much greater among patients with antecedents of CAD (13.9) than among those without a history of CAD (3.0). Table 1 shows the risk associated with use of NSAIDs. The estimate of RR of MI with current single NSAID use was 1.35 (95% CI: 1.23 to 1.48). NSAID users who stopped treatment between 3 months and 1 year in advance had a risk similar to that of nonusers (RR: 1.02, 95% CI: 0.94 to 1.12). The RR of MI was 1.13 (95% CI: 0.92 to 1.39) among current users in their first month of treatment and 1.53 (95% CI: 1.28 to 1.82) when taking NSAIDs for >3 years. There was also a greater risk among users of high dose compared with users of low-medium dose (RR: 1.28, 95% CI: 1.07 to 1.53).
Table 2 shows RRs of MI according to plasma half-life and formulation. Users of slow-release formulations had an RR of 1.22 (95% CI: 0.96 to 1.54) compared with users of the regular formulation. When daily dose was taken into account, the slightly greater risk among users of the slow-release formulation was only observed in the low-medium dose category.
Figure 1A shows the estimates of RR for individual NSAIDs. Most of them presented an RR compatible with a small increased risk or no increased risk, although the amount of information was insufficient to provide precise estimates. Only diclofenac and rofecoxib use were clearly associated with a greater risk of MI (RR: 1.67, 95% CI: 1.44 to 1.94; and RR: 1.46, 95% CI: 1.10 to 1.92, respectively).
In Figure 1B, we reported the degree of inhibition on whole blood COX-1 and -2 activities in vitro produced by therapeutic concentrations of individual NSAIDs. Except for naproxen and ibuprofen, all other NSAIDs inhibited COX-2 more profoundly than COX-1 at therapeutic concentrations. Importantly, platelet COX-1 was suppressed at the functional range, namely, ≥95% (10) by therapeutic doses of naproxen but no other NSAIDs.
In Figure 1C, we reported a statistically significant correlation (r2 = 0.7458, p = 0.0027) between the degree of inhibition of whole blood COX-2 in vitro produced by average circulating therapeutic concentrations and the RR of MI associated with individual NSAIDs: naproxen, which has an effect on platelet COX-1 activity compatible with inhibition of platelet function, was not included. When we grouped individual NSAIDs with a degree of COX-2 inhibition <90% at therapeutic dose (ibuprofen, meloxicam, celecoxib, and etoricoxib), users of these NSAIDs presented an RR of 1.18 (95% CI: 1.02 to 1.38), whereas users of rofecoxib, indomethacin, diclofenac, and piroxicam (COX-2 inhibition ≥90%) had an RR of 1.60 (95% CI: 1.41 to 1.81, p for interaction <0.01).
The effect of dose for individual NSAIDs is shown in Table 3; among users of ibuprofen, the RR for doses up to 1,200 mg daily was 1.00 (95% CI: 0.80 to 1.25), and it was 1.56 (95% CI: 0.90 to 2.71) for doses above 1,200 mg (mainly 1,800 mg). Accordingly, average drug concentrations after dosing with low and high doses of ibuprofen (39 and 111 μM, respectively) (8) inhibited whole blood COX-2 activity by 75% and 90%, respectively. Figure 2 shows in more detail the effect of dose among users of diclofenac. A clear dose-response was observed (p for trend <0.0001): users of diclofenac 50 mg daily carried an RR of 1.12 (95% CI: 0.57 to 2.19), and the corresponding estimate among users of 150 mg was 1.80 (95% CI: 1.49 to 2.18). Also, the risk was slightly greater when diclofenac was administered as slow release compared with the plain form, even after adjusting for the dose.
We looked at whether the effect of aspirin on MI was affected by concomitant use of NSAIDs (Table 4). None of the terms for interaction was statistically significant, although only concomitant use with ibuprofen and naproxen suggested a minor antagonism: the corresponding estimates of relative excess risk due to interaction were 0.16 and 0.22.
When we restricted the analysis to new users of NSAIDs, the overall estimates of risk were similar according to duration and daily dose. Also, there were only minor variations in risk among individual NSAIDs; the corresponding RR among new users of rofecoxib was 1.57 (95% CI: 1.17 to 2.11). No differences were observed in the estimates of RR with a more liberal definition of current use (supply of NSAID lasting until index date or ending in the month before index date) according to recency, duration, or daily dose of NSAIDs (data not shown). We also performed stratified analyses by sex, age, and history of CAD. The overall effect of NSAID use did not seem to be modified by sex, although the RR was slightly higher among females but not statistically significant (data not shown). The relative risk decreased with advancing age. Users of NSAIDs who were 50 to 59 years old presented an RR of 1.61 (95% CI: 1.27 to 2.04); the corresponding estimates among patients ages 60 to 74 and 75 to 84 years old were 1.34 (95% CI: 1.18 to 1.53) and 1.22 (95% CI: 1.03 to 1.45), respectively. The RRs of MI associated with NSAID use in patients with and without antecedents of CAD were 1.18 (95% CI: 0.98 to 1.42) and 1.41 (95% CI: 1.27 to 1.47), respectively. Of note, all stratum-specific CIs within each covariate were overlapping.
In this nested case-control study with 8,852 incident nonfatal MI cases, we found that patients taking NSAIDs had a 35% increased risk of MI. The excess risk of MI was observed after 1 month of treatment and appeared to slightly increase with longer treatment duration. In addition to variable pharmacodynamic features, members of the NSAID class share different pharmacokinetics, such as half-life, which was another independent predictor of the CV hazard. Also, increasing daily dose, which determines the extent of COX-1 and -2 inhibition in vivo, was a clear predictor of the corresponding risk of MI. We were able to study the effect of dose and formulation in more detail with diclofenac: a gradual increase in risk accompanied each increasing dose of diclofenac as used in the general population. Over and above the effect of dose, data for slow-release formulations of diclofenac suggested a greater risk of MI, probably as a direct consequence of prolonged drug exposure. This level of risk was greater than it was for any of the ones shared by coxibs and reinforces the approach of analyzing the CV risk of each member of the large NSAID family individually (20). The incidence of MI was much greater among patients with antecedents of CAD than it was among patients without a history of CAD. This finding is in agreement with the results of the recent celecoxib meta-analysis by Solomon et al. (21) and concurrently by Arehart et al. (22).
We found little evidence for a major effect modification of the antiplatelet effect of aspirin among users of NSAIDs (23). Data were suggestive of a potential reduction of the beneficial effect of aspirin only when taken together with ibuprofen and naproxen, although there was substantial statistical imprecision when examining the presence of this interaction with individual NSAIDs. Yet, this is consistent with the results of clinical pharmacology showing that ibuprofen and naproxen can interfere with the irreversible inhibition of platelet COX-1 by aspirin (24–26).
We showed that pharmacodynamic and pharmacokinetic features of individual NSAIDs are all important determinants of CV hazard by COX inhibitors in the general population. Among the NSAIDs studied, only naproxen and ibuprofen were more potent toward whole blood COX-1 than COX-2 in vitro (6,27) (Fig. 2, Online Figs. 1A and 1B). Thus, average therapeutic plasma concentrations of these 2 NSAIDs may be associated with a balanced inhibition of COX-1 and -2 (Fig. 1B). In contrast to low-dose ibuprofen, naproxen and high-dose ibuprofen may suppress platelet COX-1 at the functional range, namely, ≥95% (10). Complete and perpetual suppression of platelet COX-1 activity is mandatory to fulfill cardioprotection because of a nonlinear relationship between inhibition of platelet TXA2 generation and inhibition of TXA2-mediated platelet aggregation: thus, an excess of 95% inhibition of COX-1 activity is required to influence platelet function. In fact, even tiny concentrations of TXA2 can activate platelets (11). Naproxen and ibuprofen are characterized by different pharmacokinetics; in other words, naproxen has a long half-life (>12 h), whereas ibuprofen has a short-half life (≈2 h) (12). Thus, complete and persistent suppression of platelet COX-1 activity may occur in some users of naproxen (27,28), but will be absent among users of low-dose ibuprofen (the most prevalent use in the general population) and rare among users of high-dose ibuprofen. This impact on platelet COX-1 by naproxen might mitigate, even if not obliterate, the hazard conferred by the profound inhibition of COX-2–dependent prostacyclin (29,30). In fact, naproxen was associated with no increased CV risk. All other studied tNSAID and coxib results were more potent toward COX-2 in vitro, from 3- to 255-fold (rofecoxib) (Online Fig. 2). However, none presented a pattern of COX-1 inhibition compatible with prevention of CV events (10). These results showed that most tNSAIDs are as COX-2 selective as coxibs with respect to platelet function, at therapeutic concentrations. For functional COX-2 selective inhibitors, the degree of inhibition of whole blood COX-2 by plasma concentrations corresponding to the average NSAID therapeutic dose in patients (12–14), which is an index of drug potency/exposure, predicted the RR of MI. When we replaced our estimates of risk for individual NSAIDs with those reported in a recent meta-analysis of observational studies (31), the correlation was still acceptable (r2 = 0.7445, p = 0.1372), although with less power as data were reported for only 4 NSAIDs.
Analyses of previous clinical studies with rofecoxib 25 mg/day, celecoxib 200 mg twice a day, etoricoxib 90 mg/day, naproxen 220 mg twice a day, and low-dose aspirin 100 mg/day allowed us to detect a statistically significant relationship between the degree of inhibition of whole blood COX-2 ex vivo and inhibition of prostacyclin in vivo (as assessed by the urinary levels of 2,3-dinor-6-keto-PGF1α) (27,28) (P. Patrignani and S. Tacconelli, unpublished data obtained with celecoxib, rofecoxib, and etoricoxib, June 2006) (Online Fig. 3). This relationship supports the notion that the CV hazard associated with the administration of coxibs and some tNSAIDs occurs through a common mechanism involving the inhibition of COX-2–dependent prostacyclin (29,30). Failure to detect any correlation (r2 = 0.0007, p = 0.947) between the degree of inhibition of whole blood COX-1 in vitro and the risk of MI by the same NSAIDs (Online Fig. 4) lends little support to one hypothesis suggesting that stronger inhibition of endothelial COX-1 than platelet COX-1 could be the mechanism for the increased risk of thrombotic events by tNSAIDs and coxibs (32). In the same manner, after leaving out naproxen, which presents some functional suppression of platelet COX-1, the correlation observed between the achieved COX-2 selectivity (ratio of degree of inhibition of COX-2 and -1 at average plasma concentrations) and the risk of MI was very poor (r2 = 0.028, p = 0.665) (Online Fig. 5).
Another aspect to be considered is the consequence of the inhibition of renal prostanoids by NSAIDs. In susceptible patients, inhibition of prostanoid generation by tNSAIDs and coxibs causes sodium retention with resulting edema and hypertension (29). Similar to thrombogenesis, COX-1 inhibition may mitigate the consequence of COX-2 inhibition in the kidney. This is compatible with a COX-2–dependent source of vasodilatory prostacyclin and COX-1–dependent origin of vasoconstrictors, like TXA2, in the kidney (29,33).
A few considerations need to be evaluated with respect to the validity of our findings. A potential limitation of studies using computerized prescription data is underascertaining over-the-counter drug use. In our study, misclassification of NSAID exposure is expected to be small, as only ibuprofen could be purchased over the counter in the United Kingdom during the study period. This misclassification would tend to be nondifferential, which, given the small magnitude of the observed association, would have little or no effect on our results. Also, we did not have information on over-the-counter use of aspirin (mainly, short-term analgesic use). Yet the degree of underestimation of aspirin use would be limited in the current study, as prescriptions are free for patients >60 years of age in the United Kingdom. That means the majority of patients requiring cardioprotective aspirin would be likely to receive it on prescription rather than have to pay for the drug over the counter, as reflected by the high prevalence of current low-dose aspirin use in our control series (18%). Myocardial infarction cases were ascertained through review of computerized files with free text comments available in a subset. We tested the validity of our case ascertainment in a random sample of close to 500 questionnaires sent to the general practitioners, resulting in a high confirmation rate of 95%. Actually, the overall incidence of MI in our source population ages 50 to 85 years was slightly over 4 of 1,000 person-years, in line with rates reported in previous studies (17).
We propose that the extent of inhibition of COX-2–dependent prostacyclin may represent an independent key determinant of the increased risk of MI among NSAIDs with nonfunctional suppression of platelet COX-1, a property shared by most tNSAIDs and coxibs, and that the assessment of whole blood COX-2 may represent a surrogate end point to predict the CV risk of these drugs. It follows that the separation of NSAIDs into selective COX-2 and nonselective COX-2 inhibitors adds little in predicting the CV risk of NSAIDs, after taking into account the dose potency (exposure) of the specific NSAID.
The authors thank the general practitioners for their excellent collaboration. The authors also thank Ms. Elvira Massó, MSc, and Elisa Martín, BPharm, for their help in the validation process and analysis of the THIN portion, and Paola Anzellotti, PharmD, for performing the radioimmunoassays. They did not receive compensation for their contribution.
For a supplemental Methods section and supplemental Figures 1 to 5, please see the online version of this article.
The database portion (THIN) was funded by an unrestricted research grant from Pfizer to CEIFE. The corresponding author has full authorship rights to the manuscript and was not obligated to include any Pfizer comments in the submitted manuscript. The biochemistry study was supported by a grant from the European Community's Sixth Framework Program (Eicosanox, LSMH-CT-2004-005033) to Dr. Patrignani.
- Abbreviations and Acronyms
- coronary artery disease
- confidence interval
- selective inhibitors of COX-2
- myocardial infarction
- nonsteroidal anti-inflammatory drug
- relative risk
- traditional nonsteroidal anti-inflammatory drug
- Received May 2, 2008.
- Revision received July 22, 2008.
- Accepted August 6, 2008.
- American College of Cardiology Foundation
- Burke A.,
- Smyth E.,
- FitzGerald G.A.
- Patrignani P.,
- Panara M.R.,
- Greco A.,
- et al.
- Kearney P.M.,
- Baigent C.,
- Godwin J.,
- Halls H.,
- Emberson J.R.,
- Patrono C.
- Reilly I.A.,
- FitzGerald G.A.
- Minuz P.,
- Fumagalli L.,
- Gaino S.,
- et al.
- Brunton L.L.,
- Lazo J.S.,
- Parker K.L.
- Huntjens D.R.,
- Danhof M.,
- Della Pasqua O.E.
- Garcia Rodriguez L.A.,
- Varas-Lorenzo C.,
- Maguire A.,
- Gonzalez-Perez A.
- Ray W.A.
- Walker A.M.
- Solomon S.D.,
- Wittes J.,
- Finn P.V.,
- et al.
- Arehart E.,
- Stitham J.,
- Asselbergs F.W.,
- et al.
- Corman S.L.,
- Fedutes B.A.,
- Ansani N.T.
- Capone M.L.,
- Sciulli M.G.,
- Tacconelli S.,
- et al.
- Capone M.L.,
- Tacconelli S.,
- Sciulli M.G.,
- et al.
- Capone M.L.,
- Tacconelli S.,
- Sciulli M.G.,
- et al.
- Mitchell J.A.,
- Lucas R.,
- Vojnovic I.,
- Hasan K.,
- Pepper J.R.,
- Warner T.D.