Our study was performed to evaluate the implementation of laboratory monitoring in patients treated with ACEI/ARB/Diuretic plus NSAID. Our results show that despite well-known potential biochemical disturbances, serum creatinine and potassium monitoring were recorded in less than 11% of patients at risk. Monitoring occurred at day 8 and was more frequently performed to women aged over 60, treated with potassium supplements or glucose lowering drugs, but not to patients treated with ACEI, ARBs or diuretics. Monitoring was more frequent when NSAIDs' prescriber was a cardiologist or anesthesiologist.
The characteristics of NSAID prescriptions are close to the ones found in a monthly prevalence descriptive study leaded in the same area in 2006 
. First prescriptions of NSAIDs among hypertensive patients do not differ from NSAIDs prescriptions in general population, as the more frequently prescribed NSAIDs are in both studies ibuprofen, ketoprofen, diclofenac and piroxicam. Our study adds information on specificities of NSAIDs' prescriptions among prescribers. Among GPs, the rate of piroxicam prescriptions remains high, but is decreasing compared to previous studies performed in the same area 
. This phenomenon may reflect the recent recommendation from the Haute Autorité de Santé
(equivalent of the National Institute for Health and Clinical Excellence in France), underlining that piroxicam remains a second-line NSAID 
in its main indications. We also found that dentists' prescriptions were preferentially ibuprofen, the NSAID commonly prescribed for its anti-inflammatory and analgesic effect in acute dental pain 
. This information is reassuring, as low dose ibuprofen is believed to be (with naproxen) the least harmful NSAID regarding cardiovascular events 
To our knowledge, this study is the first one describing monitoring of serum creatinine and potassium in patients at risk of renal failure or hyperkalemia caused by NSAID DDIs with ACEI, ARBs or diuretics. The rate found in our study (around 11%) is unsurprisingly low. Low monitoring rates have been found in previous study whether in RASIs initiation (34% of control in the first 3 weeks 
) or with chronically prescribed ARBs/ACEIs/diuretics (68 to 72% of annual control 
). Furthermore, in our study the level of DDIs risk is not associated with a greater control. In Bootsma et al.
, being under NSAIDs was not associated to an adequate control either in patients starting ACEI/ARB therapy.
This study underlines the important lack of implementation of guidelines for DDIs between NSAIDs and antihypertensives. This finding is quite ambiguous, as GP have previously reported their concerns about NSAIDs safety of use in daily practice and claimed a caution approach in NSAID prescription 
. As an explanation to this phenomenon, two approaches can be considered focusing on guideline-related factors and GPs-related factors 
. Concerning the quality of the interaction compedia, one should underline that the main one, provided by the French Drug Agency, is available online 
. The concise information provided in this guideline is used by the main drug databases (especially the French National Formulary: Vidal
) and thus in the main medical software, which automated prompts and alerts have already demonstrated positive effects on decreasing preventable adverse drug events 
. The main limitation of the recommendations could be the absence of explicit time frame in which the monitoring should be performed. The impact of this lack of precision remains uncertain. Moreover, the recommendations are different in other compendia. Surprisingly, the British National Formulary
emphasizes on the increased risk of nephrotoxicity of the association between NSAIDs and ACEIs/ARBs/diuretics, 
but does not provide recommendations of laboratory monitoring. This lack of consistency between drug interaction compendia has already been raised 
and underlines the necessity for their standardization.
Regarding GPs-related factors for the non-implementation of drug prescribing guidelines, GPs may consider guidelines as too stringent in general. They consider laboratory monitoring as time-consuming, especially when they are uncertain that monitoring was already performed by another provider 
. GPs also raise concerns about the real impact of computerized clinical decision support to increase implementation of guidelines, as a phenomenon of alert fatigue could occur. Weingart et al.
recently emphasized on the necessity for computerized alerts to be adapted to clinicians.
In the present study, cardiologists and anesthesiologists prescribed more frequently adequate monitoring. This phenomenon can be explained by an increased prescription of flurbiprofen within these two medical specialties. Flurbiprofen is marketed in France for prevention of reinfarction and reocclusion after successful thrombolysis or angioplasty in acute myocardial infarction, in patients for whom aspirin is not recommended 
. Thus, these patients could have more frequent monitoring because of their condition. Another explanation could be that these medical specialties are more aware of the risk evaluated in the present study.
The use of the French Health Insurance Reimbursement Database in pharmacoepidemiological studies has already been fully described 
, but it implies some limitations. As for many administrative databases, we did not have access to medical characteristics of the patients. This involves using medications as proxies of morbidities (e.g. glucose lowering drugs for diabetes mellitus 
). We were not able to extract some characteristics associated to serum creatinine and potassium monitoring in a previous study 
, because of database limitations. In this study of Raebel et al.
, increasing number of outpatient visits and diagnoses of chronic heart failure or kidney disease were associated to annual monitoring. Furthermore, the disease necessitating NSAID prescription could alone be a condition implying a monitoring of serum creatinine and potassium (e.g. renal colic 
). Moreover, the database only records monitoring that have been performed and not all the ones that have been prescribed. A lot of patients-related situations (reluctance to blood test, doctor shopping, excessive self-confidence towards adverse drug reactions…) could have an impact on the realization of monitoring in a reasonable time frame.
Finally, the low prevalence of complete monitoring could have been underestimated. In our study, we only have access to ambulatory biochemical monitoring and thus could have missed the ones realized during hospitalizations. On the other hand, one could have underestimated the prevalence of ibuprofen and aspirin consumption, as these specific NSAIDs can be sold out-of-the-counter and thus not recorded in the French Health Insurance Reimbursement Database.
The low prevalence of serum creatinine and potassium monitoring shows a very poor implementation of guidelines. Further studies are required to correlate this low prevalence with a potential increased risk of severe adverse drug reactions. Moreover, intervention studies are required to improve the knowledge of this specific risk, especially among GPs.