Drug-drug interactions (DDIs) are a major cause of preventable adverse drug reactions (ADRs) in patients admitted in intensive care units (ICU)1,2. Besides possible severe toxicity and loss of an expected therapeutic effect, DDIs can prolong ICU stays and impact patient outcomes3,4. Many critically ill patients have life-threatening diseases and multiple comorbidities. Polypharmacy is highly prevalent in this population and is associated with an increased risk of DDIs4–11. Furthermore, the first 24 hours in the critical care setting are crucial for patient outcomes12,13.

The clinical relevance of DDIs can be assessed through several databases. Among these, Micromedex Drug-Reax® (DR) and Lexi-Interact® (LI) are considered the most complete and reliable compendia14–16. They contain valuable information about the severity, management and clinical effect of DDIs, as well as their grade of reliability. This information is continuously updated as new evidence appears17–19. However, discrepancies between databases are common and hinder the work of pharmacists and intensivists in their daily clinical practice. For these reasons, detection of potential DDIs (pDDIs) implies close individualized clinical assessment. Although some DDIs are desired by the clinicians, many others require close drug monitoring, dose modification, or even drug discontinuation20.

To date, the frequency and clinical relevance of pDDIs on the first calendar day of ICU admission have not been assessed. Moreover, there is little information about the drugs most frequently involved and management options4,6,7,9. DDIs in critically ill patients are therefore of concern and many questions remain unanswered.

The main objective of this study was to determine the prevalence of pDDIs on the first calendar day of admission to the ICU and to identify the drugs most frequently involved. Another aim was to assess the concordance between two highly-used drug interaction databases (DR and LI) and how the content of the two databases changed by comparing data published in 2013 with that in 2018.

This study assessed the prevalence of pDDIs in a Spanish ICU. Clinical data were collected and analysed in 2013, and assessed again in 2018 using the same databases (DR and LI) in order to find differences on pDDIs’ detection. Our results show that both compendia are continuously updating and incorporating new interactions. However, their concordance in terms of severity and level of evidence is low.

Our findings confirm a high frequency of pDDIs among critically ill patients, although their clinical relevance is normally low or even desired by the clinicians. The analysis performed in 2013 showed that 74.4%-92.1% of the patients presented at least one pDDI, depending on the database, whereas the rate was considerably higher in 2018 (86.9-94.7%). This increase in pDDI identification could be due to several factors, such as the publication of new scientific studies, the potentiation of postmarketing pharmacovigilance programs, and the introduction of data mining to discover novel and potentially harmful DDIs21–23. The pDDI rates found in our study are higher than the 54-79% reported earlier by other authors2,4,5,7–10,24. Additionally, the number of drugs prescribed per patient on the first calendar day of admission in our ICU is generally higher than that reported in previous studies. In fact, about half of the patients received more than ten drugs, whereas in the study of Vanham et al. this subgroup represented only 16% of patients9. As polypharmacy is associated with a greater likelihood of presenting pDDIs, the high number of drugs prescribed to our patients could explain our findings.

Concerning the ability of the databases used to detect pDDIs, we noticed that LI identified more pDDIs than DR (92.8% vs 34.0% in 2013; 94.1% vs 45.6% in 2018), in keeping with data previously reported by Hasan et al., who reported that 88% of pDDIs were identified by LI in comparison with the 33% detected by DR10. Additionally, we note that Vanham et al. detected only 13% of potential DDIs with the three compendia they used (Epocrates, Stockley and Micromedex)9. Likewise, our study also shows a poor percentage of concordance between DR and LI (27.4% in 2013; 38.1% in 2018). The higher percentage of concordance detected in our study could be due to the fact that we only used two compendia.

Regarding the level of evidence, we found that this was poorer in 2018 than in 2013, possibly attributable to the publication of new studies showing controversial results and the extrapolation of known data from similar drugs. Our results also show that, in 2013, DR identified fewer pDDIs than LI, but the level of evidence was higher. Conversely, in 2018, the percentage of pDDIs with excellent/good level of evidence was higher in LI. The aforementioned differences could be explained by differences between databases regarding inclusion criteria. Considering our findings, and given the lack of concordance between both databases, we would recommend using more than one database when performing the clinical assessment.

We focused specifically on the 10 most common pDDIs detected by DR and LI. These accounted for more than 25% of the total. None of them were classified as contraindicated, either by DR or by LI. In fact, some of them involved drugs used to regulate analgosedation (morphine-midazolam, morphine-propofol, midazolam-propofol), the clinical effect that clinicians aimed to achieve. It is worth noting that opioid analgesics, particularly morphine, were the main group of drugs involved in pDDIs, similarly to the findings described by Hasan et al.10. It should be borne in mind that we did not include either cardiac or cardiothoracic care unit patients. As expected therefore, the rate of pDDIs due to antiplatelet and anticoagulant drugs was low in comparison with other studies7,18. As previously mentioned, we found that the 10 most common pDDIs accounted for 26.9% (DR) and 37.8% (LI) of the total. Comparing our rates with the 17.5% reported by Smithburger et al., our percentage is markedly higher25. This difference could be due to the greater rate of drug-drug pairs/pDDIs detected in Smithburger's study. They found a rate of 65.0% (297/457), whereas the rate observed in our study was 42.7% (438/1,025) in 2013.

It should be emphasized that in 2013, LI identified 16 contraindicated pDDIs while DR detected only 9. Strikingly, only two of these of these pDDIs were typified equally in both databases (cyclosporine-simvastatin and haloperidol-metoclopramide). The concomitant use of cyclosporine and simvastatin increases the risk of myopathy or rhabdomyolysis, whereas the interaction between haloperidol and metoclopramide is associated with an enhanced risk of extrapyramidal reactions and neuroleptic malignant syndrome. If the cyclosporine-simvastatin pDDI is detected, it is advised to switch to a less sensitive statin to this interaction, such as pravastatin or fluvastatin, or to an alternative type of LDL-lowering medication. However, the patient taking simvastatin and cyclosporine concomitantly was a hearttransplanted patient who received both drugs as chronic medication. Consequently, despite the interaction detected, simvastatin was not changed. With regard to the interaction between haloperidol and metoclopramide, it should be borne in mind that haloperidol is frequently administered as a single-dose, so this interaction is normally harmless and no interventions are needed. It should be noted that although no modifications in the treatment were made, patients were continuously and prospectively followed and monitored by clinical pharmacists along with the medical team.

The major strength of our study is that we evaluated the detection of pDDIs by two databases in two different years. In 2013, we prospectively collected the medication profiles of the patients included, which were analysed to identify pDDIs in that year and reassessed in 2018. Thereby, we were able to assess how databases were updated. The study also has limitations that should be taken into account. First, some drugs, such as urapidil, did not appear in either of the two databases tested. Others were included in only one of them (such as dexketoprofen, which was only included in LI) or only in one of the years of analysis (metamizole, for example, was only found in DR in 2018). Second, we did not use all the databases available, although we selected those considered most reliable and complete. Third, we did not include cardiac or cardiothoracic patients given that they are not assigned to intensive care clinicians in our hospital. Therefore, some drugs with significant potential interactions may have been excluded. Finally, we tested only the pDDIs caused by the drugs given during the first complete calendar day of admission in the ICU. Consequently, drugs prescribed afterwards were not evaluated. However, we consider that in the first day of admission pharmacotherapy is complex enough to highlight the incidence of pDDIs. Our results are not therefore comparable to those reported from studies carried out for other intervals of time9.

In conclusion, our results show that the prevalence of pDDIs in the critical care setting is high, although their clinical relevance is generally harmless and manageable by clinicians. Our data also show a lack of concordance between DR and LI, as well as their continuous updating. Interpreting the overwhelming amount of information provided by the two compendia is a clinical challenge in daily clinical practice. To tackle this issue, it is mandatory to perform an individualized assessment of the pDDIs identified, taking into account the information given by the databases along with the clinical situation of the patient.

No funding.

We thank Carolyn Newey for her linguistic revision of the paper.

No conflict of interests.

This study provides an analysis of the information found by using two databases about drug-drug interactions in critically ill patients.

The results show the importance of using several databases for interaction analyses.