Drug–Drug Interactions in Elderly Patients with Potentially Inappropriate Medications in Primary Care, Nursing Home and Hospital Settings: A Systematic Review and a Preliminary Study

Drug–drug interactions (DDI) occurring with potentially inappropriate medications (PIM) are additional risk factors that may increase the inappropriate character of PIM. The aim of this study was (1) to describe the prevalence and severity of DDI in patients with PIM and (2) to evaluate the DDI specifically regarding PIM. This systematic review is based on a search carried out on PubMed and Web-of-Science from inception to June 30, 2020. We extracted data of original studies that assessed the prevalence of both DDI and PIM in elderly patients in primary care, nursing home and hospital settings. Four hundred and forty unique studies were identified: 91 were included in the qualitative analysis and 66 were included in the quantitative analysis. The prevalence of PIM in primary care, nursing home and hospital were 19.1% (95% confidence intervals (CI): 15.1–23.0%), 29.7% (95% CI: 27.8–31.6%) and 44.6% (95% CI: 28.3–60.9%), respectively. Clinically significant severe risk-rated DDI averaged 28.9% (95% CI: 17.2–40.6), in a hospital setting; and were approximately 7-to-9 lower in primary care and nursing home, respectively. Surprisingly, only four of these studies investigated DDI involving specifically PIM. Hence, given the high prevalence of severe DDI in patients with PIM, further investigations should be carried out on DDI involving specifically PIM which may increase their inappropriate character, and the risk of adverse drug reactions.


Introduction
A drug-drug interaction (DDI) is usually defined as a clinically significant unintended modification in the exposure and/or response to a medication (i.e., victim) that occurs with the co-administration of another medication (i.e., perpetrator) [1]. Studies in the field of DDIs usually refer to a potential DDI, defined as the co-prescription of two medications known to interact, that may occur in exposed patients. DDIs can be categorized as either of pharmacokinetic origin (i.e., modification in exposure) or of pharmacodynamic origin (i.e., modification in response) [2]. A pharmacodynamic DDI results from the concomitant administration of medications that have the same sites of action, leading to additive, synergistic or antagonistic effects altering the drug effect with usually no apparent alteration in drug exposure. A pharmacokinetic DDI results from alterations in the processes involved in drug disposition when two drugs are co-administered. Basically, these DDI are due to either metabolic enzymes and/or transporters localized in membranes and tissues involved

Materials and Methods
This review was performed according to the Preferred Reporting Items for Systematic Reviews and meta-analyses (PRISMA) statement. The PRISMA check list [11] and PRISMA flow chart are available in Supplementary material 1 and 2.

Search Strategy
We conducted a systematic search of the published peer-reviewed literature in Pubmed and Web of Science databases. Duplicate manuscripts were removed after exporting references in Zotero reference management software (www.zotero.org (accessed on 13 February 2021)).
A literature search was conducted using Medical Subject Headings (MeSH) terms and text words around the topic of drug-drug interactions and potentially inappropriate medication use in older people.
The literature search was carried out from inception to June 1, 2020 without language restrictions, and non-English publications were translated prior to data extraction. Details of the full search strategy are available in Supplementary material 3.

Eligibility Criteria
Original studies reporting the prevalence of DDI and potentially inappropriate medications. Studies were considered eligible to be included in this review if they provided the age of involved patients. Looking at patient's characteristics, most of the studies focused on patients older than 65 years old. However, two studies selected patients above 60 years old (Brazil and Pakistan). Our choice to include them in this review is based on the WHO definition of an older person in developing countries which has set the threshold at 60 years old for these countries [12]. The following exclusion criteria were applied: duplicate studies, letters, reviews.

Study Selection
Two authors (MB and PLC) independently assessed the publications for inclusion in the review. The titles and abstracts were screened to identify the potentially relevant studies aiming at determining the prevalence of DDI and of PIM in older patients. The full-text copies for the studies that apparently met the criteria as well as those for which an uncertainty existed were retrieved for an independent review by the two authors (MB and PLC). If the two authors did not reach agreement through discussion about a paper to be included, we consulted a third study author (MC) to resolve any disagreement.

Data Extraction and Synthesis
Two review authors (MB, PLC) independently extracted relevant data from full-text articles using a predetermined data extraction form. In case of disagreement, the resolution was obtained by discussion between the two review authors, and a third author (MC) was consulted if necessary.
The following items were included in the extraction form: first author, year, country (ies) of origin, age of patients (mean, threshold if mean non available), study setting (primary care or medical institution), sample size, typology of patients (general population or specific disease), study style, source of data, software or tool used for checking prevalence of PIM and DDI, mean number of medications.
Data were collected from the publications directly as reported or indirectly (i.e., after calculation from the raw data). Any missing data were requested from the study authors.

Data Analyses
The prevalence of DDI and of PIM corresponds to the percentage of patients that had at least one DDI or one PIM.
We decided to report in detail only the DDI considered as clinically significant; that is, DDI with major and moderate risk. There was a diversity of tools used to check the DDI, and the risk rating classification of DDI differed between some of them. Hence, we have carried out a mapping of the different DDI checker tools to integrate only DDI we considered as major and moderate risk according to the description that was reported by each tool.
We made the following comparisons: (a) Prevalence of each parameter (PIM and DDI with severe and moderate risks) reported by the most used tools (using reported mean) without consideration of settings, in a first time. Then, we made the same comparison using computed weighted mean and 95% CI for the most used tools to evaluate potential differences between settings. (b) Prevalence of PIM and DDI (with severe and moderate risks) in each setting (using computed weighted mean). (c) Polymedication between settings (number of prescribed medications per patient reported as the mean).

Statistics
As a result of differences in the number of patients involved the studies we computed a weighted mean for each parameter taking into account the sample size. We used the svymean and confint functions from the R survey package to compute weighted means and 95% confidence intervals (CI95%) [13]. Statistical analyses were performed on data from articles focusing on the general population using Shapiro Wilk and Kruskal Wallis tests. For pairwise comparisons of the prevalence of PIM and DDI (with severe and moderate risks) in each setting, a false discovery rate correction for alpha risk was used. Statistical analysis was performed using R software (version 3.6.3). A p-value < 0.05 was considered significant.

Mapping of DDI Risk Classification
Since there are variations in the classification of DDI among DDI softwares, we made a mapping of the DDI risk classification based on the descriptions given by the providers. We retained and classified the DDI we considered major (e.g., contra-indicated/serious/life threatening) or moderate (e.g., use with caution).

Data Extraction from Hospital Clinical Data Center
In the course of this review, we noticed that very few studies were dealing with DDI involving specifically PIM so that we decided to perform a preliminary study within our university hospital clinical data warehouse (CDW) [14]. In order to estimate the prevalence of DDI specifically with PIM, we have chosen 6 PIM from different therapeutic areas (tramadol, apixaban, digoxine, clozapine, glimepiride and quetiapine) with a narrow therapeutic index that are metabolized and/or transported by the most common pathways involved in drug pharmacokinetics (CYP450 isoenzymes: 1A2, 2C9, 2C19, 2D6, 3A4-Transporter: P-gp). For each PIM (as substrate), we have selected some interacting medications known to lead to a DDI rated as moderate or major intensity. This led to a panel of 15 DDI whose prevalence were screened in electronic health records of elderly patients from our CDW (hospitalized in 2019).

Study Selection
The literature search identified 212 studies in Pubmed and 298 in Web of Science databases leading to 440 studies after removal of duplicate studies. After titles and abstract screening, full-text articles were assessed for legibility. The flow chart is depicted in Supplementary Material 2.

Study Characteristics
Three study groups were created according to the study setting: nursing home (n = 20), primary care (n = 38) or hospital settings (n = 33). Among these 91 studies, 66 focused on general population and 25 on specific populations (population with specific medications or specific diseases, e.g., oncology, psychiatric disease, etc.).
Included studies were published from 1994 to 2019. Most of the studies from general population were conducted in Europe (n = 38) and North America (n = 13). However, other countries from Asia (n = 8) or South America (n = 7) have also published results. Further details on the characteristics of the studies and the tools used are available as Supplementary Material 4. Tables 1-3 describe the characteristics of each study evaluated in this review for nursing home, primary care and hospital settings, respectively.

Mapping of DDI Checker
Our study showed that different DDI checker tools are used in different ways in risk rating DDIs (Tables 1-3). The mapping of severity classifications of the different tools that were used is presented in Table 4. Among DDI checker tools, Micromedex and Lexicomp were most frequently used (n = 17, 18% and n = 10, 10%. respectively). The Swedish Physicians' Desk Reference tool was also frequently used, especially as a result of numerous studies (n = 9) performed in Sweden in primary care and nursing home settings. Based on these elements, comparison between studies is not straightforward as a result of difference in DDI detection.

Potentially Inappropriate Medications
We checked a potential difference between the tools used to identify PIMs by comparing Beers Criteria, STOPP/START, Swedish National Board of Health and Welfare, Fick list and NORGEP that were the most used tools (47%, 11%, 11%, 8% and 5%, respectively). The average prevalence reported by using these tools were 38.6%, 60.4%, 25.0%, 29.2% and 26%, respectively. Without consideration of settings, the distribution was not normal (p = 0.0018). We found a significant difference between the tools (p = 0.020) in their ability to detect PIMs.
The prevalence of PIM in specific populations of patients and in the different settings is presented in Tables 1-3. In the specific populations, given the diversity of diseases encountered (especially mental health diseases in nursing home and oncology in hospital settings), comparison between settings is not relevant. However, it should be noticed that PIM prevalence in the specific populations was higher than in the general populations except for hospital setting (35.0% in specific population vs. 44.6% in the general population).

Polymedication
Polymedication did not display a normal distribution (p = 0.023) and revealed no differences between settings (p = 0.6741). Indeed, the number of medications per patient in primary care, nursing home and hospital was 7.07, 7.02 and 7.44, respectively. Table 1. Prevalence (%) of potentially inappropriate medications (PIM) and drug-drug interactions (DDI) (total, severe and moderate) and polymedication (mean or median) in elderly patients in nursing home setting. At the end of the table, the prevalence of PIM and DDI is reported as the weighted mean (in bold). Footnote: Prospective (P), retrospective (R), mean (M), median (m), threshold (t), severe (S), moderate (M), not apply (NA), not determined (ND).

Author
Year

Risk Rating Reported by the DDI Checker Tools Description
Lexicomp (Lexi-interact) [111] Major X Avoid combination. Data demonstrate that the specified agents may interact with each other in a clinically significant manner. The risks associated with concomitant use of these agents usually outweigh the benefits. These agents are generally considered contraindicated. D Modify regimen. Data demonstrate that the two medications may interact with each other in a clinically significant manner. A patient-specific assessment must be conducted to determine whether the benefits of concomitant therapy outweigh the risks. Specific actions must be taken in order to realize the benefits and/or minimize the toxicity resulting from concomitant use of the agents. These actions may include aggressive monitoring, empiric dosage changes, choosing alternative agents. The interaction may be life-threatening and/or require medical intervention to minimize or prevent serious adverse effects.

Moderate Moderate
The interaction may result in exacerbation of the patient's condition and/or require an alteration in therapy.

Minor
The interaction would have limited clinical effects. Manifestations may include an increase in the frequency or severity of the side effects but generally would not require a Major alteration in therapy. Unknown Unknown MIMS drug alert [114] Major Potentially severe The interaction between these medications may be life-threatening or may cause permanent damage. These medications are not usually used concurrently; medical intervention may be required.

Moderate Moderate
These medications may interact resulting in the potential deterioration of the patient's condition. The patient should be monitored for the possible manifestations of the interaction. Medical intervention or a change in therapy may be required.

Minor
Clinical effects of the interaction are limited and may be bothersome but would not usually require a major change to therapy. The patient should be monitored for the possible manifestations of the interaction.

Caution
The interaction may occur based on the mechanism of action of the co-administered medicines. Be alert for increased or decreased effect, depending on the combination of medicines.

Not clinically significant
The interaction may occur, but the outcome is not clinically significant.  [117] Major High The interaction between these drugs can be life-threatening or cause permanently damage. These drugs are not usually used together, they require medical intervention. An alternative medicine should be used

Moderate Moderate
The clinical impact of interaction is limited, but can be disturbing. Patient should be monitored for the findings of interaction Low Caution should be taken with regard to the reduced or increased efficacy related to the combined drugs Stockley's interaction checker [119] Major Severe For interactions that could totally incapacitate a patient or result in either a permanent detrimental effect or a life-threatening event.

Moderate Moderate
For interactions that could result in an effect that may either cause considerable distress or partially incapacitate a patient. These interactions are unlikely to be life-threatening or result in long-term effects.

Mild
For interactions that could result in an effect that is mild and unlikely to unduly concern or incapacitate the majority of patients. The combination advised against should most often be avoided, except after careful consideration of the benefit/risk ratio. It requires close monitoring of the patient.

Moderate Precaution of use
The association is possible as soon as simple recommendations to avoid the occurrence of interaction (dosage adjustment, strengthening of clinical, biological monitoring, ECG, etc.) are respected, especially at the start of treatment.
To take into account

Drug-Drug Interactions
Our mapping of the different risk rating classification provided by the checker tools allowed us to report only clinically relevant DDI, i.e., considered as major and moderate risk (Table 4).
We checked for a potential difference between the tools used to identify DDI by comparing Micromedex, Lexicomp, DRUID, Swedish Physician's Desk reference and Intercheck that were the most used tools to identify DDI (in 20%, 7%, 7%, 13% and 6% of the studies, respectively). The average prevalence of DDI reported by using these tools were 19.9%, 17.5%, 1.1%, 4.6% and 48.8%, respectively for severe DDI, and 47.1%, 62.7%, 7.7%, 13.1% and 38.4% for moderate DDI. Without consideration of settings, the distribution was normal (p < 0.001) for severe DDI; and a significant difference between the tools (p = 0.018) in their ability to detect severe DDIs was found. Concerning moderate DDI, there was a normal distribution (p = 0.057) but no significant difference between the tools (p = 0.077).
Since Micromedex was the most used tool in the general population (n = 14, 20%), we made a comparison of DDI (severe and moderate) according to Micromedex between the settings. The prevalence of severe DDI was 6.1% (95% CI: 4.5-7.7), 2.5% (95% CI: 1.1-3.8) and 32.8% (95% CI: 23.0-42.7) in primary care, nursing home and hospital revealing differences between settings (p < 0.001). Concerning moderate DDI, a comparison between the three settings was not possible given that Micromedex was not used to identify such DDI in primary care. However, no difference between nursing home and hospital was found (p = 0.057). Indeed, moderate DDI prevalence was 48.3% (95% CI −23.9-120.6) and 49% (95% CI 47.6-50.4) for these settings.
The weighted mean prevalence of severe DDI (Figure 1) was in the rank order: hospital 28.9% (95% CI: 17.2-40.6), primary care 4.4% (95% CI 3.2-5.6) and nursing home 3.3% (95% CI: 3.1-3.4). The same rank order was found for moderate DDI with 54.4% (95% CI 38.8-70.0) in hospital, 29.6% (95% CI: 28.5-30.6) in primary care and 10.9% (95% CI 2.3-19.6) in nursing home, respectively. The prevalence of both PIM and DDI has been reported as a weighed mean. The surface of the circle is proportional to the total sample size of the studies used in the calculation of the parameters (the sample size, the weighed mean and the number of studies used are reported). *: significant difference between hospital and primary care settings. **: significant difference between hospital and nursing home settings. ***: significant difference between primary care and nursing home settings.
Severe and moderate DDI displayed significant differences between settings (p < 0.001 for each). Using pairwise comparisons, no significant differences were noticed for severe DDI between nursing home and primary care (p = 0.21) nor between primary care and hospital (p = 0.054). However, the prevalence of severe and moderate DDI between nursing home and hospital was significantly different (p < 0.001 and p = 0.0035 respectively). Differences in the prevalence of moderate DDI between primary care and hospital (p = 0.0035) and between primary care and nursing home (p < 0.001) were also observed.
The prevalence in DDI in specific populations of patients and in the different settings is presented in Tables 1-3.

Prevalence of Drug-Drug Interactions in the Clinical Data Warehouse
DDI with the 15 selected pairs (DDI as victim drug and inhibitor) occurred in 769 patients of our sample of 8434 patients leading to a global prevalence of 9.1% ( Table 5). The prevalence of DDI for these selected pairs ranged from 0.1% (CYP2D6 inhibition of tramadol metabolism by terbinafine) to 19.4% (P-gp transport inhibition of apixaban by amiodarone). Bubble plot of the prevalence (%) of potentially inappropriate medication (PIM) (grey) and of severe (red) and moderate (orange) drug-drug interaction (DDI) in elderly patients in primary care, nursing home and hospital settings. The prevalence of both PIM and DDI has been reported as a weighed mean. The surface of the circle is proportional to the total sample size of the studies used in the calculation of the parameters (the sample size, the weighed mean and the number of studies used are reported). *: significant difference between hospital and primary care settings. **: significant difference between hospital and nursing home settings. ***: significant difference between primary care and nursing home settings.
Severe and moderate DDI displayed significant differences between settings (p < 0.001 for each). Using pairwise comparisons, no significant differences were noticed for severe DDI between nursing home and primary care (p = 0.21) nor between primary care and hospital (p = 0.054). However, the prevalence of severe and moderate DDI between nursing home and hospital was significantly different (p < 0.001 and p = 0.0035 respectively). Differences in the prevalence of moderate DDI between primary care and hospital (p = 0.0035) and between primary care and nursing home (p < 0.001) were also observed.
The prevalence in DDI in specific populations of patients and in the different settings is presented in Tables 1-3.

Prevalence of Drug-Drug Interactions in the Clinical Data Warehouse
DDI with the 15 selected pairs (DDI as victim drug and inhibitor) occurred in 769 patients of our sample of 8434 patients leading to a global prevalence of 9.1% ( Table 5).
The prevalence of DDI for these selected pairs ranged from 0.1% (CYP2D6 inhibition of tramadol metabolism by terbinafine) to 19.4% (P-gp transport inhibition of apixaban by amiodarone). Table 5. Prevalence of DDI of moderate or major intensity (by inhibition of a metabolic or of a transport pathway) regarding six selected PIM in hospitalized patients. Data were extracted from the clinical data warehouse of Rennes University Hospital (eHop). DDI were checked by using Micromedex and literature search in Pubmed. Specific bibliographic references were added when DDI was not indexed in Micromedex. Footnote: not indexed (-).

Discussion
This systematic review showed that the mean weighted prevalence of PIM was in the following rank order: hospital > nursing home > primary care with significant differences between these settings ( Figure 1). The prevalence of severe and moderate risk DDI in hospital settings was higher than those in primary care and nursing home (Figure 1), and there were few studies dealing with DDI involving specifically PIM.

Potentially Inappropriate Medications
The polymedication rate was similar between settings, and cannot be accounted for in the differences in PIM prevalence. When considering only studies using Beers criteria, the prevalence of PIM remained the highest in the hospital setting but prevalence in primary care setting exceeded the one in nursing home setting. The highest prevalence of PIM in the hospital setting may be related to the more severe conditions of patients and to the fact that some PIM are due to drug-disease and drug-syndrome interactions (e.g., heart failure, delirium, late stage of chronic kidney disease). Moreover, as recently suggested, hospital admission per se (as a result of change in patient clinical status and of intensification of healthcare) was shown to be an important driver of PIM independently of polymedication and multimorbidity [124].
However, differences in PIM prevalence among settings may be somewhat biased by the tools used that have different criteria of detection. Indeed, only a few medications are common to all the PIM's criteria, resulting from differences in drug dosage as well as market availability [125]. If Beers criteria were the most used tool in the reported studies, NORGEP and STOPP/START were the two other most used tools. The higher prevalence observed with Beers criteria (38.6%), compared to NORGEP (26.0%) may be related to the fact that NORGEP criteria do not include medications that would be considered inappropriate for specific co-morbidities and hence led to an underestimation of the prevalence. On the other hand, the higher prevalence reported with STOPP/START (60.4%) may result from the fact that this tool identifies underuse and overuse of medications [126]. Hence, the prevalence of PIM may be dependent on the tools used in the studies.

Drug-Drug Interactions
We decided to focus only on clinically relevant DDI (i.e., considered as major and moderate risk) that are highly likely to alter the effectiveness or toxicity of one or more medications. Clinically relevant DDI can lead to measurable patient adverse events taking into account an individual patient's profile. However, there are very few studies considering DDI resulting in actual harm to the patients.
Reporting only clinically relevant DDI was made possible by a mapping of the different risk rating classification provided by the checker tools that we performed (Table 4). Besides medication severity classification, differences in DDI checker tools also result from differences in the knowledge databases they are based on. A comparison of three commercial DDI knowledge databases showed significant differences in their numbers of clinical medication pairs and a limited overlap [127]. In a data mining study performed in 10,506 elderly patients' treated statins, we showed differences in the risk rating, and in the individuals DDI among the different statins used [128].
The higher prevalence of severe and moderate risk DDI in hospital settings ( Figure 1) compared to primary care and nursing home is not unlikely given the nature of medication treatments that are prescribed to hospitalized patients as a result of more severe conditions. This higher prevalence in hospital setting was apparently not related to the polymedication rate that was very similar among the different settings.
In a recent systematic review from studies achieved in hospitalized patients [129], the rate of potential DDI (all risk rating) was estimated as 33%, with a rather large variability (95% CI: 17.3-51.3). Our estimation of clinically relevant DDI in elderly patients hospitalized from general population displayed a higher load of DDI since the prevalence of severe and moderate DDI averaged 28.9% and 54.4%, respectively.
The fact that a rather high prevalence of clinically relevant DDI occurs in a population with a high prevalence of PIM prescription (around half of the population) is of particular concern. Indeed, DDI can occur with a PIM as the victim medication, and may enhance their inappropriate character by increasing their unfavorable balance of benefits and harm. Indeed, the enzymatic inhibition of the metabolism of a PIM may enhance its systemic exposure and, by the way, may increase either its clinical effects and/or its side effects. However, the clinical impact of the increase in the inappropriate character of PIM depends on the magnitude of the increase in systemic exposure as well as on the therapeutic index of the PIM. PIM generating DDI (i.e., as a perpetrator medication) is another concern to investigate. Even though the DDI will not modify per se the inappropriate character of the PIM, it could potentially have deleterious effects related to the induction or the inhibition of the metabolic pathway and/or transporter involved in the disposition of the medication undergoing DDI (i.e., victim medication).
This review clearly showed that the issue of DDI occurring specifically with PIM has not yet been of concern. Indeed, only 4 studies out of 91 have considered this issue. In a recent study involving 368 patients in a primary care setting, the association between polypharmacy, PIM use and DDI (called "iatrogenic triad") was evidenced in around 30% of the patients. DDI and PIM were encountered in 8.2% of the patients [39]. In a study involving 1987 nursing home residents, 6.6% of PIM users were susceptible to major DDI [25]. However, these studies did not report the prevalence of DDI involving specifically PIM.
More interestingly, a study involving 120 nursing home residents showed that potential DDI in PIM occurred in 18.3% of the prescriptions with 60% of these DDI considered as potentially harmful [22]. In a study focusing on the use of benzodiazepine in 744 primary care patients, the prevalence of DDI in benzodiazepines considered as PIM occurred in 7% [68].
Ideally, in elderly patients, PIMs should always be reviewed, and DDI should be always checked. However, with regard to DDI, attention should be paid to those occurring with PIM since they could increase their inappropriate character. Even though this issue should be tackled in priority in the hospital setting, it is also relevant for patients in primary care and nursing homes even if the prevalence of clinically relevant DDI was lower, especially in nursing homes (Figure 1). Figure 1 also suggests that studies in the hospital setting involving a higher number of patients are warranted to get a better picture of DDI in PIM in this setting. If it could be difficult to reach the sample sizes reported in primary care studies, the development of prescriptome analytics from clinical data-warehouses should help in reaching this goal [130].
An overview of the future management of the burden of DDI of the healthcare system is beyond the scope of this paper, but we think that attention should be paid to DDI in PIM in elderly people given the epidemiologic transition characterized by the population aging and the emergence of chronic diseases requiring long-term medication treatment with medications classified as PIM [10]. In the hospital setting, DDIs are estimated to contribute to 3-5% of adverse events. They are also a significant cause of hospitalization (either via emergency department visits or via hospital admissions). In a retrospective study on the origin of hospital admissions, adverse drug reactions (ADRs) were rated as a significant cause and around 25% of these ADRs were related to a DDI [131].
Given the paucity of data on DDI occurring specifically with PIM in the literature mentioned above, we have performed during the course of this review a preliminary study by checking 15 DDIs involving 6 PIMs as victim medications involving 8464 elderly patients in our hospital. The global prevalence of clinically relevant DDI (i.e., 9.1%) observed in our sample should be considered with caution because only 15 potential DDIs were checked, and it is thus a real underestimation of the true prevalence. However, this short survey of a subset of 15 DDIs of moderate or major risk regarding only 6 PIMs suggests that a comprehensive investigation of DDI involving PIM should implemented. This could be facilitated by using an automatic DDI identification as we previously set up for a study on DDI with statins [128].
Furthermore, considering that some of these DDIs were risk-rated as major (Table 5), a further study would ideally investigate the potential negative outcomes of DDI in PIM on ADR, and their consequences (hospitalization as well as morbidity and mortality).

Strength and Limitations
This systematic review has several strengths. First, our study included a broad range of studies encompassing the primary care, nursing home and hospital settings. Second, given the fact that different DDI checker software are used worldwide with different classifications of risk ratings, we have performed a mapping of the DDI risk rating to allow a better classification of clinically significant DDIs, and of the estimation of their prevalence. Third, this is the first review of DDIs in elderly patients in whom PIMs are prescribed.
This study has several limitations. First, given the heterogeneity of the selected studies in terms of methodology and tools used as well as in reporting data, there are some missing data precluding a formal meta-analysis and its statistical evaluation. Nevertheless, weighted means in PIM and DDI prevalence allowed us to report some differences between settings. Second, we only performed the literature search using PubMed and Web-of-Science and did not identify grey literature so that some studies may have not been included. Third, the low number of studies in certain combinations (setting/PIM or setting/DDI) may contribute to some uncertainty in the point estimates and 95% CI. Fourth, the huge difference in the total sample size between settings (primary care >> nursing home > hospital) may constitute a bias in comparison.

Conclusions: Future Research
The current paper presents an in-depth review of the literature regarding the topic of DDI in elderly patients using PIM. It was performed with a mapping of DDI checker tools we thought to be a prerequisite for meaningful comparisons between studies. To our knowledge, this is the first systematic review of this issue.
We showed a higher prevalence of PIM, and of severe and moderate DDI, in the hospital setting compared to nursing home and primary care, unrelated to polymedication rate. Within the studies on DDI in elderly in whom PIM were prescribed, the issue of DDI occurring specifically with PIM has not yet been of concern, since only four studies involving a rather limited number of patients investigated this topic.
To help clinicians and pharmacists to identify at risk elderly patients, strategies to assess the appropriateness of medication use in elderly should additionally consider DDIs that specifically involve PIM. DDI software providers should implement alerts when a DDI occurs with PIM with a clinical decision support so that healthcare professionals can easily prevent and manage these DDIs. This would be most helpful in the hospital setting where PIMs and DDIs are highly prevalent.  Institutional Review Board Statement: According to the French Data Protection Act of 6 January 1978, this study was carried-out by the hospital care team and did not require approval from an ethics committee. Authorization to conduct research from the Clinical Data Warehouse of Hospital of Rennes was given by CNIL -Commission Nationale Informatique et Liberté (Authorization number 2020-028 issued on february 27, 2020). This study only uses data from existing medical records and did not involve interaction with patients or collection of identifiable private information. Each data sample was deidentified to ensure confidentiality.

Informed Consent Statement:
According to the French regulation, hospitalized patients are informed of the use of their hospital health data, and no specific signed consent was required.
Data Availability Statement: Not applicable.

Conflicts of Interest:
The authors declare no conflict of interest.