Drug–drug interactions of protein kinase inhibitors in chronic myeloid leukaemia patients: A study using the French health insurance database

The introduction of protein kinase inhibitors (PKIs) for chronic myeloid leukaemia (CML) has considerably improved prognosis of the disease but has also demonstrated a great potential for drug–drug interactions. Using the French health insurance databases, we aim to investigate the frequency, identify the associated factors and describe the potential consequences of potential drug–drug interactions (pPKI‐DIs) between PKIs and concurrent medications in CML. A retrospective cohort study has been performed among patients with CML identified in the French healthcare database from 2011 to 2014. A pPKI‐DI is defined as the presence of drugs listed as ‘interacting’ on the same day as PKI dispensing (co‐dispensing) or in its coverage period (co‐medication) during the first year of follow‐up. The list of interacting drugs is based on the summary of products characteristics (SPCs) and Thesaurus of interactions. We performed specific nested case–control comparisons to investigate the association between PKI‐DI and each of the three potential outcomes (death, hospitalisation for adverse drug reactions and switch to another PKI). We included 3480 patients; 1429 (41%) had a co‐dispensing pPKI‐DI, and 2153 (62%) had a co‐medication pPKI‐DI; 50% of the pPKI‐DIs were ‘to be taken into account’, and 17% were ‘not recommended’. The PKI with the most interactions was imatinib, and additional common drug classes included statins, benzodiazepines and proton pump inhibitors. Multivariate analysis demonstrated that the use of a higher number of additional drugs, comorbidities at baseline, high number of prescribers and higher ages were potential risk factors. Nilotinib and dasatinib showed a tendency towards a higher risk of pPKI‐DI compared to imatinib. Despite the fact that some PKI‐DIs were potentially clinically relevant, we did not find any significant association with death, hospitalisation for adverse drug reactions and switching. These findings should increase awareness to help reduce the prevalence of PKI–drug interactions and thereby ensure better management of CML patients.


Abstract
The introduction of protein kinase inhibitors (PKIs) for chronic myeloid leukaemia (CML) has considerably improved prognosis of the disease but has also demonstrated a great potential for drug-drug interactions. Using the French health insurance databases, we aim to investigate the frequency, identify the associated factors and describe the potential consequences of potential drug-drug interactions (pPKI-DIs) between PKIs and concurrent medications in CML. A retrospective cohort study has been performed among patients with CML identified in the French healthcare database from 2011 to 2014. A pPKI-DI is defined as the presence of drugs listed as 'interacting' on the same day as PKI dispensing (co-dispensing) or in its coverage period (co-medication) during the first year of follow-up. The list of interacting drugs is based on the summary of products characteristics (SPCs) and Thesaurus of interactions. We performed specific nested casecontrol comparisons to investigate the association between PKI-DI and each of the three potential outcomes (death, hospitalisation for adverse drug reactions and switch to another PKI). We included 3480 patients; 1429 (41%) had a co-dispensing pPKI-DI, and 2153 (62%) had a co-medication pPKI-DI; 50% of the pPKI-DIs were 'to be taken into account', and 17% were 'not recommended'. The PKI with the most interactions was imatinib, and additional common drug classes included statins, benzodiazepines and proton pump inhibitors. Multivariate analysis demonstrated that the use of a higher number of additional drugs, comorbidities at baseline, high number of prescribers and higher ages were potential risk factors. Nilotinib and dasatinib showed a tendency towards a higher risk of pPKI-DI compared to imatinib. Despite the fact that some PKI-DIs were potentially clinically relevant, we did not find any significant association with death, hospitalisation List of abbreviations: 95% CI, 95% confidence interval; ADR, adverse drug reaction; ANSM, Agence Nationale de Sécurité du Médicament et des produits de santé; ATC, Anatomical Therapeutic Chemical system classification; BCR-ABL, 'Breakpoint Cluster Region'-ABELSON gene; CCI, Charlson comorbidity index; CML, chronic myeloid leukaemia; CMU-C, Couverture Maladie Universelle Complémentaire; CYP3A4, cytochrome P3A4; DDI, drug-drug interaction; ICD-10, International Classification of Diseases 10th Revision; IQR, interquartile range; LTD, long-term disease; mOR, matched odds ratio; PKI, protein kinase inhibitors; PPI, proton pump inhibitors; pPKI-DI, potential protein kinase inhibitors-drug interaction; SNDS, Système National des Données de Santé; SPC, summary of product characteristics. for adverse drug reactions and switching. These findings should increase awareness to help reduce the prevalence of PKI-drug interactions and thereby ensure better management of CML patients.

| INTRODUCTION
Since the 2000s, protein kinase inhibitors (PKIs) mainly targeting the tyrosine-kinase activity of the BCR-ABL1 oncoprotein have been used increasingly in the treatment of chronic myeloid leukaemia (CML) [1]. Their advent has changed the prognosis of CML from hospital management in a life-threatening context to outpatient management in patients whose average survival is now only slightly different from that of the general population [1][2][3][4]. The use of these drugs (all oral) is therefore systematic in any newly diagnosed case of CML [5]. A recent study describing characteristics of incident cases in France revealed that this is an ageing population with other comorbidities at the time of management of their haematological disease and chronically in receipt of numerous medications [6]. These findings raise concerns about this polypharmacy, including lack of treatment adherence, or drug-drug interactions (DDIs) [7,8]. In fact, the theoretical clinical impact of these potential DDIs has been described in some publications [9,10].
Drug interactions can be classified into pharmacodynamics and pharmacokinetic interactions. Pharmacodynamic interactions refer to an interaction in which active compounds change each other's pharmacological effect on the same target. This effect can be synergistic, additive or antagonistic. Pharmacokinetic interactions refer to an alteration in the absorption, distribution, metabolism or excretion of drugs [11][12][13].
PKI-related drug interactions most commonly occur when pH-dependent absorption is impaired (coadministration with H 2 antagonists, proton pump inhibitors or other acid suppressive therapies), or when drug metabolism is altered either by the PKIs themselves or other interacting medications [14][15][16]. PKIs are both substrates and inhibitors of hepatic microsomal enzyme systems involved in drug metabolism, most notably through the CYP3A4 enzyme. CYP3A4 is the main CYP enzyme and is responsible for the metabolisation of more than half of all drugs. For this reason, another drug used concomitantly with a PKI is likely to affect PKI metabolisation via the CYP3A4 isozyme [17]. Use of PKIs with other drugs that decrease absorption or induce metabolisation of PKI may result in subtherapeutic levels and bring about a decrease in PKI effect [18][19][20][21]. On the other hand, drugs that inhibit the PKI metabolism may also cause supratherapeutic drug levels and toxicity [22][23][24][25]. PKIs themselves could modify the metabolisation of other drugs, which can lead to a loss of therapeutic effects in these comedications or can increase the risk of adverse drug reactions (ADRs) or toxicity in these other drugs [26,27].
Information on the frequency and actual relevance of these potential DDIs is still scarce, and only a few retrospective series have been previously published with limited information about their clinical impact. One study identified the frequency of DDIs with imatinib in the French drug reimbursement database at a regional level (southwest France) [28]. Two other studies identified the frequency of tyrosine kinase inhibitor (TKI)associated drug interactions among cohorts of oncology patients [29,30], and the last study determines DDI frequency and their clinical consequences in CML patients treated with TKI in 15 Spanish oncology centres [31].
The main objective of this study was to determine the frequency of potential PKI-drug interactions (pPKI-DIs) during the first year of treatment among a cohort of CML patients in the French health insurance databases. Secondary objectives were to identify the factors associated with this situation of potential PKI-DIs and describe their potential consequences in terms of mortality, hospitalisation for ADRs and switch of PKI.

| Data sources
Data were extracted from the French health system database, Systéme National des données de Santé (SNDS), that covers almost 98% of the French population. The SNDS has been widely used to conduct large epidemiologic studies, and further information regarding its organisation has already been described elsewhere, specifically in the field of oncology [32][33][34][35][36]. The database contains comprehensive individualised and anonymous data on the demographic characteristics of subjects (age, sex, coverage by the CMU-c [a complementary universal health coverage system for people with low incomes]) and eligibility for full reimbursement of healthcare expenses related to long-term diseases (LTDs) (in French 'Affections de Longue Durée'). The LTDs eligible for full coverage by the insurance system can be identified using the International Classification of Diseases 10th Revision (ICD-10) codes. Data concerning hospital episodes, including diagnoses (ICD-10 coded), are also available. Only information on drugs prescribed and reimbursed by the French health system is recorded in the SNDS, thus excluding drugs that are not reimbursed, dispensed during hospitalisation or sold over the counter. This information includes the name and dosage of the drugs, the date of dispensing, the quantity of the drug dispensed and the quality of the prescriber (anonymously identified) for all reimbursed drugs. Drugs are classified according to the Anatomical Therapeutic Chemical (ATC) classification system [37].

| Patient selection
We used a population-based cohort of all French incident CML adult patients initiating a treatment with one of the PKIs approved in France for CML (imatinib, dasatinib, nilotinib, bosutinib and ponatinib) between January 1, 2011 and December 31, 2014. The characteristics of this cohort have been previously published [38,39]. All patients in the cohort had a 12-month observation period before inclusion in the cohort and were followed up to 12 months after cohort entry. For the purpose of identifying potential PKI-DIs in this study, we only retained patients with a sustained exposure to PKIs defined as follows: • At least two different dispensations of any PKI (one initial + one extension) within 1 year after the initial dispensing (no dispensation of PKI within the 12 months before inclusion), • With a delay between the two dispensations of less than 30 + 10 days (for drug packages of PKIs including 30 tablets) or 90 + 10 days (for drug packages including 90 tablets).  [47]. This thesaurus provides all French health professionals with an official source of information about the main clinically meaningful DDIs (i.e., about 50 000 pairs of drugs), with specific warnings on the adverse effects they can cause and recommendations for their management. DDIs are classified into four categories:

| Exposure to pPKI-DIs
• 'Contraindicated combinations': These are absolute and must not be transgressed. • 'Not recommended combinations': These should be avoided in most cases, except after careful consideration of the benefit/harm balance. They require close monitoring of the patient. • 'Precaution for use': The combination is possible provided that the recommendations to prevent the occurrence of interactions (dose adjustment, clinical reinforcement, biological monitoring, electrocardiogram [ECG], etc.) are respected, especially in early treatment.
• 'Combinations to be taken into account': A risk of drug interaction exists. It usually corresponds to an addition of adverse effects. No practical recommendations can be proposed. The physician should assess the appropriateness of the combination before prescribing.
DDIs of PKIs indicated for CML as listed in the SPC and/or the thesaurus and their expected pharmacological consequences are summarised in Data S1. The period of exposure to one PKI was assumed to begin on the first day of PKI dispensation and to end on the 30th or 90th day (depending respectively on whether the drug packaging contains 30 or 90 tablets) following the day of the last dispensing, to which we added 10 days of PKI half-life. Successive uninterrupted dispensing of one PKI is defined as an episode of continuous treatment with a single PKI, interrupted by the end of follow-up (1 year), or by the end of treatment (30 or 90 + 10 days) after the last dispensing of the PKI, or by a switch to another PKI.
Co-medications other than PKIs were assessed from the first day of PKI dispensing to the end of followup. These concomitant drugs were examined as potential sources of PKI-DI. A potential drug interaction with a PKI was defined by superimposing a period of exposure to PKIs with an exposure to drugs with a potential to interact with them over at least 1 day (co-medication PKI-DI). Among the episodes of exposure to PKI-DI, we reported the proportion of episodes with co-dispensing, that is, where the PKI and the interacting drug were dispensed on the same day. An example of the different situations (co-medication and co-dispensing) is given in Figure 1.

| Consequences of PKI-DI
Events potentially related to the impact of PKI-DI considered in the study were death (all causes), hospitalisation for ADRs (identified by the following ICD-10 codes: Y57; Y43; Y880; T78; T887; Z036) and switch to another PKI (when a patient switched from one PKI to another during follow-up).
In order to investigate if PKI-DIs were associated with each of these outcomes, we performed three specific nested case-control comparisons for death, hospitalisation for ADRs and switch to another PKI. From the cohort, we identified cases defined as patients who died during the follow-up, patients at the time of the first hospitalisation for ADRs and patients with a first change of PKI. For each of these cases, we selected two controls from the cohort, matched on sex, age (±5 years) and duration of follow-up, without the corresponding event at the time of the matching.

| Statistical analysis
The characteristics of the cohort were described using common parameters: mean ± standard deviation (SD) or median with interquartile range (IQR) for continuous variables and frequency with percentages for qualitative variables.

| Characteristics of pPKI-DI
The frequency of the potential PKI-DIs was estimated during the first year of follow-up. The data presented were those related to co-medication PKI-DI (codispensing data were presented as sensitivity analysis).
Factors associated with at least one pPKI-DI were investigated through a model of logistic regression. Potentially relevant variables considered in univariate analysis were age at inclusion (continuous), sex, coverage by CMU-c, index of comorbidities at baseline (identified in the 12 months before inclusion) through the Charlson comorbidity index (CCI) [48,49], PKI involved in the interaction, number of concomitant drugs used during the PKI treatment (i.e., additional drugs used by patients apart from PKI) and number of different prescribers during the PKI treatment. The largest group was taken as the reference (ref.) for independent variables with more than two groups. Variables with a P value of <0.20 were entered into a multivariate logistic regression model. Interactions were investigated using the log likelihood ratio test. The goodness-of-fit of the logistic models was assessed using the Hosmer-Lemeshow test and the Akaike information criteria. Collinearity was verified by Spearman's correlation among explanatory variables. A significance level of 0.05 was retained.

| Consequences of PKI-DI
Frequencies of the three events observed during the first year following inclusion were displayed among patients with at least one PKI-DI, patients with at least one 'not recommended' PKI-DI and patients without any PKI-DI. A significance level of 0.05 was retained.
For nested case-control comparisons, associations between each event (death [all causes], hospitalisation for ADRs and switch) and any PKI-DI or 'Not recommended' PKI-DI during the month preceding the event were evaluated through multivariable conditional logistic regression models. Co-variables retained in the three models were CCI at inclusion, number of different co-medications, number of different prescribers and PKI used, and for the death event, we added F I G U R E 1 Graphical representation of different situations of PKI-drug interactions for a patient (co-medication and co-dispensing). Legend: A patient received a first dispensation of imatinib (ATC code L01XE01) in a package of 30 tablets, on June 21, 2013. The period of exposure lasted from June 21, 2013 to July 21, 2013. A co-medication PKI-DI may relate to any co-exposure to an interacting drug listed in the Thesaurus and/or the SPC, occurring on at least 1 day during that period. A co-dispensing PKI-DI may concern any co-exposure to an interacting drug listed in the Thesaurus and/or the SPC, occurring on June 21, 2013 (day of dispensation). adjustment on hospitalisation of patients in the month preceding the death. Adjusted mOR (matched odds ratio) and their 95% confidence intervals (CIs) were calculated from the parameter estimates of the conditional logistic regression models at a 0.05 significance level.
Analyses were performed using the SAS LOGISTIC Procedure of the SAS ® 9.4 software (SAS Institute Inc., Cary, NC, USA).

| Patient characteristics
Of the 3633 incident French CML patients newly treated with a PKI between January 1, 2011 and December 31, 2014 in France, 3480 patients met the inclusion criteria with at least two consecutive PKIs dispensed in the first year of follow-up. The median age was 61 years (range 48-71), with a male/female (M/F) ratio of 1:2, and one quarter of the patients (n = 748) had at least one comorbidity other than CML at baseline ( Table 1). The median duration of follow-up was 11 (range 10-11.5) months. During the time of the study, 473 patients (14%) switched to another PKI, and 96 patients (2.6%) died. We identified six patients who experienced successive events including change in PKI and death (Review Table 1). These patients all received dasatinib as second treatment, and the median time between the switch of PKI and the death for these patients was 3 months (range 2-3). Most patients (n = 3409, 98% of the cohort) were using another drug along with a PKI; the median number of concomitant medications was 2 (1-3) at the beginning of PKI treatment and 14 (8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21) over the first year of follow-up. Paracetamol (61%) and proton pump inhibitors (PPIs) (46%) were the most frequently used drugs during the first year of follow-up.
Concerning the analysis of co-dispensing data, 1429 patients (41%) presented at least one potential PKI-DI with similar characteristics compared to comedications (Data S2).

| Factors associated with potential PKI-DI
Increasing age, male sex, CCI, number of different comedications and number of prescribers were significantly associated with pPKI-DI, as well as nilotinib and dasatinib compared with imatinib (Table 3). Similar results were observed with co-dispensing PKI. Table 4 presents the frequency of each outcome in patients with at least one co-medication pPKI-DI, in patients with not recommended pPKI-DI and in patients without any pPKI-DI. Among the 66 patients who were hospitalised for an ADR, the most common diagnoses at hospitalisation for ADR were adverse reactions without precision (ICD-10 Y433/ICD-10 T887) (230), ADR with haematological disorders (10), ADR with cutaneous disorders (9), ADR with hepatic/pancreatic disorders (7), ADR with cardiac disorders (5), ADR with general alteration disorders (5), ADR with muscular disorders (3), ADR with renal disorders (2) and ADR with infectious disorders (2). Imatinib (86%) was the most frequently observed PKI for these hospitalised patients, whereas the most frequent co-medications were statins (23%), benzodiazepines (23%) and oral anticoagulants (22%).

| Outcomes of pPKI-DI
Of the 104 patients who died during the first year of follow-up, 72 patients were hospitalised in the month preceding the death. The most common main T A B L E 1 Characteristics of the study population. diagnoses at hospitalisation were palliative care (14%), medullary aplasia (6%), pleural effusion (5%), CML (4%) and congestive heart failure (4%). The main associate diagnoses were palliative care, CML and high blood pressure. Similar observations appeared when comparing codispensing pPKI-DI patients with patients without any pPKI-DI.
The nested case-control comparison included 104 patients who died during the follow-up and their 208 controls, 66 patients with a first hospitalisation for an ADR and 132 matched controls and 473 patients with a first change of PKI and 946 controls. We did not find any significant association with PKI-DI (

| DISCUSSION
In this French population-based cohort of 3480 incident CML patients, almost all of them were using at least one additional drug alongside their treatment with PKI in the first year of follow-up. Consequently, potential DDIs are very frequent with PKIs, concerning 62% of patients. Fortunately, contraindicated PKI-DIs were not observed, half of PKI-DIs were 'to be taken into account' and 17% were 'not recommended'. These potential DDIs were associated with ageing, male sex, comorbidities, polypharmacy and number of different prescribers. The most frequent pharmacological impact T A B L E 3 Baseline factors associated with a potential PKI-drug interaction (co-medication) in incident CML patients during the first year of follow-up. Results are presented as odds ratios (OR) with their 95% confidence interval. T A B L E 4 Frequencies of surrogate markers of the impact of potential PKI-DIs in patients with at least one co-medication pPKI-DI, those with at least a 'not recommended' PKI-DI and patients without any PKI-DI during the first year of follow-up. Comparisons were made between all patients with pPKI-DI and without PKI-DI and between patients with not recommended PKI-DI and without PKI-DI. includes increase of PKI or other drug toxicities and decrease of PKI efficacy. The main PKIs involved were imatinib and nilotinib, and the most frequently interacting additional drugs were benzodiazepines, statins and anti-secretory agents (proton pump inhibitors and histamine 2 receptor antagonists) and oral anticoagulants. The frequency of potential PKI-DIs observed in this study was in the same range as that observed in previous studies performed among clinical series or from electronic healthcare database samples [28,29,31,50]. In a study focusing on DDI with imatinib identified in a drug reimbursement database at a regional level between 2012 and 2015 (southwest France), 89% of treated patients were exposed to a potential interaction, involving mainly paracetamol, omeprazole, dexamethasone and levothyroxine. In the same period, an analysis of ADRs reported to the pharmacovigilance centre for the corresponding geographical area identified 25 individual safety reports with imatinib, of which 10 could be related to a DDI between imatinib and PPI (n = 7), atorvastatin, paracetamol and levothyroxine (n = 1 for each drug) [26]. Osorio et al. identified 60% of 105 CML patients in 15 Spanish oncology centres with a potential PKI-DI; the main drugs involved being proton pump inhibitors, statins and antidepressants; 20% of these potential PKI-DIs could induce a decrease in PKI efficacy and 1% an increased PKI toxicity. Finally, a DDI-related clinical effect (toxicity and/or lack of efficacy) was suspected in only five patients (4.7%) [28]. In a case series of 310 patients treated by several PKIs between 2007 and 2017 (for several haematological or oncological diseases), 47% of patients were exposed to a potential PKI-DI, with 3% of contraindicated interactions [29]. One third of the patients were treated with imatinib, and in that study, imatinib was the PKI the most frequently involved, and most frequently interacting additional drugs were antibiotics and PPI. The prevalence of exposure to acid suppressant drugs was investigated in the large US Market Scan database and found that 22% of CML patients were exposed to PPIs, leading to potential PKI-DI [46]. A recent study described a potential DDI between TKIs and oral anticoagulants mediated by P-glycoprotein during an in-vitro evaluation [21]. As observed in other studies, ageing, comorbidities and polypharmacy were expected as factors associated with pPKI-DI, as well as exposure to dasatinib or nilotinib compared with imatinib [29,31].
To the best of our knowledge, this is one of the few studies in which the clinical impact of potential PKI-DI was investigated. Despite the fact that some PKI-DIs were potentially clinically relevant ('not recommended'), we did not find any significant association with death, hospitalisation for ADRs and switch in PKI in the first year of follow-up. When we analysed the 66 patients hospitalised for ADRs, we identified some specific adverse events, such as ADR with haematological disorders (medullary aplasia), ADR with hepatic disorders, ADR with pleural disorders or ADR with muscular disorders (myopathy) that can be related to the toxicities of the interacting drugs. Considering the 72 patients hospitalised 1 month before their death, hospitalisation for palliative care or CML management can be part of the progression/degradation of the disease, while management for medullary aplasia or pleural effusion and heart failure can be seen as complications or adverse drug effects. Despite the high prevalence of potential PKI-DI, we did not find any association between exposure to these PKI-DIs and adverse outcomes. Conversely, a potential PKI-DI was associated with exposure to second-line PKI, as well as nilotinib (OR adj = 1.72 [1.30-2.29]) and dasatinib (OR adj = 1.59 [1.08-2.32]) compared to imatinib. These findings seem consistent since other expected factors were associated with the investigated outcomes: The risk of hospitalisation for ADRs was independently associated with dasatinib and nilotinib compared with imatinib as observed in other studies [39,51,52].
T A B L E 5 Results of the multivariate conditional logistic regression models for three events: death (all causes), adverse drug reaction (hospitalisation) and switch, in three nested case-control comparisons. Association between these events and any PKI-DI or 'not recommended' PKI-DI in the month before index date was estimated by the calculation of a matched odds ratio (mOR) and its 95% confidence interval (CI), adjusted on type of PKI used, number of comorbidities in the Charlson comorbidities index (CCI) and number of prescribers. Imatinib is more often chosen as first-line therapy versus other PKIs since it is the first to appear and the most widely used PKI. Second-line PKIs (nilotinib and dasatinib) were initially used for imatinib-resistant or intolerant patients and are progressively used as firstline therapy. We did not find any association between PKI-DI and switch of PKI, which could be considered partly as a proxy of a lack of efficacy of the individual treatment without the DDI. We considered a potential PKI-DI with at least a day of overlap, and in the case-control comparison, we investigated the exposure to this PKI-DI within the month before the index date, regardless of the dose of each drug and the duration of overlap. Due to the uncertainty of dose intake in the database, we have not explored further. Finally, because we did not identify any contraindicated interactions in this study, and because 'not recommended' PKI-DIs were not frequent, we can hypothesise that health professionals in charge of CML patients are aware of the risk of DDI in the context of chronic use of PKIs and that they take precautions through adapted monitoring and an individual appraisal of the benefit-risk balance when prescribing these drugs.
Our study presents several limitations. First, this is an observational cohort study based on electronic healthcare data, with a lack of detailed clinical or biological data. All PKI-DIs are theoretical and have been determined using the European SPC and in the ANSM Thesaurus, as these two sources are the most widely available to clinicians in our setting, and the information they provide is often the basis for the way DDIs are managed in clinical practice. Consultation of different databases and interpretation of the available evidence on a case-by-case basis by an expert in pharmacology may be the best option, whenever possible.
The French SNDS has already been used efficiently in several pharmacoepidemiological studies in the field of oncology [35,36,38,53]. The use of SNDS has several benefits, such as power and representativeness as the sample size is so large [34]. It also covers practically the entire population of France (including those whose income is insufficient to pay for their own medication). However, similar to other health insurance database, SNDS does not record all drugs dispensed during hospitalisation, nor those not reimbursed or sold over the counter (e.g., St. John's wort). Therefore, the number of drugs taken by the CML patients may have been underestimated. Conversely, there might be an overestimating factor, as we are not sure that the patients included in the study really took the prescribed drugs. In addition, we described 'potential' DDIs because it was not possible to know if prescribers have taken any precautions to avoid them (measurement of plasma levels, dose adjustment, biological monitoring and so on).
The outcomes that we used to analyse the consequences of PKI-DI are certainly questionable, but since we cannot directly capture markers of PKI efficacy due to a lack of clinical data (such as BCR-ABL level or frequency of major molecular response), we considered some indirect markers of response in first-line treatment patients during the month that followed the interaction, that is, patients who switch to a second-line PKI (even if it could be caused by a lack of efficacy of the individual treatment without the DDI), patients who were hospitalised with a code of ADRs as the cause of hospitalisation (the codes that we used in our selection are ICD-10 codes of ADR as described in the ICD-10 coding software) and patients who died (even if we did not have the causes of death). However, a study based on data from SEER-Medicare database for the years 2007-2012 has found that the interaction between PPIs and PKIs significantly decreased the survival of patients at 90 days and 1 year but was not associated with discontinuation of PKI [19]. In that study, the authors defined the PKI-DI for an overlap of at least 30 days between PKI and PPI.
We investigated PKI-DI as a whole, whereas it could have been interesting to focus on specific outcomes of interactions with specific drugs, such as statins and muscle damage or PPI and PKI lack of efficacy. For the first example, we did not identify any ICD-10 code suggesting muscle damage among the 66 patients hospitalised with an ADR. For the later example, without any clinical and biological data about CML, it was not possible to investigate a lack of clinical efficacy. Finally, at the time of the data extraction, we did not have access to causes of death, which could be a proxy of treatment failure.

| CONCLUSION
Our study is one of the few that analyses the frequency and consequences of DDI with PKI for the treatment of CML in the French CML population. This study underlines the high prevalence of pPKI-DIs in patients newly treated for CML in France. The use of PKI for a chronic condition in patients with other comorbidities increases the risk of potential interactions because of polypharmacy in this population. Despite this high prevalence of potential PKI-DI, the risk of ADRs, change of PKI or death remains low. The outputs of this study are intended to increase knowledge and raise awareness on the contribution to therapy of comorbidities in the occurrence of potential drug interactions with PKI and their potential to cause pharmacologically relevant effects and, moreover, clinically relevant adverse effects.

AUTHOR CONTRIBUTIONS
MLM and FD conceived the study. MP participated in the study design, data management and statistical analysis plan. MP performed analysis and wrote the first draft. All the authors interpreted results and validated the manuscript's content. MLM is the scientific guarantor for the study. All authors contributed to the article and approved the submitted version.

SUPPORTING INFORMATION
Additional supporting information can be found online in the Supporting Information section at the end of this article.