In Belgium, linezolid is exclusively prescribed by authorized doctors in hospitals (infectious diseases specialists), and dispensed to outpatients by the hospital pharmacy. Hospitalized and/or ambulatory adult (≥18 years) patients treated with linezolid (600 mg q12h orally or intravenously) for a period of at least 3 days (to achieve steady-state concentrations) were included after signing an informed consent form. Patients with baseline hematological disorders (defined as hemoglobin level <8 g/dL and/or platelet counts <75 10⁹/L) were excluded.

The required sample size for this clinical study was estimated based on a statistical power calculation (using a two-sided two-sample equal-variance t-test) and on the results of a previous study (Dong, Xie et al., 2014). Power calculations indicated the need for a sample size of at least 20 patients per group (thrombocytopenia vs. no thrombocytopenia) to be able to observe a statically significant difference in linezolid C_min in patients with and without hematotoxicity (see Supplementary Table S1 for more details).

Data were collected from the first to the last day of treatment. If ADRs occurred during the treatment, patients were followed during one-month post-treatment to assess recovery. Depending on the status of the patient (in- or outpatient) data were extracted from hospitalization reports, electronic medical records (EMR), visit reports (in general every 15 days) and a weekly phone call for outpatients. The following data were extracted from the EMR: demographic data, comorbidities [to calculate the Charlson comorbidity index (Charlson, Pompei et al., 1987)], body weight, renal function [creatinine level and glomerular filtration rate estimated with CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) formula (Levey, Stevens et al., 2009)], hemogram (performed concomitantly to linezolid TDM in most of the cases and including hemoglobin concentration, platelets, neutrophils, white blood cell counts), type of infection, microbiological data, previous antibiotic treatments, reason for linezolid prescription, dosage, treatment duration, route of administration, any change during linezolid treatment (change of dosage or early stop based on TDM, toxicity or any other reason), and comedications that could lead to drug-drug interaction with linezolid (increasing the risk of serotonin syndrome or thrombocytopenia or modifying linezolid pharmacokinetics) (Pea, Furlanut et al., 2010; Rondina, Walker et al., 2010; Buckley, Dawson et al., 2014).

If ADRs occurred during the treatment, the following data were collected: ADR description, time of onset, management, recovery, potential alternative cause. Thrombocytopenia was defined as a platelet count <150 10⁹/L and a reduction of 30% from baseline (Erkurt, Kaya et al., 2012; Hanai, Matsuo et al., 2016). Anemia was defined as a hemoglobin value <12 g/dL and a reduction of 30% from baseline (Cappellini and Motta, 2015; Hanai, Matsuo et al., 2016). The potential association between linezolid usage and the observed ADR was calculated using the Naranjo adverse drug reaction probability scale (Naranjo, Busto et al., 1981). Two predictive scores for linezolid toxicity development were applied to the whole population to verify their applicability. The first refers to the prediction of ADRs with linezolid (Buzele, Lemaignen et al., 2015). It includes an age-adjusted Charlson comorbidity index, and the treatment duration. A score ≥7 is proposed as associated with ADRs. The second concerns the prediction of linezolid hematological toxicity through a score including the presence of a moderate to severe liver disease, cerebrovascular disease, a baseline platelet count <90 10⁹/L and a baseline glomerular filtration rate (GFR) < 50 mL/min (Gonzalez-Del Castillo, Candel et al., 2017). A score between 0 and 4 represents a low risk, between 5 and 10 an intermediate risk and >10, a high risk of developing hematological toxicity.

Linezolid monitoring was performed every 7 days on average for hospitalized patients and the day of the follow-up consultation for ambulatory patients (on average every 15 days for treatments longer than 28 days). As the administered linezolid dose was 600 mg q12h, venous blood samples were collected at the 12 h time point right before administration of the next dose. In case the 12 h sampling time could not be respected (i.e., sampling >2 h before or after the target time for ambulatory patients taking linezolid orally), the trough values at 12 h were simulated based on the actual measured sample values using an established population pharmacokinetic model for linezolid (Plock, Buerger et al., 2007). This pharmacometric approach allowed comparison of all patient trough values at 12 h despite varying sampling times.

Blood samples were collected without any anticoagulant and centrifuged for 10 min at 1968 g and sera were frozen at −80°C until analysis. Linezolid quantification was performed according to the published method (Fage, Deprez et al., 2021). Briefly, the extraction and the sample deproteinization were done with organic solvent. After centrifugation, the supernatant was recovered, evaporated and the residue containing linezolid was reconstituted with 200 μL of 2% acetonitrile and 0.6% phosphate solution at pH 5. The mix was then injected onto an ultra-performance liquid chromatography system coupled to a photodiode array detector (UPLC-PDA). The assay was linear in the 0.75–50.00 mg/L range.

A literature research was performed to identify available oral linezolid pop-PK models derived from patient populations comparable to the one in this study, excluding those using covariates that were not assessed in our patients. The predictive performance and fit of five eligible models (Plock, Buerger et al., 2007; Abe, Chiba et al., 2009; Boak, Rayner et al., 2014; Matsumoto, Shigemi et al., 2014; Tsuji, Holford et al., 2017) was evaluated and Bland-Altman-plots were used to compare agreement between the model results (see Methods and Results in Supplementary Material). The Plock et al. model (Plock, Buerger et al., 2007) was selected to simulate trough concentrations at 12 h, as the studied population was the most similar to our patients and it showed the best fit with the data observed in our study. All simulations were performed in NONMEM (version 7.4.3, ICON, Gaithersburg, MD, United States).

Statistical analyses were performed using IBM SPSS statistics version 26.0 (IBM, Armonk, NY) or Graphpad prism version 9.5.1 (GraphPad Software, San Diego, CA). Distribution and normality of quantitative data were analyzed with Shapiro-Wilk test (n < 100). As data distribution was predominantly not normal, data are shown as median and range. Continuous variables were analyzed with Mann-Whitney U test. Pearson’s Chi-squared test or Fisher exact test was used to compare categorical variables (according to the conditions of each test). Correlation between two parameters were assessed with Spearman correlation test and presented with the Spearman correlation coefficient r_s and correlation was significant when p-value < 0.05. Correlation was positive with a positive r_s and negative with a negative r_s. Scores sensitivity was calculated with the formula true positive divided by the observed patients with ADR. Specificity was calculated by the division of true negatives divided by the total without ADR. Time of onset of ADR were assessed with a Kaplan-Meier survival curve. Identification of risk factors for hematological disorders was performed through a univariate analysis and multivariate logistic regression. Variables with a p-value < 0.1 were included in the final multivariate model (instead of 0.2 based on the small number of patients and high number of variables) but had to be limited in view of the small sample size. This was notably performed by testing the collinearity between variables using the variance inflation factor (VIF): when VIF was >2, the corresponding parameter was removed from the analysis and not included in the multivariate analysis. A backward procedure was used to select final interesting variables. The goodness-of-fit of the final model was evaluated with the Hosmer-Lemeshow test. Odds ratio (OR) with 95% confidence interval was used to present the results of the final model. ROC curves, a widely used method to figure out the accuracy of a diagnostic test and establish a cut-off value for the diagnosis of the studied disease, were used to draw plots of sensitivity (true positive rate) by 1-specificity (false positive rate) at every test value, by dichotomizing patients into having the disease (here thrombocytopenia) or not. The cut-off point was determined as the value where the sensitivity and specificity are highest (Hajian-Tilaki, 2013).

In total, 59 patients met the inclusion criteria and were included in the study, representing 63 treatments (3 patients received 2 or 3 linezolid treatments during the study period). Patient baseline characteristics are described in Table 1. On average, the patients were old (median age = 65 years), slightly overweight (BMI = 28.1 kg/m²), and presented with mild renal insufficiency (median GFR = 69 mL/min/1.73 m²). Median of GFR was significantly lower among inpatients (62 mL/min/1.73 m²) than outpatients (82 mL/min/1.73 m²) [p-value, 0.037]. The majority of patients received their treatment at the hospital (76.2%), with a minority in the intensive care unit (12.5% of inpatients).

Linezolid was administered as 600 mg twice daily via the oral (n = 54/63) or intravenous (IV, n = 9/63, with one switch from IV to oral) routes. It was prescribed as a first line in 12 patients (19%) or second line after vancomycin in 21 patients (33.3%), or after another antibiotic in 30 patients (47.7%).

The reasons for selecting linezolid were: oral availability [switch to an oral therapy (14.3%) or only oral option (7.9%)], efficacy [susceptibility profile of the isolated strain (49.2% including 4.8% of VRE) or no effect of the first prescribed antimicrobial (4.8%)], safety reasons [renal insufficiency (6.3%), allergies to other drugs (3.2%), or ADRs of the first prescribed antimicrobial (4.8%)] or not specified (9.6%).

The type of infections treated with linezolid and the corresponding median treatment duration are summarized in Table 2. A majority of patients were treated for BJI, SSSI, secondary bacteremia or endocarditis. Treatment durations for the different types of infections were in the limit recommended in the SmPC (≤28 days), but longer for 10/22 (45%) patients with BJIs (range = 9–121 days). These infections were mainly caused by enterococci (25.4% E. faecium and 4.8% vancomycin-resistant enterococci), methicillin-resistant Staphylococcus epidermidis (20.6%), methicillin-resistant Staphylococcus aureus (19%).

Linezolid concentrations were measured in 120 blood samples. On average, 2 samples were collected per patient (range 1–6 depending on treatment duration). 16/120 samples were not collected at the target 12 h of sampling (C_min) due to practical reasons; in these cases, the C_min at 12 h was estimated based on simulations using the population-PK model described by Plock et al. (Plock, Buerger et al., 2007) and included in the final calculations. High interpatient PK variability for linezolid was observed, with C_min values ranging from the lower limit of quantification to 46.4 mg/L for the same 600 mg dose (Supplementary Figure S1). Renal function was identified as the main factor affecting C_min (Supplementary Figures S1, S2; Supplementary Table S2). Patients with C_min > 8 mg/L had a lower GFR (median value: 58.5 mL/min/1.73 m²) than those with C_min ≤ 8 mg/L (median value: 96 mL/min/1.73 m²; p = 0.004 for this difference), so that patients with mild to severe renal insufficiency were at higher risk of overdosing (Figure 1). Age was also positively correlated with C_min (r_s = 0.5509; p-value <0.0001), although this association is likely the result of the confounding effect of age on GFR (negative correlation, r_s = −0.6123; p-value < 0.0001).

Subgroup analyses were performed in order to identify specific subpopulations that may be at higher risk of under- or overexposure (Supplementary Table S2). Obese patients (n = 24) showed a median C_min of 8.3 mg/L, which is actually at the upper limit of the target interval. As expected, patients with GFR >100 mL/min/1.73 m² (n = 14) had significantly lower C_min than those with GFR <100 mL/min/1.73 m² (median C_min value: 3.8 mg/L vs. 10.8 mg/L). Importantly, all 6 ICU patients had low C_min values (0–5.7 mg/L). Conversely, patients with hepatic disorders (n = 8) tended to have elevated C_min (median value: 8.7 mg/L). Of note, among 4 patients with GFR between 60 and 90 mL/min/1.73 m² and C_min > 20 mg/L, two had hepatic disorders (Supplementary Figures S1, S2).

Overall, only 38.3% (46/120) C_min values were within the recommended 2–8 mg/L therapeutic range, with 48.3% (58/120) values >8 mg/L (supratherapeutic) and 13.3% (16/120) values <2 mg/L (subtherapeutic). Correcting for the number of samples per patient, these results indicate that more than half of the patients (35/63) were exposed to supratherapeutic concentrations while 5 were underdosed. Linezolid dose adaptation based on TDM results was applied in 5 patients with C_min > 8 mg/L (reduction from 600 mg twice to once daily) and in 1 patient with C_min < 2 mg/L (increase from 600 mg twice to trice daily) (Supplementary Figure S3). Dose reduction brought C_min back in the target values for 4/5 patients but did not prevent thrombocytopenia in 2 of them, either because platelets were already decreasing before dose readjustment (patient B.15) or because they were already low from day 1 (patient B.26). Dose increase in patient B.10 brought C_min back into the desired range.

ADRs occurred in 52.4% of linezolid treatments, with up to 4 ADRs per patient. These ADRs and their corresponding Naranjo probability score are shown in Table 3. According to this score, the majority of ADRs were probably associated with linezolid. Hematological disorders were detected in 20 patients, among whom 11 developed thrombocytopenia; 2 anemia; and 7 both anemia and thrombocytopenia. Thrombocytopenia and anemia appeared after a median of 13 days and 18 days, respectively. Five patients had a thrombocytopenia of grade 2 (50–75 10⁹ platelets/L) when their treatment was stopped. Among patients developing anemia, 6 received a blood transfusion, and among those with thrombocytopenia, 2 received a platelet transfusion. Importantly, among the 20 patients with hematological toxicity, the median trough value was 12.4 mg/L and 15 showed trough values above the recommended therapeutic range (2–8 mg/L, Figure 2). Patients developing thrombocytopenia had a treatment duration ≥9 days and the majority (15/17 with C_min values) showed supratherapeutic linezolid levels (>8 mg/L).

Gastrointestinal disorders were noticed in 18 patients and included diarrhea, nausea, vomiting, or loss of appetite. Metallic taste occurred in 9 patients and appeared between 9 and 17 days. Six patients experiencing metallic taste also developed thrombocytopenia, with metallic taste reported earlier in 5/6 patients. Higher C_min (median >8 mg/L) were observed in patients developing thrombocytopenia, anemia or metallic taste, but the difference with patients without ADR reached significance only for thrombocytopenia; Supplementary Figure S4).

No serotonin syndrome was observed during the study, despite the mention of comedications that can increase the risk in patients’ files (Supplementary Table S3). Twenty-five patients took at least 1 serotonergic agent; 10 patients, 2 serotonergic agents; and 1 patient, 3 serotonergic agents. Antidepressants were prescribed in 16 patients (7 patients with 2 antidepressants).

Eleven patients had to stop their linezolid treatment because of toxicity, among whom, 10 for hematological toxicity (after 9–34 days of treatment) and one, for paresthesia after 109 days. In this last patient, the trough value was 4.9 mg/L when the treatment was stopped (Figure 2).

Supplementary Tables S4, S5 shows individual Gonzalez-Del Castillo and Buzelé scores for patients developing hematological toxicity or any ADR. The predictive score for hematological toxicity developed by Gonzalez-Del Castillo et al. (Gonzalez-Del Castillo, Candel et al., 2017) did not prove useful to predict the risk of adverse reaction in our population, as all patients had a score <5 (Table 4). Conversely, the Buzelé et al. score (Buzele, Lemaignen et al., 2015) was ≥7 and significantly higher for patients developing any type of ADR [p-value, 0.001] (Table 4). The specificity and sensitivity of the Buzelé test for the detection of ADR are 76.7% (95% CI = 59.7–89.2%) and 66.6% (95% CI = 49.8–81.1%), respectively (Supplementary Figure S5).

As thrombocytopenia was the most frequent ADR in our population, our further analyses focused on this ADR. A significant difference among patients with (n = 18) and without (n = 45) thrombocytopenia was observed for the following parameters (Supplementary Table S6): median linezolid C_min (15.3 mg/L vs. 7.6 mg/L), treatment duration (21 days vs. 10 days), GFR (50 mL/min/1.73 m² vs. 74 mL/min/1.73 m²), basal platelet count (253 vs. 338 10⁹/L), and previous vancomycin treatment (55.6% vs. 24.4% of patients). A positive correlation was found between the decrease in platelet count during treatment compared to baseline levels and average trough concentrations (r_s = 0.3642; p-value = 0.0046) (Supplementary Figure S2). Interestingly, while the risk of thrombocytopenia increased as treatment duration was extended, 3 patients completed >80 days of linezolid therapy without signs of hematological toxicity. Of note, their trough values remained within the 2–8 mg/L therapeutic range (Figure 2).

Table 5 shows the results of the univariate and multivariate analysis for the development of thrombocytopenia (see Supplementary Figure S6 for the establishment of cut-ff values to <150 10⁹/L for basal platelet count, <60 mL/min/1.73 m² for GFR [similar cut-off value if using absolute GFR; see Supplementary Figure S7], ≥10 days for treatment duration, and >8 mg/L for C_min). Based on the results of the univariate analysis, we decided to exclude outpatients from the multivariate analysis as this parameter showed a VIF >2 and is generally associated with a longer treatment duration (median treatment duration in hospitalized patient = 10 days vs. 32 days in ambulatory patients). We also excluded age, which was negatively correlated with renal function. In the resulting multivariate analysis, factors significantly associated with thrombocytopenia were treatment duration longer or equal to 10 days and C_min > 8 mg/L.

Although this study was not designed to perform systematic dose readjustment based on TDM data, we exploited the data from patients with a sample collected for platelet counts 7 days after the determination of linezolid C_min (at day 7 or 10) in order the determine whether elevated C_min values could predict the risk of developing thrombocytopenia in the forthcoming days. Figure 3 shows the change in platelets counts from the pretreatment value (day 0) at day 7 (day at which C_min was measured) or 1 week later (day 14). In most of the cases, platelet counts at day 7 were close to the values at day 0 with only 2/19 patients showing >30% decrease in platelet counts. Contrarily, 1 week later, 10/19 patients showed >30% decrease in platelet counts, among whom 5 with C_min > 8 mg/L. Seven out of these 10 patients had also platelets counts <150 10⁹/L (not shown), corresponding to the definition of thrombocytopenia adopted in this study.

Supplementary Figure S8 shows the evolution over time of platelets counts in patients for whom at least 2 TDM values were available (panel A) or having developed thrombocytopenia (panel B). Although these data are fragmental, they seem to confirm that elevated C_min may precede platelet count reduction. In four patients where linezolid was stopped after 7–14 days and experiencing high C_min without thrombocytopenia, platelet numbers continued to decrease over the next week, suggesting the slow reversibility of the process in these patients (Supplementary Figure S9).