The risk factors for linezolid-induced lactic acidosis in patients older than 85 years

Lactate is considered a prognostic indicator in critically ill patients; however, studies on linezolid-induced lactic acidosis (LILA) are limited, and data from patients older than 85 are even more scarce, therefore we evaluated the risk factors for LILA in patients older than 85 years and established a risk prediction model. In a retrospective cohort study, patients older than 85 years who were monitored for blood gas analysis and arterial lactate levels during the use of teicoplanin or linezolid were enrolled. After using propensity score-matched analyses we compared the incidence of lactic acidosis between teicoplanin or linezolid therapy and identified the risk factors of LILA. with high stability.


Introduction
Lactate is produced by anaerobic glycolysis, mainly in the skeletal muscles, skin, erythrocytes and central nervous system [1] . Clinically, elevated lactate levels often represent hypoxia in tissues, so lactate is commonly used to evaluate tissue perfusion and the prognosis for critically ill patients [2,3] . It has been reported that a slight increase in lactate levels leads to a higher mortality rate [4,5] .
However, elevated lactate levels caused by drugs do not necessarily indicate hypoxia, and the lactate levels gradually decrease back to the normal range after drug withdrawal.
However, there have been no large-sample studies on the risk factors for LILA and no relevant data exist pertaining to the extremely elderly population. Hence, we analysed the risk factors for LILA and established a risk prediction model.

Study Design and Population
This was a retrospective cohort study using a case-control study conducted at the Second teaching hospital located in Beijing, China.
We included patients older than 85 years who were monitored for blood gas analysis and arterial lactate levels during the use of teicoplanin or linezolid from October 2016 to April 2019 in our hospital. To compare the incidence of lactic acidosis between teicoplanin and linezolid therapy, the baseline characteristics of patients were adjusted using the propensity score matching. Patients with linezolid therapy were invided into lactic acidosis group and non-lactic acidosis group, and then we evaluated the risk factors of LILA.
Patients with shock; those who used drugs that affect lactate levels, such as metformin, salicylates, and nucleotide reverse transcriptase inhibitors; patients with respiratory failure or liver failure (Child-Pugh classification C); and those receiving renal replacement therapy were excluded.
Lactic acidosis was defined as a serum pH < 7.35 and arterial lactate ≥ 3.5 mmol/L in our study. The baseline and the end-point lactate levels were obtained in blood gas analyses performed when teicoplanin or linezolid therapy was started and stopped.

Data Collection
The following baseline clinical and laboratory variables were collected retrospectively from the electronic medical record system: sex; age; duration of linezolid therapy; infection site; the use of invasive ventilation; comorbid diseases such as chronic obstructive pulmonary disease (COPD), pulmonary fibrosis, coronary heart disease, hypertension, atrial fibrillation (AF), diabetes mellitus (DM), chronic kidney disease (CKD), neurological disease, malignant tumour, and thyroid hypofunction; laboratory indexes including the levels of serum creatinine, albumin, haemoglobin, creatine kinase and lactate dehydrogenase, alanine dehydrogenase (ALT), aspartate transaminase (AST), troponin I (TNI), pro-brain natriuretic peptide (pro-BNP), D-dimer, and arterial blood sugar and the estimated glomerular filtration rate (eGFR); the sequential organ failure assessment (SOFA) score; and arterial lactate levels at baseline and the end-point.

Statistical Analysis
Quantitative datas with normal distribution were expressed as mean and standard deviation and analysed by t tests. Quantitative datas with abnormal distribution were expressed as medians and interquartile range (IQR) assessing by Mann-Whitney U tests.
Categorical variables were described as frequency, and comparison was performed by the Chi-square test. To adjust for the significant differences in baseline characteristics of patients, we used propensity score matching by implementing nearest neighbor matching in a 1:1 ratio. Factors with significant differences in the univariate analysis were entered into a multivariate binary logistic regression model (forward: LR) to determine their independent effects. The results of the binary logistic regression model are presented as odds ratios (ORs) and the associated 95% confidence intervals (CIs). The sensitivity and specificity of the risk prediction model were tested by receiver operating characteristic (ROC) curve analysis. Variables with p-values less than 0.05 were considered statistically significant. All analyses were performed using the IBM SPSS statistical software package version 23.0 (SPSS, Chicago, IL, USA).

Patient characteristics and clinical factors
In this retrospective cohort study, 199 and 108 patients were administered teicoplanin or linezolid therapy, and the patient characteristics and clinical factors were shown in Table   1. Infection site, underlying disease (coronary heart disease, atrial fibrillation and neurological disease) and sequential organ failure assessmen (SOFA) were significant different between the two groups(p<0.05, Table 1). we used propensity score matching to adjust the significant differences in baseline characteristics of the two groups. As a result, 98 patients each in the teicoplanin or linezolid therapy groups were matched. The balance in baseline characteristics between two groups was improved considerably (Table 2).

Arterial lactate at baseline and the end-point of patients in teicoplanin and linezolid therapy groups
In the matched pairs, the incidence of lactic acidosis between teicoplanin and linezolid therapy groups were 0%(0/98) vs 35.7%(35/98) with significantly difference (p<0.0001, Table 3). No significant difference was found at baseline arterial lactate between the two groups, while the arterial lactate at end-point in linezolid therapy groups were significantly higher than that in teicoplanin therapy groups (p<0.0001, Table 3, Figure   1).

Univariate analysis of risk factors of linezolid-induced lactic acidosis
In linezolid therapy group, 35 and 63 patients were included in the lactic acidosis group and non-lactic acidosis group, respectively. The characteristics of the patients are described in Table 4. No significant differences were observed between the two groups in terms of sex, age, infection site, use of invasive ventilation and prevalence of comorbid diseases (COPD, pulmonary fibrosis, coronary heart disease, coronary heart disease, atrial fibrillation, diabetes mellitus, neurological disease, malignant tumour, thyroid hypofunction). The median duration of linezolid therapy was 10 [7,12] days in the lactic acidosis group and 8 [5,11] days in the non-lactic acidosis group (p=0.053, Table 4). The numbers of patients with CKD in the lactic acidosis group and the non-lactic acidosis group were 24 (68.6%) and 31 (49.2%) (p=0.064,  Table 4). The clinical parameters such as serum creatinine, haemoglobin, TNI, pro-BNP, D-dimer, and arterial blood glucose levels and the eGFR were significantly different between the two groups (p<0.05, Table 4). The SOFA scores of the two groups were 10 [9,15] and 6 [5,9] (p<0.0001, Table 4), respectively. The 30-day mortality rates were 48.6% and 28.6%, respectively, which were significantly different (p=0.015, Table 4, Figure 2 Table 5) and a high sequential organ failure assessment (SOFA) score (OR, 1.429; 95% CI, 1.213-1.685; p<0.0001, Table 5) were associated with linezolid-induced lactic acidosis.

Establishment of the risk prediction model
The risk of LILA can be predicted by three factors: the duration of linezolid therapy, the arterial blood glucose level and the SOFA score. According to the fourfold table (Table 6), the sensitivity of the risk prediction model was 100%, the specificity was 80%, the negative predictive value was 100%, and the positive predictive value was 58.3%.

Discussion
Lactate is an important product of cell metabolism during anaerobic glycolysis. Clinically, lactic acidosis can occur due to either excessive production or impaired metabolism. In addition to type A lactic acidosis caused by tissue hypoxia, common drugs such as nucleotide reverse transcriptase inhibitors, salicylates, and metformin cause type B lactic acidosis by interfering with oxidative phosphorylation when there is no obvious tissue hypoxia [1] .
Linezolid is a major tool in the treatment of MDR gram-positive pathogens and MDR-TB and can also cause type B lactic acidosis. We found that all patients showed different degrees of elevated lactate levels after using linezolid (1.2 [0.9,1.4] vs. 2.6 [1.8,3.7], p < 0.0001), and 35.7% had lactic acidosis; this percentage was higher than those in previous reports [17][18][19][20] . This may be because patients with mild disease were excluded due to the absence of blood gas monitoring, while the included patients had relatively severe disease, with multiple comorbid diseases in our study. In addition, patients who were not monitored for lactate in some previous studies were included in the non-lactic acidosis group, which led to the underestimation of LILA [20] .
The 30-day mortality rate was 48.6% in the lactic acidosis group, which was significantly higher than that in the control group (p < 0.05). In a systematic review and meta-analysis of 47 cases of LILA retrieved from PubMed, 25.5% of the patients died, indicating a high risk of death with LILA [21] .
Previous reports have shown that LILA is associated with a longer duration of medication [15,16,[22][23][24] . In a retrospective study, a duration of linezolid therapy > 6 weeks was a risk factor for LILA [19] , but LILA has been reported to occur after a shorter duration of linezolid therapy (4 hours-7 days) [21,[25][26][27][28] and to occur even earlier in children, with a median time of 2 (1,13) days [29] . Del Pozo showed that the median duration of the administration of linezolid in lactic acidosis patients was 8 days [20] . We found that a duration of linezolid therapy ≥ 9 days was a risk factor for LILA in the superelderly population, which suggested that LILA in the super-elderly population could occur after a short course of medication; therefore, the early and routine monitoring of lactate levels and blood gas is necessary.
We found that when the arterial blood glucose level was ≥ 8 mmol/L, the risk of lactic acidosis was increased. A high SOFA score is also a risk factor for LILA, and the risk of lactic acidosis increased 0.429 times for every 1 point increase in the SOFA score, suggesting that super-elderly patients with high blood glucose or sequential organ failure are more prone to lactic acidosis. One study of 10 cases of LILA found that a SOFA score ≥ 11 and duration of linezolid therapy ≥ 7 days were not risk factors for LILA [20] . This was inconsistent with our results, which may be related to the univariate analysis employed in the other study. Del Pozo also found that an eGFR ≤ 30 mL/min (OR, 7.4; 95% CI, 1.0-84.4, p = 0.02) was a risk factor for LILA; however, we did not find that the eGFR was associated with LILA [20] .This difference may have occurred because 30% of lactate metabolism occurs in the kidney, and only when lactate is above 6-10 mmol/L can it be excreted by the kidney [1] . Therefore, the eGFR may only be associated with severe hyperlactic acidaemia.
In our study, the lactate levels were mostly mildly to moderately elevated, so no correlation was found between LILA and the eGFR.
Based on the multivariate logistic regression analyses, we established a risk prediction model with high sensitivity and specificity (91.4% and 65.1%, respectively) for the occurrence of LILA, and the cut-off value was 0.2825. After verifying the model in 32 patients from another medical centre, it showed high stability. Therefore, the risk prediction model can be applied in the super-elderly population.
The mechanism of LILA is unclear. Linezolid inhibits 23 s rRNA from the 50S subunit of the bacterial ribosome, which is similar to human mitochondrial 16S rRNA. Hence, linezolid may produce toxic mitochondrial effects by binding to human mitochondrial 16S rRNA and inhibiting mitochondrial protein synthesis [24] . Human mitochondrial DNA polymorphisms (A2706G and G3010A) have been reported to be associated with LILA [30,31] , although this finding is controversial due to the high frequency (up to 80%) of these polymorphisms and the relatively rare occurrence of LILA [32] .
To the best of our knowledge, we are the first to analyse the occurrence of LILA in patients older than 85 years with a large sample size. Moreover, we established a risk prediction model to predict the occurrence of LILA. There are some limitations. First, the sample size was relatively small. Second, the incidence of and mortality from LILA may have been overestimated due to the relatively severe illness of the included patients.
Third, linezolid plasma concentrations was not tested. The mechanism underlying LILA remains to be further studied.       Figure 1 Lactic acidosis at baseline and the end-point  Screening of patients usinging linezolid during the study period