Increased red cell distribution width predicts severity of drug-induced liver injury: a retrospective study

We used a retrospective study design to evaluated the predictive value of red cell distribution width (RDW) for drug-induced liver injury (DILI) severity in Chinese patients with liver biopsy to assist with early DILI management. We included 164 DILI patients with complete laboratory information and medical history. We compared outcomes of 36 patients with severe DILI with outcomes of a control group of 128 patients with mild-to-moderate DILI. Multivariate analyses of risk factors for severe liver injury in Chinese patients with DILI revealed an estimated adjusted odds ratio (AOR) (95% CI) of 4.938 (1.088–22.412) in patients with drinking. Risk for serious liver injury was also increased significantly in patients with dyslipidemia [AOR (95% CI) 3.926 (1.282–12.026)], higher serum total bile acid (TBA) levels [AOR (95% CI) 1.014 (1.009–1.020)] and higher RDW [AOR (95% CI) 1.582 (1.261–1.986)]. The result for area under the curve of 0.905 for TBA levels indicated this variable had high diagnostic performance for predicting DILI severity. Based on an area under the curve value of 0.855, RDW also had superior diagnostic performance in prediction of DILI severity. This performance was not significantly different compared with TBA and was superior compared with other variables, which had area under values ranging from poor to failure (0.527–0.714).The risk for severe DILI was associated with drinking, dyslipidemia, higher TBA levels and RDW values. This study found that RDW and TBA levels were predictors of DILI severity in Chinese patients.

Clinical characteristics of patients with DILI are shown in Table 2. The most prevalent clinical manifestations were fatigue (61.6%) and jaundice (42.1%). Abdominal distension and nausea accounted for 32.3% and 14.6%, respectively. A total of 12.2% of the patients had no symptoms or clinical signs.
Therapeutic classes and uses of drugs. The therapeutic classes of drugs used by participants in the DILI group are listed in Table 3. Of note, 79 DILI patients (48.2%) used Chinese herbal medicines, 56 (34.1%) used Western medicines, and 29 (17.7%) used a combination of the two. These drugs are suspected to be the cause of liver injury in DILI patients. To further evaluate the indication of the herbal drugs, we subdivided the 79 patients Table 1. Baseline characteristics of the study population. Continuous variables are expressed as median (25th, 75th percentiles) values. Categorical variables were displayed as numbers and percentages. WBC white blood cell, RDW red blood cell distribution width, MCV mean corpuscular volume, AST aspartate aminotransferase, ALT alanine aminotransferase, TBIL total bilirubin, ALP alkaline phosphatase, GGT gamma-glutamyltransferase, TBA total bile acid, PT prothrombin time, INR international normalized ratio.   Fig. 1).
Maximizing the sum of sensitivity and specificity, the optimal cut-off for TBA was 45.850 for diagnosis of severe DILI. The results indicated that the sensitivity was 100% and the specificity was 74.2% for diagnosis of severe DILI. The correct classified percentage was 79.3%. The optimal cut-off for RDW was 14.650 for diagnosis of severe DILI. The sensitivity was 86.1% and the specificity was 73.4% for diagnosis of severe DILI. The correct classified percentage was 76.2%.

Discussion
The most important findings of this study were that the results further supported the hypotheses that there was a positive correlation between TBA levels and DILI severity and that RDW predicted DILI severity. At a cutoff of 14.650, RDW predicted severe DILI with a sensitivity of 86.1% and a specificity of 73.4%. The percent correctly classified could be as high as 76.2% with an AUROC of 0.855. These diagnostic capabilities were similar to TBA and were higher than those for other non-invasive markers.
In hepatic diseases, such as viral hepatitis, fatty liver disease, and liver cirrhosis 21,22 , serum bile acid (BA) levels can increase from 10-to 100-fold [21][22][23][24][25] . Compared with serum bilirubin, BA measurement may be a more sensitive method to assess hepatic function 23 . Changes in serum BAs correlate with the severity of hepatic dysfunction. Horvatits et al. 's results indicated that TBA levels can be used as markers to stratify risk in patients with cirrhosis with respect to new onset of acute decompensation and acute-on-chronic liver failure 21 .
Numerous studies have examined associations between RDW levels and liver-related disease progression and outcomes. One large sample size cohort study found that RDW is predictive marker for advanced fibrosis in Table 3. Therapeutic classes of drugs that caused liver injury in 164 Chinese patients. www.nature.com/scientificreports/ patients with nonalcoholic fatty liver disease 26 . Fan et al. found that RDW is elevated in patients with CHB and that there is a positive correlation between these changes and CHB severity 27 . RDW is significantly increased in patients with HCC; it represents an easily measured marker of prognosis in these patients 28,29 . Abnormalities including inflammation, erythrocyte fragmentation, oxidative stress, poor nutritional condition, and abnormality of erythropoietin function can cause significant variation in RDW 5,30-34 . Because these disorders and anemia correlate with liver disease severity, elevated RDW values might also be associated with the severity of liver disease. Inflammation results in impairment of erythrocyte maturation and entry of immature erythrocytes into the systemic circulation that results in elevated RDW values 19,35 . Study results suggest that inflammatory cytokines (e.g., tumor necrosis factor-α, interleukin (IL)-1β, and IL-6) inhibit iron metabolism and erythropoietin production. This process results in disorders of RBC synthesis and abnormal erythropoietin production [35][36][37] . Impairment of the balance between oxidant and antioxidant defenses are characteristic of the oxidative stress that occurs in liver disease. Erythrocyte homeostasis and survival is strongly affected by oxidative stress, and low serum antioxidant concentrations are correlated with increased RDW levels 38,39 . Taken together, this suggests oxidative stress results in increased RDW levels in patients with liver disease. A liver disease-associated poor nutritional status (e.g., iron deficiency, folate deficiency, vitamin B12 deficiency) can also result in abnormal RBC production and increased RDW levels 40 . Patients with chronic liver disease can experience the common complication of portal hypertension. This condition can cause splenomegaly with an associated increase in the rate of RBC destruction and release of immature RBCs to the systemic circulation and consequent increase in RDW 38,39 . Table 5. Univariate and multivariate analyses of variables associated with severe drug-induced liver injury. Continuous variables are expressed as median (25th, 75th percentiles). Categorical variables were displayed as numbers and percentages. INR international normalized ratio, RLR red blood cell distribution width to lymphocyte ratio, RPR red blood cell distribution width to platelet ratio, AAR aspartate aminotransferase to alanine aminotransferase ratio. # P value for univariate analysis. *Adjusted for sex, age, smoking, drinking, allergic history, DM, hyperlipemia, history of liver disease, TBA, RDW, RPR, RLR, and AAR. **P value for multivariate analysis.

Variables
Level 0-2 (N = 128) Level ≥ 3 (N = 36) P # AOR (95% CI)* P** www.nature.com/scientificreports/ Our study did not find relationships between RPR, RLR, and severe DILI. One explanation for this result is that the patients were not further analyzed (because of the sample size) to determine whether they had liver cirrhosis, which would have influenced PLT levels. Innate/adaptive immune responses have key roles in the biological mechanisms of DILI 41 and could also could affect blood cell levels. Consistent with the results of other studies, we found associations between dyslipidemia, drinking, and the severity of DILI 2 . The detailed mechanisms have been described in our previous study 2,41 .
There are some limitations to our study. First, the study's retrospective design might have caused selection bias that resulted in underestimated sensitivity and overestimated specificity values in the serum models 42 . Second, detailed information about lipid-lowering drugs and hypoglycemic agents was not available. More study is needed to understand associations between drugs used for metabolic disease and severe DILI. Third, the patients Table 6. Diagnostic performance of serum models for severe drug-induced liver injury. The bold values were considered statistically significant (P < 0.05). Se, sensitivity; Sp, specificity. Cut-offs were established by maximizing the sum of sensitivity and specificity. AUROC area under the receiver operating characteristic curve, 95% CI 95% confidence interval.  www.nature.com/scientificreports/ included in our study were not combined with other liver disease to tried to avoid factors which might influence RDW value, however we cannot avoid other system diseases because that might the reason patients use drugs. Fourth, because the study did not include a large sample size, subgroup analyses by acute and chronic DILI could not be performed. The number of cases was limited by the requirements to include patients with a diagnosis of DILI with liver biopsy and exclude patients without complete medical information.
In conclusion, risk for severe DILI was associated with drinking, dyslipidemia, higher TBA levels, and higher RDW. This study found that RDW and TBA levels may be valid for non-invasive, real-time monitoring of DILI severity in Chinese patients.

Patients. Data from 2097 patients who underwent standard laboratory tests and liver biopsies (The First
Hospital of Jilin University, China) between January 2010 and December 2019 were assessed for inclusion in this retrospective study. Test and biopsy results indicated that 187 patients had a diagnosis of DILI. After excluding the data from 23 patients who had incomplete medical information, the data from 164 patients with complete medical histories and laboratory information were included in the analysis.
Subjects were excluded due to the following The exclusion criteria: adopted were (1) co-infection with human immunodeficiency virus or hepatitis B virus (HBV) or hepatitis C virus (HCV); (2) the evidence or history or other evidence of hepatitis (3) presence of other types of liver disease, such as (e.g., non-alcoholic fatty liver disease or, alcoholic liver disease, et al.
The Independent Institutional Review Board of The First Hospital of Jilin University approved the study protocol and the use of data from human subjects. All participants in the study signed informed consent.
Liver biopsy. Each ultrasound-guided percutaneous liver biopsy was performed using the Menghini technique. All liver samples were preserved in phosphate-buffered formalin and then paraffin-embedded and sectioned and stained for histology. Liver activity and fibrosis were scored using the Metavir system 43 . We excluded tissue sections with fewer than three portal tracts (i.e., as poor quality). The pathologists were blinded to all clinical data. If their conclusions were different, the slides were examined by a another experienced hepatopathologist in the UK who was blinded to all data.

Diagnosis of DILI.
Dyslipidemia was defined using criteria from the 2015 Chinese Medical Association guidelines for DILI diagnosis and treatment 44 : All patients' s RUCAM score more and equal to 6, and the presence of liver injury was finally diagnosed based on liver histology of percutaneous liver biopsies.
Diagnosis of DILI severity. The 2015 Chinese Medical Association guidelines for DILI diagnosis and treatment were used to classify severity into five categories (level 0: exposure to drug, but no liver injury; level 1: mild, typically reversible elevations of serum enzyme activities only, total bilirubin (TBil) < 2.5 the upper limit of normal (ULN) value and international normalized ratio (INR) < 1.5; level 2: more extensive injury with early impairment of liver function indicated by increases in alanine aminotransferase (ALT) and/or alkaline phosphatase (ALP), TBil ≥ 2.5 ULN or INR ≥ 1.5; level 3: serious clinical illness in conjunction with obvious jaundice and disabling symptoms (TBil ≥ 5 ULN and/or INR ≥ 1.5); level 4: increases in ALT and/or ALP, TBil ≥ 10 ULN or TBil increases ≥ 17.1 μmol/L per day, INR ≥ 2.0 or prothrombin activity < 40% in addition to secondary loss of other organ functions (e.g., encephalopathy or hepatorenal syndrome); level 5: liver required transplant or death of patient 44 . Diagnosis of dyslipidemia. Dyslipidemia was defined using criteria from the National Cholesterol Education Program Adult Treatment Panel III (ATPIII), which were total cholesterol greater than 240 mg/dL, HDL cholesterol less than 40 mg/dL, LDL cholesterol greater than or equal to 160 mg/dL, or triglycerides greater than or equal to 200 mg/dL 45 .

Study variables.
Demographic characteristics (e.g., age, sex, drinking, smoking) and variables associated with the clinical presentation (presence of hypertension, history of hypersensitivity, presence of dyslipidemia, type 2 diabetes) were included in this study.
At the time of liver biopsy, fasting blood samples were obtained for standard laboratory tests. Retrospective data on white blood cell (WBC), neutrophil, and lymphocyte counts, and on hemoglobin, RDW, platelet counts, mean corpuscular volume, serum aspartate aminotransferase (AST), ALT, gamma glutamyl transferase, ALP, TBil, TBA, and globulin levels were obtained from the patient's medical record.
The INR, prothrombin activity, and prothrombin time were also analyzed in the patients with DILI and were used to classify severity of liver injury.

Statistical analysis.
The results for continuous variables were calculated as median, and 25th and 75th percentile values and analyzed using two-tailed independent sample t-tests. Categorical variables were summarized as numbers/percentages and analyzed using Chi-squared tests. To adjust for potential confounding effects, www.nature.com/scientificreports/ multivariate logistic regression analyses were performed, including adjusted odds ratios (AORs) and 95% confidence intervals (CIs). We used SPSS software (version 13, SPSS Inc., Chicago, IL, USA) to perform the statistical analyses. All tests were two-tailed. We considered P values < 0.05 to be statistically significant. Receiver operating characteristic (ROC) curves and the area under the ROC (AUROC) curve were used to evaluate and compare the accuracy of AAR, RPR, RLR, TBA levels, and RDW for the diagnosis of DILI severity. ROC curve analysis and Z-tests were used to compute and compare AUROCs, respectively (MedCalc Software bvba, version 16, Ostend, Belgium). Maximizing the sum of sensitivity and specificity or optimizing a specificity of at least 95% were used to obtain cut-offs. All methods were carried out in accordance with relevant guidelines and regulations.