A retrospective cross-sectional study for predicting 72-h mortality in patients with serum aspartate aminotransferase levels ≥ 3000 U/L

Risk factors associated with 72-h mortality in patients with extremely high serum aspartate aminotransferase levels (AST; ≥ 3000 U/L) are unknown. This single-centre, retrospective, case-controlled, cross-sectional study obtained data from medical records of adult patients treated at Saitama Medical Center, Japan, from 2005 to 2019. We conducted a multivariate logistic after adjusting for age, sex, height, weight, body mass index, Brinkman Index, vital signs, biochemical values, updated Charlson Comorbidity Index (CCI) score, CCI components, and underlying causes. A logistic regression model with selected validity risks and higher C-statistic for predicting 72-h mortality was established. During the 15-year period, 428 patients (133 non-survivors and 295 survivors [cases and controls by survival < 72 and ≥ 72 h, respectively]) with AST levels ≥ 3000 U/L were identified. The 72-h mortality rate was 133/428 (31.1%). The model used for predicting 72-h mortality through the assessment of alkaline phosphatase, creatine kinase, serum sodium, potassium, and phosphorus levels had a C-statistic value of 0.852 (sensitivity and specificity, 76.6%). The main independent risk factors associated with 72-h mortality among patients with AST levels ≥ 3000 U/L included higher serum values of alkaline phosphatase, creatine kinase, serum sodium, potassium, and phosphorus.

www.nature.com/scientificreports/ www.nature.com/scientificreports/ Prediction model. Five covariates (i.e., ALP, CK, Na, K, and P) that were determined to be the most significant predictors were selected for stepwise regression analysis. The best combination of covariates that predicted the 72-h mortality risk with the highest C-statistic comprised serum ALP, CK, Na, K, and P (C-statistic: 0.852), and surpassed the performance of K alone obtained during multivariate logistic regression analysis (MLRA) ( Table 4). The predictive probability (p) for 72-h mortality of the final prediction model was calculated using the following formula: This model had 76.6% sensitivity and specificity for predicting 72-h mortality among patients with extremely high outlier values of AST (Table 5; Fig. 2).

Discussion
This study revealed that an extremely high outlier value of AST (≥ 3000 U/L) was a critical value. Because the 72-h mortality rate among 133 patients with AST ≥ 3000 U/L was 31.1%. The mortality risk was associated with higher values of ALP, CK, Na, K, and P, regardless of serum AST levels and age. The reason for the association between the 72-h mortality rate among the participants with extremely high outlier values of AST and high serum ALP, CK, Na, K, and P remains unknown. The distinctive profile of underlying causes among non-survivors (i.e., significantly higher proportions of sepsis and significantly lower proportions of hepatic disease and heart failure) may be one of the reasons. However, this does not imply that patients with cardiac or liver disease have a better prognosis and, instead, may indicate that advances in treatments and interventions for these two disorders have decreased the likelihood of short-term mortality within 72 h in patients with markedly increased AST. The long-term prognosis of these patients still needs to be closely monitored.
This study showed a significant difference in AST levels between survivors and non-survivors. One study showed that, among patients with severe hypoxic hepatitis, indicated by mean AST levels of 4896 U/L, low serum albumin levels predicted early mortality 14 . Another study reported a strong relationship between aminotransferase concentrations, including AST, and mortality from liver disease 15 . Koehler et al. reported that AST was positively associated with all-cause mortality among the elderly population 16 , whereas a higher AST level (> 20 U/L) was reported to be incrementally associated with higher mortality among patients receiving maintenance haemodialysis 17 . However, the previously mentioned studies did not include extremely high outlier values for p = 1/ 1 + exp −19.98 + 0.290 ALP 0.2 + 0.348log (CK) + 1.766 Na 0.2 + 1.568 K 0.2 + 0.474 P 0.8 . www.nature.com/scientificreports/ www.nature.com/scientificreports/ AST, and the follow-up period in such studies was more than 1 year. Srivatsan et al. reported that the urinary albumin-creatinine ratio, ALT, AST, and prothrombin time/international normalized ratio were significantly correlated with APACHE II scores and mortality among patients with systemic inflammatory response syndrome 18 , suggesting that extremely high outlier values of AST are associated with short-or long-term mortality. The serum ALP level was associated with 72-h mortality among patients with extremely high outlier values of AST. Increased ALP levels could be a risk factor for mortality among patients with acute coronary syndromes 19,20 . Furthermore, ALP was related to long-term mortality in an elderly population 16 , suggesting that high ALP levels may affect short-term prognosis.
The median serum sodium value among the non-survivors included in this study was 139.0 mEq/L, which may be considered a normal value; however, a 1.0 mEq/L increase in the serum sodium concentration increased the odds ratio for 72-h mortality by 2.30. Hypernatremia > 150 mEq/L is uncommon 21,22 and it is an independent risk factor for inpatient mortality and poor outcomes among patients with acute subarachnoid haemorrhage 21 and those admitted to the intensive care unit with comorbid organ dysfunction 23 . Persistent hypernatremia is a strong risk factor for in-hospital mortality among patients with various diseases 22 . These reports support the association between higher sodium levels and poor outcomes among patients with extremely high outlier values of AST.
The non-survivors included here had a median serum phosphorus value of 6.60 mg/dL. A serum phosphorus level greater than 5.50 mg/dL was associated with a further increase in the risk of cardiovascular and all-cause mortality among patients with mild or moderate renal dysfunction 24 . In contrast, survivors in this study had a median phosphorus value of 5.00 mg/dL, which was significantly lower than that for non-survivors. This difference might have affected the early prognosis. Dhingra et al. reported an association between all-cause mortality within 45 months among patients with chronic kidney disease and hyperphosphataemia 25 . High serum phosphorus levels are associated with the risk of cardiovascular disease, which leads to high mortality rates 17 .  26 . Another study conducted at our hospital that included patients with extremely high CRP levels suggested that serum phosphorus might serve as a good predictor of 72-h mortality 27 . These findings suggest that phosphorus is associated with various types of diseases and might be an important variable for predicting short-term prognosis. Both the increase in RPG, T-Bili, AST, ALT, ALP, BUN, Cre, Na, K, Mg, P, and lactate levels as well as the decrease in haemoglobin, platelets, ALB, Ca, and bicarbonate levels were reported as end-of-life laboratory values 28,29 . Regarding other risk variables that were identified by the MLRA, K and CK are well-documented mortality risk factors. Hyperkalaemia prior to death can be explained by renal failure, apoptotic release from cells in multi-organ failure, or increased intravascular haemolysis. The elevation of serum CK levels is known to be commonly caused by hypoxaemia in muscle circulatory failure and leakage from damaged cells in skeletal and cardiac muscles. High CK and hyperkalaemia are linked to each other and are associated with renal failure.
Patients with a marked increase in aminotransferase levels (> 10 times the upper reference limit) typically have acute viral hepatitis or toxic or ischemic liver injury 6 . However, a previous study showed that extremely high outlier values of AST were not always attributable to ischemic hepatitis 7 . In this study, the major underlying causes of high AST value with a prevalence of 10% or more included sepsis, malignancy, heart failure, cerebrovascular disease such as aortic aneurysm or dissection, myocardial infarction, acute arterial obstruction, and stroke. As hypoxic hepatitis and ischemic hepatitis are pathological rather than clinical terms, it remains unclear as to what proportion of patients with malignant disease had hypoxic hepatitis among the non-survivors.

Limitations
Several limitations of this study need to be mentioned. First, considering that this was a single-centre retrospective study conducted at Saitama Medical Center in Japan, the findings presented herein may not be generalisable to other patients. Second, given that patients with AST levels equal to greater than 3000 U/L were selected, risk factors for 72-h mortality may be different among patients with elevated AST levels that are lower than the set threshold.

Conclusions and future research
This study found that an extremely high outlier value of AST (≥ 3000 U/L) could be a critical value that resulted in higher mortality rates and was associated with higher serum values of ALP, CK, Na, K, and P, regardless of AST values and age. Therefore, physicians treating patients with AST levels ≥ 3000 U/L can easily estimate the probability of 72-h mortality by focusing on ALP, CK, Na, K, and P levels to anticipate disease conditions, explain the prognosis to the patient and their family, and make appropriate decisions on subsequent treatments.
This study could not determine why 72-h mortality among those with extremely high outlier values of AST was associated with high serum ALP, CK, Na, K, and P levels. Moreover, the formula presented herein remains untested. Thus, these preliminary findings warrant testing in prospective studies to validate the performance of the mortality prediction model among patients with extremely high outlier values of AST to identify the mechanisms whereby higher ALP, CK, Na, and P levels lead to increased 72-h mortality, and to prevent mortality associated with extremely high outlier values of AST.
We believe that 72-h mortality prediction among patients with extremely high outlier values of AST can help physicians make prompt decisions and provide therapeutic and management options to patients and their families, which should, in turn, improve the quality of initial medical management.

Methods
The study protocol was designed in accordance with the tenets of the Declaration of Helsinki 30 . The Institutional Clinical Research Ethics Review Board of Saitama Medical Center, Jichi Medical University, Saitama, Japan approved this study (Clinical #10-79 and #S20-025) and waived the requirement to obtain informed consent because of the retrospective design.

Study design and patient selection.
This retrospective, single-centre, case-controlled cross-sectional study enrolled participants were selected through a chart review of the medical records of Saitama Medical Centre, Japan. We identified a cohort of 1,879,239 patients aged > 18 years who underwent blood biochemical examinations, including AST, at the hospital's clinical laboratory within a 15-year period from 2005 to 2019. Among the identified patients, 1197 had extremely high outlier values of AST, defined as levels more than 100 times the upper limit of normal (reference range: 13-30 U/L), that is, ≥ 3000 U/L (occurrence rate ~ 0.0637%), as previously reported 7 . Patients who were metachronous duplicates of the same patient (only one highest AST value from each patient was considered), those who experienced cardiopulmonary arrest on arrival, and those with unknown outcomes were excluded. After applying the exclusion criteria, a sample of 428 patients was selected for further analysis. A flow chart of the cohort selection process is presented in Fig. 1.
The primary outcome was defined as 72-h mortality 11,12 following the AST test, regardless of symptom onset, hospitalisation, or setting (i.e., emergency or outpatient). Cases were defined as patients with extremely high outlier values of AST who died in the hospital during the first 72 h after the test, whereas controls were defined as patients with extremely high outlier values of AST who survived beyond 72 h.
The www.nature.com/scientificreports/ defined according to the updated CCI 13 , and underlying causes. The vital signs were recorded either at the time or as close to the time of blood sample collection as possible.
Statistical analysis. Sample size. From a preliminary analysis, we obtained a dead survival ratio of 30:70 (mortality rate of 30%) among patients with AST levels ≥ 3000 U/L. We calculated the required sample size using G*Power software 31 . In order to compute the required sample size given α, power, and effect size, we input the parameters as follows: the platforms of Test family, Exact; Statistical test, Proportions: Inequality, two independent groups (Fisher's exact test); and type of power analysis, A priori. Assuming the need to test the utility of a binary risk variable by assessing the proportions of the two groups, the sample size that was required to detect a difference of 0.15 in proportion was calculated as 236 (71 for the dead group vs. 165 for the surviving group) by setting tails of two, a power of 80%, an alpha error of 5%, and an allocation ratio of 0.429. Thus, we expanded the actual data size to 428 (with expected data sizes of 133 vs. 295 for the non-survivor and survivor groups, respectively) to ensure attainment of a higher power.
Descriptive statistics. All  where x and X are the test results before and after the transformation, and λ is the power. The power used for the major laboratory tests was λ = 0.0 (log-transformation) for BW, CK, T-C, LDL-C, and TG; λ = 0.2 for T-Bili, D-Bili, AST, ALT, ALP, γ-GTP, amylase, CRP, and Na; λ = 0.4 for LDH and RPG; λ = 0.6 for HDL-C; and λ = 0.6 for BUN, Cre, Cl, and P.
First, potential risk factors for 72-h mortality were determined using univariate logistic regression analysis. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated without adjustment (crude OR in univariate regression analysis). C-statistics for covariates with significant crude ORs were determined using receiver operating characteristic (ROC) analysis.
The MLRA was subsequently performed using the listwise deletion, which is a method for handling missing data. In this method, an entire record was excluded from the analysis if any single value was missing. ORs and 95% CI were then determined after adjusting for age and AST level for all covariates that were significant during univariate logistic regression analysis and the C-statistic using ROC analysis. Clinically important risk factors determined to be significant during MLRA (including association analysis for age and AST level) were included in the final prediction model. The final prediction model was estimated using MLRA with a stepwise selection method, which was used to obtain an optimal combination of risk variables while simultaneously removing those deemed unimportant. In the final prediction model, the logit function reflected the probability of death, defined as 72-h mortality.
The following statistical variables were estimated during MLRA and ROC analyses: intercept, regression coefficient, standard error, z value, P-value, adjusted ORs, 95% CIs, variance inflation factor (VIF), C-statistic, sensitivity, and specificity. A value of P < 0.05 was considered statistically significant. The C-statistic was defined as the area under the ROC curve. In addition, multiple collinearities between covariates were assessed using the VIF. Multiple collinearities were defined as positive when the VIF was ≥ 5.
Statistical software. The statistical package for the StatFlex software version 7.0.11 (Artech Co. Ltd., Osaka, Japan) was used for data analysis, while G*Power version 3.1.9.4 (Heinrich Heine University Düsseldorf, Düsseldorf, Germany) 31 was used for sample size calculation.

Data availability
The datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.