Serum Transaminases at Presentation and Association with Acute Dialysis in Children with Hemolytic Uremic Syndrome

Visual Abstract

Abstract

Background

To determine whether serum transaminases at presentation predict the need for dialysis in children with hemolytic uremic syndrome (HUS).

Methods

Single-center, retrospective chart review of pediatric patients with HUS. Data collected included demographics, clinical and laboratory parameters, and need for dialysis. These factors were compared between two groups: “dialysis” versus “no dialysis.” Continuous data were compared using a t test whereas categoric data were compared by the chi-squared test. Multivariate logistic regression was performed on a prior set of variables to determine if serum transaminases independently predict the need for dialysis.

Results

A total of 70 children were included in the study, of which, 39 (27%) received dialysis. The no-dialysis group had a higher proportion of white patients compared with the dialysis group (74% dialysis versus 94% no dialysis). The only clinical sign at admission associated with dialysis was reduced urine output (56% versus 16%, P<0.001). Univariate logistic regression identified admission serum creatinine, aspartate transaminase (AST), and alanine transaminase (ALT) to be associated with the need for dialysis. Multivariate logistic regression showed serum AST and ALT to be independent predictors of the need for dialysis, with both improving the performance of the regression model. Sensitivity analysis showed a cutoff of 129 U/L for AST and 83 U/L for ALT with high specificity.

Conclusions

Serum transaminases at presentation are independently associated with the subsequent need for dialysis in patients with HUS. Our study suggests that when both serum ALT and AST are normal, the likelihood to need dialysis is very low; alternatively, when both serum ALT and AST are more than two times the upper level of normal, the need for dialysis is very high.

Introduction

Hemolytic uremic syndrome (HUS) is one of the most common causes of AKI in children worldwide (1–3). It is characterized by a triad of microangiopathic hemolytic anemia, thrombocytopenia, and injury from microthrombi to the kidney and other organs (4). In the United States, the incidence of HUS ranges from 0.8 to 2 cases per 100,000 children, with a majority of cases caused by Shiga toxin–producing Escherichia coli (STEC) (5,6).

Studies on predictors of the severity of HUS are limited and the utility of various predictors is not completely understood. Studies have reported that at presentation leukocytosis, shorter duration of prodromal illness, higher hemoglobin, neurologic involvement, hyponatremia, and presence of blood in stools, among others, are predictors of poor outcomes, including death (7–9). Studies looking at the predictors of need for dialysis in patients with HUS have identified additional factors including presence of oliguria, anuria, azotemia, and severe electrolyte abnormalities as the major predictors for dialysis (10). Elevated transaminases are often seen in HUS (as much as 60% in one study); however, their utility in independently predicting the severity of HUS, with respect to the need for dialysis, is seldom studied (11). A nationwide study from Japan reported that high alanine transaminase (ALT) could predict the need for dialysis; however, this study was limited by study design (use of questionnaires to conduct a national survey), sample size, as well as the lack of reproducibility in other countries (12).

To determine whether serum transaminases at presentation are independently associated with the need for dialysis in children with HUS, we conducted a retrospective study of 70 patients admitted to Children’s of Alabama (COA) between January 2000 and December 2017 with the presumed diagnosis of STEC (i.e., HUS in patients with prodromal symptoms like diarrhea and hematochezia, and no concerns for atypical HUS).

Materials and Methods

This is a retrospective study of patients with HUS admitted to our institution between January 2000 and December 2017. Patients with a diagnosis of HUS (International Classification of Diseases, Ninth Edition [ICD-9] code 283.11 or ICD-10 code D59.3) at admission or discharge were screened and HUS was confirmed by the presence of hemolytic anemia, AKI/failure, and thrombocytopenia from the medical records. To meet inclusion criteria, subjects had to have the presence of the triad of hemolytic anemia (hemoglobin less than the lower level of normal for age with evidence of hemolysis on peripheral blood smear), thrombocytopenia (platelet count <150,000/mm³), and AKI based on the Kidney Disease Improving Global Outcomes for AKI criterion by either urine output <1 cc/kg per hour for infants and <0.5 cc/kg per hour for children, or a rise in serum creatinine of 0.3 mg/dl or 50% increase from baseline (13). Other inclusion criteria included age <18 years and serum transaminases (AST and ALT) drawn at presentation of their illness. Exclusion criteria included presence of CKD; history of renal transplant; preexisting liver condition leading to abnormal liver enzymes; a diagnosis of atypical HUS; HUS related to known infections other that STEC including pneumococcus, Streptococcus, or influenza; and HUS secondary to chronic systemic diseases like SLE and HIV.

Following the screening, subjects were classified by the primary outcome of the receipt of RRT (either hemodialysis, peritoneal dialysis, or continuous RRT at any time during the index hospitalization versus no dialysis). The decision to initiate dialysis was not protocolized and was made by the nephrologist in discussion with the family. Data collection included demographics (age, sex, ethnicity), symptoms at presentation (diarrhea, bloody stools, vomiting, abdominal pain, fever, reduced urination), and presence of any comorbid condition that could affect hospital stay (including but not limited to asthma, chronic lung disease, allergies, sickle cell disease, hematologic malignancy). Information on laboratory markers at presentation (serum transaminases, electrolytes, bilirubin, BUN, creatinine, albumin, hemoglobin, white blood cell [WBC] count, and platelets), presence or absence of STEC in stool studies (including stool culture or immunoassay), any antibiotics received before admission, number of days of dialysis (for those who received dialysis), and hospital length of stay were captured from the medical record. Presence of any comorbidity including asthma, obesity, sickle cell anemia, seizure disorder, undernutrition, and cerebral palsy was also including after careful review of diagnosis codes in the charts. For patients transferred from another institution, laboratory markers at presentation to that institution were considered. The patients were followed until their discharge from the hospital. The study was approved by the Institutional Review Board at the University of Alabama at Birmingham (UAB).

Statistical Analyses

Continuous data are expressed as mean±SD and compared between the two groups using the t test. Categoric data were represented as percentages of total in each group. These percentages were compared between the two groups using chi-squared or Fisher exact tests as appropriate. Univariate logistic regression was carried out with dialysis as the dependent variable and a priori set of variables (including serum levels of transaminases, total bilirubin, BUN, creatinine, albumin, hemoglobin, WBC count, and platelets) as the predictor variables to obtain crude odds ratios (ORs). Multivariate logistic regression was then conducted, incorporating the univariate variables significantly associated with the need for dialysis to determine if AST and ALT independently predict the need for dialysis. Furthermore, a receiver operating characteristic (ROC) graph was obtained for AST as well as ALT to determine an optimal cutoff value that would predict the need for dialysis. Once sensitivity and specificity to predict dialysis using these optimal cutoffs were derived, the normative values and a doubling of the normative values for AST/ALT were assessed. All statistics were done using JMP Pro, version 14.0 (SAS Institute, Cary, NC). A P value <0.05 was considered statistically significant.

Results

Of the 91 patients who had HUS, 70 met inclusion criteria (Figure 1). Reasons for exclusion include: five who did not meet criteria for HUS, four who had atypical HUS, and 11 who did not have transaminases measured at the time of admission. Among patients who met inclusion/exclusion criteria, 39 of 70 (27%) received dialysis during their hospital stay.

Figure 1.

Consort diagram of patient selection. HUS, hemolytic uremic syndrome.

Demographics

Table 1 provides a comparison of the demographics and baseline characteristics of our patient population among the two groups. There was no difference between the two groups with respect to age and sex. Interestingly, a significantly higher proportion of patients in the no-dialysis group were white (94% versus 74%, P<0.05). The clinical features at presentation in our patient population consisted of diarrhea (94%), bloody diarrhea (81%), vomiting (64%), abdominal pain (50%), fever (37%), reduced urination (39%), and hematuria (11%). Among the nine patients without diarrhea, all had hematochezia. There was no significant difference between the two comparison groups with respect to symptoms at presentation, except for reduced urination (56% in dialysis group versus 16% in no-dialysis group, P<0.001). Comorbidities were common in the cohort: 34 patients (49%) had at least one comorbidity. The comorbidities that affected the cohort include 20 (29%) patients that had asthma, eight (11%) had obesity, four (6%) had undernutrition, and two (3%) had the sickle cell trait. There was no difference among the two groups with respect to any specific comorbidities; however, a significantly higher proportion of patients who received dialysis had at least one comorbidity (54% versus 29%, P<0.05).

Table 1.

Demographics and clinical features of patients with hemolytic uremic syndrome

Features	Dialysis (n=39)	No Dialysis (n=31)	OR (95% CI)	P Value
Age, yr (SD)	6.1 (3.9)	6.0 (4.7)		0.93
Gender, N (%)
Male	20 (51)	18 (58)	0.76 (0.3 to 3)	0.57
Ethnicity, N (%)
White	28 (72)	29 (94)	0.18 (0.1 to 0.9)	<0.05a
Symptoms, N (%)
Diarrhea	38 (97)	28 (90)	4 (0.4 to 41.2)	0.32b
Fever	15 (39)	11 (36)	1.1 (0.4 to 3)	0.80
Vomiting	27 (69)	18 (58)	1.6 (0.6 to 4.4)	0.33
Abdominal pain	19 (49)	16 (52)	0.9 (0.3 to 2.3)	0.81
Jaundice	2 (5)	2 (7)	0.8 (0.1 to 5.9)	>0.99b
Hematuria	5 (13)	3 (10)	1.4 (0.3 to 6.3)	>0.99b
Reduced urination	22 (56)	5 (16)	6.7 (2.1 to 21.2)	<0.05a
Hematochezia	33 (85)	24 (77)	1.6 (0.48 to 5.4)	0.44
Comorbidity, N (%)	21 (54)	9 (29)	2.9 (1.1 to 7.7)	<0.05a
Asthma	12 (31)	8 (26)	1.3 (0.5 to 3.7)	0.65
Obesity	5 (13)	3 (10)	1.4 (0.3 to 6.3)	0.73b
Undernutrition	2 (0.05)	2 (0.07)	0.8 (0.1 to 5.9)	>0.99
Sickle cell trait	0 (0)	2 (0.07)	—
Stool culture positive for STEC, N (%)	18 (47)	16 (51)	0.8 (0.3 to 2.1)	0.65
Antibiotics, N (%)	15 (39)	9 (29)	1.5 (0.6 to 4.2)	0.41
DOI, mean (SD)	5 (3)	5.8 (2.6)		0.23

OR, odds ratio; STEC, Shiga toxin–producing E. coli; DOI, day of illness.

^aP<0.05.

^bP value based on Fisher exact test.

There was no difference in the proportion of patients receiving antibiotics before development of HUS or the proportion of patients who tested positive for STEC among the two groups. The average day of illness at the time of presentation was 5±3 days with no statistical difference between the two groups.

Laboratory Findings at Presentation and the Need for Dialysis

Table 2 provides laboratory parameters at presentation among the two groups and the results of univariate logistic regression analysis. Those who had higher serum creatinine were more likely to need dialysis. For each rise in serum creatinine of 1 mg/dl, there was a twofold increased odds of needing dialysis (OR, 2.1; 95% CI, 1.3 to 3.3; P<0.05). Those with a higher AST had higher odds need dialysis. For each rise in AST of 100 U/L, there was a 3.6 times increased odds of needing dialysis (OR, 3.6; 95% CI, 1.4 to 8.9; P<0.01). Those with a higher ALT also had higher odds to need dialysis. For each rise in ALT of 100 U/L, there was a seven times increased odds of needing dialysis (OR, 7.1; 95% CI, 1.7 to 31; P<0.01). Those who had higher serum albumin had a lower odds to need dialysis. For each rise in serum albumin 1 g/dl, there were reduced odds of needing dialysis (OR, 0.2; 95% CI, 0.1 to 0.6; P<0.01). There was no difference in the mean values of other laboratory parameters that were evaluated between the two groups.

Table 2.

The mean (SD) of laboratory markers at presentation among the two groups and the crude odds ratio from univariate logistic regression

Laboratory Markers	Dialysis (n=39)	No Dialysis (n=31)	Crude OR (95% CI)	P Value
AST, 100 U/L	167.2 (150.3)	75.3 (53.2)	3.6 (1.4 to 8.9)	<0.01a
ALT, 100 U/L	104.1 (103.5)	39.7 (27.6)	7.1 (1.7 to 31)	<0.01a
Total bilirubin, mg/dl	1.8 (2.8)	2.2 (2.6)	0.95 (0.8 to 1.1)	0.60
BUN, mg/dl	53 (42.3)	35.9 (30.5)	1.01 (0.99 to 1.03)	0.07
Creatinine, mg/dl	3 (2.8)	1.1 (0.97)	2.1 (1.3 to 3.3)	<0.01a
Albumin, g/dl	2.9 (0.6)	3.4 (0.6)	0.3 (0.1 to 0.6)	<0.01a
Hemoglobin, g/dl	10.2 (2)	10.1 (2.9)	1.1 (0.8 to 1.2)	0.87
WBC, ×103/μl	16.9 (9.6)	14 (5.5)	1.1 (0.98 to 1.1)	0.15
Platelets, ×103/μl	96 (99)	125 (110)	1.0 (0.99 to 1.002)	0.24

OR, odds ratio; AST, aspartate transaminase; ALT, alanine transaminase; WBC, white blood cell.

^aP<0.05.

Multivariate Logistic Regression

Table 3 shows the performance of different models to predict the receipt of dialysis using multivariate logistic regression analysis. After controlling for serum albumin, reduced urine output, and serum creatinine, for every 100 U/L rise in AST, there were 2.4 times higher independent odds to need dialysis (adjusted OR, 2.4; 95% CI, 1.1 to 5.7; P<0.05); whereas with a rise of 100 U/L in ALT, there was a four times higher independent odds of receiving dialysis (adjusted OR, 4.2; 95% CI, 1.1 to 16; P<0.05).

Table 3.

Results of different multivariate logistic regression models to determine the performance of serum transaminases at presentation as predictors of the need for dialysis in children with HUS

Predictors	Adjusted Odds Ratio (95% CI)	P Value
Model 1: whole model fit ROC 0.82
Reduced urine output	4.3 (1.2 to 15.4)	0.02
Serum albumin	0.5 (0.2 to 1.3)	0.14
Serum creatinine	1.8 (1.1 to 2.7)	0.01
Model 2: whole model fit ROC 0.86
AST (100 U/L)	2.4 (1.04 to 5.7)	0.04
Reduced urine output	4.7 (1.2 to 18.1)	0.03
Serum albumin	0.5 (0.2 to 1.5)	0.24
Serum creatinine	1.5 (1.01 to 2.3)	0.045
Model 3: whole model fit ROC 0.86
ALT (100 U/L)	4.2 (1.1 to 16)	0.03
Reduced urine output	4.85 (1.2 to 19)	0.02
Serum albumin	0.52 (0.2 to 1.4)	0.20
Serum creatinine	1.6 (1.02 to 2.6)	0.04

Adjusting for serum albumin, serum creatinine and reduced urine output. HUS, hemolytic uremic syndrome; ROC, receiver operating characteristic; AST, aspartate transaminase; ALT, alanine transaminase.

To determine whether incorporation of AST and/or ALT improved the ability to predict dialysis, we compared the performance of the models with and without these variables. Overall performance of the model when neither serum AST nor ALT was used was 0.82. This improved to 0.86 when incorporating either AST or ALT. A model using AST and ALT together did not improve the performance of the model.

Sensitivity Analysis to Determine Cutoffs

Using ROC analysis, we found that the optimal cutoff to maximize the area under the curve for AST was 129 U/L, which gave a sensitivity for the need for dialysis of 54% and a specificity of 94%. Similarly, the optimal ALT cutoff to maximize area under the curve for the need of dialysis was 83 U/L, which gave a sensitivity and specificity for the need for dialysis of 51% and 97%, respectively.

We then evaluated how different simple-to-remember combinations could be used to provide the clinician with a “set of rules” about the use of AST/ALT that could provide high sensitivity and specificity. Table 4 provides sensitivity and specificities of different cutoffs for AST and/or ALT and the need for dialysis. Normal AST and ALT at presentation have a sensitivity of 84%; whereas when both AST and ALT at presentation are more than two times the upper level of normal for age and sex, the specificity of these criteria for the need for dialysis is 94%.

Table 4.

Sensitivity and specificity of various AST and/or ALT cutoffs in predicting the need for dialysis

Criteria	Sensitivity (%)	Specificity (%)
AST and ALT normal for age/gender	84	10
AST and ALT both abnormal	77	48
AST >2× ULN	69	58
ALT >2× ULN	64	81
AST and ALT >2× ULN	59	94

AST, aspartate transaminase; ALT, alanine transaminase; ULN, upper level of normal.

Discussion

Our study suggests that elevated serum ALT and AST at hospital admission are independently associated with the receipt of dialysis in children with HUS, even after controlling for other known risk factors of dialysis receipt. Using either ALT or AST along with other parameters improved the performance of the multivariate model with a ROC increase from 0.82 to 0.86. In our study, the sensitivity to predict dialysis when both serum ALT and AST was normal was almost 90%, whereas the specificity when both serum ALT and AST were more than two times the upper level of normal was 93%. The information from this study can serve valuable to clinicians as they manage patients and counsel families whose child presents with HUS with the valuable information that if both transaminases at the presentation of HUS are normal, the patient is unlikely to receive a dialysis, whereas if both are doubled at presentation, the likelihood of receiving dialysis is high.

Serum transaminases have been previously shown to be associated with HUS and are often considered as a consequence of microangiopathic systemic involvement. Binding of Shiga toxin to the surface of the glomerular endothelium leads to activation of endothelial cells, the release of cytokines, and increased adherence of platelets to endothelial cells, eventually leading to platelet activation (14). This leads to the microvessels being obstructed with microangiopathic complexes of activated platelets. When microangiopathic hemolysis occurs in the liver, serum transaminases are elevated. Studies have shown that serum liver enzymes either independently or in combination with other parameters can be useful in predicting the severity of conditions associated with thrombotic microangiopathy, including HELLP syndrome (15,16). Given that the pathophysiology of HUS is similar, it is reasonable to think that serum liver transaminases would help in predicting the severity of HUS.

However, data on the utility of serum transaminases in predicting the severity of HUS is limited. A study by Balestracci et al. (10) in 153 patients with postdiarrhea HUS of which 88 received dialysis demonstrated a higher serum ALT among patients on dialysis compared with those not on dialysis (85 IU/L versus 33 IU/L, P<0.05). However, unlike our study, serum ALT was not found to be an independent predictor of dialysis upon multivariate analysis in that study. Another study by Kawasaki et al. (17) in patients with HUS reported elevated serum ALT in those receiving dialysis, with a cutoff of 70 IU/L predicting the need for dialysis. However, this study was limited in its study design and had a smaller sample size (overall 24 patients with HUS, with 11 requiring dialysis) compared to our cohort which has 70 patients with HUS. Moreover, the study also included two patients with HUS who did not have prodromal diarrhea, both of which received dialysis.

Other presenting signs that have been associated with HUS severity have been reported, including leukocytosis, hemoconcentration, lower platelet count, hyponatremia, elevated serum lactate dehydrogenase (LDH), BUN, and creatinine. Leukocytosis at presentation has been associated with the risk of developing HUS and, in a few studies, has been shown to predict the severity of HUS (18,19). Similar to previous studies, in our cohort, serum creatinine was associated with the need for dialysis, even when adjusted for other factors; however, we did not find any statistically significant difference between WBC count, hemoglobin, platelet counts, serum BUN, or total bilirubin in those who received dialysis versus those who did not. Some of these findings could be due to the smaller sample size because there was a higher WBC count and serum BUN, and lower platelet count observed in those patients who received dialysis, albeit these were statistically NS. Although serum albumin was found to be associated with the need for dialysis on univariate analysis, multivariate analysis incorporating these predictors did not show any independent association. A reason for some of these findings could be the possible effect of dehydration. Patients with HUS are often dehydrated, which could lead to concentration and thus mask hypoalbuminemia or thrombocytopenia. Unfortunately, due to the retrospective nature of our study, it was difficult to assess for dehydration which is known to be associated with the need for dialysis (20).

With respect to presenting symptoms, our study also supports previously reported observations that there is no association of bloody diarrhea with the severity of HUS (21). Obviously, oliguria at presentation was associated with dialysis needs. We did not observe any association between the need for dialysis and receiving antibiotics before presentation. Also, although comorbidity and white race were associated with dialysis in our cohort, we did not use these variables in multivariate regression models. This was due to limited sample size of our cohorts and we chose to use dynamic and modifiable variables as opposed to the ones that were not modifiable.

Our study indeed has some limitations. A major issue with our study is that we did were not able to assess the effect of serum LDH as a predictor of dialysis. Few of our patients did not have serum LDH, making it difficult to use this marker, and we did not incorporate this because serum LDH can be both a marker of hemolysis and liver injury. Also, due to its retrospective nature, there is a possibility of potential unknown confounders being missed. Due to our inclusion criteria, some patients who did not have serum transaminases at presentation were excluded. It is possible that these patients have a milder form of HUS. We also did not have data on degree of proteinuria in our patients. This would have bolstered this study’s findings to some extent. However, we chose to use serum albumin, which could be considered as a proxy to determine the degree of proteinuria for our study. Another limitation of the study is that less than half of our patient cohort was positive for STEC based on stool cultures. Our institution did not use immunoassays or PCR to diagnose STEC widely until a few years ago. This could possibly explain why a significant proportion of our cohort did not test positive. Another explanation for this would be the fact that a significant proportion of our patients are transferred from other institutions. By the time they present to us, their diarrhea might have already stopped and testing is often not possible. Thus, a negative test would not necessarily mean this was not STEC-associated HUS. Another possible concern is that some patients with preexisting liver disease including nonalcoholic fatty liver disease could account for some of our findings, especially patients who were not diagnosed or not followed/screened for these conditions. We do acknowledge that this is a possibility, albeit in a very low number of patients, and given that the proportion of obesity was not different among the two groups, we do not think this would have affected our results. Finally, this study was limited to a single tertiary center in a developed country and thus may not be generalizable to other pediatric communities. Thus, the findings and the cutoffs we propose should be validated in a larger, multicentered cohort. Serum transaminases at presentation are associated with the receipt of dialysis in children with HUS. Additional studies are necessary to further evaluate the role of these levels in predicting the severity of the disease.

Disclosures

D. Askenazi serves on the speaker board for the AKI Foundation (Cincinnati, OH) and Baxter. D. Askenazi is a consultant for Baxter, CHF solutions, and Medtronic. D. Askenazi also receives grant funding for studies not related to this project from Baxter, CHF solutions, National Institutes of Health (NIH)–Food and Drug Administration (R01 FD005092), and the Pediatric and Infant Center for Acute Nephrology (PICAN). PICAN is part of the Department of Pediatrics at the UAB and is funded by COA Hospital, the Department of Pediatrics, UAB School of Medicine, and UAB’s Center for Clinical and Translational Sciences (NIH grant UL1TR001417). I. Aban, M. Barnes, R. Dimmit, and S. Talathi have nothing to disclose. All authors declare no real or perceived conflicts of interest that could affect the study design, collection, analysis and interpretation of data, writing of the report, or the decision to submit for publication.

Funding

None.