A Clinical Model to Predict Fibrosis on Liver Biopsy in Pediatric Subjects with Nonalcoholic Fatty Liver Disease

Abstract

Background:

The incidence of Nonalcoholic fatty liver disease (NAFLD) in children is rapidly increasing. Liver fibrosis is a poor prognostic feature that independently predicts cirrhosis. The time that intercedes the first medical encounter and biopsy is rate-limiting to multi-modal treatment. This study aimed to identify non-invasive parameters to predict advanced NAFLD and fibrosis.

Methods:

We conducted a single-center, retrospective 10-year analysis of 640 pediatric patients who underwent liver biopsy. Fifty-five patients, age 3-21 years, had biopsy-confirmed NAFLD. We assessed primary outcomes, NAFLD activity score (NAS) and fibrosis scores, against non-invasive parameters by linear regression, by using binary cutoff values, and by a multivariate logistic regression fibrosis prediction model.

Results:

NAS correlated with platelets and female sex. Fibrosis scores correlated with platelet counts, gamma glutamyl transferase (GGT), and ultrasound shear wave velocity. 25-hydroxy-vitamin D and GGT differentiated mild vs moderate-to-advanced fibrosis. Our multivariate logistical regression model-based scoring system predicted F2 or higher (parameters: BMI%, Vitamin D, Platelets, GGT), with sensitivity and specificity of 0.83 and 0.95 (area under the ROC curve, 0.944).

Conclusion:

We identify a clinical model to identify high-risk patients for expedited biopsy. Stratifying patients to abbreviate time-to-biopsy can attenuate delays in aggressive therapy for high-risk patients.

INTRODUCTION

The rise in prevalence of childhood obesity over the past 2 decades coincided with a rise in pediatric nonalcoholic fatty liver disease (NAFLD) ^1,2. Due to the clinical and pathophysiological relationship between obesity and NAFLD, NAFLD is now the most common form of chronic liver disease in both children and adults, and is quickly becoming the leading cause of liver transplantation worldwide ^3,4. Amongst adolescents and young adults, the incidence of NAFLD is higher in mals; however post-menopausal female NAFLD patients may be at a higher risk of NASH progression^4,5. Pre-pubertal children have distinct histologic features (predominant zone 3 inflammation), and higher histologic severity of disease compared to post-pubertal children⁶. Insulin resistance, which is associated with obesity and metabolic syndrome, leads to hepatic steatosis. In a proportion of individuals with steatosis, hepatic inflammation develops, and this is known as nonalcoholic steatohepatitis (NASH). Uncontrolled hepatic inflammation along with certain genetic (e.g. PNPLA3 gene variants) and environmental modifiers (alcohol, intestinal dysbiosis) result in liver fibrosis and eventually end-stage liver disease⁷.

A diagnosis of pediatric NAFLD is suspected in subjects with obesity who have elevated liver enzymes (alanine transaminase or ALT, aspartate transaminase or AST and gamma-glutamyl-transferase or GGT) ⁸. GGT is an enzyme present in the cell membrane of bile duct epithelial cells, and has been shown to be associated with worse prognosis in pediatric NAFLD subjects⁹. Ultrasound is used to demonstrate hepatic steatosis, while elastography can provide a non-invasive estimate of the degree of liver fibrosis^10,11. Despite significant progress toward imaging and biomarker modalities, liver biopsy remains the gold-standard method for diagnosis, stage and assess extent of liver fibrosis in pediatric subjects with NAFLD ^12,13. Current treatment strategies for pediatric NAFLD include lifestyle modifications. Currently, the only pharmacological agent approved for use in pediatric NAFLD is Vitamin E⁸.

NAS is a scoring system used to assess NAFLD severity and progression using the severity of hepatic steatosis, ballooning and inflammation (but not fibrosis), on a liver biopsy¹⁴. Even though NAS can reliably be used in the diagnosis of NASH, which in turn is predictive of overall mortality in NAFLD subjects, the score itself has limitations in directly predicting prognosis of NAFLD^15,16. At diagnosis 10–25% of children can have advanced fibrosis on liver biopsy ¹⁷. Even though the natural history of fibrosis present on liver biopsy in pediatric NAFLD is not completely elucidated, longitudinal pediatric studies have suggested progression of liver fibrosis in 25% of subjects within 2 years of diagnosis, with 20% progressing to advanced fibrosis¹⁸. Rapid progression of NASH in children has been previously described ^18–20. Pediatric obesity is a risk factor for developing hepatocellular carcinoma²¹. Data from adult cohorts suggest that the presence of fibrosis predicts severe liver disease, cirrhosis and liver mortality ^15,22–24.

Early, aggressive, multidisciplinary treatment targeted at reducing obesity in high-risk pediatric NAFLD subjects is of primary significance^25,26. It follows, that a point-of-care stratification model to rapidly identify patients with the greatest risk for advanced NAFLD and fibrosis would thus be of clinical value²⁷. The utility of this is to accelerate the time that intercedes first encounter, liver biopsy, and subsequent referral to a center with multidisciplinary weight management clinic²⁸. This could also identify patients at most risk for advanced NASH and fibrosis for possible enrollment in future pediatric clinical trials testing medications which have been approved in adult NAFLD²⁷. Sensitive and specific clinical scoring models have already been developed and to predict fibrosis in adult patients ²⁹ and in children ^30–32. However, adult stratification methods are not readily applicable to pediatric populations, and no clear pediatric stratification modality has yet achieved widespread utility. Here, we define the predictive value of several non-invasive clinical, biochemical and imaging parameters, in an effort to develop a clinical model to predict fibrosis, in a 10-year retrospective single center cohort of patients with biopsy-confirmed NAFLD.

METHODS:

Study Population:

Retrospective data were analyzed for all pediatric patients who presented to St. Louis Children’s Hospital and received liver biopsy (640 patients) from October 2009 to March 2020. Patients were identified by international statistical classification for diseases (ICD)-9 and -10 codes for liver biopsy. Inclusion criteria were: (I) A histopathological diagnosis of nonalcoholic fatty liver disease, (II) Age > 3 years and less than 21 years, and (III) clinical, imaging, biochemical, and histopathological exclusion of any other liver diseases, including, but not limited to: autoimmune hepatitis, viral hepatitis, Wilson disease, α1 anti-trypsin disease and drug induced liver disease. 55 patients were included in our analysis.

Histological Features:

NAFLD severity was graded using NAS as proposed by Kleiner et al. ³³ Which is briefly described; steatosis grade 0-3, 0 (<5% steatosis), 1 (5-33% steatosis), 2 (34-66% steatosis) and 3 (>66% steatosis). Lobular inflammation grade 0-3; 0 (no foci), 1 (1 focus per 200X field), 2 (2-4 foci per 200X field) and 3 (>4 foci per 200X field). Liver cell injury/ballooning grade 0-2; 0 (no ballooning), 1 (few balloon cells) and 2 (many/prominent balloon cells). The grade of steatosis (0-3), lobular inflammation (0-3), and ballooning (0-2) were then combined to determine the NAS (0-8) as proposed. Fibrosis score was assessed 0-4. Stage 0 (no fibrosis); Stage 1a (zone 3 perisinusoidal fibrosis delicate); Stage 1b (zone 3 perisinusoidal fibrosis dense); Stage 1c (portal-only fibrosis, without perisinusoidal fibrosis); Stage 2 (as above with portal fibrosis); Stage 3 (as above with bridging fibrosis); and Stage 4 (cirrhosis).

Patterns of Nonalcoholic Steatohepatitis (NASH):

Previous studies demonstrated distinct histopathological patterns when comparing nonalcoholic steatohepatitis in pediatric and adult patients. Type 1 NASH (adult pattern) is characterized as the presence of steatosis, hepatocyte ballooning, degeneration, and perisinusoidal fibrosis in the absence of portal features. Type 2 NASH (pediatric pattern) is defined as presence of steatosis with portal inflammation, and portal fibrosis in the absence of ballooning degeneration and perisinusoidal fibrosis. We stratified nonalcoholic steatohepatitis type in our population by concordant readings of two independent blinded reviewers of histopathology readings.

Patient Demographic Characteristics:

Patient demographic data were reviewed and collected from patient charts, including age, sex, race, weight, height and blood pressure. BMI%ile was calculated for their age and sex using extended CDC curves.

Clinical Parameters:

Laboratory data were reviewed and collected included alanine aminotransferase, aspartate aminotransferase, gamma glutamyl transferase, alkaline phosphatase, albumin level, prothrombin time, white blood cell count, hematocrit, platelets count, hemoglobin A1C, vitamin D level, lipid profile (total cholesterol, triglyceride, high density lipoproteins, low density lipoproteins) thyroid stimulating hormone, free T4 and urine analysis. Hypothyroidism, which is a strong driver of NASH, was ruled out in all subjects in our study.

Imaging Assessment:

Patient images including abdominal ultrasound and elastography were reviewed. Elastography values are expressed in meters/second.

Statistical Analysis, derivation of predictive model and validation in an external cohort:

Data was analyzed by SAS® (SAS Institute Inc., Cary, NC, USA) version 9.4. A P < 0.05 was considered as significant. Pearson’s correlation coefficient was used to assess correlation between each continuous variable and NAS or Fibrosis. One-way ANOVA was performed between each categorical variable and NAS or Fibrosis. Univariate logistic regression was performed for NAS ≥ 4 and Fibrosis>=2 on each potential continuous factor. A cut-off value was selected based on optimized sensitivity and specificity. Two-sample t-testing was performed to detect the difference of continuous variables between males and females. Chi-square test or Fisher’s exact test (for any expected value less than 5) was used to assess the relationship between categorical variables and sex. Multivariate logistic regression model was used to calculate score for predicting Fibrosis ≥ 2. The liver firosis threshold of ≥ 2 was based on prior pediatric NAFLD studies ^34,35. We validated this predictive model in subjects (N=12), who met the inclusion criteria outlined above ages 3-21 years, during the time period April 2020 and April 2021.

RESULTS

Patient Demographics:

Clinical and laboratory data for 55 patients are demonstrated in supplemental tables I–III. Patient age range is 6 years, 7 months to 19 years, 5 months, with a mean age of 13 years 5 months. 65% (36) of study subjects were male. BMI ranged from 20.4 to 52.2 kg / m². BMI %ile, as assessed by using the extended CDC curve, ranged from 88^th to 190^th %ile. Patients were categorized into obesity classes I (BMI > 95 to 120%ile) II (BMI >120-140%ile and III (BMI >140%ile). Our BMI distribution was evenly distributed, with 18 (33.6%), 22 (40%) and 13 (23.6%) patients belonging to classes I, II, and III, respectively (Table S1). Two patients were categorized as “overweight”, with BMI of (20.4 and 27.6 kg / m²). The majority of our population was Caucasian (83.6%). Based on liver biopsy, patients were stratified into stages of fibrosis ranging from 0-4 as shown in Table S2. 13 (23.6%) had stage 0 fibrosis, 25 patients (45.4%) had stage 1 fibrosis, 11 (20%) had stage 2 fibrosis and 4(7.3%) had stage 3 fibrosis. 2 (3.6%) patients were found to have cirrhosis. We stratified nonalcoholic steatohepatitis in our population into type 1 NASH, type 2 NASH and combination of both based on histopathology. Consistent with prior data by Takahashi, Nobili, and Carter-Kent, et al ^36–38, the preponderance of patients (27/55) exhibited mixed-type NASH 1 & 2 patterning, 29% (16/55) of patients had type 2 NASH and 12.7% (7/55) type 1 NASH. 5 patients were excluded from this analysis, because 2 had NAS = 0, 2 had cirrhosis and 1 had simple steatosis.

Non-invasive parameters that correlate with NAS:

Pearson’s correlation coefficient was calculated between each continuous variable demonstrated (Table S3) and NAS. We tested all clinical and laboratory variables of interest. Two variables, platelet count and sex were statistically correlated an elevated NAS. There was significant negative correlation between platelets and NAS (Pearson’s correlation coefficient, p = 0.042, Fig 1A). The odds ratio was 1.0, (95% CI 0.99-1.01) with an area-under the receiver-operating characteristic (ROC) curve of 0.4931. In contrast, we observed a significant positive association between female sex and NAS score at biopsy. None of the patients in our study had a repeat liver biopsy. Females had significantly higher NAS at first biopsy, (p = 0.017, (Fig. 1B)), and in particular, females specifically had a greater likelihood to have a NAS in advanced states ≥ 4 (OR 3.51, 95% CI 1.04-11.8). The area-under the ROC curve for female sex in predicting NAS ≥ 4 was 0.6357 (Fig. 1B).

Figure 1:

Significant correlations between NAS and platelets, patient sex and albumin. A. Left, inverse linear regression relationship between platelets and NAS. Middle, ROC curve for the event NAS ≥ 4 vs. platelet count. Right, poor predictive value of platelets for NAS ≥ 4. B. Increased likelihood for advanced NAS scoring on first biopsy in females versus males. Left, mean NAS in male and female subjects. Diamond, group mean NAS. Mid-line, mean value. Boxes, standard deviation from the mean. Whiskers, maxima and minima. Dots – outliers. Middle, ROC curve for NAS ≥ 4 and female sex. Right probability for NAS ≥ 4 in male and female subjects.

Several non-invasive parameters correlate with fibrosis:

We therefore next evaluated non-invasive predictors for fibrosis. Three biochemical variables and one imaging variable significantly correlated with fibrosis score. Additionally, gamma glutamyl transferase (GGT) correlated positively with fibrosis (p = 0.012) (Fig. 2A), whereas platelet count and 25-OH-vitamin D correlated negatively with fibrosis as continuous variables (p = 0.009, OR 0.99, 95% CI 0.98-1.003, AUC 0.5949) (Fig 2B and C). Ultrasound elastography-based quantification positively correlated with fibrosis (p = 0.032, OR 228.2, 95% CI 0.197-> 999.9, AUC 0.775) (Fig. 2D).

Figure 2:

Significant correlations between biopsy fibrosis score and GGT, platelets, 25-OH-vitamin D and ultrasound elastography (USE) scores in m/s. Left graphs, linear regression relationship between (A) GGT, (B) platelets and (C) 25-OH-vitamin D and (D) USE. vs. fibrosis score. Middle graphs, ROC curve for the event fibrosis ≥ 2 vs. (A) GGT, (B) platelets, (C) 25-OH-vitamin D and (D) USE. Right, predictive value for fibrosis score ≥ 2 vs. (A) GGT, (B) platelets and (C) 25-OH-vitamin D and (D) USE.

We then performed univariate logistic regression for the events NAS ≥ 4 and Fibrosis ≥ 2 against each continuous variable: US elastography, platelets, albumin, GGT and 25-OH-Vitamin D. No predictors for NAS ≥ 4 were identified, however the data revealed GGT and 25-OH-Vitamin D as two predictors for the event of fibrosis ≥ 2. For the effect of GGT in predicting fibrosis ≥ 2 (p = 0.010), the odds ratio was 1.02 with the 95% confidence interval of 1.01 - 1.04. ROC curve analysis, (AUC = 0.774) indicated a 77.4% chance that the model can distinguish between fibrosis ≥ 2 and fibrosis < 2. A GGT cutoff of 65 U/L yielded a sensitivity 66% and specificity 76% to predict fibrosis ≥ 2. For the significant effect of 25-OH-Vitamin D for the eventFibrosis ≥ 2 (p = 0.035), the odds ratio was 0.832 (95% CI 0.701-0.987). The 25-OH-Vitamin D ROC curve, (AUC = 0.759) indicated a 75.9% chance that the model can identify fibrosis ≥ 2 on biopsy. A 25-OH-Vitamin D threshold of 22 ng/mL yielded 72% sensitivity 69% and specificity to predict fibrosis.

A novel scoring system predicts fibrosis

Multivariate logistic regression using the current data provides a scoring system to predict fibrosis. This optimized model includes BMI, GGT, 25-OH-Vitamin D and platelet count in the following relationship:

$$\Sigma \left[\mathrm{BMI}\%*0.07,\mathrm{GGT}*0.04,25-OH-\mathrm{Vitamin\; D}*\left(-0.5\right),\mathrm{Platelets}*\left(-0.03\right)\right]$$

The combination of these individual predictors yielded a total score that significantly predicted fibrosis ≥ 2 (p = 0.0082, ROC AUC = 0.945, Figure 5). Analysis of all cut-off scores determined an optimized sensitivity and specificity using a score of −6.13 (Fig 3). The sensitivity and specificity of this predictive score was 83.3% and 94.6%. The scoring system misclassified 2 patients as falsely negative, and 3 patients as falsely positive. We further validated this predictive model in 12 subjects who met inclusion criteria for our study during the time period between April 2020 and April 2021. The sensitivity and specificity of the predictive score in this external dataset was 75% and 87.5%. The scoring system misclassified 1 patient as falsely negative, and 1 patient as falsely positive.

Figure 3:

A novel multivariate model predicts fibrosis ≥ 2 on biopsy. Left, ROC area-under-curve for threshold score > −6.13 versus fibrosis ≥ 2. Right, probability of fibrosis ≥ 2 as a function of model score.

Comparing existing scoring systems with new model

We examined the performance of this model (ROC 0.945) in comparison with previously developed models (PNFS, AST: PLAT, APRI and FIB-4), which we applied to this dataset. As described in Alkhouri et al, PNFS ROC AUC was 0.741 to predict advanced fibrosis ³⁰. Similarly, our calculated ROC for other models to predict significant fibrosis (fibrosis score ≥ 2) for AST: PLAT, APRI and FIB-4 were: 0.689, 0.688, and 0.72 respectively (Fig 4A, 4B and 4C).

Figure 4:

Areas-under-ROC curve comparison in this cohort using previously reported predictive models to predict fibrosis score ≥ 2 using (A) AST:PLAT, (B) APRI, and (C) FIB-4.

DISCUSSION

Our study shows that higher GGT values and US elastography liver-stiffness values, as well as lower Vitamin D level and platelet counts correlated with a higher fibrosis score. We identified an equation based on patient’s BMI, GGT level, Vitamin D level and platelet count to reliably predict liver fibrosis ≥2 (AUC = 0.945). The sensitivity and specificity of this equation in an external cohort was 75% and 87.5% respectively. Our study revealed that female gender and lower platelet counts positively correlated with higher NAS.

In this context, significant interest has recently emerged to develop predictive models for NAFLD, which can reliably (albeit not perfectly) stratify children with obesity who develop higher NAFLD severity, and can be used at point of care by clinicians ³⁹. Patton and co-workers demonstrated that increasing AST and GGT correlates with NASH severity ⁴⁰, whereas fibrosis correlates with increasing AST, increasing white blood cell count, and decreasing hematocrit. Similarly, Carter-Kent and co-workers reported that AST independently predicted liver fibrosis ³⁷. Alkhouri and colleagues devised the pediatric NFS (PNFS), which includes ALT, alkaline phosphatase, platelet count, and GGT ³⁰. Nobili and his colleagues developed PNFI (pediatric NAFLD fibrosis index) to predict fibrosis in children, using age, waist circumference and circulating triglycerides ³¹. Alkhouri et al then investigated the performance of using both, PFNI with enhanced liver fibrosis test (combination of 3 extracellular matrix components, namely hyaluronic acid, amino terminal propeptide of type III collagen, and inhibitor of metalloproteinase) to predict NAFLD induced fibrosis ³². The combination of these yielded an area under the receiver operating curve of 0.944. However, enhanced liver fibrosis test requires specialized tests which are not readily available and incur extra costs.

Here, our data corroborate and extend prior pediatric data. Specifically, we demonstrate that GGT, BMI percentile and platelet count predict fibrosis. The common findings across cohorts point us to the pathophysiology of pediatric NAFLD and NASH. For example, GGT maintains intracellular glutathione stores to protect against intracellular and extracellular oxidant stress ⁴¹. Increased serum GGT activity may reflect increased insulin resistance, oxidant load, anti-oxidant insufficiency, or outright biliary ductular inflammation ^42,43. Histopathological data generally support the contribution of bile duct injury and proliferation, as ductular proliferation precede the periportal fibrosis that is characteristically observed of pediatric NAFLD ⁴⁴. Diminishing platelet counts also consistently correlate with advanced NAS and fibrosis in ours and other studies. The remarkable consistency of this finding across multiple populations indicates that periportal fibrosis in NASH may induce subclinical portal hypertension and hypersplenism. The inverse correlation between 25-OH-Vitamin D in our dataset corroborate similar observations by Nobili et al ⁴⁵. Vitamin D deficiency has been reported with pediatric obesity and NAFLD, and has been inversely associated with NASH and fibrosis in children with NAFLD ^8,46,47. In light of growing evidence that indicate anti-oxidant and anti-inflammatory functions of 25-OH-Vitamin D in rodent models and in humans ^48,49, our clinical data corroborate a pathophysiological sequence of fat accumulation, oxidant stress, ductular injury and proliferation, periportal fibrosis, and subclinical portal hypertension in pediatric NAFLD. In our study, higher NAS scores were correlated with female gender. The overall incidence of NAFLD is higher in male gender⁵⁰. However, postmenopausal women, especially those who experience early menarche, have been shown to have increased incidence of NAFLD as compared to premenopausal women^51,52. Prior studies have shown that, compared to men, women are at a lower risk of steatosis, but a higher risk of NASH severity and progression^5,53.

The retrospective nature of the study, along with the small number of subjects with high NAS and fibrosis scores are the study’s main limitations. One patient in our study was 19 years of age, which is strictly above the pediatric age range. We do follow NAFLD patients till the age of 21 at our center. Our study did not investigate the role of polymorphisms in genes which have been shown to impact NASH progression (e.g. PNPLA3, TM6SF2) as non-invasive markers of NASH progression⁵⁴. At our center, SNPs in genes associated with NASH progression are not included in the diagnostic evaluation of NAFLD patients. There were significant variations in the study design, inclusion criteria, number of subjects, gender and racial composition of subjects, NAS thresholds and fibrosis score thresholds between our study and prior similar pediatric studies.

In conclusion, our study identified parameters which correlated with higher NAS and fibrosis grade ≥ 2 on liver biopsy in pediatric NAFLD subjects. A clinical model based on a combination of BMI percentile, GGT, platelet count and Vitamin D levels, which are data commonly obtained in pediatric subjects with obesity yielded a score which could reliably predict NAFLD induced fibrosis ≥ 2 in pediatric subjects seen in our clinic. Many of the parameters which made up this score are similar to the parameters which were integrated in predictive models by previously published pediatric studies. However, our model will require prospective validation. Additionally, future multi-center prospective studies using a larger sample size, more racial diversity, inclusion of SNPs as non-invasive markers of NASH progression, and uniform outcome thresholds for NAS and fibrosis score may identify a more widely applicable predictive model for fibrosis in pediatric NAFLD subjects. Such pragmatic, reliable and reproducible predictive scores developed from future multicenter prospective efforts may be incorporated into electronic health record of NAFLD patients, and used by clinicians at point of care to make a decision on whether to proceed with a liver biopsy.

What is known?

• Nonalcoholic fatty liver disease represents a spectrum of liver histopathologic abnormalities ranging from simple steatosis to steatohepatitis to fibrosis

• Fibrosis is a histopathological feature that predicts progression to cirrhosis and mortality.

• Adult-oriented models for predicting liver fibrosis are not readily applicable to pediatric populations, and no clear pediatric stratification modality has yet achieved widespread utility

What is new?

• We developed a novel multivariate clinical prediction model for fibrosis on first biopsy

• This can be used at point-of-care to accelerate time-to-biopsy in children at high risk of liver fibrosis

• After an expedited biopsy diagnosis of NASH and fibrosis is obtained, we anticipate that this prediction model will ultimately permit early, aggressive multi-disciplinary therapy to halt the progression of NASH and fibrosis in children