ALT Poorly Predicts Nonalcoholic Fatty Liver Disease (NAFLD) and Liver Fibrosis as Determined by Vibration-Controlled Transient Elastography in Adult National Health and Nutrition Examination Survey 2017–2018

Abstract

Background:

Non-alcoholic fatty liver disease is a growing problem in the United States, contributing to a range of liver disease as well as cardiovascular disease. ALT is the most widely used liver chemistry for NAFLD evaluation. We hypothesized that the normal range many laboratories use was too high, missing many patients with clinically important steatosis and/or fibrosis.

Methods:

This study utilized 2017–2018 NHANES data including 9,254 participants. We compared four different upper limits of normal for ALT with specific measurements of steatosis and liver stiffness as determined by liver elastography with FibroScan^®. Liver stiffness was further characterized as showing any fibrosis or advanced fibrosis. After exclusions, our final pool was 4,184 for liver stiffness measurement and 4,183 for steatosis grade as measured by Controlled Attenuation Parameter (CAP). Using these variables, we performed logistic regression between ALT and CAP, and ALT and fibrosis/advanced fibrosis, and did a Receiver Operating Characteristic curve.

Results:

Based on three of the most widely used cut off values for ALT, we found that ALT does not reliably rule out NAFLD in over 50% of cases. It also missed 45.9–64.2% of patients with liver fibrosis.

Conclusions:

Our study revealed that ALT is an inaccurate marker for NAFLD as measured by FibroScan^® with CAP greater than or equal to 300 dB/m. Accuracy improved specific risk factors were considered. These data also showed that ALT was a poor marker for liver fibrosis. We conclude that there is no single ALT level that accurately predicts hepatic steatosis or fibrosis.

1. Introduction:

Non-alcoholic fatty liver disease (NAFLD) is a growing problem in the United States, and world-wide. Before 2017, the National Health and Nutrition Examination Survey (NHANES) data showed an estimated prevalence of NAFLD of 32.3%.¹ One meta-analysis of the epidemiology of NAFLD reported the global prevalence as 25.24%, with the highest prevalence in the Middle East and South America and the lowest in Africa.² NAFLD exists on a spectrum from simple steatosis to non-alcoholic steatohepatitis (NASH) to advanced fibrosis to cirrhosis, as well as, potentially, to hepatocellular carcinoma (HCC). NAFLD is defined as ≥5% hepatic steatosis without evidence of hepatocellular injury in the form of hepatocyte ballooning. NASH is defined as ≥5% hepatic steatosis along with inflammation with hepatocyte injury such as ballooning with or without fibrosis.³

The metabolic syndrome is associated with and a risk factor for NALFD. Common comorbid conditions include obesity, type 2 diabetes mellitus, and dyslipidemia. Emerging conditions associated with NAFLD include polycystic ovarian disease, obstructive sleep apnea, hypothyroidism, hypogonadism, hypopituitarism, psoriasis, and pancreaticoduodenal resection.³ Age and sex are also risk factors. Men are twice as likely to have NAFLD.⁴ Both the prevalence and fibrosis stage increase with age.⁵ Ethnicity also plays a major role. In the 2004 Dallas Heart Study, Hispanic individuals had a significantly higher rate of NAFLD, non-Hispanic black individuals had significantly lower rates, and Caucasians were somewhere in the middle.

These differences weren’t explained by differences in body mass index (BMI), insulin resistance, medication use, or alcohol ingestion. ^6,7

The ability to diagnose NAFLD accurately is important not only due to the risk of liver disease and dysfunction, but also to predict cardiovascular risk -the most common cause of death in patients with NAFLD.⁸ In a Finnish study of over 10,000 patients with NAFLD, the risk of liver-related outcomes was 1%, including hospital admissions related to advanced liver disease, HCC, and liver-related death. However, the 20 year cumulative incidence of cardiovascular disease was 38% and 23% for men and women, respectively. For non-hepatic cancer, it was 22% and 15%, respectively--many times higher than the incidence of liver-related outcomes.⁹

Checking liver chemistries is generally the first step in assessing liver disease, including alanine aminotransferase (ALT) and aspartate aminotransferase (AST), alkaline phosphatase, and bilirubin. An abnormal ALT/AST ratio prompts further investigation into possible liver dysfunction. However, there isn’t a standard normal range for ALT and AST due to differences in healthy control populations used to establish normal reference ranges across different institutions and locations.¹⁰ Both practicing clinicians and patients have a need to determine an ALT level that is normal, as many clinical decisions for diagnosis and treatment are based on this level being abnormal. Indeed, ALT is probably the most widely used single liver chemistry employed in the diagnosis of NAFLD. Several studies have proposed a standardized upper limit of normal (ULN). We hypothesize that the current cutoff values for ALT in commercial laboratories are too high and miss a significant number of patients developing clinically significant steatosis. Vibration-controlled transient elastography (VCTE) using FibroScan^® can assess steatosis by measuring the increased attenuation ultrasound waves when travelling through steatotic hepatic tissue. These results are expressed numerically as the controlled attenuation parameter (CAP). The interpretation varies, depending on etiology of liver disease and may be less accurate in the case of liver function tests (LFTs) over 5 times the ULN, excessive alcohol drinking, sinusoidal congestion, extrahepatic cholestasis, and severely high BMI.^10,11 Likewise, liver fibrosis can be assessed by FibroScan^® using the liver stiffness measurement (LSM).

ALT has previously been shown to be normal in many patients with NAFLD.¹² However, this hasn’t been replicated using the large sample size of NHANES along with FibroScan^® data. VCTE was performed by FibroScan^® in the NHANES 2017–2018. This provides an opportunity for more accurate measurement of liver fibrosis and steatosis which can be correlated with aminotransferase levels. Our study compared each stage of fibrosis and steatosis seen on FibroScan^® with sequential analyses of ALT and AST level cutoffs based on the following: ACG 2017 clinical guideline¹³, NHANES¹⁴, and Prati et al¹⁵, and our own ALT cut-off determined by Youden Index.

1. Materials and Methods:

2.1. Study population and exclusion criteria

This study utilized 2017–2018 NHANES data, the latest iteration of the National Health and Nutrition Examination Survey series. NHANES surveys are administered annually by the National Center for Health Statistics of the CDC to noninstitutionalized civilian U.S. residents to gather health and nutrition data. 9,254 individuals participated in the 2017–2018 NHANES. To investigate whether ALT is a good marker predicting NAFLD and fibrosis, we wanted to identify alcohol use and determine whether heavy drinkers affected our results. We used the Alcohol Use Questionnaire (ALQ) to identify alcohol users. There were 4,124 ALQ non-respondents, which were excluded. The ALQ survey included respondents of 18 years or older. Therefore, our study excluded all individuals under 18. We also excluded subjects with liver elastography partial exam (n=642), those with positive Hepatitis B surface antigen (n=21) or positive Hepatitis C antibody (n=42), and those without CAP measurement (n=1) or without ALT measurements (n=241), which left 4,183 cases in our study cohort. A flowchart forming the study cohort is shown in Figure 1.

Figure 1:

A flowchart for creating the study cohort based on 2017–2018 NHANES data, stratified by alcohol use, steatosis scores (S0-S3), fibrosis, and advanced fibrosis.

We further categorized the participants into drinking groups. Based on questionnaire responses, ALQ respondents were subdivided into four drinking groups: never drinkers, non-drinkers, moderate drinkers, and heavy drinkers. “Never drinkers” (n = 451) were defined as individuals who responded “no” when asked if they had consumed at least one alcohol drink during their lifetime, excluding small sips. “Non-drinkers” (n = 834) were defined as individuals who reported some degree of alcohol consumption in their lifetime but responded that they had “never in the last year” consumed alcohol when asked how often alcoholic drinks were consumed during the past 12 months. Based on Dietary Guidelines for Americans (DGA), “moderate drinkers” (“moderate” for short, n = 2,523) were defined as women who reported an average consumption of one or fewer drinks per day and men who reported an average of two or fewer drinks per day. Individuals consuming alcohol moderately based on this definition could still be classified as “heavy drinkers” if they reported binge drinking one or more times per month on a separate survey question (survey question ALQ270), as at least monthly binge drinking is considered heavy alcohol consumption by the National Institute on Alcohol and Alcohol Abuse (NIAAA) criteria.¹⁶ “Heavy drinkers” (n = 375), then, were individuals who reported one binge or more per month, and/or women who drank an average of more than one drink per day, and men who drank an average more than two drinks per day. Alcohol consumption by drinks per day was calculated by multiplying the amount of alcohol consumed during drinking occasions (survey question ALQ130) by drinking frequency (survey question ALQ121) divided by seven days. We conducted the following analyses in the dataset without heavy drinkers, resulting a total sample size n=3,808, including 1,842 males and 1,966 females. Heavy drinkers were excluded because this study aimed to look at steatosis and fibrosis associated with NAFLD as the primary liver disease. Moreover, active heavy alcohol use has been shown to overestimate the liver fibrosis stage in transient elastography.¹⁷

2.2. Analysis of Alanine Aminotransferase

When comparing ALT versus CAP, we analyzed males and females separately. We compared four different upper limits of normal for ALT to our analysis of CAP, as defined below. For males, we used ALT of 33 U/L (American College of Gastroenterology (ACG) cut-off)¹³, 29 U/L (NHANES cut-off)¹⁴, 30 U/L (from Prati, et al.)¹⁵, and 24.5 U/L. 24.5 U/L was the cut-off value determined using Youden’s Index, in order to maximize the sum of the sensitivity and specificity. For females, we used ALT of 19 U/L (from Prati, et al.)¹⁵, 22 U/L (NHANES cut-off)¹⁴, 25 U/L (ACG cut-off)¹³, and 17.5 U/L. Again, 17.5 U/L was the value calculated by Youden’s Index which would be most accurate in diagnosis for women.

When comparing ALT versus any fibrosis/advanced fibrosis, the above ALT values for males remained the same except for the value calculated using Youden’s index. We used an ALT cutoff of 25.5 U/L for any fibrosis and a cut-off of 30.5 U/L for advanced fibrosis. The ALT values, including Youden’s index, for females remained the same.

1.3. Analysis of Controlled Attenuation Parameter

When analyzing CAP as compared to ALT, we used steatosis score of zero (S0) as compared to steatosis score on non-zero (not S0), which encompasses stages 1–3. To define the different stages of steatosis, S0 is <300 dB/m (no steatosis), S1 is 300–331 dB/m (5–33% steatosis), S2 is 332–337 dB/m (34–66% steatosis), and S3 is >337 dB/m (≥67% steatosis).¹⁸

1.4. Analysis of Liver Stiffness Measurement

We analyzed LSM as compared to ALT in two groups. We first analyzed no fibrosis (F0–1) versus any fibrosis (F2–4). Secondly, we analyzed no-to-mild fibrosis (F0–2) against advanced fibrosis (F3–4). Liver stiffness values were categorized as F0–1 if LSM< 8.2 kPa (no fibrosis), and further fibrosis stages as F2 if LSM is between 8.2–9.6 kPa, F3 if LSM is between 9.7–13.5 kPa, and F4 if LSM is ≥13.6 kPa.¹⁸

1.5. Assessment of covariates

We obtained the demographic variables (age, sex, education) from the demographics data, obtained body measures such as body mass index (BMI) and waist circumferences (BMIWAIST) from the examination data, and obtained blood lipids (glycohemoglobin (LBXGH), total cholesterol (LBXTC), triglyceride (LBXTR), LDL-cholesterol (LBDLDL), Direct HDL cholesterol (LBDHDD)) and liver chemistries (say, ALT and AST) from the laboratory data. Diabetic status was assessed by glycohemoglobin and the questionnaire data on diabetes (DIQ). If the participant was diagnosed as diabetes based on the questionnaire data or hemoglobin A1c (HgA1c) > 6.5, it was considered diabetes. If HgA1c < 5.7, it was normal; otherwise, it was prediabetes. BMI was classified into under or normal weight, overweight, obese based on the cut points 25 and 30 kg/m².

1.6. Statistical Analysis

First, we assessed the descriptive statistics of the demographic variables, BMI, waist circumferences, diabetic status, blood lipids, and liver chemistries ALT and AST. The descriptive statistics were summarized in the total population and subgroups by steatosis score (S0 vs. S1-S3) (see Tables 1 and 2). For a categorical variable, the frequency and percentage at each level were summarized, and Chi-square test was used to test whether this variable was independent of the steatosis scores. For a continuous variable, the mean and standard deviation (SD) were reported, and the Kruskal–Wallis test was used to test whether this variable was significantly different between the subgroups stratified by steatosis score (Tables 1 and 2)¹⁹

Table 1:

Demographics stratified by steatosis scores: S0 when CAP < 300 dB/m, and S1-S3 when CAP ≥300 dB/m

Demographic variables		Overall N(%)	S0 N(%)	S1-S3 N(%)	P-value
Gender	Male	1842(48.4%)	1223(45.1%)	619(56.5%)	<0.001
	Female	1966(51.6%)	1490(54.9%)	476(43.5%)
Ethnicity	Non-Hispanic White	1339(35.2%)	920(33.9%)	419(38.3%)	<0.001
	Non-Hispanic Black	826(21.7%)	652(24%)	174(15.9%)
	Non-Hispanic Asian	533(14%)	399(14.7%)	134(12.2%)
	Other Hispanic	365(9.6%)	259(9.5%)	106(9.7%)
	Mexican American	538(14.1%)	338(12.5%)	200(18.3%)
	Other race-including Multi-Racial	207(5.4%)	145(5.3%)	62(5.7%)
Education	Less than 9th grade	291(7.7%)	196(7.3%)	95(8.8%)	0.002
	9–12 grade	456(12.1%)	335(12.5%)	121(11.2%)
	High school graduate	900(24.0%)	623(23.3%)	277(25.5%)
	College or AA	1181(31.4%)	814(30.5%)	367(33.8%)
	College graduate or above	928(24.7%)	703(26.3%)	225(20.7%)
BMI	Under or normal weight	1039(27.4%)	983(36.4%)	56(5.1%)	<0.001
	Overweight	1214(32.0%)	938(34.7%)	276(25.3%)
	Obese	1538(40.6%)	779(28.9%)	759(69.6%)
Diabetes	Prediabetes	933(24.8%)	613(22.8%)	320(29.6%)	<0.001
	Normal	2127(56.5%)	1744(65%)	383(35.4%)
	Diabetes	706(18.7%)	328(12.2%)	378(35%)
Age	Mean (SD)	50.0(18.4)	48.4(18.9)	53.9 (16.4)	<0.001
Waist	Mean (SD)	99.7(16.8)	94.6(14.8)	112.2(15.0)	<0.001

Table 2:

Blood lipids stratified by steatosis scores: S0 when CAP < 300 dB/m, and S1-S3 when CAP ≥300 dB/m

Variables	Overall mean(SD)	S0 Mean (SD)	> S0 Mean (SD)	P-value
LBXGH-Glycohemoglobin (%)	5.83(1.04)	5.65(0.83)	6.28(1.33)	<0.001
LBXTC – Total Cholesterol (mg/dL)	187.05(40.99)	185.75(40.37)	190.27(42.34)	0.001
LBXTR – Triglyceride (mg/dL)	110.88(87.82)	94.64(59.12)	152.54(127.05)	<0.001
LBDLDL – LDL – Cholesterol (mg/dL)	110.3(35.73)	109.5(35.07)	112.39(37.38)	0.096
LBDHDD – Direct HDL – Cholesterol (mg/dL)	52.6(14.59)	55.16(14.81)	46.25(11.84)	<0.001
LBXSATSI-Alanine Aminotransferase (ALT) (IU/L)	22.09(17.28)	19.75(15.57)	27.88(19.77)	<0.001
LBXSASSI-Aspartate Aminotransferase (AST) (IU/L)	21.52(11.28)	20.82(10.53)	23.23(12.8)	<0.001

We also did scatter plots and calculated Pearson correlation coefficients between BMI and WAIST (Figure A1 in Appendix), also between ALT and AST (Figure A2 in Appendix). We found that there was a significantly positive correlation between BMI and WAIST (Pearson’s correlation coefficient 0.9, p<0.001), also between ALT and AST (Pearson’s correlation coefficient 0.8, p<0.001). To avoid collinearity, we ignored WAIST and AST, and we chose BMI and ALT for further analysis.

We carried out logistic regression models in males and females to examine which factors predict steatosis scores. Table A1 in Appendix provided the odds ratio, 95% confidence interval, and P-value. It showed no significant differences in the variables Direct HDL-cholesterol and total cholesterol and CAP; thus, we ignored these two variables in the subsequent analyses. In the final analyses, we included the demographic variables (age, race, education), the blood lipids triglyceride, BMI and diabetic status as predict variables for steatosis (S0 vs. S1-S3) by using logistic regression models. Further we used the Receiver Operating Characteristic (ROC) curves to visualize the predictive capabilities of these variables on steatosis (Figure 2 for males and Figure 3 for females). Youden’s index, predictive accuracy, and area under curves (AUC) were used to capture the performance of these predictive models.²⁰ Youden index is defined as the summation of sensitivity and specificity minus 1. The larger are these metrics, the better the predictive models are. The method by Delong et al²¹ was used to compare two ROC curves.

Figure 2:

ROC curves for diagnosis of steatosis in males based on ALT alone, ALT along with BMI, diabetes status, demographics, and blood lipids.

Figure 3:

ROC curves for diagnosis of steatosis in females based on ALT alone, ALT along with BMI, diabetes status, demographics, and blood lipids.

A test with p-value less than 0.05 was considered significant. All the analyses were carried out in the statistical software R (https://www.r-project.org/).

3. Results:

3.1. Characterization of our study population

Our study population of 4,183 individuals (prior to applying exclusion criteria mentioned above) was divided into 4 categories based on drinking behavior. There were 10.8% never drinkers (lifetime abstainers), 19.9% non-drinkers for the past year, 60.3% moderate drinkers (1 or 2 drinks per day for females and males, respectively, with no binge drinking), and 9.0% heavy drinkers (>1 or >2 drinks per day for females and males, respectively, and/or binge drinking). This is shown in Figure 1. The analyses below were carried out based on the data excluding heavy drinkers.

Demographic characteristics of our study participants including sex, age, ethnicity, and education. We also included BMI and diabetes status as shown in Table 1. Our population included 48.4% males and 51.6% females. The mean age for our study population was 50.0 years of age with a standard deviation of 18.4. For ethnicity, our study population was 35.2% non-Hispanic white, 21.7% non-Hispanic black, 14.0% non-Hispanic Asian, 9.6% other Hispanic, 14.1% Mexican American, and 5.4% other-including multi-racial. Compared to the US population: this study population had a higher percentage of non-Hispanic black ethnicity (21.7 vs 13.4%) and a higher percentage of non-Hispanic Asian (14.0 vs 5.9%).²² 7.7% of the population had less than a 9^th grade education, 12.1% had 9–12^th grade education, 24.% were high school graduates, 31.4% went to college or had an associate’s degree, and 24.7% had graduated college or further education. Demographics of BMI were 27.4% under or normal weight subjects, 32.0% overweight, and 40.6% obese. 56.5% of the study population had a normal hemoglobin A1c, while 24.8% were pre-diabetic and 18.7% had diabetes.

3.2. Analysis of CAP as compared to ALT

As shown in Table 3A, for males using the ULN of ALT as 29 U/L, comparing CAP S0 vs S1-S3, there was 66.3% accuracy in predicting steatosis or no steatosis. However, 61.7% of patients had false negatives for steatosis using 29 U/L as an upper limit of normal, meaning 61.7% of patients with NAFLD had a normal ALT by that standard. Using the ULN of 30 U/L, the prediction accuracy was 66.6%. This cutoff missed 64.3% of patients with significant steatosis. Using the ACG cutoff of 33 U/L, the accuracy is 67.4%. This ULN missed 70.0% of patients with steatosis. Using the strictest ULN, 24.5 U/L, the accuracy was still only 64.6%, and this cutoff missed 45.7% of patients with S1–3 steatosis. This was the only ULN which was associated with NAFLD correctly >50% of the time. Youden index at this ULN was the largest among the four ALT cutoff values. The false positive rate was 30.2%, as compared to 19.5%, 17.7%, and 13.7% for the above ULNs, respectively. Of note, even when the ALT is high it does not predict NAFLD accurately. For ULN 24.5, only 47.7% of male subjects with ALT ≥23.5 U/L had S1-S3 steatosis.

Table 3:

The sensitivity, specificity, accuracy, and Youden index with different ALT cut points based on the ROC curves of diagnosis of steatosis.

A: Results based on data on males only
		ALT >29				ALT>30				ALT>33				ALT>24.5 (Maximizing Youden index)
	Tot al N (%)	Yes N (%)	No N (%)	Accuracy	Youden ind ex	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden ind ex	Yes N (%)	No N (%)	Accuracy	YouDen index
S1-S3	619 (33.6%)	237 (38.3%)	382 (61.7%)	66.3 %	0.188	221 (35.7%)	398 (64.3%)	66.6 %	0.181	186 (30%)	433 (70%)	67.4 %	0.163	336 (54.3%)	283 (45.7%)	64.6 %	0.241
S0	1223 (66.4%)	238 (19.5%)	985 (80.5%)			217 (17.7%)	1006 (82.3%)			168 (13.7%)	1055 (86.3%)			369 (30.2%)	854 (69.8%)
B: Results based on data on females only
		ALT >19				ALT>22				ALT>25				ALT>17.5 (Maximizing Youden index)
	Total N (%)	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index
S1-S3	476 (24.2%)	235 (49.4%)	241 (50.6%)	70.5 %	0.267	172 (36.1%)	304 (63.9%)	72.9 %	0.208	137 (28.8%)	339 (71.2%)	75.1 %	0.181	273 (57.4%)	203 (42.6%)	68.0 %	0.287
S0	1490 (75.8%)	338 (22.7%)	1152 (77.3%)			228 (15.3%)	1262 (84.7%)			150 (10.1%)	1340 (89.9%)			427 (28.7%)	1063 (71.3%)

In females, as shown in Table 3B, using the ULN of ALT as 19 U/L, and comparing CAP S0 vs S1-S3, there is 70.5% accuracy in predicting steatosis or no steatosis. However, 50.6% of patients had false negatives for steatosis using 19 U/L as an upper limit of normal, meaning 50.6% of patients with NAFLD had a normal ALT by that standard. Using the ULN of 22 U/L, the prediction accuracy was 72.9%, and this cutoff missed 63.9% of patients with significant steatosis. Using the ACG cutoff of 25 U/L, the accuracy was 75.1%, and this ULN missed 71.2% of patients with steatosis. With the strictest ULN, 17.5 U/L, the accuracy was still only 68.0%, and this cutoff missed 42.6% of patients with S1-S3 steatosis. Youden index at this ULN was the largest among the four ALT cutoff values. The false positive rate was 28.7%, as compared to 22.7%, 15.3%, and 10.1% for the above ULNs, respectively. Of note, even when the ALT is high it does not predict NAFLD accurately. For ULN 17.5 U/L, only 39% of female subjects with ALT ≥17.5 had S1-S3 steatosis.

3.3. Analysis of Liver stiffness measurement as compared to ALT

As shown in Table 4A, for males using the ULN of ALT as 29 U/L, comparing Fibrosis (F2–4) vs no fibrosis (F0–1), there is 72.1% accuracy in predicting fibrosis or no fibrosis. However, 55% of patients had false negatives for fibrosis using 29 U/L as an upper limit of normal, meaning 59.7% of patients with fibrosis had a normal ALT by that standard. Using the ULN of 30 U/L, the prediction accuracy was 73.5%. This cutoff missed 62.7% of patients with fibrosis. Using the ACG cutoff of 33 U/L, the accuracy is 76.7%. This ULN missed 68.7% of patients with fibrosis. Using the strictest ULN, 25.5, the accuracy was only 65.5%, and this cutoff missed 48.3% of patients with fibrosis. This was the only ULN with which subjects with fibrosis had true positives >50% of the time. The false positive rate was 32.8%, as compared to 24.0%, 22.1%, and 17.7% for the above ULNs, respectively. Of note, even when the ALT is high it does not predict fibrosis accurately. For ULN 25.5 U/L, only 16.2 % of male subjects with ALT ≥25.5 had fibrosis.

Table 4:

The sensitivity, specificity, accuracy, and Youden index with different ALT cut points based on the ROC curves of diagnosis of fibrosis.

A: Results based on data on males only
		ALT >29				ALT>30				ALT>33				ALT>25.5 (Maximizing Youden index)
	Tot al N (%)	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index
F2–4	201 (10. 9%)	81 (40. 3%)	120 (59. 7%)	72.1 %	0.1 63	75 (37. 3%)	126 (62. 7%)	73.5 %	0.1 52	63 (31. 3%)	138 (68. 7%)	76.7 %	0.1 36	104 (51. 7%)	97 (48. 3%)	65.5 %	0.1 89
F0–1	164 2 (89. 1%)	394 (24 %)	124 8 (76 %)			363 (22. 1%)	127 9 (77. 9%)			291 (17. 7%)	135 1 (82. 3%)			538 (32. 8%)	110 4 (67. 2%)
B: Results based on data on females only
		ALT >19				ALT>22				ALT>25				ALT>17.5 (Maximizing Youden index)
	Tot al N (%)	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index
F2 −4	156 (7.9 %)	84 (53. 8%)	72 (46. 2%)	71.5 %	0.2 68	65 (41. 7%)	91 (58. 3%)	78.3 %	0.2 32	56 (35. 9%)	100 (64. 1%)	83.2 %	0.2 31	100 (64. 1%)	56 (35. 9%)	66.6 %	0.3 10
F 0–1	181 0 (92. 1%)	489 (27 %)	132 1 (73 %)			335 (18. 5%)	147 5 (81. 5%)			231 (12. 8%)	157 9 (87. 2%)			600 (33. 1%)	121 0 (66. 9%)

As shown in Table 4B, for females using the ULN of ALT as 19 U/L, comparing Fibrosis (F2–4) vs no fibrosis (F0–1), there is 71.5% accuracy in predicting fibrosis or no fibrosis. However, 46.2% of patients had false negatives for fibrosis using 29 U/L as an upper limit of normal, meaning 46.2% of patients with fibrosis had a normal ALT by that standard. Using the ULN of 22, the prediction accuracy was 78.3%. This cutoff missed 58.3% of patients with fibrosis. Using the ACG cutoff of 25 U/L, the accuracy is 83.2%. This ULN missed 64.1% of patients with fibrosis. Using the strictest ULN, 17.5 U/L, the accuracy was only 66.6%, and this cutoff missed 35.9% of patients with fibrosis. The false positive rate was 33.1%, as compared to 27.0%, 18.5%, and 12.8% for the above ULNs, respectively. Of note, even when the ALT is high it does not predict fibrosis accurately. For ULN 17.5 U/L, only 14.3% of female subjects with ALT >17.5 U/L had fibrosis.

As shown in Table 5A, for males using the ULN of ALT as 29 U/L, comparing advanced fibrosis (F3–4) vs none to mild fibrosis (F0–2), there is 72.8% accuracy in predicting advanced fibrosis or none to mild fibrosis. However, 59.7% of patients had false negatives for advanced fibrosis using 29 U/L as an upper limit of normal, meaning 59.7% of patients with advanced fibrosis had a normal ALT by that standard. Using the ULN of 30 U/L, the prediction accuracy was 74.6%. This cutoff missed 61.2% of patients with advanced fibrosis. Using the ACG cutoff of 33 U/L, the accuracy was 77.9%. This ULN missed 70.1% of patients with advanced fibrosis. Using the Youden Index ULN, 30.5 U/L, the accuracy was still only 65.3%, and this cutoff missed 49.3% of patients with advanced fibrosis. The false positive rate was 33.6%, as compared to 24.6%, 22.6%, and 18.4% for the above ULNs, respectively. Of note, even when the ALT was high it did not predict fibrosis accurately. For ULN 30.5 U/L, only 10.6% of male subjects with ALT >30.5 U/L had advanced fibrosis.

Table 5:

The sensitivity, specificity, accuracy, and Youden index with different ALT cut points based on the ROC curves of diagnosis of advanced fibrosis.

A: Results based on data on males only
		ALT >29				ALT>30				ALT>33				ALT>30.5 (Maximizing Youden index)
	Tot al N (%)	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index
F 3–4	134 (7.3 %)	54 (40. 3%)	80 (59. 7%)	72.8 %	0.1 57	52 (38. 8%)	82 (61. 2%)	74.6 %	0.1 62	40 (29. 9%)	94 (70. 1%)	77.9 %	0.1 15	68 (50. 7%)	66 (49. 3%)	65.3 %	0.1 71
F 0–2	1709 (92.7%)	421 (24.6%)	1288 (75.4%)			386 (22.6%)	1323 (77.4%)			314 (18.4%)	1395 (81.6%)			574 (33.6%)	1135 (66.4%)
B: 6Results based on data on females only
		ALT >19				ALT>22				ALT>25				ALT>17.5 (Maximizing Youden index)
	Tot al N (%)	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index	Yes N (%)	No N (%)	Accuracy	Youden index
F3 – 4	101 (5.1 %)	61 (60. 4%)	40 (39. 6%)	71.9 %	0.3 29	47 (46. 5%)	54 (53. 5%)	79.3 %	0.2 76	42 (41. 6%)	59 (58. 4%)	84.5 %	0.2 85	68 (67. 3%)	33 (32. 7%)	66.2 %	0.3 34
F0 – 2	186 5 (94. 9%)	512 (27. 5%)	135 3 (72. 5%)			353 (18. 9%)	151 2 (81. 1%)			245 (13. 1%)	162 0 (86. 9%)			632 (33. 9%)	123 3 (66. 1%)

As shown in Table 5B, for females using the ULN of ALT as 19 U/L, comparing advanced fibrosis (F3–4) vs none to mild fibrosis (F0–2), there is 71.9% accuracy in predicting advanced fibrosis or none to mild fibrosis. 39.6% of patients had false negatives for advanced fibrosis using 19 U/L as an upper limit of normal, meaning 39.6% of patients with advanced fibrosis had a normal ALT by that standard. Using the ULN of 22 U/L, the prediction accuracy was 79.3%. This cutoff missed 53.5% of patients with advanced fibrosis. Using the ACG cutoff of 25 U/L, the accuracy is 84.5%. This ULN missed 58.4% of patients with advanced fibrosis. Using the strictest ULN, 17.5 U/L, the accuracy was only 66.2%, and this cutoff missed 32.7% of patients with advanced fibrosis. The false positive rate was 33.9%, as compared to 27.5%, 18.9%, and 13.1% for the above ULNs, respectively. Of note, even when the ALT is high it does not predict advanced fibrosis accurately. For ULN 17.5 U/L, only 9.7 % of female subjects with ALT >17.5 had advanced fibrosis.

3.4. Analysis of CAP as compared to ALT with the addition of BMI and Diabetes status

ALT became more of a helpful marker when combined with BMI, diabetes status, blood lipids, and demographics. Diabetes is defined at HbA1c ≥6.5 or diagnosed based on DIQ_J survey data. The ROC curves for steatosis diagnosis based on ALT in combination with these factors are shown in Figures 2 and 3 for males and females, respectively.

In males, when comparing the area under the ROC curve (AUC), using ALT alone as a predictor for significant steatosis, the AUC was 0.662, with specificity of 0.683, sensitivity of 0.563, and accuracy of 0.675. Using ALT + BMI + diabetes, the AUC was 0.807, with specificity of 0.660, sensitivity of 0.814, and accuracy of 0.75. When using ALT +BMI + diabetes + blood lipids, the AUC is 0.829, with specificity of 0.728, sensitivity of 0.788, and accuracy of 0.767. When using ALT + BMI + diabetes + demographics, the AUC was 0.821, with specificity of 0.730, sensitivity of 0.76, and accuracy of 0.76. When using ALT + BMI + diabetes + blood lipids + demographics, the AUC was 0.845, with specificity of 0.75, sensitivity of 0.802, and accuracy of 0.787 (Table 6).

Table 6:

AUC, Specificity, Sensitivity, and Accuracy of ALT in predicting CAP in males, in combination with the following factors: BMI, Diabetes, Blood lipids, demographics

					P-Value
	AUC	Specificity	Sensitivity	Accuracy	ROC 1	ROC 2	ROC 3	ROC 4	ROC 5
ROC 1: ALT only	0.681	0.706	0.577	0.758		<0.001	<0.001	<0.001	<0.001
ROC 2: ALT + BMI + Diabetes	0.811	0.710	0.773	0.781			0.239	0.380	0.045
ROC 3: ALT + BMI + Diabetes + Blood lipids	0.831	0.664	0.885	0.790				0.655	0.480
ROC 4: ALT + BMI + Diabetes + Demo	0.823	0.733	0.771	0.792					0.205
ROC 5: ALT + BMI + Diabetes + Blood lipids	0.843	0.678	0.901	0.793

In females, when comparing the AUC, using ALT alone as a predictor for significant steatosis, the AUC was 0.681, with specificity of 0.706, sensitivity of 0.577, and accuracy of 0.758. Using ALT + BMI + diabetes, the AUC was 0.811, with specificity of 0.710, sensitivity of 0.773, and accuracy of 0.781. When using ALT + BMI + diabetes+ blood lipids, the AUC was 0.831, with specificity of 0.664, sensitivity of 0.885, and accuracy of 0.790. When using ALT + BMI + diabetes + demographics, the AUC was 0.823, with specificity of 0.733, sensitivity of 0.771, and accuracy of 0.792. When using ALT + BMI + diabetes + blood lipids + demographics, the AUC was 0.843, with specificity of 0.678, sensitivity of 0.901, and accuracy of 0.793 (Table 7). For both female and males, the AUC with ALT alone was significantly different from the other four AUCs with diabetes and BMI as additional predictors, while the other four AUCs were not significantly different, indicating that diabetes and BMI are important predictors for steatosis.

Table 7:

AUC, Specificity, Sensitivity, and Accuracy of ALT in predicting CAP in females, in combination with the following factors: BMI, Diabetes, Blood lipids, demographics

					P-Value
	AUC	Specificity	Sensitivity	Accuracy	ROC 1	ROC 2	ROC 3	ROC 4	ROC 5
ROC 1: ALT only	0.662	0.683	0.563	0.675		<0.001	<0.001	<0.001	<0.001
ROC 2: ALT + BMI + Diabetes	0.807	0.660	0.814	0.750			0.177	0.287	0.020
ROC 3: ALT + BMI + Diabetes + Blood lipids	0.829	0.728	0.788	0.767				0.632	0.406
ROC 4: ALT + BMI + Diabetes + Demo	0.821	0.730	0.760	0.760					0.146
ROC 5: ALT + BMI + Diabetes + Demo + Blood lipids	0.845	0.750	0.802	0.787

4. Discussion:

ACG guidelines suggest an upper limit of normal for ALT for men of 33 U/L, and for women of 25 U/L.¹³ In an examination of the NHANES 1999–2002 and 2005–2008 databases, the ULN of ALT was 29 U/L for men and 22 U/L for women.¹⁴ Prati, et al. proposed a ULN in men of 30 U/L and in women of 19 U/L, based on a study where all subjects had normal viral serologies and BMI <25 kg/m².¹⁵ Our results concerning these three of the most widely used cut off values for ALT, show that in greater than 50% of cases, ALT does not reliably rule out NAFLD and is an unreliable screening marker to assess hepatic steatosis. Even when ALT is high, it does not consistently rule in steatosis. In female subjects with high ALT, less than 50% had significant steatosis greater than S0. In male subjects with high ALT, about 50% had steatosis. These are important findings given that many doctors follow ALT as a measure of improvement or progression of NAFLD, with high ALT often being the factor that prompts further investigation into liver disease.

Our study used the strictest cutoff value for CAP of 300 dB/m, which was derived from two studies. Eddowes et al. compared CAP measured by Fibroscan^® with liver biopsy analysis by blinded expert pathologists in a multicenter study. Their value to distinguish S0 from S1 was 302.¹⁸ The other study reported 294 as the cut off for S0 vs. S1-S3, as determined by a meta-analysis of biopsy proven NAFLD patients.²³

Because ALT alone is a poor marker, algorithms have been proposed to predict NAFLD that include ALT as a component. NAFLD Fibrosis score and Fib-4 are established non-invasive methods to predict fibrosis in NAFLD patients; however, there is not a widely used algorithm to predict the earlier stages of NAFLD when interventions may more easily resolve/attenuate this condition. Other proposed algorithms include the NAFLD liver fat score (NLFS) which uses the presence of metabolic syndrome, type 2 diabetes mellitus, fasting serum insulin, fasting serum AST and AST/ALT ratio.²⁴ The hepatic steatosis index (HIS) uses sex, type 2 diabetes mellitus, BMI, ALT and AST.²⁵ The SteatoTest includes ten markers: Serum α2-macroglobulin, apo A1, haptoglobin, total bilirubin, gamma-glutamyl transferase (GGT), ALT, BMI, serum cholesterol, triglycerides, and glucose.²⁶ The NAFL screening score uses age, fasting blood glucose, BMI, triglycerides, ALT/AST, and uric acid.²⁷ Each of these algorithms has a common thread of diagnosing NAFLD more accurately when factoring in other components of the metabolic syndrome. Type 2 diabetes is the highest risk factor for developing NAFLD and is an independent risk factor in fibrosis progression, due to the role insulin resistance plays in the pathogenesis of steatotic liver disease. Obesity- especially central obesity- adds significant risk of NAFLD and disease progression.²⁸ Our study was consistent with this ideology, as an elevated ALT became a more helpful marker in regard to steatosis when combined with elevated BMI and diabetes status, significantly improving the AUC of our ROC in both males and females. Consistent with data showing NAFLD patients are twice as likely to exhibit dyslipidemia, further improvement was seen with addition of blood lipids and demographics.²⁸ Demographics being important, as older adults, males and Hispanic individuals have higher risk for NAFLD.²⁹

A factor important to gastroenterologists, hepatologists, and other health care providers is whether ALT is an accurate marker to monitor an established patient with NAFLD, to track improvement or progression. Our study was not able to evaluate this. Several studies have suggested that ALT may be inaccurate in tracking disease progression in patients with established NAFLD. A French review article shows evidence that although many practitioners may be reassured by improving ALT levels, it may not a helpful marker. They reviewed 70 patients with untreated NAFLD and with two biopsies performed more than one year apart. After a mean follow-up of 3.7 years, they found that progression of ballooning hepatocytes and bridging fibrosis on biopsy usually co-existed with a reduction in ALT, along with higher weight gain and higher incidence of diabetes. This led them to conclude that a spontaneous decline in ALT should not reassure patients and health care providers unless concomitant with significant weight reduction, in which case ALT may be more accurate.³⁰ However, another study showed that although ALT was not an independent factor associated with worsened NAFLD activity score or fibrosis stage, patients with worsening of either steatosis or fibrosis on liver biopsy tended to have less reduction in ALT.³¹ This is an important clinical issue that require further investigation.

Our study also indicated that ALT is not a reliable marker to indicate the presence of fibrosis or advanced fibrosis. This finding has been observed in other studies. A 2013 study that attempted to determine the ALT value that would accurately predict NASH and advanced fibrosis analyzed 222 patients and found no difference in the rate of advanced fibrosis between normal and elevated ALT.³² A 2018 retrospective study on 771 liver biopsies did not find that ALT was statistically significant for any stage of fibrosis.³³ This highlights a key difficulty in identifying patients at risk for morbidity and mortality from liver disease prior to developing complications. Often patients are only referred for evaluation when their liver function tests are abnormal and many of these patients with advanced fibrosis have normal laboratory tests. Not only do our steatosis and fibrosis results have clinical implications, they are also relevant to epidemiology studies using ALT or unexplained ALT elevation to diagnose NAFLD or its severity. Clearly, more epidemiological data utilizing VCTE are required for NAFLD.

There are limitations to our study. CAP cut off is not standardized, thus, these results may not be representative of patients in all facilities. CAP may also be inaccurate for reasons mentioned above. The gold standard for diagnosis is liver biopsy³⁴, but liver biopsy is not standard-of-care in the U.S. Also the demographics of the NHANES group have a higher percentage of non-Hispanic black and non-Hispanic Asian subjects than the general population of the United States, as well as a higher percentage of Hispanics. In our study, there were 22.6% non-Hispanic black participants and 14.2% non-Hispanic Asians. According to the U.S census, the population of the United States consists of 13.4% non-Hispanic black and 5.9% non-Hispanic Asian individuals. Our study included 23.1% Hispanic subjects compared to 18.5 % in the general population.²² As noted previously, longitudinal data are not available in this large cohort.

5. Conclusion:

In conclusion, our study shows that ALT is not an accurate measure to diagnose or exclude the diagnosis of NAFLD. The false negative rates are extremely high. It is, however, useful when used as part of a comprehensive evaluation, taking into account BMI, diabetes status and lipid profile. Validation studies are required before we recommend specific diagnostic equations. As the prevalence of NAFLD continues to grow and become an even greater problem in the United States and elsewhere, it is important to have a method for predicting whether a patient is at risk. Further studies are needed to clarify an ALT cut off and an algorithm for predicting the presence of NAFLD.

Supplementary Material

1

Figure A1: Correlation between BMI and waist circumference (BMXWAIST).

2

Figure A2: Correlation between AST and ALT

3

Table A1: Odds ratio and P-value from Logistic regression model for CAP in male and female.

Abbreviations:

NAFLD

Non-alcoholic fatty liver disease

NASH

non-alcoholic steatohepatitis

NHANES

National Health and Nutrition Examination Survey

HCC

hepatocellular carcinoma

BMI

body mass index

ALT

alanine aminotransferase

AST

aspartate aminotransferase

ULN

upper limit of normal

VCTE

Vibration-controlled transient elastography

CAP

controlled attenuation parameter

LFT

liver function test

LSM

liver stiffness measurement

ALQ

Alcohol Use Questionnaire

DGA

Dietary Guidelines for Americans

NIAAA

National Institute on Alcohol and Alcohol Abuse

ACG

American College of Gastroenterology

ROC

Receiver Operating Characteristic

AUC

Area under curves

GGT

gamma-glutamyl transferase