Glycated haemoglobin is a major predictor of disease severity in patients with NAFLD

Abstract

Objectives

Currently, non-invasive scoring systems to stage the severity of non-alcoholic fatty liver disease (NAFLD) do not consider markers of glucose control (glycated haemoglobin, HbA1c); this study aimed to define the relationship between HbA1c and NAFLD severity in patients with and without type 2 diabetes.

Research Design and Methods

Data were obtained from 857 patients with liver biopsy staged NAFLD. Generalized-linear models and binomial regression analysis were used to define the relationships between histological NAFLD severity, age, HbA1c, and BMI. Paired biopsies from interventional studies (n=421) were used to assess the impact of change in weight, HbA1c and active vs. placebo treatment on improvements in steatosis, non-alcoholic steatohepatitis (NASH), and fibrosis.

Results

In the discovery cohort (n=687), risk of severe steatosis, NASH and advanced fibrosis correlated positively with HbA1c, after adjustment for obesity and age. These data were endorsed in a separate validation cohort (n=170). Predictive modelling using HbA1c and age was non-inferior to the established non-invasive biomarker, Fib-4, and allowed the generation of HbA1c, age, and BMI adjusted risk charts to predict NAFLD severity. Following intervention, reduction in HbA1c was associated with improvements in steatosis and NASH after adjustment for weight change and treatment, whilst fibrosis change was only associated with weight change and treatment.

Conclusions

HbA1c is highly informative in predicting NAFLD severity and contributes more than BMI. Assessments of HbA1c must be a fundamental part of the holistic assessment of patients with NAFLD and, alongside age, can be used to identify patients with highest risk of advanced disease.

Introduction

The global burden of metabolic disease continues to escalate (1), including its hepatic manifestation, non-alcoholic fatty liver disease (NAFLD). NAFLD is the most prevalent chronic liver condition, affecting approximately 25% of the global population. It is a spectrum of disease that ranges from lipid accumulation within hepatocytes (steatosis), through to inflammation (non-alcoholic steatohepatitis, NASH) and fibrosis, with the potential to develop cirrhosis and its associated risk of decompensation and hepatocellular carcinoma (HCC). NAFLD is strongly associated with type 2 diabetes (T2D) where prevalence rates increase to 56% and 37% for NAFLD and NASH respectively (2). The development of T2D is established as a risk factor driving disease progression (3) based on paired interval liver biopsies (4) or non-invasive liver stiffness measurements (LSM) (5).

Although liver biopsy is the gold-standard for diagnosing NASH and assessing fibrosis severity, it is an invasive, expensive, labour-intensive, and time-consuming procedure with the potential for significant complications and poor patient acceptability. In addition, it is prone to sampling error. As a result, clinical prediction tools and blood-based biomarkers have been developed to identify those individuals with NAFLD (e.g. fatty liver index, FLI) and also to identify those at high risk for NASH and advanced fibrosis (e.g. Fib-4, NAFLD fibrosis score, BARD) (6).

Relatively few studies have examined the link between blood glucose control and NAFLD severity in detail, often using non-invasive assessments without liver biopsies (7–9). In a study including 713 participants with NAFLD who had undergone a liver biopsy, higher HbA1c levels were associated with worsening steatosis, hepatocyte ballooning, and hepatic fibrosis (10). Data examining the impact of change in HbA1c on liver outcomes are sparse. In our published data from 637 patients with T2D, we identified a strong association between baseline HbA1c and FLI, as well as improvement in NAFLD and fibrosis using non-invasive markers that specifically related to improvements in blood glucose control as a result of glucose lowering therapy; improvements that were independent of weight loss and choice of glucose-lowering medication (11). In a study evaluating the thresholds of AST in the screening for liver fibrosis, an increase in HbA1c of 0.3% was associated with an elevated risk of liver fibrosis (RR=1.27 C.I. 1.06-1.53, p=0.01)(9).

Despite the strong relationship between T2D and NAFLD, it is surprising that, currently, no non-invasive assessment score has included HbA1c. Whilst various classes of glucose-lowering agents have been shown to improve NAFLD, including glucagon-like peptide-1 receptor agonists (GLP-1 RA)(12, 13), pioglitazone (14, 15), sodium-glucose cotransporter-2 (SGLT2) inhibitors (16), the focus has often been on the contribution of weight loss, with relatively little attention paid to the potential impact of improvements in HbA1c. Importantly, there are several potential mechanisms by which high circulating glucose levels might exacerbate and worsen NAFLD, but these have not been explored in interventional clinical studies(17, 18).

Our aims were to establish the relationship between HbA1c and histological disease severity, evaluate its ability to accurately predict liver biopsy findings, and to assess whether changes in HbA1c in interventional studies influence histological improvement or worsening on liver biopsy.

Research Design and Methods

Study Design

This analysis combined data from well phenotyped cohorts of patients, all of whom had clinical data collected as well as liver biopsies performed that were scored according to the Kleiner classification(19).

Discovery cohort

consisted of cross-sectional baseline data from the following cohorts; PIVENS (Pioglitazone vs Vitamin E vs Placebo for the Treatment of Nondiabetic Patients with Non-alcoholic Steatohepatitis)(14), FLINT (Farnesoid X Receptor Ligand Obeticholic Acid in NASH Treatment) (20) and TrUSt-NAFLD (Translating the potential of the urine steroid metabolome to stage NAFLD, including pilot and development phases, ref. 21/WM/0177). Data from the PIVENS, FLINT and pilot phase of the TrUSt-NAFLD study are already published (14, 20, 21). The University of Oxford Medical Sciences Interdivisional Research Ethics Committee approved this study (Ref: R69921/RE002).

The PIVENS trial recruited 240 adults without T2D who had histologically confirmed NASH. Participants were required to have definite or possible steatohepatitis with a NAFLD activity score (NAS) of 5, or definite steatohepatitis with a score of 4. All participants had a ballooning score of 1 and patients with cirrhosis were excluded. They were randomly assigned to receive pioglitazone (30 mg, once daily), vitamin E (800IU, once daily), or matched placebo for 96-weeks. Liver biopsies and clinical data collection were repeated at the end of the study.

The FLINT trial recruited 283 adults with histologically proven NASH. Inclusion criteria were definite or borderline NASH with a NAS score of 4, with1 point in each of steatosis, ballooning, and lobular inflammation scores; 53% of participants had coexistent T2D and patients with cirrhosis were excluded. Once enrolled, participants were randomized to receive obeticholic acid or matched placebo for 72 weeks. Liver biopsies and clinical data collection were repeated at the end of the study.

The Trust-NAFLD study (pilot and development phases) was an observational, non-interventional cohort study in which participants with an established diagnosis of NAFLD were recruited and underwent a NAFLD staging liver biopsy as part of their ongoing clinical care. All stages of NAFLD severity were eligible for recruitment and research data and sample collections were undertaken on a single occasion to coincide with a routine clinic visit (21).

The validatory cohort

comprised follow-up clinical and liver biopsy data from participants randomized to the placebo arms of both the PIVENS and FLINT trials and included no data that have been used in the discovery cohort.

For continuous variables, histological grading was defined according to the Kleiner classification (19). For categorical variables, severe steatosis was defined as a histological score ≥2; NASH was defined as a NAS ≥5, or NAS≥4 with evidence of ballooning; advanced fibrosis was defined as stages F3 and F4 inclusive. For the interventional studies (PIVENS, FLINT), NASH resolution was defined as a change in diagnosis from definite or borderline NASH at baseline to a diagnosis of fatty liver disease without steatohepatitis or no fatty liver disease at the end of treatment. No worsening of NASH was defined as no increase in any of ballooning, lobular inflammation, or steatosis. Fibrosis improvement was defined as a reduction of one stage (14, 20).

In subgroup analyses, patients with hyperglycaemia were defined as having Hb1Ac ≥6% (≥42mmol/mol) at baseline. Those patients with HbA1c between 6-6.5% (42-48 mmol/mol) were defined as people with pre-diabetes and those with HbA1c ≥6.5 (≥48 mmol/mol) as people living with T2D.

Statistical Analysis - cross-sectional

Generalised linear regression models were used to study the association of histological features of NAFLD with the relevant metabolic variables available across all the datasets. Each feature was a dependent variable and the regression models included HbA1c at baseline, body mass index (BMI) at baseline and age, all as continuous independent variables. Once dichotomized, we studied severe steatosis, NASH diagnosis and advanced fibrosis using binomial regression modelling. Age (<45, 45-60, or >60 years) and BMI (<30 or ≥30 kg/m²) were included as independent categorical variables to make the results of the regression transferable to heatmaps or risk tables, with HbA1c at baseline as an independent continuous variable. Follow-up data from the placebo-treated participants from the FLINT and PIVENS studies were used as a validatory cohort and identical binomial regression models were used. Regression co-efficient estimates were generated for each variable within the respective models.

Statistical Analysis - longitudinal

To study change in the features of NAFLD, we used dichotomised variables and performed binomial regression analysis with the following independent variables: change in HbA1c and weight change as continuous variables, and treatment arm (active vs. placebo) as a categorical variable. Alongside improvements in steatosis, NASH, and fibrosis, FDA regulatory end-points for phase 3 registration trials (22) were included in the binomial regression analysis (resolution of NASH and no worsening of liver fibrosis; improvement in fibrosis and no worsening of NASH). A post-hoc analysis in patients with hyperglycaemia (Hb1Ac ≥6%, ≥42mmol/mol) was performed using an identical binomial regression approach.

All statistical analysis was performed using Jamovi (version 2.3.21, The Jamovi project, Sydney, Australia) and replicated with JASP (Version 0.17.1, JASP Team, 2023) and SPSS (IBM Corp. Released 2021. IBM SPSS Statistics for Mac, Version 28.0. Armonk, NY: IBM Corp).

Results

Discovery cohort analysis

Clinical data and liver biopsy findings were obtained from a total of 687 participants (301 males) enrolled into the studies described above (Supplementary Table 1). There was a degree of heterogeneity amongst the clinical characteristics of the recruited population reflecting the inclusion criteria for the individual studies. Of note, the age of participants within the PIVENS study was significantly younger, with few patients with co-existent T2D and a lower prevalence of advanced fibrosis on liver biopsy.

Cross-sectional analysis using multivariate linear regression models

Multivariate linear regression models were used to assess the relationships between liver biopsy findings (steatosis, NAS and fibrosis scores) and HbA1c. There was a significant correlation between histologically graded steatosis and HbA1c (estimate 0.14, p<0.001 for HbA1c, R=0.24, R²=0.05, p<0.001 for overall model) even after adjustment for age, which inversely correlated with steatosis (p<0.001), and BMI, which was unrelated to the degree of steatosis (p=0.95) (supplementary Fig. S1 A-C). Similarly, NAS score showed a significant association with HbA1c (estimate 0.25, p<0.001 for HbA1c, R=0.19, R²=0.04, p<0.001 for overall model), after adjustment for age (p<0.008) which again was inversely correlated, and BMI, which showed no significant relationship (p=0.08) (supplementary Fig. S1 D-F), but when looking specifically at lobular inflammation in isolation, it was not significantly associated with HbA1C. In addition, histological fibrosis correlated with HbA1c (estimate 0.24, p<0.001 for HbA1c, R=0.43, R²=0.188, p<0.001 for overall model), after adjustment for age (p<0.001) and BMI; BMI did not impact fibrosis stage (p=0.16) (supplementary Fig. S1 G-I).

Cross-sectional analysis using binomial regression models

Binomial regression models were used to assess the probability of severe steatosis, NASH and advanced fibrosis with levels of HbA1c. Analysis was stratified and adjusted for age (<45, 45-60 and >60 years) and obesity (<30kg/m² and ≥30kg/m²). HbA1c was significantly associated with the probability of severe steatosis (estimate 0.253, OR=2.57, p=0.01 for HbA1c, X²=14.9, p=0.005 for overall model) when adjusted for age class (p>0.1) and obesity (p=0.36); neither age or obesity impacted on the likelihood of severe steatosis (Fig. 1A and B, Table 2). Endorsing the data from the linear regression models, the probability of identifying NASH on a liver biopsy significantly increased with progressive elevations in HbA1c (estimate 0.30, OR=1.36, p=0.011 for HbA1c, X²=10.4, p=0.035 for overall model) after adjustment for obesity (p=0.135), age class (p=0.64) and T2D diagnosis (p=0.34). Obesity and age did not influence the likelihood of identifying NASH (Fig. 1C and D, Table 1). Finally, the probability of having advanced fibrosis (F3-F4) on liver biopsy was significantly associated with HbA1c at baseline (estimate 0.40, OR=1.49, p<0.001 for HbA1c, X²=83.8, p<0.001 for overall model) after adjustment for age class (increasing age associated with advanced fibrosis, p<0.001) and obesity, which did not influence the likelihood of identifying advanced fibrosis (p=0.54) (Fig. 1E and F, Table 1). When T2D diagnosis was added the model, it changes the significance of the association of fibrosis and HbA1c (p=0.181), but a threshold effect can be appreciated for HbA1c greater than 42 mmol/mol (p=0.004 for HbA1c, and p=0.022 for HbA1c>42mmol/mol). Inclusion of sex in all the models tested did not alter statistical significance and therefore it was not included in the final reported models.

Figure 1.

HbA1c can predict the severity of NAFLD based on liver biopsy assessment in a discovery cohort of 687 patients after adjustment for age and BMI (Probability of severe steatosis A and B; NASH C and D; advanced fibrosis E and F). NAFLD severity risk charts (severe steatosis; NASH C; advanced fibrosis) derived from a combined cohort of 857 patients with liver-biopsy staged NAFLD. G shows risk charts for people with BMI<30 kg/m²; H shows risk charts for people with BMI ≥ 30 kg/m².

Table. 2. Binomial regression analysis determining the impact of change in HbA1c, weight change, and active vs. placebo intervention on the probability of improvement in steatosis, NASH and fibrosis on liver biopsy in the whole longitudinal cohort and in the subgroup of patients with pre-existing hyperglycaemia (HbA1c >6%) (est.=estimates).

	Steatosis improvement				NASH improvement, no worsening of fibrosis				Fibrosis Improvement, no worsening of NASH
Whole cohort (n=421)
Model coefficients	χ2= 57.3, p<0.001				χ2=41.4, p<0.001				χ2= 20.6, p<0.001
Predictor	Est.	SE	p	OR (95% CI)	Est.	SE	p	OR (95% CI)	Est.	SE	p	OR (95% CI)
1% change in HbA1c	-0.49	0.16	0.003	0.616 (0.45 – 0.85)	-0.30	0.14	0.034	0.74 (0.56 – 0.98)	-0.08	0.14	0.57	0.93 (0.71 – 1.21)
1kg change in Weight	-0.06	0.02	<0.001	0.94 (0.91 – 0.97)	-0.06	0.02	0.001	0.94 (0.91 – 0.97)	-0.05	0.0.18	0.007	0.95 (0.92 – 0.99)
Treatment; Placebo - Active	-1.11	0.22	<0.001	0.33 (0.22 – 0.50)	-1.03	0.23	0.001	0.36 (0.23 – 0.56)	-0.87	0.24	0.001	0.42 (0.26 – 0.68)
Patients with hyperglycaemia (HbA1c >6%) (n=185)
Model coefficients	χ2= 18.6, p=0.005				χ2=31.2, p<0.001				χ2= 16.1, p=0.013
1% change in HbA1c	-0.34	0.17	0.047	0.71 (0.51 – 0.99)	-0.37	0.18	0.039	0.69 (0.49 – 0.98)	-0.08	0.17	0.63	0.92 (0.66 – 1.29)
1kg change in Weight	-0.05	0.03	0.078	0.95 (0.90 – 1.01)	-0.09	0.03	0.004	0.92 (0.86 – 0.97)	-0.04	0.03	0.13	0.96 (0.91 – 1.01)
Treatment; Placebo – Active	-0.87	0.32	0.006	0.42 (0.23 – 0.78)	-1.23	0.35	0.001	0.29 (0.15 – 0.58)	-1.24	0.38	0.001	0.29 (0.14 – 0.61)
Pre-Diabetes – T2D	-0.22	0.33	0.51	0.80 (0.41 – 1-55)	-0.18	0.36	0.62	0.84 (0.42 – 1.68)	0.10	0.37	0.78	1.11 (0.54 – 2.30)
Age: 45+ - <45	0.09	0.40	0.82	1.10 (0.49 – 2.43)	0.49	0.45	0.28	1.63 (0.68-3.91)	-0.20	0.44	0.65	0.82 (0.35 – 1.93)
Age: >60 - <45	-0.32	0.46	0.49	0.73 (0.30 – 1.80)	0.44	0.50	0.38	1.56 (0.58 – 4.16)	-0.60	0.51	0.243	0.55 (0.20 – 1.50)

Table. 1. Binomial Regression Analysis determining the impact of HbA1c, age and obesity on the probability of identifying severe steatosis, NASH and advanced fibrosis on liver biopsy in the discovery and validatory cohorts (est.=estimates).

	Severe Steatosis				NASH				Advanced Fibrosis
Discovery cohort (n=687)
Model coefficients	χ2= 14.9, p=0.005				χ2=10.4, p=0.035				χ2= 83.8, p<0.001
Predictor	Est.	SE	p	OR (95% CI)	Est.	SE	p	OR (95% CI)	Est.	SE	p	OR (95% CI)
HbA1c (%)	0.25	0.0981	0.01	1.29 (1.06 – 1.56)	0.30	0.12	0.011	1.36 (1.07 – 1.72)	0.40	0.08	<0.001	1.49 (1.27 – 1.75)
BMI<30kg/m2	-0.42	0.20	0.036	0.66 (0.44 – 0.97)	0.34	0.23	0.13	1.41 (0.90 – 2.21)	-0.13	0.20	0.54	0.88 (0.59 – 1.32)
Age (years); <45- 45+	0.25	0.23	0.23	1.31 (0.84 – 2.07)	-0.07	0.26	0.8	0.94 (0.56 – 1.56)	-0.58	0.24	0.0016	0.56 (0.35 – 0.90)
Age (years); 45+- >60	-0.16	0.23	0.47	0.85 (0.55 – 1.32)	-0.15	0.31	0.64	0.86 (0.47 – 1.60)	1.09	0.21	<0.001	2.97 (1.96 – 4.48)
Validatory cohort (n=170)
Model coefficients	χ2= 10.1, p=0.038				χ2=20.6, p<0.001				χ2= 18.0, p=0.001
HbA1c (%)	0.53	0.192	0.006	1.70 (1.17 – 2.48)	0.61	0.22	0.006	1.83 (1.19 – 2.82)	0.56	0.18	0.002	1.75 (1.24 – 2.46)
BMI<30kg/m2	-0.06	0.35	0.87	0.94 (0.48 – 1.87)	1.02	0.36	0.005	2.77 (1.36 – 5.65)	0.52	0.41	0.21	1.68 (0.75 – 3.77)
Age (years); <45- 45+	-0.22	0.37	0.55	0.80 (0.38 – 1.67)	-0.28	0.39	0.48	0.76 (0.35 – 1.63)	0.68	0.45	0.13	1.98 (0.83 – 4.74)
Age (years); 45+- >60	-0.50	0.44	0.25	0.60 (0.25 – 1.43)	0.62	0.49	0.21	1.85 (0.71 – 4.84)	1.05	0.51	0.04	2.85 (1.04 – 7.80)

HbA1c with age modelling compared to Fib-4 to predict advanced fibrosis

The Fib-4 score is an established non-invasive biomarker that is used to assess the likelihood of advanced fibrosis on liver biopsy, with emerging data suggesting that is has utility in predicting long-term clinical outcome (23). Extending the binomial regression analyses, we developed a predictive model for advanced fibrosis risk based on either Fib-4 or HbA1c with age. The area under the curve (AUC) for the receiver operator characteristic (ROC) curves for Fib-4 (AUROC 0.77, 95% CI 0.73-0.81) and for HbA1c with age (AUROC 0.73, 95% CI 0.69-0.77) were not significantly different.

Validatory cohort analysis

As described above, the validatory cohort comprised follow-up clinical and liver biopsy data from participants randomized to the placebo arms of both the PIVENS and FLINT trials. A total of 170 participants were included in the validatory analysis and clinical and liver biopsy findings are presented in table 1. There were no significant differences in demographics between the validatory and discovery cohorts, however, NAFLD disease severity on liver biopsy was milder with a lower prevalence of severe steatosis, NASH and advanced fibrosis (Supplementary Table 1).

Overall, using binomial regression analysis in the validatory cohort, the patterns identified with regards to the contribution of HbA1c to NAFLD disease severity were similar to those observed in the discovery cohort. HbA1c alone (and neither age nor obesity) contributed significantly to the likelihood of severe steatosis (Fig. S2A and B, Table 1). As in the discovery cohort, HbA1c was associated with the presence of NASH, but in addition, in the validatory cohort there was a significant contribution from obesity, but not age. As in the discovery cohort, HbA1c, alongside age, but not obesity, was associated with the presence of advanced fibrosis (Fig. S2C-F, Table 1).

Having replicated observations from the discovery cohort in the validatory cohort, data from the binomial regression models from the combined cohorts (discovery and validatory, n=840) were used to generate HbA1c, age and BMI dependent risk charts demonstrating the potent impact of HbA1c and age (but not BMI) on NAFLD severity (Fig. 1G-H).

Change in HbA1c; impact on histological severity of NAFLD

Prospective, longitudinal changes in the histological severity of NAFLD were interrogated using clinical data and liver biopsy samples from the PIVENS and FLINT cohorts (pre-and post-intervention, n=421). Using a binomial regression model (X²=57.3, p<0.001 for model), the probability of improvement of steatosis was significantly associated with the change HbA1c (estimate=-0.48, OR= 0.62, p=0.003) after adjustment for weight change (estimate=-0.06, p<0.001) and treatment (active drug vs. placebo, p<0.001), both of which contributed significantly (Fig. 2A and B, Table 2). Similarly, the probability of NASH improvement (reduction in NAS by >=2), was associated with HbA1c change (estimate=-0.30, OR=0.742, p=0.034 for HbA1c, X²=41.4, p<0.001 for overall model) even after adjustment for weight change (estimate=-0.06, p<0.001) and treatment (p<0.001) (Fig. 2C and D, Table 2).The probability of NASH resolution with no fibrosis worsening did not correlate with HbA1c change (p=0.189), nor with weight change (p=0.086), but only with treatment (active drug vs. placebo, p<0.001) (supplementary Fig. S3A and B). Specifically, ballooning improvement was associated with treatment (p<0.001), weight change (p=0.007) and also with change in HbA1c (p=0.024). The likelihood of fibrosis improvement without worsening of NASH was significantly associated with treatment (p<0.001) and weight change (p=0.007), but not with change in HbA1c (p=0.57) (Fig. 2E and F, Table 2). Adjustment for baseline histology did not affect the overall analyses; HbA1c reduction was significantly associated with steatosis and ballooning improvement with a borderline significant association with NASH resolution. Changes with fibrosis remained non-significant (Supplementary Table 2).

Figure 2.

In a cohort of 421 patients with NAFLD, with and without hyperglycaemia and liver biopsies performed before and after intervention, change in both HbA1c and weight predict improvements in steatosis (A and B) and NASH (C and D). Change in weight, but not HbA1c predicts improvement in severity of fibrosis (E and F). Data are presented after adjustment for active vs. placebo treatment which predicted improvements in steatosis, NASH and fibrosis (Table 2).

Subgroup analysis; participants with hyperglycaemia

Because the lack of association with HbA1c change in the initial longitudinal interventional analysis could have been confounded by individuals with normal HbA1c values, we therefore undertook a subgroup analysis to include only patients with evidence of hyperglycaemia at baseline (pre-intervention), defined as HbA1c≥6% (n=185). The associations between change in HbA1c and steatosis and NASH improvement (but not with fibrosis) were similar to those observed in the whole cohort (Figure S4, Table 2).

Conclusions

The analysis from this study has shown that HbA1c is highly informative at predicting stage of NAFLD severity based on liver biopsy findings and may make a more significant contribution than BMI. A reduction of HbA1c by 1% had the equivalent impact to 8kg of weight loss in conferring a 50% increase in the probability of steatosis improvement; a reduction of HbA1c by 1%, equated to a reduction of 5kg in body weight driving a 30% increase in the probability of NASH improvement. The combination of age and HbA1c was able to predict the presence of advanced fibrosis with a similar accuracy to the established non-invasive marker, Fib-4. Within the LITMUS cohort including 966 patients with biopsy proven NAFLD, a comprehensive array of non-invasive biomarker strategies were compared to detect advanced fibrosis and Fib-4 data in that cohort was comparable to the current study (AUROC 0.75, 95% CI 0.70-0.80) (24). HbA1c data was not presented as part of that analysis. In a further individual patient data meta-analysis of 1,780 patients with biopsy-proven NAFLD and T2D, similar AUROC values for Fib-4 to the current were also reported (0.754) (25).

Fibrotic-NASH Index (FNI) has been recently proposed as a specific liver fibrosis score in NASH and it includes HbA1c alongside HDL and AST (26). The same was validated only on a cohort of subjects with high risk of NASH fibrosis assessed with established surrogate marker Fib-4 but without liver biopsy; a definitive validation is therefore expected, based on histology-proven fibrosis.

When compared to the FNI, our model, which relies on HbA1c and Age, seems to provide compatible predictability of fibrosis, based on the AUROC (Supplementary Figure 5). Therefore, it could prove beneficial to incorporate this approach as an adjunctive tool within our clinical practice, given its user-friendly nature and the accessibility of pertinent variables. It is worth highlighting that this proposed method obviates the need for evaluating AST and HDL, potentially leading to a more streamlined and cost-effective approach to our endeavours.

We have been able to generate risk charts that include HbA1c (alongside age and BMI) that may guide clinicians as to the likelihood of advanced NAFLD and this could help to risk stratify patients who may benefit from additional investigations (including liver biopsy) or treatments. There is well-established evidence linking liver biopsy findings to outcome, but the data using non-invasive scoring systems is only beginning to emerge (27). Whilst the data that we have generated does not allow us to link our biopsy predictions to clinical outcome, the same approach could be used on a much broader scale and with point-of-care testing, an accurate risk assessment can be generated that would allow immediate, targeted, and aggressive management of those individuals at highest risk. Whilst there are currently no licenced therapies for NAFLD, it also has the potential to identify patients who are most at risk of advanced disease and therefore perhaps most likely to benefit from liver-targeted therapy once these become available for clinical use.

There is a high degree of biological plausibility in the role of HbA1c impacting on NAFLD disease severity and multiple mechanisms that are likely to be involved. Although free fatty acid delivery is the major contributor to hepatic lipid accumulation, de novo lipogenesis (DNL), the synthesis of new lipid from non-lipid precursors, is highly regulated and makes a much greater contribution to those individuals with higher liver fat content (28). In patients with T2D, the principal substrate for DNL is glucose and, as a result, chronic hyperglycemia will drive DNL. Kupffer cells, resident macrophages within the liver, are critical in mediating the inflammatory response that defines the development of NASH. In combined Kupffer cell and hepatocyte models, high glucose levels promote lipid accumulation, glucose output and impaired insulin sensitivity (18). In addition, glucose metabolism increases oxidative damage within hepatocytes.

There is now a considerable body of evidence demonstrating that glucose lowering therapies are beneficial in NAFLD. There has been a significant focus on the impact of weight loss, notably with respect to the GLP-1RAs(13, 29), SGLT2 inhibitors (16), as well as low-energy diets. Interestingly, our own work failed to show an improvement in liver steatosis in patients with NAFLD, but without T2D treated with an SGLT2 inhibitor, despite 4.4kg of weight loss, perhaps suggesting that glucose lowering may be fundamentally important (30). Pioglitazone has been shown to improve NAFLD liver histology in some studies (in patients with and without T2D), but the precise contribution to the benefits that it delivers in terms of HbA1c lowering have not been dissected (15, 31, 32). Finally, although there are concerns with weight gain following the commencement of insulin therapy and the potential to drive DNL, published data would suggest that starting insulin treatment decreases hepatic steatosis (33) and we would argue that this largely reflects improvements in glucose control. The specific impact of insulin therapy on liver histology has to be evaluated.

There are limitations in the current analysis. Data were combined from several studies and although within each individual study data were rigorously collected, protocols differed, and histological grading was performed by different histopathologists. However, it is reassuring that our findings with regards to other non-invasive assessments of NAFLD severity (e.g., Fib-4) are consistent with the published literature (24, 25). We only have data for HbA1c at the time of the biopsy. We are therefore unable to assess the impact of prolonged hyperglycaemia for which duration of T2D alongside historical assessments of blood glucose control would be required. There are additional metabolic variables that could influence the severity and progression of NALFD (inlcuding cholesterol, HDL, tryglicerides, uric acid, duration of diabetes, waist circumference), but data for these was not uniformly available across. Confounding medications is also an important consideration. Whilst SGLT2i and GLP-1RA might have an impact on NAFLD, they were not available at the time of PIVENS recruitment, and were still experimental at the time of FLINT recruitment. Specifically with regards to statins, insulin, metformin and pioglitazone, doses of these mediactions remained unchanged per protocol throughout the time of observation. Despite these shortcomings, the relationships that we observed with HbA1c were robust and reproducible in the discovery and validatory cohorts. Our interventional analysis was limited by relatively small numbers as well as the fact that some of the interventions (obeticholic acid and vitamin E) do not directly target HbA1c and therefore improvements in glucose control were modest. This further supports the need to establish another replication cohort that is not sourced from clinical trials. Moreover, the impact of cirrhosis on heamoglobin levels might have limited the accuracy of HbA1c as a marker of fibrosis and fibrosis change. Along the same line, variables such as glucose-independent variation in HbA1c values and patient-specific glycation kinetics can add further bias.

The management of T2D has become progressively more holistic, however, glucose control in the context of NAFLD has been somewhat overlooked. Data from this study and others (7, 10) would argue that an assessment of glucose control is crucial for accurate estimations of disease severity. Clinical guidance from major hepatology, diabetes, metabolic, and endocrine professional organizations within the US and Europe advocates case finding and risk stratification to identify individuals at hat the highest risk of advanced fibrotic NAFLD (31, 34). The analyses and tools that we have developed may offer significant clinical utility to identify those patients most at risk of advanced disease using simple point-of-care tests that may be immediately available to the practicing clinician. The multidisciplinary and holistic care of the individual with NAFLD should include liver assessment and specific treatment where available, weight loss and cardiovascular risk reduction (35). In addition, based on our data, we would advocate aggressive glucose lowering in those individuals with hyperglycemia in order to reach and maintain tight glycaemic control.

Research in context.

Evidence before this study

Prior to the study analysis, we performed a review of the literature to identify studies that have utilised assessments of glucose control (HbA1c) in the non-invasive assessment of NAFLD severity. The following databases were searched (latest search 31/3/2023); PubMed, the Cochrane registry of studies and Embase. We reviewed all studies that were published in full in English and included data on NAFLD severity (but not including other chronic liver conditions). Studies were review that included both liver biopsy and non-invasive assessments of NAFLD severity in which measurement of HbA1c had been performed and reported.

Added value of this study

Despite the tight relationship between NAFLD and type 2 diabetes, few studies have determined the impact of HbA1c on NAFLD disease severity. Furthermore, we are not aware of any currently used clinical non-invasive biomarker strategy to stage NAFLD severity incorporates an assessment of glucose control or HbA1c. We have shown, in the largest cohort published to date, that HbA1c is a major contributor in predicting disease severity and contributes more than weight; in absolute terms, a 1% difference in HbA1c conveys the same risk / benefit to the liver as 8kg in weight. Extending these observations, simply using age and HbA1c, can predict NAFLD severity as accurately as other markers such as Fib-4. In addition, our interventional analysis has shown that improving HbA1c are independently associated with improvements in NAFLD.

A simple assessment using HbA1c, age, and BMI can predict the likelihood of NAFLD severity, and we have produced risk stratification charts to facilitate this. Such an assessment can be undertaken cheaply, effectively, and immediately at the point-of-care. This could be hugely beneficial for both the clinician (primary and secondary care) and for the patient, to identify those at the highest risk of advanced NAFLD, helping to guide the need for further investigations and potential treatments.

Highlights.

• Previous studies have shown that HbA1c is a risk factor for NAFLD, but its impact on NAFLD severity has not been well studied.

• This analysis shows that HbA1c is a major contributor to NAFLD severity; a 1% difference in HbA1c is associated with the same risk to the liver as 8kg in weight.

• HbA1c was independently associated with NAFLD severity, even after adjusting for other factors including age, BMI, and waist circumference.

• A simple assessment using HbA1c, age, and BMI can predict the likelihood of NAFLD severity.

Disclosure summary.

JWT has been an advisory board member for Novo Nordisk, Pfizer and Poxel and a member of a data and safety monitoring committee for Novartis; JFC has been an advisory board member for Novo Nordisk and Intercept; consultancy for Alnylam; PNN discloses the following financial relationship(s) on behalf of the University of Birmingham with a commercial interest: Grant/research support from Novo Nordisk; Consulting fees from Astra Zeneca, Boehringer Ingelheim, BMS, Gilead, GSK, Intercept, Madrigal, Novo Nordisk, Pfizer and Sun Pharma. GDT has received funding from Novo Nordisk, Lilly, Sanofi. JWT, DAK, and GDT were supported by National Institute of Research (NIHR) Oxford Biomedical Research Centre (BRC). The views expressed are those the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. DAK, JWT, and JFC have been investigators in two publicly-funded (NIHR) trials where the weight loss intervention was donated by Nestle Health Sciences to University of Oxford outside the submitted work. GPA is supported by the NIHR Nottingham Biomedical Research Centre [BRC-1215-20003]. GPA contracted through University of Nottingham contracting team has received consulting fees from Abbott Products Operations AG, Albireo Pharma Inc, Amryth, AstraZeneca, BenevolentAI Bio, BerGenBio ASA, Clinipace, GSK, DNDi, Merck, Pfizer Inc and PureTech. WA is in receipt of a Medical Research Council project grant, has received consultancy fees from Novo Nordisk, Glaxosmithkline, Gilead Science and Janssen and competitive grant funding from Gilead Science and Glaxosmithkline.

Data Availability

PIVENS and FLINT data are publicly available upon request to the CRN. Trust-NAFLD data are available upon reasonable request.