Association between hepatic steatosis index and glucose status conversion from impaired fasting glucose base on Chinese adults: a cohort study from real-world

Abstract

Objective

Evidence regarding the relationship between the hepatic steatosis index (HSI) and glycemic conversion outcomes in individuals with impaired fasting glucose (IFG) is still limited. Our study aims to explore the role of HSI in the reversion to normoglycemia or the progression to diabetes among Chinese IFG individuals.

Methods

We conducted a retrospective analysis using data from 11,327 IFG individuals who had undergone wellness examinations at Rich Healthcare Group. To analyze the association between the baseline HSI and glucose status conversion, a Cox regression model was used, and the hazard ratio (HR) and 95% confidence interval (CI) were computed. A generalized additive model was used to examine non-linear relationships. A two-piecewise binary logistic regression model was employed to further elucidate the non-linearity. Sensitivity and subgroup analyses were also conducted.

Results

Over an observation period spanning 33,892 person-years, the rate of normoglycemia reversion was found to be 41.75%, whereas the rate of progression to diabetes was 11.63%. After accounting for potential confounding variables, our analysis demonstrated that among IFG individuals, there was an inverse relationship between HSI and the likelihood of returning to normoglycemia (HR = 0.93, 95% CI: 0.90–0.96, P < 0.001), and a positive association between the HSI and progression to diabetes (HR = 1.49, 95% CI: 1.40–1.58, P < 0.001). The smooth curve-fitting plot revealed a nonlinear association between the HSI and diabetes progression, with inflection points at 26.55 and 40.74. Sensitivity analysis and subgroup analysis confirmed the stability of the study’s findings.

Conclusion

HSI was significantly linked to normoglycemia reversion and diabetes progression in IFG individuals, indicating its potential as a risk indicator for diabetes and a guide for prevention strategies. However, further research is needed to confirm this.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40001-025-02354-4.

Background

Prediabetes represents an early stage of type 2 diabetes (T2DM). It is characterized by elevated blood glucose levels, but not the diagnostic threshold for diabetes. Approximately one-third of the American population has abnormal blood glucose levels, a group that accounts for $237 billion in annual healthcare expenditures [1]. It was estimated that by 2030, over 470 million individuals worldwide would suffer from prediabetes [2]. In China, about 30% of adults suffered from prediabetes [3]. Furthermore, there was an increased risk of stroke, cardiovascular disease, and T2DM in people with prediabetes [4, 5]. Prediabetes encompasses two distinct conditions: impaired fasting glucose (IFG) and impaired glucose tolerance. Individuals may also present with a combined form of these conditions, referred to as IFG combined with impaired glucose tolerance. According to current research data, IFG was ranked as the fifth leading cause of disease burden globally in 2017 [6]. Preventing the development of T2DM and reducing the complications associated with impaired glucose metabolism are critical health objectives. To achieve these, understanding the factors that affect the glucose status conversion in pre-diabetic individuals is essential.

Nonalcoholic fatty liver disease (NAFLD), is closely linked to insulin resistance, obesity, and dyslipidemia, all of which are risk factors for prediabetes [7]. A liver biopsy is traditionally considered the gold standard for diagnosing NAFLD, but it is invasive, expensive, and can cause complications [8, 9]. To address these issues, abdominal ultrasound is frequently employed to diagnose fatty liver diseases. Nevertheless, its sensitivity was restricted and failed to offer details regarding the degree of fibrosis, posing a significant difficulty in individuals with steatosis below 20–30% [10]. As diagnostic methods for NAFLD continue to evolve, simple and cost-effective predictors have been developed, including the fatty liver index (FLI) [11], the hepatic steatosis index (HSI) [12], and the lipid accumulation products index [13]. Compared to FLI and lipid accumulation products index, the calculation of HSI is relatively straightforward as it integrates several factors including sex, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and body mass index (BMI). AST and ALT are widely recognized as cost-effective biochemical markers in routine health assessments, which could promote the broader implementation of HSI. The HSI has been demonstrated to be a robust indicator for NAFLD, exhibiting high sensitivity and specificity [14]. It has a strong association with metabolic syndrome, insulin resistance and T2DM [15–18]. It serves not only as a potential indicator for the development of T2DM, but also as a valuable tool for monitoring its progression. Therefore, we proposed the hypothesis that the HSI may serve as a potential predictive marker for the glycemic outcomes in IFG individuals.

Studies have shown that prediabetic populations have different predictors of progression to diabetes and regression to normoglycemia [19]. Therefore, further investigation is essential to elucidate the specific determinants that influence the glycemic trajectory in individuals with prediabetes. The HSI has been shown to be a robust indicator of NAFLD, with a significant positive association with the risk of T2DM [18]. However, its association with glycemic transition in IFG individuals has not been widely reported in the literature. Further research is needed to determine its role in monitoring and forecasting the reversion of glycemic levels and the progression to diabetes.

Material and methods

Data source

Our study utilized original data from the publication by Chen et al. [20]. The authors have graciously relinquished their proprietary rights to this data set, which allows us to carry out secondary analyses without any concerns about violating copyright regulations. The dataset is readily available for download at no charge from the official Dryad database website (https://datadryad.org/stash/dataset/doi:10.5061/dryad.ft8750v) [21].

Population study and data collection

It was initially compiled from a database developed by Rich Healthcare Group, which contained all medical records of Chinese individuals who underwent health checks between 2010 and 2016. The original study population of 685,277 individuals was narrowed down to 211,833 individuals based on the filtering criteria presented in Fig. 1. In our study, we excluded 185,586 participants with fasting plasma glucose (FPG) levels below 5.6 mmol/L at baseline. The exhaustive criteria for excluding subjects were based on the guidelines showed in Fig. 1. In our study, 11,327 subjects were included. In the initial study, demographic information and laboratory data were collected by specialized technicians. Demographic information included information on age, sex, BMI, diastolic blood pressure (DBP), systolic blood pressure (SBP), smoking and drinking status, and family history of diabetes. The measured laboratory parameters included FPG, high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), blood urea nitrogen (BUN), triglycerides (TG), AST, ALT and serum creatinine (Scr).

Fig. 1.

Flowchart of study participant

Definition

The criteria for determining impaired fasting glucose in our study were based on the FPG levels at the first follow-up visit (5.6 mmol/L ≤ FPG < 7 mmol/L) [22]. To evaluate the return to normoglycemia, FPG < 5.6 mmol/L at the second follow-up visit without self-reported incident diabetes was considered indicative of reversion to normoglycemia. The formula for calculating the HSI was as follows: HSI = 8 × ALT/AST ratio + BMI (+2 if female) [10].

Missing data treatment

Our study observed missing data for the following measured indicators: SBP, DBP, TC, TG, BUN, Scr, LDL-C, HDL-C, alcohol consumption status, and smoking status. Specifically, the number of individuals with missing data for these variables was 5 (0.04%), 5 (0.04%), 89 (0.79%), 89 (0.79%), 1271 (11.22%), 524 (4.63%), 3606 (31.84%), 4166 (36.78%), 7884 (69.60%) and 7884 (69.60%), respectively. To safeguard the integrity of our analysis, we employed a variety of imputation and simple interpolation techniques to address the missing data points. For categorical variables such as alcohol consumption and smoking status, we treated the missing values as ‘NA’ and analyzed them as a separate category. For continuous variables such as SBP, DBP, TC, TG, Scr, BUN, LDL-C, and HDL-C, we employed a multiple imputation technique with five iterations to handle missing data. The imputation was conducted using the chained equations method from the mice package in the R programming language, a widely recognized approach for managing missing data [23, 24]. Our analysis demonstrated that the core results from the imputed data were consistent with those from the original dataset (Supplementary Table S3).

Statistical analysis

The Kolmogorov–Smirnov test was used to determine the normality of continuous variables. Variables with skewed distributions were described using medians and interquartile ranges. In the case of variables following a normal distribution, means and standard deviations (SD) were calculated. Categorical variables were described using percentages. In our study, we examined inter-group differences across various variables. For categorical variables, we utilized the Chi-square test. Continuous variables with a normal distribution were analyzed using one-way ANOVA, whereas the Kruskal–Wallis H test was applied to those with a skewed distribution. The analyses were conducted within the framework of HSI quartile categorization. Upon identifying significant differences among at least two groups, post hoc multiple comparisons were performed to further investigate these findings. For continuous variables, the Student–Newman–Keuls method was employed, while the Bonferroni correction was applied to categorical variables to adjust for multiple comparisons. The objective of these comparisons was to ascertain significant differences relative to the first quartile, thereby elucidating the relative standings within the quartile distribution. Incidence was reported as cumulative incidence and per 1000 person-years. Kaplan–Meier survival curves were utilized to evaluate the likelihood of IFG normalization or progression to diabetes across each HSI quartile, with the proportional hazard assumption being subsequently validated. A comprehensive multivariate Cox regression analysis was then conducted to examine the relationship between HSI and glycemic outcomes in subjects with IFG. The selection of covariates was informed by baseline variables exhibiting a significant correlation with HSI or glycemic outcomes, as presented in Table 1 and Supplementary Table S1. Guided by clinical expertise and supported by existing literature [18], these variables were incorporated into a multivariable regression model to adjust for potential confounding factors: Crude Model adjusted without covariates. Model I minimally adjusted by sex and age. Model II partially adjusted by age, SBP, DBP, TG, HDL-C and FPG. Model III was further adjusted for sex, LDL-C, TC, BUN, Scr, smoking status and drinking status in addition to the variables included in Model II. Cox regression analysis was utilized to calculate the hazard ratio (HR) and their corresponding 95% confidence interval (CI) for the HSI in relation to glucose conversion within the IFG population. We stratified the population into four distinct groups based on the quartiles of HSI, with the lowest quartile group (Quartile 1) serving as the reference. The P value for the trend was calculated to evaluate the dose–response relationship and to substantiate our findings.

Table 1.

Baseline characteristics of the participants stratified by HSI quartiles

HSI quartiles	Total	Quartile 1	Quartile 2	Quartile 3	Quartile 4	P value
		≤29.85	>29.85, ≤33.08	>33.08, ≤36.77	>36.77
Participants (numbers)	11,327	2832	2831	2832	2832
Age (years)	49.98 ± 14.02	49.40 ± 15.69	52.34 ± 13.88*	51.29 ± 13.22*	46.88 ± 12.46*	<0.001
Sex			*	*	*	<0.001
Male	7574 (66.87)	1708 (60.31)	1787 (63.12)	1918 (67.73)*	2161 (76.31)*
Female	3753 (33.13)	1124 (39.69)	1044 (36.88)	914 (32.27)*	671 (23.69)*
BMI (kg/m2)	24.83 ± 3.34	21.24 ± 1.84	23.97 ± 1.62*	25.77 ± 1.88*	28.35 ± 2.88*	<0.001
SBP (mmHg)	127.65 ± 17.75	123.90 ± 17.90	127.26 ± 18.10*	128.67 ± 17.45*	130.75 ± 16.83*	<0.001
DBP (mmHg)	78.46 ± 11.20	75.37 ± 10.78	77.84 ± 11.22*	79.26 ± 11.06*	81.35 ± 10.88*	<0.001
FPG (mmol/L)	5.96 ± 0.33	5.91 ± 0.30	5.94 ± 0.32*	5.97 ± 0.34*	6.01 ± 0.34*	<0.001
TC (mmol/L)	4.99 ± 0.95	4.80 ± 0.91	4.99 ± 0.96*	5.04 ± 0.93*	5.13 ± 0.97*	<0.001
HDL-C (mmol/L)	1.34 ± 0.30	1.44 ± 0.29	1.36 ± 0.33*	1.31 ± 0.27*	1.25 ± 0.27*	<0.001
LDL-C (mmol/L)	2.89 ± 0.71	2.76 ± 0.67	2.90 ± 0.71*	2.92 ± 0.69*	2.98 ± 0.73*	<0.001
BUN (mmol/L)	5.04 ± 1.26	4.95 ± 1.31	5.04 ± 1.23*	5.14 ± 1.30*	5.02 ± 1.20*	<0.001
Scr (μmol/L)	74.07 ± 16.20	72.57 ± 16.41	73.75 ± 16.26*	74.43 ± 16.45*	75.53 ± 15.54*	<0.001
ALT (U/L)	21.80 (15.10, 32.65)	14.10 (11.10, 18.00)	19.00 (14.90, 24.00)*	25.00 (19.00, 32.32)*	40.00 (29.00, 59.00)*	<0.001
AST (U/L)	24.00 (20.00, 29.40)	22.00 (19.00, 26.00)	23.00 (19.25, 27.00)*	24.00 (20.00, 29.13)*	28.00 (23.00, 36.00)*	<0.001
TG (mmol/L)	1.40 (0.96, 2.13)	1.00 (0.71, 1.41)	1.31 (0.92, 1.98)*	1.56 (1.09, 2.30)*	1.88 (1.30, 2.73)*	<0.001
Smoking status				*	*	<0.001
Current smoker	777 ( 6.86)	179 (6.32)	169 (5.97)	194 (6.85)	235 (8.3)
Ever smoker	165 ( 1.46)	42 (1.48)	39 (1.38)	44 (1.55)	40 (1.41)
Never smoker	2501 (22.08)	716 (25.28)	639 (22.57)	613 (21.65)	533 (18.82)
Not recorded	7884 (69.60)	1895 (66.91)	1984 (70.08)	1981 (69.95)	2024 (71.47)
Drinking status				*	*	<0.001
Current drinker	165 ( 1.46)	42 (1.48)	39 (1.38)	47 (1.66)	37 (1.31)
Ever drinker	660 ( 5.83)	151 (5.33)	145 (5.12)	182 (6.43)	182 (6.43)
Never drinker	2618 (23.11)	744 (26.27)	663 (23.42)	622 (21.96)	589 (20.8)
Not recorded	7884 (69.60)	1895 (66.91)	1984 (70.08)	1981 (69.95)	2024 (71.47)
Family history of diabetes			*		*	0.005
No	11,060 (97.64)	2783 (98.27)	2756 (97.35)	2775 (97.99)	2746 (96.96)
Yes	267 (2.36)	49 (1.73)	75 (2.65)	57 (2.01)	86 (3.04)*
Outcomes (n)			*	*	*	<0.001
Normoglycemia	4729 (41.75)	1488 (52.54)	1202 (42.46)	1074 (37.92)	965 (34.07)
Prediabetes	5281 (46.62)	1189 (41.98)	1378 (48.68)	1362 (48.09)	1352 (47.74)
Diabetes	1317 (11.63)	155 (5.47)	251 (8.87)	396 (13.98)	515 (18.19)

Continuous variables were summarized using mean ± SD or median (quartile 1, quartile 3), while categorical variables were expressed as n (%). An asterisk (*) indicated values that showed statistically significant differences compared to Quartile 1

We used a restricted cubic spline model to examine potential non-linear relationships between HSI and glycemic progression outcomes in the IFG population. HSI was treated as a continuous variable with four knots (5th, 35th, 65th, and 95th). Non-linearity was assessed using a likelihood ratio test to compare linear and cubic spline terms. Based on the resulting curve, we employed a two-piecewise linear regression model to identify any threshold effects, while adjusting for potential confounding variables.

To verified the robustness of our study, we performed sensitivity analyses to assess the effect of excluding participants with certain characteristics, such as those who smoke, consume alcohol, have a family history of diabetes, or have triglyceride levels ≥1.7 mmol/L [25]. We also calculated E-values to evaluate the potential influence of unmeasured confounders on our results. Furthermore, we utilized the Cox proportional hazards model to analyze subgroups defined by clinical thresholds, converting variables such as age (<50 years, ≥50 years), BMI [18] (<24 kg/m², ≥24 kg/m²), and blood pressure (BP) [26] (normal BP: SBP < 120 mmHg and DBP < 80 mmHg) into categorical variables. Each stratum was fully adjusted for covariates beyond the stratification variables. Subgroup interactions were confirmed through a likelihood ratio test (P ≤ 0.05).

All results are presented by the STROBE statement. In our study, there was no initial assessment of statistical power, as the sample size was determined entirely based on available data. All statistical analyses were conducted utilizing the free statistical software version 1.9 (http://www.clinicalscientists.cn/freestatistics), which was developed based on R programming language version 4.2.2 (http://www.R-project.org).

Results

Baseline characteristics of participants

Our study included 11,327 IFG participants (Table 1). The HSI ranged from 18.88 to 59.49 with a mean of 33.57. After a median follow-up period of 2.85 years (IQR, 2.09–3.83), 4729 (41.75%) participants achieved normoglycemia, while 1317 (11.63%) progressed to diabetes. Compared with the Quartile 1 groups, as the HSI quartiles ascended, there was a notable upward trend in the metrics of BMI, BP, FPG, Scr and lipid profiles among the participants. In stark contrast, HDL-C levels demonstrated a downward trajectory. Compared to the other three quartiles, Quartile 4 has the highest proportion of current smokers, as well as a higher percentage of males. In the analysis of glycemic outcomes, Quartile 1 demonstrated a notably higher proportion of participants (52.54%) who regained normoglycemia. While the incidence of diabetes progression was most pronounced in Quartile 4, affecting 18.19% of the participants. The baseline characteristics were also categorized by regression and progression status of IFG (Supplementary Table S1). Participants who experienced regression to normoglycemia exhibited notably lower levels of age, BMI, blood pressure, FPG, HSI, TC, TG, BUN and Scr. They also had significantly higher levels of HDL-C than participants with persistent IFG or diabetes.

The incidence rate of glucose status conversion

Supplementary Table S2 shows that participants in the lowest HSI quartile (Quartile 1) had the highest rate of achieving normoglycemia and the lowest rate of diabetes progression. The incidence rates were 52.54% for normoglycemia in Quartile 1 and 18.19% for diabetes in Quartile 4. A statistically significant trend (P for trend <0.001) was observed, suggesting that higher HSI quartiles were associated with a decreased likelihood of normoglycemia and an increased risk of diabetes. The Kaplan–Meier curves (Supplementary Fig. S1) also exhibited the same trend (Log-rank test P < 0.001), further substantiating the close association between HSI and the glycemic outcomes in participants with IFG.

The association between HSI and glucose status conversion

We employed four Cox regression models to evaluate the association between the HSI and the conversion of glycemic status in individuals with IFG, as shown in Table 2. Our analysis revealed a negative association between HSI and the likelihood of regression to normoglycemia, along with a positive association with the progression to diabetes. These relationships were consistently observed even after full adjustment for potential confounders. In Model III, for every one SD increased in the HSI, the likelihood of an individual with IFG reverting to normoglycemia was reduced by 7% (HR = 0.93, 95% CI: 0.90–0.96; P < 0.001). Conversely, the risk of progressing to diabetes increased by 49% (HR = 1.49, 95% CI: 1.40–1.58; P < 0.001). Furthermore, we assessed the relationship between HSI quartiles and the conversion of glycemic status among the IFG population across all models. The results indicated that as the HSI quartiles increase, there was a progressively stronger negative association with the regression to normoglycemia and a correspondingly stronger positive association with the progression to diabetes (P for trend <0.05). The results of the multivariable regression analysis were further validated using the original dataset without addressing missing values, confirming that the relationship remains robust (Supplementary Table S3).

Table 2.

Relationship between HSI and glucose conversion outcome among IFG populations in different models

Variable	Crude model		Model I		Model II		Model III
	HR (95%CI)	P value	HR (95%CI)	P value	HR (95%CI)	P value	HR (95%CI)	P value
Prediabetes to normoglycemia
HSI (per SD increase)	0.84 (0.82–0.87)	<0.001	0.84 (0.81–0.86)	<0 .001	0.92 (0.89–0.96)	<0.001	0.93 (0.90–0.96)	<0.001
HSI (quartiles)
Quartile 1	1 (Ref)		1 (Ref)		1 (Ref)		1 (Ref)
Quartile 2	0.75 (0.70–0.81)	<0.001	0.83 (0.77–0.89)	<0.001	0.91 (0.84–0.99)	0.026	0.90 (0.83–0.98)	0.014
Quartile 3	0.70 (0.65–0.76)	<0.001	0.76 (0.70–0.82)	<0.001	0.91 (0.83–0.99)	0.032	0.90 (0.82–1.00)	0.046
Quartile 4	0.63 (0.58–0.68)	<0.001	0.62 (0.57–0.68)	<0.001	0.82 (0.74–0.90)	<0.001	0.82 (0.73–0.91)	0.001
P for trend		<0.001		<0.001		<0.001		<0.001
Prediabetes to diabetes
HSI (per SD increase)	1.47 (1.40–1.54)	<0.001	1.61 (1.53–1.70)	<0.001	1.48 (1.39–1.56)	<0.001	1.49 (1.40–1.58)	<0.001
HSI (quartiles)
Quartile 1	1 (Ref)		1 (Ref)		1 (Ref)		1 (Ref)
Quartile 2	1.47 (1.20–1.79)	<0.001	1.40 (1.14–1.71)	0.001	1.38 (1.13–1.69)	0.002	1.34 (1.09–1.66)	0.007
Quartile 3	2.48 (2.06–2.98)	<0.001	2.46 (2.04–2.96)	<0.001	2.18 (1.80–2.65)	<0.001	2.13 (1.75–2.60)	<0.001
Quartile 4	3.14 (2.63–3.76)	<0.001	3.52 (2.94–4.22)	<0.001	2.81 (2.32–3.40)	<0.001	2.79 (2.28–3.40)	<0.001
P for trend		<0.001		<0.001		<0.001		<0.001

The Crude Model was not adjusted for any variables. Model I was adjusted for age and sex. Model II was adjusted for age, SBP, DBP, TG, HDL-C and FPG. Model III was further adjusted for sex, LDL-C, TC, BUN, Scr, smoking status, and drinking status in addition to the variables included in Model II

Curve fitting and inflection point analysis

To more precisely elucidate the link between HSI and glucose status transition, we undertook a curve fitting analysis. The results, as illustrated in Fig. 2, indicated that after adjusting for potential confounding factors, HSI exhibited a linear inverse relationship with the likelihood of normoglycemia restoration (P for nonlinearity >0.05), and a non-linear relationship with the advancement to diabetes (P for nonlinearity <0.05). For the non-linear relationship, an inflection point analysis was performed. As showed in Table 3, our analysis disclosed that within the HSI range of 26.55–40.74, there existed a positive association between HSI and the risk of IFG transitioning to diabetes (HR: 1.09, 95% CI: 1.07–1.11, P < 0.001). Beyond an HSI value of 40.74, this association was significantly attenuated, and the relationship between the HSI and the advancement to diabetes from IFG was not significant (HR: 0.99, 95% CI: 0.95–1.05, P > 0.05).

Fig. 2.

Association between between HSI and A reversion to normoglycemia B progression to diabetes among IFG populations. The model adjusted for age, sex, SBP, DBP, TG, LDL-C, HDL-C, BUN, Scr, FPG, smoking status. The light blue histograms illustrated the percentage density distribution of HSI across the study’s participant group. The bold central lines denote the estimated adjusted hazard ratios, with the surrounding shaded areas representing the 95% confidence intervals for these estimates. The horizontal dotted lines serve as a benchmark, indicating a reference HR value of 1.0. The highest 0.1% of HSI measure was trimmed

Table 3.

Analysis of the threshold effect of HSI on the progression to diabetes in IFG individuals

The inflection point of HSI	HR (95% CI)	P value
<26.55	1.02 (0.76–1.36)	0.899
26.55–40.74	1.09 (1.07–1.11)	<0.001
>40.74	0.99 (0.95–1.05)	0.898
Likelihood ratio test		0.043

The model adjusted for age, sex, FPG, LDL-C, HDL-C, TC, TG, BUN, Scr, SBP, DBP, smoking status. The highest 0.1% of HSI measure was trimmed

Sensitivity analyses

In the sensitivity analyses, we conducted comprehensive sensitivity analyses across diverse subpopulations: non-smokers (n = 2501), individuals without a family history of diabetes (n = 11,060), non-drinker (n = 2618), and those with TG levels below 1.7 mmol/L (n = 7017). Throughout our analyses, a consistent inverse association was observed between HSI and the likelihood of reverting from prediabetes, alongside a positive association with the risk of diabetes progression. High HSI values in the IFG population were associated with a reduced likelihood of restoring normal blood glucose levels, regardless of smoking and alcohol abstinence, maintaining triglycerides below 1.7 mmol/L, or lacking a family history of diabetes. Conversely, these individuals faced an increased risk of diabetes. Treating HSI as a categorical variable, we found that individuals in higher quartiles had a significantly reduced chance of transitioning from IFG to normal glucose levels compared to the first quartile, with diabetes progression risk escalating incrementally with higher quartile ranks (Supplementary Table S4).

Additionally, to assess the sensitivity of our findings to potential unobserved confounding variables, we calculated E-values. The resulting E-value for incident prediabetes reversal was 1.28 and for diabetes progression was 2.34 (Supplementary Fig. S2). In our research, an E-value of 1.28 for the reversal of prediabetes suggests that an unmeasured confounding factor with a risk ratio of at least 1.28 would be needed to significantly impact the study’s findings. Likewise, for the progression to diabetes, the presence of an unmeasured confounding factor with a risk ratio of at least 2.34 would be essential to alter the results of our study.

The consequence of subgroup analysis

Subgroup analysis examined whether age, sex, BMI, BP and family history of diabetes modified the association between HSI and glucose status conversion. Across all strata defined by these factors, IFG reversion to normoglycemia was consistently negatively associated with HSI. No significant interactions were observed in the stratified analyses. In the subgroup with a BMI below 24 kg/m², there was no statistical significance for association between HSI and reversion to normoglycemia. Notably, we identified significant interactions between age, BP, BMI, and HSI with respect to the progression from prediabetes to diabetes. Individuals under the age of 50 with IFG demonstrated a more pronounced positive association between HSI and the risk of progressing to diabetes compared to those aged 50 and above (HR: 1.67 vs. 1.30, P for interaction <0.001). This trend was also observed in individuals with normal blood pressure, who exhibited a stronger association than those with abnormal blood pressure (HR: 1.74 vs. 1.41, P for interaction = 0.023). Additionally, there was a stronger link between HSI and diabetes progression in individuals with a BMI below 24 kg/m² compared to those with a BMI of 24 kg/m² or higher (HR: 1.85 vs. 1.42, P for interaction = 0.015). In the subgroup with a family history of diabetes, there was no statistically significant association between HSI and glucose conversion outcome among IFG populations (P > 0.05). The multivariate-adjusted HRs with 95% CIs were presented as a forest plot (Fig. 3).

Fig. 3.

Subgroup analysis of association between HSI and A reversion to normoglycemia B progression to diabetes among IFG populations. Models adjusted for age, sex, SBP, DBP, TG, FPG, LDL-C, HDL-C, BUN, Scr, FPG, smoking status, drinking status, if not stratified

Discussion

Our study aimed to examine the relationship between transitions in glucose status and the HSI in individuals with IFG. The findings indicated a linear negative association between HSI and the normalization of glucose status from IFG, whereas a nonlinear positive association was observed with the progression to diabetes. Sensitivity analyses confirmed the robustness of these results. Specifically, within the HSI range of 26.55–40.74, each unit increase in HSI corresponded to a 9% higher risk of developing diabetes. Beyond an HSI of 40.74, the estimated dose–response curve gradually plateaued. Furthermore, significant interactions were identified between HSI and diabetes progression across various subgroups stratified by age, BP, and BMI. Notably, the association between HSI and the progression from prediabetes to diabetes was more pronounced in individuals under 50 years of age, with normal BP, and a BMI below 24 kg/m², compared to those over 50 years of age, with elevated BP, or a BMI exceeding 24 kg/m².

Diabetes is a metabolic disease typically caused by insulin resistance or insufficient insulin production. Previous studies have established a close relationship between NAFLD and diabetes [27, 28]. Prediabetes is a significant stage in the development of diabetes, as it often precedes the onset of diabetes [27, 29]. Most studies have traditionally focused on the relationship between conventional markers and diabetes within the general population, with relatively less attention paid to individuals with prediabetes and the factors that influence the normalization of blood glucose levels [20, 30, 31]. Our study serves to complement existing research. HSI, which is calculated based on ALT, AST, and BMI, incorporates the impact of sex, making it a more comprehensive and systematic score compared to traditional indicators. The ALT/AST ratio and BMI have been considered closely related to glucose metabolism [20, 32–34]. Among them, an increase in BMI was widely recognized as a risk factor for the development of diabetes [34–36]. As reported by Han et al., BMI was negatively related to normoglycemia regression in Chinese people with IFG [29]. In a study conducted among Japanese, it was discovered that the AST/ALT ratio was significantly negatively associated with the risk of T2DM in male NAFLD patients [37]. Xie et al. also found that in a non-diabetic population, that the lower the AST/ALT ratio, the higher the risk of developing diabetes [33]. Similarly, AN and Song et al. consistently found that AST/ALT levels in early pregnancy were negatively associated with the risk of developing gestational diabetes [38, 39]. These findings collectively indicated that a high ALT/AST ratio was a potential risk factor for diabetes. However, the impact of their combined metric, the HSI, on glucose metabolism has not been extensively studied. Cai et al. were the first to report that in a non-diabetic Chinese population, higher HSI levels were associated with an increased risk of developing type 2 diabetes [18]. In our study, conducted among Chinese IFG individuals, we investigated the association between the HSI and glycemic outcomes, which included the reversal to normal blood glucose levels and the progression to diabetes. The findings revealed that changes in HSI had a relatively weak effect on the ability to restore impaired fasting glucose to normal levels; for every one SD increase in HSI, the likelihood of normalizing blood glucose levels decreased by only 7%. Under the same conditions, for every one SD increase in HSI, the risk of progressing to diabetes increased by 49%. The reasons for this significant disparity are not well understood and may be due to the more complex factors and biological mechanisms involved in reversing abnormal blood glucose levels to normal. While the present investigation revealed an inverse relationship between HSI and the reversion from IFG, the veracity of whether reduction HSI can genuinely facilitates the reversion to a normoglycemic state from prediabetes was yet to be elucidated. It is of particular interest to highlight that lifestyle modifications have been unequivocally identified as efficacious in facilitating the reversion of prediabetes [40, 41]. In the context of lifestyle interventions, HSI may be employed as a prospective biomarker, aiding in the appraisal of the probability of a favorable glycemic normalization.

We conducted a retrospective cohort analysis utilizing historical data to explore the relationship between HSI and the progression of glycemic status among people with IFG. While this observational study design did not permit direct causal inferences, we propose plausible biological pathways that may explain the observed relationship between HSI and changes in glycemic status. NAFLD may contribute to the dysregulation of glucose metabolism through various pathways, including insulin resistance, oxidative stress, lipid metabolism disorders, and increased hepatic gluconeogenesis [42]. Drawing upon prior literature, mouse models of fatty liver disease have linked NAFLD to diabetes development through insulin resistance and lipid metabolism abnormalities [43]. Epidemiology and pathology have indeed highlighted the connection between NAFLD and diabetes [44]. As a laboratory marker for NAFLD, a robust association has been observed between HSI and lipid metabolism risk indicators in previous studies [45, 46]. Simultaneously, HSI has been associated with β-cell dysfunction and insulin resistance in individuals without diabetes [17]. Furthermore, NAFLD development involves activated inflammatory cytokines, which can affect insulin signaling through multiple complex pathways [47]. These underlying mechanisms indeed help us understand the connection between the HSI and the progression of prediabetes from a pathophysiological perspective.

Our study revealed that across various age groups, sexes, blood pressure levels, and BMI categories, the association between the HSI and the normalization of IFG was consistently negatively associated, with no interaction effects observed among different subgroups. Regarding the progression of IFG to diabetes, in individuals with a BMI less than 24 kg/m², normal blood pressure, and under the age of 50, an increase in HSI was found to have a relatively greater effect on the risk of diabetes compared to its effect in the corresponding subgroups. In individuals with a BMI of less than 24 kg/m², the association between HSI and the normalization of blood glucose levels did not reach statistical significance. This finding implied that an elevation in HSI may exert minimal influence on the restoration of blood glucose levels within this demographic. Furthermore, among individuals possessing a family history of diabetes, the relationship between HSI and glycemic outcomes in those with IFG was similarly not statistically significant. This lack of significance may be attributed to the reduced sample size, potentially introducing statistical bias. Cai et al.’s study in a non-diabetic population also found similar results, where the impact of increased HSI on the risk of diabetes development was more pronounced in individuals under the age of 40 compared to those over 40 (HR: 2.17 vs. 1.47) [18]. At present, there is no consensus on the impact of HSI in assessing the risk of diabetes across different populations. However, we have formulated some hypotheses regarding this issue. In traditional perspectives, advanced age and obesity are commonly recognized as factors that increase the risk of metabolic diseases such as diabetes. However, recent studies have revealed that, akin to the HSI, changes in the AIP have a more significant impact on blood glucose variability in relatively younger populations in China [31, 48]. In individuals with lower BMI or lower SBP, the increase in AIP has a more pronounced effect on the risk of developing prediabetes compared to those with higher BMI (HR: 1.76 vs. 1.23) or higher SBP (HR: 1.43 vs. 1.18) [31]. In our study, compared to individuals over 50 years old, those under 50 had significantly lower average body mass index (BMI: 24.65 kg/m² vs. 25.01 kg/m²) and blood pressure (SBP: 122.99 mmHg vs. 132.29 mmHg; DBP: 76.93 mmHg vs. 79.97 mmHg). In this younger demographic with relatively normal BMI and blood pressure, they had fewer underlying diseases, and therefore, changes in metabolic indices such as HSI or AIP might have a more pronounced effect in this population. The reasons for this phenomenon may be related to China’s social context and relevant policies. Since the early 1990s, China’s fertility rate has been on a declining trend [49]. The phenomenon was not only accompanied by a lack of labor supply but also triggered issues related to an aging population. In China, the relatively young demographic bears significant work and life pressures. Unhealthy dietary habits, such as frequent takeout consumption and staying up late, can also increase the risk of metabolic diseases like diabetes and cardiovascular diseases in this population [50]. This has also been corroborated in animal models, where exposure to social stress predisposes mice to develop insulin resistance induced by high-fat diet feeding [51]. Therefore, managing diabetes health in younger populations with normal blood pressure and body mass index is equally significant. It is not exclusive to high-risk groups. However, since the precise mechanisms underlying these results are not yet clear, we have interpreted the interaction analysis results with caution. We hope that future large-scale prospective cohort studies will further validate these findings.

The study’s strengths are as follows: firstly, this was a comprehensive multicenter investigation from China with a significant sample size. Secondly, sensitivity analyses have been carried out to assess the robustness of the results. To enhance statistical accuracy and minimize potential biases from missing covariate data, multiple interpolations were employed. The HSI was utilized as continuous and categorical variables to explore the relationship with glucose status transition in our study. This approach minimized information loss and enabled the quantification of their relationship with the outcome. Lastly, we utilized FPG as the primary criterion for identifying prediabetes and diabetes. This approach ensured that the study’s findings are widely applicable in clinical settings.

Nevertheless, several limitations should be noted: Firstly, we did not identify individuals with impaired glucose tolerance in this investigation, potentially underestimating the population with prediabetes at baseline. To enhance the validity of our findings, we will gather data on hemoglobin A1c and the 2-h oral glucose tolerance test in the future. Secondly, we could not account for uncontrolled and unmeasured confounders in this study due to the use of previously published data. However, we estimated the potential impact of these confounders by calculating the E-value. Given that this risk ratio (1.28) was much higher than any observed for the known risk factors for prediabetes reversal examined in the current study, such as FLI [52], urinary albumin [53], and BMI [54], unmeasured confounders were unlikely to overcome the prediabetes reversal effect observed in our study. Similarly, the presence of an unmeasured confounding factor with a risk ratio greater than 2.34 could potentially invalidate the research hypothesis that links HIS to the progression from IFG to diabetes within the scope of our study. Thirdly, due to substantial missing data in our study’s original dataset, we used simple and multiple imputation methods to address this issue. Although these methods can introduce bias, ignoring missing data could result in similar biases [23]. To evaluate the potential bias introduced by multiple imputation, we performed sensitivity analyses using the original dataset to assess the accuracy of the results derived from this approach. To enhance reliability and reduce bias, we plan to use a prospective study design with more confounder adjustments and better variable control. Lastly, our retrospective cohort study indicated a potential association between HIS and glucose conversion outcome among IFG populations, although it did not establish a causal relationship. The utilization of Kaplan–Meier curves and hazard ratios provided significant insights into survival times and associated event risks. It is crucial to acknowledge that retrospective studies may be subject to unmeasured or uncontrolled confounding variables, which could affect the observed relationship between HIS and glucose conversion outcome, thereby posing a risk of misinterpretation. Future research should aim to investigate this causal relationship through prospective cohort study designs.

Conclusion

Our investigation has delineated a significant association between the HSI and glucose status conversion in prediabetic states. In addition, we have conducted a meticulous quantification of the propensity for normoglycemia restoration and the escalated risk of advancing to diabetes mellitus within the cohort of individuals exhibiting IFG in the Chinese population. These findings will provide clinicians a valuable reference for assessing prediabetes reversal, enabling earlier intervention and treatment of diabetes. Further research and dedicated efforts are still required to improve interventions in IFG populations.

Abbreviations

HSI

Hepatic steatosis index

IFG

Impaired fasting glucose

HR

Hazard ratio

CI

Confidence interval

SD

Standard deviation

Ref

Reference

T2DM

Type 2 diabetes mellitus

NAFLD

Nonalcoholic fatty liver disease

FLI

Fatty liver index

FPG

Fasting plasma glucose

BP

Blood pressure

SBP

Systolic blood pressure

DBP

Diastolic blood pressure

HDL-C

High-density lipoprotein cholesterol

TC

Total cholesterol

LDL-C

Low-density lipoprotein cholesterol

BUN

Blood urea nitrogen

TG

Triglycerides

AST

Aspartate aminotransferase

ALT

Alanine aminotransferase

Scr

Serum creatinine

BMI

Body mass index

Author contributions

JW and XDH contributed to the initial design of the study. JZZ were primarily responsible for assembling and analyzing the data, ensuring its statistical validity. YYC, YYL, and HHC provided valuable insights in data interpretation. XDH oversaw all aspects of the study, ensuring the integrity and accuracy of the data analysis. The final manuscript was approved by all contributing authors.

Funding

The authors have no relevant financial or non-financial interests to disclose.

Data availability

The dataset can be downloaded from the official website of the Dryad database (https://datadryad.org/stash/dataset/doi:10.5061/dryad.ft8750v).

Declarations

Ethics approval and consent to participate

Our study utilized data from an existing database. Ethical approval for the use of this database was obtained from the Rich Healthcare Group Review Board, following the guidelines outlined by the Declaration of Helsinki. The Research Board approved the use of the database for the current retrospective study and waived the requirement for informed consent because the study involved the use of preexisting de-identified data.

Competing interests

The authors declare no competing interests.