Study participants and design

This study used data from the Beijing Health Management Cohort (BHMC) collected from January 2016 to December 2021, with annual follow-up including questionnaires, physical examinations, and laboratory tests. The study was approved by the Ethics Committee of Capital Medical University (No. 2019SY088), and all participants provided informed consent.

Inclusion criteria: (1) willingness to participate in the project study and sign the informed consent form. Cooperation in completing survey questionnaires, undergoing physical examinations, providing blood samples, and obtaining a unique identification code; (2) age ≥ 18 years; (3) at least two follow-up data points, including comprehensive information on NAFLD-related indicators.

Exclusion criteria: (1) excessive alcohol consumption: ≥ 140 g/week for males and ≥ 70g/week for females; (2) self-reported viral hepatitis, history of cancer, or autoimmune diseases; (3) long-term medication use; (4) lack of important indicators such as BMI, height, and weight data; (5) lack of follow-up; (6) individuals who developed NAFLD during follow-up but whose BMI remained < 24 kg/m² throughout the study period were excluded, in order to avoid potential bias from lean NAFLD trajectories that may involve distinct etiological mechanisms unrelated to metabolic factors. The study inclusion flow chart is shown in Figure 1 [27].

Display full size

Figure 1. Flow chart of participant selection in this study. NAFLD: non-alcoholic fatty liver disease; BMI: body mass index. Adapted from [27]. © Author(s) 2021. CC BY 4.0.

Participants with severe cognitive impairment, inability to complete the questionnaire independently, or major health conditions affecting liver health (such as advanced malignancy or severe organ failure) were excluded.

Data collection

Standardized self-report questionnaires collected information on demographics, health behaviors (smoking, alcohol consumption, dietary habits, exercise), medication use, and disease history. Long-term medication was defined as daily use for more than 6 months.

Smoking: never, former, current. Alcohol consumption: none or moderate (≤ 20 g/day for men; ≤ 10 g/day for women). Dietary habits: balanced, vegetable-based, meat-based, vegetarian-based. Exercise: never, occasional, active.

Anthropometric measurements included height, weight, waist circumference (WC), systolic and diastolic blood pressure (SBP, DBP). BMI was calculated as weight (kg)/height² (m²).

Laboratory tests were performed after a 12 h fast and included total cholesterol, triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), fasting plasma glucose (FPG), aspartate aminotransferase (AST), alanine aminotransferase (ALT), and serum uric acid (SUA), measured using a Beckman LX20 chemistry analyzer. Hepatic ultrasound examinations were performed by experienced radiologists.

Quality control: participants were selected according to predefined eligibility criteria, with ineligible individuals excluded. Baseline and follow-up surveys were conducted by professionals. Double data entry and checking were performed to ensure data accuracy. Ultrasonographic examinations were conducted by experienced physicians following standard procedures to minimize variability. Participants were censored at their last available follow-up if lost to follow-up or at the end of the study period. Since death or severe liver outcomes (cirrhosis/HCC) were not systematically recorded in this routine health check cohort, these potential absorbing states were not included in the model. We acknowledge that the absence of absorbing states may result in potential informative censoring.

Diagnostic criteria

NAFLD was diagnosed based on the following criteria: (1) enhanced echogenicity in the near field and attenuated echogenicity in the far field of the liver parenchyma; (2) presence of intra-hepatic ducts that are relatively obscure in structure and rounded liver margins with hepatomegaly; (3) alcohol consumption of 140 g/week for men and 70 g/week for women; and (4) absence of other diseases and conditions that may cause fatty liver disease, such as viral hepatitis, total parenteral nutrition, hepatomegaly, and drug-related liver disease [28].

Diagnosis of hyperuricemia (HUA): refer to the China HUA-related disease diagnosis and treatment of multidisciplinary expert consensus, male fasting SUA > 420 μmol/L, female fasting SUA > 360 μmol/L diagnosed as HUA [29].

According to the Chinese guidelines recommended by the National Health Commission of China, adult BMI categories differ from the WHO classification. For Chinese adults, BMI < 18.5 kg/m² is classified as underweight, 18.5–23.9 kg/m² as normal weight, 24–27.9 kg/m² as overweight, and a BMI ≥ 28 kg/m² as obese [18]. This classification is based on evidence showing that Asian populations may experience obesity-related health risks at lower BMI levels compared to Western populations.

This study classifies NAFLD patients into three categories based on BMI:

• S1 (healthy state): This state represents healthy individuals without NAFLD, serving as the baseline control group.

• S2 (NAFLD, BMI < 18.5 kg/m² or 18.5–23.9 kg/m²): This group includes NAFLD patients with a BMI below 18.5 kg/m² (underweight) or within the range of 18.5–23.9 kg/m² (normal weight).

• S3 (NAFLD, BMI ≥ 24 kg/m²): This group includes NAFLD patients with a BMI between 24–27.9 kg/m² (overweight) and BMI ≥ 28 kg/m² (obesity).

Underweight NAFLD may represent a distinct ‘lean NAFLD’ phenotype with potential genetic or non-metabolic etiologies. However, in our cohort, the proportion of underweight individuals in S2 was extremely low (n = 5 out of 1,139 at baseline, 0.4%). Given this negligible fraction, merging them with normal-weight NAFLD is unlikely to materially affect transition intensity estimates or covariate effects. Therefore, we retained the combined definition of S2.

Model introduction

A multistate Markov model can effectively process longitudinal follow-up data. The disease process is divided into several states. According to the observation interval, the number and frequency of transitions between states are calculated, and the probability transition matrix between states is determined. Based on the probability transition matrix, the state starting from different states after several years is predicted.

The transition intensity is an index describing the instantaneous transition risk between states. The probability of an individual in state “H” before time “t” transferring to state “J” at time “t + 1”. Transition probability is a key indicator of the Multistate Markov model for long-term prediction. The probability of starting from a state and passing through a transition to another state is called the transition probability. Sojourn time refers to the length of time that the research object is in a transient state [30].

The fitted Multistate Markov model in this study contains three states: healthy state (S1), normal weight NAFLD state (S2), and overweight or obesity NAFLD state (S3). According to previous studies, NAFLD is reversible in the early stage. Liver steatosis can be eliminated by lifestyle changes such as exercise or drug regulation, and abdominal ultrasound results can be restored to normal. Therefore, the three states involved in this study are transient; this study does not set an absorption state. The transition between states is shown in Figure 2, which shows the probability of transfer between states in the total population after 1 year.

Display full size

Figure 2. Structure of the multi-state Markov model showing transitions between healthy state (S1), NAFLD with BMI < 24 kg/m² (S2), and NAFLD with BMI ≥ 24 kg/m² (S3). NAFLD: non-alcoholic fatty liver disease; BMI: body mass index.

Statistical methods

Data management and statistical analyses were performed using R software (version 4.2; R Project for Statistical Computing, Vienna, Austria, http://www.R-project.org/). Continuous variables were expressed as mean ± standard deviation (SD) and compared between groups using ANOVA. Categorical variables were expressed as numbers (percentages) and compared using the chi-square test. A two-sided P < 0.05 was considered statistically significant.

A multi-state Markov model was constructed using the msm package in R. The model included three states: S1 (healthy state, reference group), S2 (NAFLD with BMI < 24 kg/m²), and S3 (NAFLD with BMI ≥ 24 kg/m²). The Q-matrix allowed all possible transitions between the three states (S1↔S2, S1↔S3, and S2↔S3). Transition intensities were modeled as functions of covariates (age, sex, BMI, and metabolic factors). Time was treated as interval censored according to follow-up visits. Model optimization was performed using maximum likelihood estimation with the default convergence settings of the msm package. Given the sex-specific incidence of NAFLD, subgroup analyses were further conducted separately for men and women.

Baseline characteristics of the study population

A total of 24,893 participants were included, with 17,906 (71.9%) in the healthy group (S1), 1,139 (4.6%) in the NAFLD group with BMI < 24 kg/m² (S2), and 5,848 (23.5%) in the NAFLD group with BMI ≥ 24 kg/m² (S3). Compared with S1, participants in S2 and S3 were generally older, had a higher proportion of males, and showed markedly higher prevalence of metabolic abnormalities, including dyslipidemia, hypertension, hyperglycemia, elevated liver enzymes, HUA, and abdominal obesity (all P < 0.001). Detailed characteristics are presented in Table 1.

Transition frequencies between different NAFLD states: overall and gender-stratified analysis

In the study population, most individuals remained in their original state during follow-up. The S1 group exhibited the highest stability, with 89.61% remaining in S1, while only 2.77% progressed to S2 and 7.61% to S3. The S2 group showed greater variability, with 48.36% remaining in S2, 28.02% reverting to S1, and 23.61% progressing to S3. The S3 group was relatively stable, with 77.30% staying in S3, 18.98% reverting to S1, and 3.72% transitioning to S2.

There were notable differences in state transitions between genders. A higher proportion of females in the S1 group remained healthy compared to males (93.76% vs. 83.52%), whereas males in S1 were more likely to progress to S3 (12.59% vs. 4.23%). Additionally, females in the S3 group were more likely to revert to S1 compared to males (21.33% vs. 18.18%). As shown in Table 2.

Average residence time in different NAFLD states: overall and gender-stratified analysis

The average residence time in S1 for the total population was 11.12 years (95% CI: 10.77–11.47). For males, it was 6.97 years (95% CI: 6.69–7.25), while females had a significantly longer average of 18.58 years (95% CI: 17.61–19.55). In S2, the average residence time for the total population was 2.22 years (95% CI: 2.11–2.34). Males had an average of 2.20 years (95% CI: 2.06–2.35), and females had a slightly longer average of 2.26 years (95% CI: 2.08–2.46). For S3, the average residence time for the total population was 5.01 years (95% CI: 4.83–5.19). Males had an average of 5.37 years (95% CI: 5.14–5.60), while females had a shorter average of 4.18 years (95% CI: 3.91–4.46). As shown in Table 3.

Influence factors of the transition process between different NAFLD states

Risk prediction of the Markov model

This study employed a multi-state Markov model to predict the transition probabilities between S1, S2, and S3, focusing on the transition trends over 1, 3, and 5 years. The results showed that in the male population, the probabilities of transitioning from S1 to S2 were 4.1%, 8.1%, and 9.4% at 1, 3, and 5 years, respectively, showing a gradual increase. The probabilities of transitioning from S1 to S3 were 9.4%, 22.5%, and 30.6%, also increasing over time. Similarly, the probabilities of transitioning from S2 to S3 were 12.9%, 27.8%, and 35.0%, showing a consistent upward trend. As shown in Figure 3.

Display full size

Figure 3. Predicted transition probabilities for different states in the male cohort based on the multi-state Markov model.

In contrast, in the female population, the probabilities of transitioning from S1 to S2 were 2.9%, 5.8%, and 7.4%, showing a steady increase. The probabilities of transitioning from S1 to S3 were 2.7%, 7.4%, and 10.5%, also increasing gradually. The probabilities of transitioning from S2 to S3 in females were 7.6%, 15.1%, and 17.6%, demonstrating a consistent increase as well. As shown in Figure 4.

Display full size

Figure 4. Predicted transition probabilities for different states in the female cohort based on the multi-state Markov model.

Overall, males were more likely to transition from S1 to S3, with a faster transition rate, while females tended to remain in S1. Furthermore, although both males and females showed an increasing probability of transitioning from S2 to S3 over time, females transitioned back to S1 relatively faster.

Gender differences in NAFLD progression

Firstly, gender differences are particularly evident in NAFLD progression. It was found that NAFLD progresses faster in males, especially in the transition from S1 to S3, with males showing a much faster progression than females. In males, abdominal obesity, elevated TG, low HDL-C, and elevated FPG are major contributing factors. Abdominal obesity, as an independent risk factor for NAFLD progression, is particularly prominent in males. Visceral fat accumulation not only increases insulin resistance but also exacerbates liver injury through inflammatory responses, thus promoting the worsening of NAFLD [31]. Elevated TG and low HDL-C, representing lipid metabolism abnormalities, are also key drivers of NAFLD progression, leading to fat accumulation in the liver and further exacerbating hepatic metabolic stress [32]. In contrast, NAFLD progresses more slowly in females. Although factors like elevated TG, low HDL-C, and elevated FPG also promote NAFLD progression in females, their impact is weaker compared to males. We further performed an age-stratified analysis in females (< 50 vs. ≥ 50 years, roughly corresponding to pre-menopause and post-menopause), and the results showed that premenopausal females had the longest S1 sojourn time (21.3 years), whereas postmenopausal females had a significantly shorter S1 sojourn time of 10.2 years, with males generally showing shorter S1 durations. These findings suggest that hormonal changes and other physiological alterations after menopause may shorten the S1 duration and accelerate NAFLD progression in females. Furthermore, females exhibit lower transition probabilities from S1 to S3, which may be related to more favorable fat distribution and higher estrogen levels. Studies have shown that fat in females is primarily distributed subcutaneously, whereas males tend to accumulate fat in the visceral region [33]. This difference in fat distribution may influence the progression of NAFLD. Additionally, estrogen is considered to have a protective effect, as it can improve lipid metabolism and reduce inflammatory responses, thereby slowing the progression of NAFLD. Postmenopausal estrogen decline may lead to increased central adiposity and less favorable lipid and glucose profiles, further shortening S1 duration and increasing disease progression risk. Estrogen may inhibit fat accumulation in the liver by improving lipid metabolism and enhancing insulin sensitivity, thus delaying NAFLD progression [34, 35].

Gender differences in NAFLD recovery

Gender differences are also significant in the recovery process of NAFLD. The study suggests that females show a higher recovery probability when transitioning from S3 (advanced stage) to S1 (healthy state). This result may be linked to females’ metabolic characteristics, hormone levels, and fat distribution. Estrogen not only aids in improving lipid metabolism but may also provide protective effects against hepatic fat deposition and inflammatory responses, giving females an advantage over males in NAFLD recovery [36]. However, significant metabolic disturbances remain during recovery. Both males and females are affected by abdominal obesity, a key factor inhibiting NAFLD recovery. Abdominal fat, especially visceral fat, is not only an independent risk factor for NAFLD but also one of the strongest barriers during recovery [37]. The study also found that elevated TG, low HDL-C, and elevated FPG negatively impact NAFLD recovery in both males and females. Elevated TG and low HDL-C, important markers of lipid metabolism disorders, may impede NAFLD reversal by exacerbating hepatic fat deposition and inflammation [38, 39]. As for liver function indicators, ALT and AST play significant roles in NAFLD recovery. For males, ALT is negatively correlated, while AST shows a positive correlation with recovery from S2 to S1. ALT and AST are biomarkers of liver injury, reflecting liver function status and repair processes [40]. High ALT may indicate ongoing liver damage, hindering NAFLD recovery, while elevated AST may be associated with improved liver metabolism and repair capacity.

The role of age and metabolic abnormalities in NAFLD progression and recovery

This study further reveals the dual role of age and metabolic abnormalities (such as elevated FPG, elevated TG, low HDL-C, and abdominal obesity) in both NAFLD progression and recovery. Age, as a significant influencing factor, shows different effects in males and females regarding NAFLD progression. In males, age promotes the transition from S1 to S2; however, in females, age has a more pronounced effect, particularly in the transition from S1 to S3. Increasing age may be accompanied by a decline in metabolic capacity, thus promoting disease progression [41]. Abdominal obesity, a core risk factor for NAFLD, exerts a strong impact on both males and females [42]. In males, abdominal obesity not only accelerates the transition from S1 to S3 but also induces a series of metabolic abnormalities via visceral fat, further worsening NAFLD progression. In females, abdominal obesity is also a negative factor, with its inhibitory effect on NAFLD recovery being most evident during the transition from S3 to S2. This highlights the pivotal role of abdominal obesity in gender differences, influencing both NAFLD progression and recovery.

Recently, glucagon-like peptide-1 receptor agonists (GLP-1 RAs) have shown significant therapeutic potential in the management of type 2 diabetes and obesity. Emerging evidence suggests that GLP-1 RAs may also have beneficial effects in patients with NAFLD. For example, a systematic review and network meta-analysis reported that GLP-1 RAs significantly reduced hepatic fat content, improved liver function parameters, and positively affected body weight and insulin sensitivity in NAFLD patients [43]. Furthermore, GLP-1 RAs may exert beneficial effects through mechanisms such as improving hepatic insulin signaling, reducing oxidative stress, and modulating the gut-liver axis [44]. These findings provide a theoretical basis for exploring GLP-1 RAs in NAFLD treatment. However, the mechanisms and long-term effects of GLP-1 RAs in NAFLD remain incompletely understood, and there is a lack of Markov-state model studies evaluating this therapy in NAFLD patients. Therefore, future studies should investigate the efficacy and safety of GLP-1 RAs in NAFLD, particularly in subgroups defined by age, gender, and obesity status, to provide more robust evidence for clinical application.

Limitations and future directions

This study is primarily based on a population from Beijing, China, which may limit the generalizability of the results to other ethnic or regional populations. There may be differences in NAFLD prevalence, risk factors, and disease progression patterns across various populations and geographical regions. Although the term “NAFLD” is used throughout this study, our cohort largely corresponds to the recently proposed MASLD (Metabolic Dysfunction-Associated Steatotic Liver Disease) and MASH (Metabolic Dysfunction-Associated Steatohepatitis) criteria, which emphasize the role of metabolic dysfunction as the underlying cause. Furthermore, given that our study population is primarily from Beijing, China, caution should be taken when generalizing these findings to other regions or ethnic groups. Comparison with other East Asian cohorts suggests that the prevalence and risk factors of NAFLD/MASLD may differ due to regional lifestyle, dietary habits, and genetic backgrounds. Future studies are warranted to validate these findings in broader populations and to explore potential variations in disease progression and recovery.

Although potential confounding factors were controlled as much as possible, studies based on follow-up data cannot completely eliminate the influence of unmeasured confounders such as diet, physical activity, and genetic factors. Moreover, as the study population included individuals undergoing regular health checks and no liver biopsies were performed, the lack of invasive procedures reduces the ability to fully understand NAFLD progression, particularly in terms of directly assessing hepatic histological changes. Another limitation is that the diagnosis of NAFLD was based on ultrasonography, which is operator-dependent and has limited sensitivity for mild steatosis. Although standardized protocols and strict quality control measures were applied, some degree of non-differential misclassification was unavoidable. This may have resulted in underestimation of NAFLD incidence and attenuation of the observed associations. Finally, although the multi-state Markov model effectively captured the transition processes between NAFLD states, the assumption that transition probabilities depend solely on the current state, without considering early disease history or potential future interventions, may limit the model’s applicability in some cases. In addition, potential absorbing states such as death or severe liver outcomes (cirrhosis/HCC) were not included due to a lack of systematic data in this routine health check cohort. Participants who experienced these events would likely be lost to follow-up. Therefore, potential informative censoring may exist, which could affect the estimated transition probabilities between NAFLD states. Future studies with mortality and liver complication data could implement competing-risk models for a more comprehensive evaluation of NAFLD progression. Additionally, the study population is limited to Beijing, which may restrict external validity. Future studies should include other regions and ethnic groups in East Asia to confirm the generalizability of these findings.

Conclusion

This study reveals significant gender differences in NAFLD progression and recovery, further confirming the crucial role of gender in the disease course through the prediction of transition probabilities between different NAFLD states. Males tend to progress more rapidly from the healthy state (S1) to the severe state (S3) and show poorer recovery potential, whereas females exhibit a higher recovery probability. Abdominal obesity, metabolic abnormalities (such as elevated TG, low HDL-C, and elevated FPG), and age play vital roles in NAFLD progression and recovery. Gender differences not only affect the disease course but also provide a theoretical basis for future gender-specific, personalized treatments. Future research should further explore the interactions between these factors to offer more precise theoretical support for early intervention and personalized treatment of NAFLD.