We performed a post-hoc analysis of the data from the NAGALA study (the original data were obtained from the Murakami Memorial Hospital Physical Examination Center in Gifu, Japan, and the available data were uploaded to the Dryad database for public sharing) (29). NAGALA study was a longitudinal cohort study that continuously recruited 20,944 participants who received health checkups between 1994 and 2016 to assess risk factors for common chronic diseases with the aim of promoting public health, and it was authorized by the Ethics Committee of Murakami Memorial Hospital and informed consent was provided by all participants. The study design and part of the study results have been previously published (30). Based on the previous research, the current study used the de-identified public data in the NAGALA study for post-hoc analysis, aiming to analyze and explore the association between MAP and NAFLD in different populations. The new study protocol has been submitted to the Ethics Committee of Jiangxi Provincial People's Hospital for review and has been approved (IRB2021-066), and the STROBE reporting guidelines have been followed throughout the study. In addition, because the current data set has been de-identified, the Ethics Committee of Jiangxi Provincial People's Hospital exempted the informed consent of the subjects. According to the new study objectives we excluded participants with the following characteristics: (i) participants with diagnosed hepatitis (viral/alcoholic) or diabetes or fasting plasma glucose (FPG) above 6.1 mmol/L at baseline (n = 1,547); (ii) participants with alcohol intake ≥ 210 g/w for men or ≥ 140 g/w for women at baseline (n = 1,952) (31); (iii) participants who were taking medications at baseline (n = 2,321); (iv) participants with missing data at baseline (n = 863); and (v) participants who withdrew from the study (n = 10). Finally, we included 14,251 eligible participants, and the flowchart for selecting participants was shown in Figure 1.

Data on participants' demographic characteristics (sex, age), lifestyle (smoking and drinking status, habits of exercise), disease history, and medication history were recorded through face-to-face interviews using a standardized questionnaire. Height, weight, waist circumference, and blood pressure were measured using standard methods in a standard setting, and BMI was calculated as weight/height². MAP was calculated as 1/3 SBP + 2/3 DBP (19). Smoking status was divided into non-smoking, former smoking, and current smoking; drinking status was divided into non, light, and moderate based on the amount and type of weekly alcohol consumption in the past month (32). Habits of exercise were evaluated based on the frequency of exercise per week, and those who exercised more than once were considered to have exercise habits. Venous blood specimens were drawn after participants had fasted for at least 8 h, and then glycosylated hemoglobin (HbA1c), alanine aminotransferase (ALT), high-density lipoprotein cholesterol (HDL-C), aspartate aminotransferase (AST), total cholesterol (TC), triglycerides (TG), and gamma-glutamyl transferase (GGT), FPG were measured using an automated biochemical analyzer according to standard procedures.

During the physical examination, the diagnosis of NAFLD was determined based on imaging; trained ultrasound technicians performed the abdominal ultrasound procedure, and then an experienced gastroenterologist reviewed the ultrasound images, scored them, and made the diagnosis of NAFLD based on the following four characteristics: hepatorenal echo contrast, liver brightness, deep attenuation and vascular blurring (33).

Considering the large sample size included in this study, we proposed to characterize the data of participants by MAP quintiles. Baseline characteristics of participants were reported as mean with standard deviation or median with interquartile range, and comparisons between groups were made using one-way ANOVA or Kruskal-Wallis H test, depending on whether the data distribution was normal (assessed using Q-Q plots). Categorical data were then expressed as frequencies and percentages, and comparisons between groups were made using the chi-square test.

To analyze the association between MAP and NAFLD, we calculated the ORs and 95% CIs using logistic regression analysis; in addition, we calculated variance inflation factors (Supplementary Table S1) for checking potential multicollinearity for all covariates before modeling (34). Four stepwise adjusted models were constructed according to the methodology suggested by the STROBE statement (35). Model 1 was adjusted for age, BMI, and height; model 2 further considered the effects of smoking and drinking status and habits of exercise based on model 1; model 3 continued with adjustments for AST, ALT, and GGT based on model 2; model 4 was adjusted for all non-collinear variables as the final model, i.e., it additionally considered the effects of glycemic lipid factors (TG, HDL-C, TC, FPG, and HbA1c) based on model 3. We also assessed the dose-response relationship between MAP and NAFLD using the RCS on the basis of model 4, with nodes located at the 5th, 35th, 65th, and 95th percentiles of the exposure distribution (36).

To test the robustness of the association between MAP and NAFLD, we also performed 4 sensitivity analyses. Based on model 4, we performed the same analyses in the population without exercise habits, in the young and middle-aged population, in the population with normal blood pressure, and in the non-obese population, respectively.

To explore whether the association between MAP and NAFLD differed among different populations, we further performed stratified analyses according to age, sex, BMI, habits of exercise, smoking status, and drinking status, and used likelihood ratio tests to assess differences between different subgroups.

Data analysis was performed in November 2022 using R version 3.4.3 (The R Foundation for Statistical Computing) and Empower(R) version 2.0. Statistical tests were two-tailed, and P values less than 0.05 were considered statistically significant.

The study population was 14,251 general people undergoing health screening, with a median (interquartile range) age of 42 (36–50) years and 48% being women. Table 1 shows the baseline characteristics of participants across MAP quintiles. The results showed that the prevalence rates of NAFLD in the MAP quintiles were 3.75%, 9.82%, 15.37, 22.85 and 35.81%, respectively. The higher MAP group was older, had higher height, weight, BMI, and WC, more people who smoked and drank alcohol, and had higher levels of liver enzymes, blood sugar, blood pressure, and blood lipids except for HDL-C. Additionally, it is worth noting that with the MAP quartile increasing, the number of men gradually increased and the number of women gradually decreased.

At baseline, 2,507 (17.59%) participants were diagnosed with NAFLD. In the multivariate logistic analysis (Table 2), after gradually adjusting for potential confounders, we found that MAP and NAFLD always maintained a stable positive correlation, with a large OR variation between the highest MAP quintile and the other quintiles; overall, a positive correlation trend was maintained between MAP and NAFLD in all four models. In model 4, the risk of NAFLD increased by 15% for every 10 mmHg increase in MAP, and the risk of NAFLD was 1.75 times higher in those with MAP at greater than or equal to 94.83 mmHg (Q5) than in those with MAP at 51.17–75.33 mmHg (Q1).

Table 3 shows the results of the association between MAP and NAFLD in multiple relatively low-risk populations, and the four sensitivity analyses based on model 4 provided similar findings to the main analysis, which underscored the robustness of the current study results.

We proceeded to assess the association between MAP and NAFLD in different subgroups of age, sex, BMI, habits of exercise, smoking status, and drinking status by stratified analysis and interaction tests (Table 4). We found that MAP-related NAFLD risk appeared to be associated only with age (P-interaction = 0.002), but not with sex, BMI, habits of exercise, drinking status, or smoking status (P-interaction > 0.05).

In the RCS regression analysis, we more intuitively observed a positive association between MAP and NAFLD, although it is worth noting that the association between MAP and NAFLD was not straight linear (P for non-linearity = 0.024). As seen in Figure 2, in the general population, the risk of NAFLD neither increased nor decreased when MAP was between 85 and 95 mmHg, with a transient saturation effect interval, which was similar to the results in model 4 of Table 2 [OR: (Q3)1.48 vs. (Q4)1.50]. In addition, based on the results of stratified analysis, we continued to fit the dose-response curve of the association between MAP and NAFLD in different age groups (Figure 3); the results showed that the aforementioned saturation effect interval remained in the young and middle-aged population (18–44 years old) and in the middle-aged and elderly population (45–59 years old), while in the elderly population (≥60 years old), there was a U-shaped curve association between MAP and NAFLD, where the lowest point was about 90 mmHg, and both low and high MAP would increase the risk of NAFLD.

Several strengths of the current study are worth mentioning: first, MAP is a simple and easily understood measurement parameter, and for the first time we have identified the potential clinical use of this parameter in assessing NAFLD risk in the general population. Second, the current study assessed for the first time the non-linear association between MAP and NAFLD and the differences across different populations, and these findings provided useful reference material for subsequent studies and clinical applications. Third, the participants in the current study came from physical examination centers and had a large sample size, so the results of this study were more suitable for popularization among the general public, had good external applicability, and had a high reference value.

There are some limitations of the current study that need to be mentioned: The first and most important point was that the study cannot make a causal inference between MAP and NAFLD, because it was a cross-sectional design. Second, it is about the diagnosis of NAFLD. Combined with previous experience, the use of color Doppler ultrasound for the diagnosis of NAFLD may miss approximately 30% of patients with mild hepatic steatosis (43); in the current study, only 2,507 (17.59%) subjects were diagnosed with NAFLD at baseline, significantly lower than the global or Asian average prevalence of NAFLD (1, 3). It should be noted, however, that the current study analyzed the association between MAP and NAFLD at a much lower prevalence of NAFLD, which rather further validated the stability of the findings of the current study. Third, the current study population is from Japan, and regional and ethnic limitations exist (3, 44), and more studies based on different ethnicities and countries are needed. Fourth, as with all observational studies, the current study must inevitably be subject to residual confounding because the observed covariates are finite and the risk and protective factors are relatively infinite (45). However, the current study performed rigorous statistical adjustments as well as sensitivity analyses whenever possible, and the findings can be considered relatively reliable at this point. Fifth, calculating non-invasive fibrosis scores (such as NAFLD fibrosis score, Fibrosis-4 index, aspartate transaminase-to-platelet ratio index, etc.) to determine the characteristics of the patient cohort is highly beneficial for the current study in the absence of elastography imaging; However, due to the lack of some parameters for calculating the non-invasive fibrosis score in this dataset, further staging information of liver fibrosis could not be provided in the current study. Sixth, only patients with alcoholic fatty liver disease and viral hepatitis were excluded from the current study, which may lead to potential selection bias because there are many competing causes of liver disease, including autoimmune hepatitis, Wilson's disease, alpha-1-antitrypsin disease, and hemochromatosis.