Association between Activity and Brain-Derived Neurotrophic Factor in Patients with Non-Alcoholic Fatty Liver Disease: A Data-Mining Analysis

Reduction in activity links to the development and progression of non-alcoholic fatty liver disease (NAFLD). Brain-derived neurotrophic factor (BDNF) is known to regulate an activity. We aimed to investigate the association between reduction in activity and BDNF in patients with NAFLD using data-mining analysis. We enrolled 48 NAFLD patients. Patients were classified into reduced (n = 21) or normal activity groups (n = 27) based on the activity score of the Chronic Liver Disease Questionnaire-NAFLD/non-alcoholic steatohepatitis. Circulating BDNF levels were measured using an enzyme-linked immunoassay. Factors associated with reduced activity were analyzed using decision-tree and random forest analyses. A reduction in activity was seen in 43.8% of patients. Hemoglobin A1c and BDNF were identified as negative independent factors for reduced activity (hemoglobin A1c, OR 0.012, p = 0.012; BDNF, OR 0.041, p = 0.039). Decision-tree analysis showed that “BDNF levels ≥ 19.1 ng/mL” was the most important classifier for reduced activity. In random forest analysis, serum BDNF level was the highest-ranked variable for distinguishing between the reduced and normal activity groups (158 valuable importance). Reduced activity was commonly seen in patients with NAFLD. Data-mining analyses revealed that BNDF was the most important independent factor corresponding with the reduction in activity. BDNF may be an important target for the prevention and treatment of NAFLD.


Introduction
Non-alcoholic fatty liver disease (NAFLD) is a major health problem worldwide [1]. NAFLD is a risk factor for liver cirrhosis, liver cancer, extrahepatic malignancies, and cardiovascular diseases [1]. A sedentary lifestyle is a well-known risk factor for NAFLD [2], and lifestyle modification, including maintaining physical activity, is the first-line therapy for NAFLD [3]. In fact, exercise improves insulin sensitivity, lipid metabolism, hepatic steatosis, and hepatic fibrosis in patients with NAFLD [4,5]. Moreover, maintaining physical activity is related to increased survival rates in patients with NAFLD [6,7]. Accordingly, activity assessment is important for the prevention and management of NAFLD.

Patient Characteristics
Patient characteristics are summarized in Table 1. There was no significant difference in age or BMI between the reduced and normal activity groups. Although the prevalence of women was significantly higher in the reduced activity group than in the normal activity group, no significant difference was seen in the liver frailty index, the prevalence of type 2 diabetes mellitus, or liver stiffness between the two groups (Table 1).
There was no significant difference in hemoglobin level, hepatic and renal function tests, or hemoglobin A1c (HbA1c) levels between the two groups. No significant difference was observed in serum levels of C-reactive protein, BDNF, myostatin, or inflammatory cytokines, such as IL-1, IL-6, IL-10, and IL-15, between the two groups (Table 1).

Patient Characteristics
Patient characteristics are summarized in Table 1. There was no significant difference in age or BMI between the reduced and normal activity groups. Although the prevalence of women was significantly higher in the reduced activity group than in the normal activity group, no significant difference was seen in the liver frailty index, the prevalence of type 2 diabetes mellitus, or liver stiffness between the two groups (Table 1). There was no significant difference in hemoglobin level, hepatic and renal function tests, or hemoglobin A1c (HbA1c) levels between the two groups. No significant difference was observed in serum levels of C-reactive protein, BDNF, myostatin, or inflammatory cytokines, such as IL-1, IL-6, IL-10, and IL-15, between the two groups (Table 1).

Independent Factors Associated with the Reduction in Activity in Patients with NAFLD
In a stepwise procedure, the following five factors were selected as explanatory variables: HbA1c, BDNF, liver frailty index, BMI, and sex. In a logistic regression analysis, HbA1c and BDNF levels were identified as independent negative factors associated with the reduction in activity in patients with NAFLD ( Figure 2).

Decision-Tree Analysis for the Reduction in Activity
A decision-tree algorithm was created by using two variables to classify three groups of subjects for the reduction in activity ( Figure 3). BDNF was selected as the most important classifier with an optimal cut-off of 19.1 ng/mL. In patients with BDNF levels < 19.1 ng/mL, all patients showed normal activity. On the other hand, in patients with BDNF levels ≥ 19.1 ng/mL, 53.6% of patients showed a reduction in activity. In such patients, HbA1c was selected as the second classifier with an optimal cut-off of 7.1%. Reduction in activity was seen in all patients with BDNF levels ≥ 19.1 ng/mL and HbA1c > 7.1%. Figure 3. Decision-tree analysis for the reduction in activity in patients with non-alcoholic fatty liver disease. The subjects were classified according to the indicated cut-off values of the variables. The pie graphs indicate the proportion of patients with reduced activity (black) and normal activity (white). Abbreviations: HbA1c, hemoglobin A1c; BDNF, brain-derived neurotrophic factor.

Decision-Tree Analysis for the Reduction in Activity
A decision-tree algorithm was created by using two variables to classify three groups of subjects for the reduction in activity ( Figure 3). BDNF was selected as the most important classifier with an optimal cut-off of 19.1 ng/mL. In patients with BDNF levels < 19.1 ng/mL, all patients showed normal activity. On the other hand, in patients with BDNF levels ≥ 19.1 ng/mL, 53.6% of patients showed a reduction in activity. In such patients, HbA1c was selected as the second classifier with an optimal cut-off of 7.1%. Reduction in activity was seen in all patients with BDNF levels ≥ 19.1 ng/mL and HbA1c > 7.1%.

Decision-Tree Analysis for the Reduction in Activity
A decision-tree algorithm was created by using two variables to classify three groups of subjects for the reduction in activity (Figure 3). BDNF was selected as the most important classifier with an optimal cut-off of 19.1 ng/mL. In patients with BDNF levels < 19.1 ng/mL, all patients showed normal activity. On the other hand, in patients with BDNF levels ≥ 19.1 ng/mL, 53.6% of patients showed a reduction in activity. In such patients, HbA1c was selected as the second classifier with an optimal cut-off of 7.1%. Reduction in activity was seen in all patients with BDNF levels ≥ 19.1 ng/mL and HbA1c > 7.1%. Figure 3. Decision-tree analysis for the reduction in activity in patients with non-alcoholic fatty liver disease. The subjects were classified according to the indicated cut-off values of the variables. The pie graphs indicate the proportion of patients with reduced activity (black) and normal activity (white). Abbreviations: HbA1c, hemoglobin A1c; BDNF, brain-derived neurotrophic factor.

Random Forest Analysis for Distinguishing between Reduced and Normal Activity Groups
The results of the random forest analysis are summarized in rank order in Figure 4. The serum BDNF level was the highest-ranked variable for distinguishing between the reduced and normal activity groups. This was followed by HbA1c, IL-6, and liver stiffness ( Figure 4). Decision-tree analysis for the reduction in activity in patients with non-alcoholic fatty liver disease. The subjects were classified according to the indicated cut-off values of the variables. The pie graphs indicate the proportion of patients with reduced activity (black) and normal activity (white). Abbreviations: HbA1c, hemoglobin A1c; BDNF, brain-derived neurotrophic factor.

Random Forest Analysis for Distinguishing between Reduced and Normal Activity Groups
The results of the random forest analysis are summarized in rank order in Figure 4. The serum BDNF level was the highest-ranked variable for distinguishing between the reduced and normal activity groups. This was followed by HbA1c, IL-6, and liver stiffness (Figure 4).
Life 2021, 11, x FOR PEER REVIEW 6 of 12 Figure 4. Random forest analysis for distinguishing between the reduced and normal activity groups. Variable importance is a general measure of the contribution of each variable in distinguishing the classes. Abbreviations: HbA1c, hemoglobin A1c; BDNF, brain-derived neurotrophic factor; IL, interleukin; BMI, body mass index.

Discussion
In this study, we demonstrated that reduction in the activity domain of CLDQ-NAFLD/NASH was seen in 43.8% of patients with NAFLD. In addition, we found that BDNF and HbA1c were independent negative factors for the reduction in activity. The data-mining analysis also revealed that the BNDF was the most important classifier for the reduction in activity. A higher serum level of BDNF was associated with a reduction in activity in patients with NAFLD.
The male-to-female ratio was significantly higher in the reduced activity group than in the normal activity group in this study. Sex was not an independent factor for reduced activity; this may be related to the low statistical power. However, Cohen et al. recently investigated gender differences in physical activity of general subjects and demonstrated that physical activity level is significantly higher in men than in women [27]. Moreover, in decision-tree analysis, sex was not identified as a classifier with the reduced activity. Furthermore, in random forest analysis, sex was the fifth-ranked variable for distinguishing between the reduced and normal activity groups in patients with NAFLD. Thus, in patients with NAFLD, the impact of sex on physical activity may be different from the general population and may have less impact on physical activity level.
In this study, we investigated factors associated with the reduced activity using various factors, including age, sex, BMI, HbA1c, liver stiffness, and inflammatory and antiinflammatory cytokines, including IL-6, IL-1, IL-10, and IL-15. However, none of these inflammatory cytokines was identified as an independent factor for reduced activity. Taken together, our findings suggest that activity may be mainly regulated by factors other than hepatic fibrosis and inflammatory cytokines in patients with NAFLD.
We employed exploratory analyses to eliminate possible bias and found that BDNF and HbA1c were independent factors associated with the reduction in activity among patients with NAFLD. The data-mining analysis further revealed that HbA1c was the second classifier for reduced activity in this study. Although there was no significant difference in HbA1c value and BDNF levels at the baseline between the normal and reduced activity

Discussion
In this study, we demonstrated that reduction in the activity domain of CLDQ-NAFLD/NASH was seen in 43.8% of patients with NAFLD. In addition, we found that BDNF and HbA1c were independent negative factors for the reduction in activity. The data-mining analysis also revealed that the BNDF was the most important classifier for the reduction in activity. A higher serum level of BDNF was associated with a reduction in activity in patients with NAFLD.
The male-to-female ratio was significantly higher in the reduced activity group than in the normal activity group in this study. Sex was not an independent factor for reduced activity; this may be related to the low statistical power. However, Cohen et al. recently investigated gender differences in physical activity of general subjects and demonstrated that physical activity level is significantly higher in men than in women [27]. Moreover, in decision-tree analysis, sex was not identified as a classifier with the reduced activity. Furthermore, in random forest analysis, sex was the fifth-ranked variable for distinguishing between the reduced and normal activity groups in patients with NAFLD. Thus, in patients with NAFLD, the impact of sex on physical activity may be different from the general population and may have less impact on physical activity level.
In this study, we investigated factors associated with the reduced activity using various factors, including age, sex, BMI, HbA1c, liver stiffness, and inflammatory and anti-inflammatory cytokines, including IL-6, IL-1, IL-10, and IL-15. However, none of these inflammatory cytokines was identified as an independent factor for reduced activity. Taken together, our findings suggest that activity may be mainly regulated by factors other than hepatic fibrosis and inflammatory cytokines in patients with NAFLD.
We employed exploratory analyses to eliminate possible bias and found that BDNF and HbA1c were independent factors associated with the reduction in activity among patients with NAFLD. The data-mining analysis further revealed that HbA1c was the second classifier for reduced activity in this study. Although there was no significant difference in HbA1c value and BDNF levels at the baseline between the normal and reduced activity groups, the difference in the results between univariate and multivariate analyses can be explained by confounding factors [28]. Serum BDNF levels are known to be modulated by various factors, including age [29] and sex [30], and the presence of hypertension [31]. A previous study has also reported that type 2 diabetes mellitus is associated with unhealthy behaviors, including reduced physical activity and increased sedentary behaviors [32]. In patients with NAFLD, type 2 diabetes mellitus has been reported to be associated with a reduction in activity, which was evaluated by the CLDQ-NAFLD/NASH [8]. Thus, our findings were in good agreement with previous reports. Although the causal relationship between HbA1c and reduction in activity remains unclear, the following are possible explanations. Physical inactivity may cause a reduction in energy consumption, leading to a reduction in glucose metabolism [33]. Alternatively, an association between HbA1c and reduction in activity may be affected by confounding factors, such as sarcopenia and depression [34,35].
BDNF is reported to be a regulator of physical activity [19]. Previous animal studies have demonstrated a positive association between brain BDNF levels and physical activity in rats [36,37]. These findings led us to hypothesize a positive association between serum BDNF levels and activity in patients with NAFLD. However, serum BDNF levels were negatively associated with the activity level of patients with NAFLD in our study. We also investigated the correlation of serum BNDF levels and FIB-4 index, M2BPGi levels, and liver stiffness values. However, no strong correlation was seen between serum BNDF levels and these hepatic fibrosis markers (data not shown). Chan et al. previously reported that the serum BDNF level is higher in healthy subjects with sedentary behaviors, such as watching television [38]. Huang et al. reported that higher serum BDNF levels were associated with lower physical activity levels in adolescents [39]. Moreover, Suwa et al. reported that serum BDNF levels were high in patients who are obese and have type 2 diabetes mellitus [40]. Levinger et al. also reported that serum BDNF levels were positively correlated with serum triglyceride levels, insulin resistance, and the number of metabolic risk factors [41]. Furthermore, de Avila et al. reported that, in patients with hepatitis C virus infection, BDNF was negatively correlated with SF-36 physical functioning [42], which is significantly associated with the activity domain of the CLDQ-NAFLD/NASH [8].
We could not examine the causal relationship between BNDF level and reduction in activity in patients with NAFLD in this study. Although the biological significance of BNDF in the reduction in activity remains unclear, BDNF has been reported to be modulated by various factors, including race [43], depressive symptoms [44], cognitive function [45], and genetic polymorphism [46,47]. These factors might be involved in the negative association between serum BDNF levels and activity in this study. In addition, BDNF has been reported to directly affect various metabolic pathways. BDNF reduces appetite through the regulation of serotonergic neurons in animal models [48,49]. BDNF also improves insulin resistance and lipid metabolism through the upregulation of energy metabolism in a mouse model of diabetes mellitus [50]. Taken together, BDNF may have been upregulated in this study as a compensatory mechanism to improve metabolic dysfunction and insulin resistance in patients with NAFLD who have a reduction in activity.
We employed a data mining analysis, an artificial intelligence approach, and revealed that "BDNF level ≥ 19.1 ng/mL and HbA1c ≥ 7.1%" is a profile for the high prevalence of reduced activity in patients with NAFLD. Since a reduction in activity is a risk factor for disease progression and mortality of patients with NAFLD [6,7], patients with such a profile may be an important target for the prevention and treatment of NAFLD.
This study has several limitations. First, this was a single-center study with small sample size. Second, this was a nested case-control study lacking a healthy subjects group and time-course changes in BDNF level. Third, we stratified subjects into two groups. Fourth, activity was evaluated by the CLDQ-NAFLD/NASH alone and not by other questionnaires or activity monitoring, including the International Physical Activity Questionnaire or an accelerometer. Thus, further multicenter prospective studies with healthy control, multiple questionnaires, and activity monitoring are required to elucidate the association between BDNF and reduction in activity in patients with NAFLD.
In this study, we showed that a reduction in activity was seen in 43.8% of patients with NAFLD. Furthermore, data-mining analyses revealed that BDNF was the most important negative factor associated with the reduction in activity in patients with NAFLD. Since a reduction in activity is involved in the development and progression of NAFLD, BDNF may be an important target for the prevention and treatment of NAFLD.

Study Design and Ethics
We conducted a single-center, nested case-control study in Japan. The protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki, as reflected by prior approval from the institutional review board of the Kurume University School of Medicine (approval number 20226). This research was performed under the relevant guidelines and regulations. An opt-out approach was used to obtain informed consent from patients, and personal information was protected during data collection.

Subjects
We enrolled 48 consecutive outpatients with NAFLD who visited Kurume University Hospital between May 2019 and April 2020. The inclusion criteria were as follows: (1) age > 18 years, (2) answered the CLDQ-NAFLD/NASH, and (3) underwent liver stiffness measurement. Exclusion criteria were as follows: (1) complications of other major diseases that might bias their activity, including congestive heart failure, renal failure, respiratory disease, cancer, and orthopedic disease, and (2) any condition that could interfere with answering survey queries, including psychiatric or emotional problems, language, or cognitive difficulties.

Diagnosis of NAFLD
The diagnosis of NAFLD was established according to the Clinical Practice Guidelines for NAFLD/NASH [3,51,52]: (1) hepatic steatosis evaluated by liver biopsy, ultrasonography, computed tomography, or magnetic resonance imaging; (2) ethanol intake <20 g/day in women or <30 g/day in men; and (3) exclusion of other liver diseases, including hepatitis B virus, hepatitis C virus, autoimmune hepatitis, drug-induced liver disease, primary biliary cholangitis, primary sclerosing cholangitis, biliary obstruction, Wilson's disease, and hemochromatosis.

Data Collection
All data were collected prospectively at the patients' regular visits. The following information was obtained using a self-reported questionnaire: age, sex, and amount of daily alcohol intake. In the clinical review, we obtained the following data: body mass index (BMI), waist circumference, presence/absence of type 2 diabetes mellitus, hypertension, and dyslipidemia; these were diagnosed according to standard criteria [53]. We also assessed physical frailty using the liver frailty index, as previously reported [54].

Disease-Specific Patient-Reported Outcome Measure for Patients with NAFLD
The CLDQ-NAFLD/NASH is a disease-specific questionnaire that consists of 36 items grouped into 6 domains: fatigue, systemic symptoms, emotional health, worry, abdominal symptoms, and activity [8,10,11]. All participants were asked to complete the Japanese version of the CLDQ-NAFLD, which has previously been validated in Japan [9]. The questions were closed-ended, and the answers were provided using a Likert scale ranging from 1 (all the time) to 7 (none of the time), representing the frequency of clinical symptoms and emotional problems. The average score was calculated for each domain, and higher scores indicated a minimal frequency of symptoms and, consequently, a better healthrelated quality of life.

Assessment of Activity and Definition for Reduction in Activity
Activity was assessed by the activity domain of the CLDQ-NAFLD/NASH. Patients with activity domain scores < 6 were classified into the reduced activity group. Patients with activity domain scores ≥ 6 were classified into the normal activity group.
Serum levels of BDNF were measured using a Human BDNF enzyme-linked immunoassay (ELISA) kit according to the manufacturer's instructions (R&D Systems, Inc., Minneapolis, MN, USA).
Serum levels of IL-1, IL-6, IL-10, IL-15, and myostatin were measured using a Human IL-1 ELISA Kit, a Human IL-6 ELISA Kit, a Human IL-10 ELISA Kit, a Human IL-15 ELISA Kit, or a Myostatin Quantikine ELISA Kit from R&D Systems, Inc. (Minneapolis, MN, USA), according to the manufacturer's instructions.

Measurement for Liver Stiffness and Controlled and Hepatic Steatosis
Liver stiffness and hepatic steatosis were evaluated under fasting conditions for more than 4 h by vibration-controlled transient elastography and controlled attenuation parameters using FibroScan ® (Echosens, Paris, France), respectively. The median value of 10 valid measurements was calculated, and the results of elasticity and controlled attenuation parameters were expressed in kiloPascals (kPa) and dB/m, respectively. A valid result included at least 10 valid measurements with an interquartile range (IQR)/median value of <30%, as previously described [55].

Statistical Analysis
Data are expressed as median (interquartile range), range, or number. The differences between the reduction and no reduction in activity groups were analyzed using Wilcoxon rank-sum tests. Factors associated with the reduction in activity were analyzed using logistic regression analysis and data mining techniques. The statistical methods are described in detail below.
Independent factors associated with reduced activity levels were analyzed using logistic regression analysis, as previously described. Explanatory variables were selected in a stepwise manner, minimizing the Bayesian information criterion, as previously described [26]. Data were expressed as odds ratios and 95% confidence intervals.
A decision-tree algorithm was constructed to reveal profiles associated with reduced activity, as previously described [26]. The initial classifier was the most important factor associated with reduced activity. Subjects were classified according to the cut-off values indicated for each variable. The cut-off value was determined by the decision-tree analysis, which gives the best split of valuables [56,57].
A random forest analysis was used to identify factors distinguishing between the reduced and normal activity groups on an ordinal scale, as previously described [23]. The variable importance value, which reflects the relative contribution of each variable to the model, was estimated by randomly permuting its values and recalculating the predictive accuracy of the model. All statistical analyses were performed using JMP Pro ® 15 (SAS Institute Inc., Cary, NC, USA). All p values were 2-tailed, and a value < 0.05 was considered statistically significant.  Informed Consent Statement: An opt-out approach was used to obtain informed consent from patients, and personal information was protected during data collection.
Data Availability Statement: Data are available upon reasonable request.