Association between new dietary index for gut microbiota and lung cancer: A population-based study

Abstract

The dietary index for gut microbiota (DI-GM), which reflects microbiota diversity, has not yet been linked to lung cancer. To explore this, we analyzed the relationship between DI-GM and lung cancer using data from the National Health and Nutrition Examination Survey. This analysis used National Health and Nutrition Examination Survey data from 2007 to 2018. We assessed multicollinearity among the independent variables with variance inflation factors, then performed weighted logistic regression, subgroup analysis, and applied a restricted cubic spline model to explore the relationship between DI-GM and lung cancer. The study included 22,473 adults aged 20 and older. The adjusted findings indicated that a higher DI-GM was associated with a reduced likelihood of lung cancer. variance inflation factors confirmed the absence of multicollinearity. After adjusting for potential confounders, participants with the highest DI-GM had a 72% lower risk of developing lung cancer compared to those with the lowest scores (OR = 0.28; 95% CI: 0.09–0.84, P = .02). A linear negative correlation was observed between DI-GM and the risk of lung cancer. This cross-sectional investigation revealed a negative relationship between the DI-GM and the likelihood of lung cancer.

1. Introduction

Lung cancer remains one of the primary causes of cancer-related mortality globally, representing a major public health challenge. Recent cancer statistics from the United States reveal an alarming trend: the incidence of lung cancer in adults under 50 is rising by approximately 2% annually, highlighting a shift that warrants immediate attention.^[1,2] Although significant progress has been achieved in targeted therapies and immunotherapies, these advancements have yet to meaningfully improve overall survival rates, which persist at a low level, with <20% of lung cancer patients surviving long-term.^[3,4] This underscores the urgent need for further research and innovation in lung cancer prevention, early detection, and treatment strategies. Moreover, it is projected that lung cancer will emerge as the most costly cancer in terms of diagnosis and treatment over the next 30 years, with estimated expenses potentially reaching a staggering USD 3.9 trillion.^[5] Consequently, it becomes increasingly crucial to understand the root causes and risk factors associated with lung cancer, as this knowledge is essential for informing effective public health initiatives. To effectively address this ongoing public health crisis, comprehensive insights into all underlying causes of lung cancer are not only beneficial but essential for formulating targeted prevention strategies and interventions.

The human microbiome encompasses all microbes residing on and within the human body, and it is intricately linked to various aspects of individual health, playing a vital role in maintaining overall well-being.^[6] Over the past 2 decades, interest in the microbiome’s role in host physiology has expanded significantly, evolving from initial descriptive studies into more detailed mechanistic explorations that focus on causality and molecular insights.^[7] The gut microbiome is instrumental in regulating metabolism and physiology, as it aids in processes such as digestion, nutrient availability, and the modulation of immune responses. Its composition is shaped by a variety of factors, including genetic predispositions and environmental influences, with diet emerging as a key regulator of intestinal microbiota.^[8] In the context of cancer, the intricate relationships between the microbiome – often referred to as oncobiosis – are recognized as significant contributors to both disease development and treatment efficacy. The oncobiome can influence cancer progression through several mechanisms, including direct interactions with cancer cells, alterations in the tumor microenvironment, or by modulating the immune response.^[9]

Kase et al conducted an extensive review of 106 studies focused on diet and gut microbiota, through which they identified 14 specific dietary components that significantly influence gut health.^[10] From these findings, they developed the DI-GM, an index designed to evaluate how diet quality impacts microbiome health. This index showed a positive correlation with biomarkers indicating microbiota diversity, such as urinary enterodiol and enterolactone, suggesting its effectiveness in reflecting the richness of gut microbial communities.^[10] DI-GM is particularly valuable as it differentiates between dietary patterns that are beneficial versus those that are detrimental to gut microbiota health, offering a reliable metric for assessing balanced diets. The creation of this index not only provides a tool for gauging dietary impact on microbiota but also has the potential to become a standardized measure for promoting diet quality. Moreover, DI-GM encourages interdisciplinary collaboration by bridging the fields of nutrition, microbiology, and medicine, fostering a holistic approach to diet-related health studies. Despite its potential, it is important to acknowledge that research specifically examining the association between DI-GM and lung cancer risk remains notably absent from the current body of literature, highlighting an area ripe for future investigation.

By utilizing the large and diverse National Health and Nutrition Examination Survey dataset, we aim to gain insights into whether variations in DI-GM are associated with lung cancer risk among different demographic groups within the U.S. population. This analysis will contribute to filling the current gap in research on DI-GM and lung cancer, offering new perspectives on diet and microbiome health as possible factors in lung cancer prevention.

2. Methods

2.1. Data source

The NHANES 2007 to 2018, a publicly accessible database in the United States, provided the data for this study.^[11]

2.2. Study population

This study specifically targeted individuals aged 20 years and older. Participants were selected from a comprehensive pool of respondents in the NHANES over 12 years, from 2007 to 2018. The initial dataset included 34,770 participants, providing a broad basis for analyzing health metrics. The exclusion criteria for this study included participants missing lung cancer data, as well as those lacking information on DI-GM components and covariates. After applying these criteria, a total of 22,473 eligible participants were retained for the final analysis, as illustrated in Figure 1.

Figure 1.

The flow chart of the included participants in this study.

2.3. Assessment of DI-GM

Fourteen food items or nutrients were identified as components of DI-GM. These included beneficial components such as avocado, broccoli, chickpeas, coffee, cranberries, fermented dairy, fiber, soybean, whole grains and green tea, while red meat, processed meat, refined grains, and a high-fat diet (≥ 40% energy from fat) were considered adverse components.^[10] Dietary recall data from NHANES 2007 to 2018 were analyzed to calculate DI-GM scores, with detailed methods available in the Table S1, Supplemental Digital Content, https://links.lww.com/MD/Q837. DI-GM values, which ranged from 0 to 14, were categorized into 4 groups: 0 to 3, 4, 5, and ≥6. This grouping facilitates a clearer understanding of the relationship between DI-GM levels and health risks.

2.4. Diagnosis of lung cancer

Lung cancer was classified based on the “mcq230” variable from the NHANES “mcq” questionnaire, which inquired, “What kind of cancer did you have?” Participants who reported lung cancer were included in the analysis.

2.5. Covariables

The variables examined in this analysis included demographic and health-related factors such as age, sex, marital status, and ethnicity, which was classified into Mexican American, Non-Hispanic Black, Non-Hispanic White, and Other. Education level was grouped as “less than high school” or “high school or above,” to distinguish educational attainment levels.^[12] The PIR was classified into ≤1, 1 to 3, and ≥3 based on household income relative to the PIR threshold.^[13] Smoking status was defined by lifetime cigarette use, with categories of never (fewer than 100 cigarettes), former (over 100 cigarettes but not currently smoking), and now (over 100 cigarettes and actively smoking). The self-reported drinking status was classified into the following categories: never (<12 drinks consumed in a lifetime), former (12 or more drinks consumed in 1 year but not in the last year, or no drinking in the last year but 12 or more drinks consumed in a lifetime), mild (one or fewer drinks per day for females and 2 or fewer drinks per day for males), moderate (two or fewer drinks per day for females and 3 or fewer drinks per day for males), and heavy (three or more drinks per day for females and 4 or more drinks per day for males).^[14] BMI was calculated using the standard formula of weight over height squared, with categories for normal weight (<25 kg/m²), overweight (25–29.9 kg/m²), and obesity (>29.9 kg/m²). Diabetes status was determined by self-reported diagnosis or current use of antihyperglycemic medications.^[15] Hypertension diagnosis was based on self-reported history, current antihypertensive medication use, or an average of 3 blood pressure readings showing systolic pressure ≥130 mm Hg or diastolic pressure ≥80 mm Hg.^[16,17] An affirmative answer to any of the following statements indicated CVD: having been informed of congestive heart failure, coronary heart disease, angina, a heart attack, or a stroke.^[18] Depression is diagnosed based on whether the PHQ9 score is >10.^[19]

2.6. Statistical analysis

All statistical analyses followed the NHANES guidelines for analysis and reporting, which take into account sample weights, stratification, and clustering to ensure representative and reliable results.^[20] By following these guidelines, we incorporated the complex sampling design and the sample weights from the mobile examination center into our study, thereby ensuring that our data sample accurately represents the diverse and broad U.S. adult population.

Baseline characteristics of the study population were evaluated using appropriate statistical tests: T-tests were applied to continuous variables, with results presented as mean ± standard error, and chi-square tests were used for categorical variables, presented in percentage form. This approach ensured that both continuous and categorical data were analyzed accurately, facilitating a comprehensive understanding of the baseline characteristics.

The independent association between DI-GM and lung cancer was thoroughly investigated using a survey-weighted multivariable logistic regression model. This approach allowed for a robust examination of the link between DI-GM and lung cancer risk, while controlling for a range of potential confounding variables that might otherwise impact the results. By adjusting for these confounders, the analysis aimed to isolate the specific influence of DI-GM on lung cancer risk, thereby providing a clearer understanding of any potential associations. Variance inflation factors (VIF) were calculated to assess multicollinearity, with VIF >10 indicating severe collinearity. To further explore the nature of this relationship, a restricted cubic spline (RCS) analysis was conducted. The RCS analysis was specifically employed to detect and assess any nonlinear associations between DI-GM levels and lung cancer risk. This method enables a more nuanced examination of the data, capturing potential fluctuations in risk across different DI-GM levels, which could reveal important insights about how changes in DI-GM might correlate with varying levels of lung cancer risk.

To investigate the relationship between DI-GM and lung cancer in various populations, stratified analyses were performed considering factors such as age, sex, ethnicity, BMI, DM, and CVD. The significance of interactions was assessed using P-values for the interaction coefficients between DI-GM and subgroup populations.

All statistical analyses were performed using R software. Statistical tests were 2-sided, with significance set at P <.05.

3. Results

3.1. Basic characteristics of participants

Our sample represents 159, 340, 876 people. Table 1 The baseline characteristics of the 22,473 participants are outlined, with an average age of 45.6 ± 0.3 years. The prevalence of lung cancer was 0.24%. Significant differences were noted in DI-GM, age, total calories intake, ethnicity, smoking status, drinking status, and levels of CVD, DM, and hypertension between participants with and without lung cancer.

Table 1.

Participant characteristics by presence of lung cancer, NHANES, 2007 to 2018.

Variable	Total	Without lung cancer	Lung cancer	P-value
DI_GM, mean ± SE	4.7 ± 0.0	4.7 ± 0.0	4.1 ± 0.3	.04
Age, mean ± SE	45.6 ± 0.3	45.5 ± 0.3	66.9 ± 1.7	<.0001
Total calories intake (kcal), mean ± SE	2209.9 ± 8.9	2210.6 ± 9.0	1875.3 ± 158.3	.04
Sex (N, weighted %)
Female	11226 (50.0)	11205 (50.1)	21 (42.6)	.4
Male	11247 (50.0)	11214 (49.9)	33 (57.4)
Ethnicity (N, weighted %)
Mexican American	3455 (8.6)	3454 (8.6)	1 (0.4)	.01
Non-Hispanic Black	4848 (11.2)	4838 (11.2)	10 (9.2)
Non-Hispanic White	9284 (67.1)	9251 (67.1)	33 (81.6)
Other	4886 (13.1)	4876 (13.2)	10 (8.8)
PIR (N, weighted %)
≤1	4907 (14.6)	4898 (14.7)	9 (9.6)	.5
≥3	8198 (49.7)	8179 (49.7)	19 (53.1)
1–3	9368 (35.6)	9342 (35.6)	26 (37.3)
Education (N, weighted %)
>High school	12162 (62.1)	12140 (62.1)	22 (45.7)	.1
≤High school	10311 (37.9)	10279 (37.9)	32 (54.3)
Marital status (N, weighted %)
Coupled	13337 (63.1)	13309 (63.1)	28 (60.2)	.7
Single or separated	9136 (36.9)	9110 (36.9)	26 (39.8)
Smoke (N, weighted %)
Former	5132 (23.5)	5095 (23.4)	37 (65.9)	<.0001
Never	12565 (56.5)	12560 (56.6)	5 (11.1)
Now	4776 (20.0)	4764 (20.0)	12 (23.0)
Drinking status (N, weighted %)
Current	16020 (77.5)	15992 (77.5)	28 (57.7)	<.0001
Former	3389 (12.2)	3367 (12.1)	22 (35.9)
Never	3064 (10.4)	3060 (10.4)	4 (6.5)
BMI (N, weighted %)
Normal weight	6386 (29.4)	6364 (29.3)	22 (42.3)	0.2
Obese	8832 (38.4)	8814 (38.4)	18 (36.2)
Overweight	7255 (32.3)	7241 (32.3)	14 (21.5)
CVD (N, weighted %)
No	20363 (92.7)	20329 (92.7)	34 (68.6)	<.0001
Yes	2110 (7.3)	2090 (7.3)	20 (31.4)
Hypertension (N, weighted %)
No	13372 (64.4)	13353 (64.5)	19 (46.3)	.03
Yes	9101 (35.6)	9066 (35.5)	35 (53.7)
DM (N, weighted %)
DM	3562 (11.6)	3544 (11.5)	18 (28.8)	.003
No	18911 (88.4)	18875 (88.5)	36 (71.2)
Depression (N, weighted %)
No	20473 (92.1)	20424 (92.1)	49 (92.6)	0.9
Yes	2000 (7.9)	1995 (7.9)	5 (7.4)

Data are presented as weighted mean (±SE) or weighted frequencies (weighted percentages).

BMI = body mass index, CVD = cardiovascular disease, DI-GM = dietary index for gut microbiota, DM = diabetes mellitus, NHANES = National Health and Nutrition Examination Survey, PIR = poverty to income ratio.

3.2. The association between DI-GM and lung cancer

To assess multicollinearity among the covariates in the multivariable logistic regression model, we calculated the VIF for each variable (Table S2, Supplemental Digital Content, https://links.lww.com/MD/Q837). All VIF values were below the accepted threshold of 10. Furthermore, the tolerance values exceeded 0.1, further confirming the absence of significant collinearity among the covariates.

The relationship between DI-GM and lung cancer risk was examined through 3 models: a crude model, Model 1, and Model 2. As presented in Table 2, the crude model results indicated no significant association between DI-GM (treated as a continuous variable) and lung cancer risk. However, both Model 1 and Model 2 revealed an inverse relationship between DI-GM and lung cancer risk. These findings suggest that, after adjusting for potential confounders, DI-GM may be negatively associated with lung cancer risk. In the fully adjusted Model 2, higher DI-GM scores were significantly associated with a lower risk of lung cancer compared to the lowest DI-GM scores, with an odds ratio of 0.28 (95% CI = 0.09, 0.84, P = .02). This finding underscores the potential protective effect of higher DI-GM levels against lung cancer risk.

Table 2.

Association between DI-GM and lung cancer in different models.

Variables	Crude model		Model 1		Model 2
	OR (95% CI)	P	OR (95% CI)	P	OR (95% CI)	P
DI-GM (continuous)	0.80 (0.65–1.00)	.05	0.76 (0.62–0.95)	.01	0.75 (0.61–0.93)	.01
DI-GM (multi-category)
0–3	Ref	–	ref	–	ref	–
4	1.28 (0.49–3.29)	.61	1.24 (0.50–3.10)	0.64	1.19 (0.46–3.08)	.72
5	0.52 (0.16–1.68)	.27	0.46 (0.14–1.58)	.22	0.43 (0.13–1.41)	.16
≥6	0.39 (0.13–1.19)	.10	0.31 (0.10–0.94)	.04	0.28 (0.09–0.84)	.02
P for trend\	–	.03	–	.01	–	.005

Crudel model: No covariates adjusted.

Model 1: Adjusted for age, sex, ethnicity, education, PIR, marital status.

Model 2: Adjusted for sex, age, ethnicity, education, PIR, marital status, drinking status, smoke, BMI, CVD, DM, hypertension, total calories intake (kcal), depression.

BMI = body mass index, CI = confidence interval, CVD = cardiovascular disease, DI-GM = dietary index for gut microbiota, DM = diabetes mellitus, OR = odd ratio, PIR = poverty to income ratio.

3.3. Subgroup analysis

The stratified analysis revealed a significant correlation between DI-GM levels and lung cancer risk across various demographic and health-related subgroups. As detailed in Figure 2, each unit increase in DI-GM was associated with a reduced risk of lung cancer among participants aged ≥ 60 years [OR: 0.78, 95% CI = 0.65–0.94, P = .01], those with a BMI 25 to 30 kg/m² [OR: 0.69, 95% CI = 0.52–0.92, P = .01], individuals diagnosed with DM [OR: 0.64, 95% CI = 0.43–0.95, P = .03], and those with CVD [OR: 0.79, 95% CI = 0.68–0.91, P = .001]. These findings indicate that within these specific subgroups, higher DI-GM values are linked to a reduced likelihood of developing lung cancer, highlighting the potential protective effect of DI-GM in individuals with these characteristics.

Figure 2.

Subgroup analysis for the association between DI-GM and lung cancer. Models were adjusted for education, PIR, marital status, drinking status, smoke, hypertension, total calories intake (kcal), depression. BMI = body mass index, CI = confidence interval, CVD = cardiovascular disease, DI-GM = dietary index for gut microbiota, DM = diabetes mellitus, OR = odd ratio, PIR = poverty to income ratio.

3.4. Dose-response association between DI-GM and the risk of lung cancer

The RCS analysis, conducted using weighted multivariable logistic regression, revealed a negative linear relationship between DI-GM and lung cancer risk (P for non-linearity >.05) (Fig. 3). Specifically, higher levels of DI-GM were associated with a reduced risk of lung cancer. These findings suggest that an increase in DI-GM is linked to a lower likelihood of developing lung cancer.

Figure 3.

The dose-response relationship of continuous DI-GM with lung cancer in the overall sample. Adjusted for sex, age, ethnicity, education, PIR, marital status, drinking status, smoke, BMI, CVD, DM, hypertension, total calories intake (kcal), depression. BMI = body mass index, CVD = cardiovascular disease, DI-GM = dietary index for gut microbiota, DM = diabetes mellitus, PIR = poverty to income ratio.

4. Discussion

In this study, we have identified a significant inverse relationship between increases in DI-GM and the prevalence of lung cancer, with the effect most pronounced in individuals within the DI-GM ≥6 group. This finding indicates that higher DI-GM values may correlate with a decreased likelihood of lung cancer, suggesting a potential protective association that warrants further exploration. Through rigorous interaction testing, we have also validated the reliability and consistency of this association across multiple analytical frameworks, underscoring the robustness of our results and their applicability in varied analytical contexts. The RCS analysis further highlighted a consistent linear negative correlation between DI-GM levels and lung cancer risk, illustrating that as DI-GM increases, the risk of lung cancer decreases in a manner that aligns with our hypothesis and reinforces the credibility of our findings.

To the best of our knowledge, this study is pioneering as it represents the first extensive, retrospective analysis to investigate the association between DI-GM and lung cancer risk across a broad population sample. This positions DI-GM as a potentially valuable indicator that can be used in risk assessment models for lung cancer, offering new insights into how DI-GM levels might inform predictive frameworks in clinical settings. Our stratified analyses further support these findings, revealing a statistically significant reduction in lung cancer risk with each incremental unit increase in DI-GM for specific subpopulations. In particular, individuals aged ≥ 60, those with a BMI between 25 and 30 kg/m², and patients diagnosed with DM or CVD showed notable reductions in lung cancer risk, suggesting that certain demographic and health factors may enhance the protective effect of elevated DI-GM levels. Nonetheless, further in-depth research is essential to clarify the biological and clinical mechanisms that underlie these observed associations, which could ultimately guide future risk stratification strategies and potential preventive interventions in lung cancer.

The impact of diet on gut microbiota and its possible contribution to lung cancer development has been a focal point in historical research and is receiving growing attention in recent studies. Notably, dietary modifications have the potential to modulate gut microbiota composition, which can, in turn, play a crucial role in enhancing the effectiveness of immunotherapy treatments for cancer patients. Recent research highlights that specific dietary interventions targeting gut health may not only support general health but also contribute to improved therapeutic outcomes. For instance, several studies have shown that the administration of probiotics is correlated with positive clinical responses in patients diagnosed with advanced or recurrent non-small cell lung cancer undergoing anti-PD-1 monotherapy. These findings suggest that probiotics may have an adjunctive role in optimizing treatment efficacy for patients with non-small cell lung cancer, offering a potentially beneficial approach to supplementing immunotherapy protocols.^[21] Thus, the reasonable use of probiotics, prebiotics, and targeted dietary interventions aimed at gut microbiota may represent a promising strategy for enhancing the clinical efficacy of immune checkpoint inhibitor treatment.^[22] Although the gastrointestinal and respiratory tracts are physically distant from one another, they share the same embryonic origin and exhibit significant structural similarities, which is an important consideration in understanding their interconnectedness.^[23] Recent findings supporting various pathways involving their respective microbiota highlight the existence of the gut–lung axis,^[24] This concept suggests that gut microbiota can influence the development of lung cancer through multiple mechanisms (Fig. 4).^[25]

Figure 4.

Proposed mechanism of bidirectional influence of gastrointestinal microbiome and lung carcinogenesis. Alterations in the gut microbiome lead to the processing of specific antigenic cells by gastrointestinal dendritic cells, which results in the proliferation and expansion of various T-cell subsets that target them. In turn, these T cells produce cytokines that migrate to other organs, including the lungs, using immune homing molecules, thereby causing alterations in systemic inflammation leading to lung cancer. APC = antigen-presenting cell.

Dried fruits, abundant in various bioactive compounds with strong antioxidant properties, offer numerous potential health benefits that warrant further exploration.^[26] Research suggests that specific types of dried fruits may positively influence the composition of human gut microbiota, indicating a possible mechanism by which these fruits could impact the risk or progression of non-small cell lung cancer.^[27] This potential link highlights the importance of understanding how dietary components like dried fruits might contribute to cancer prevention strategies, particularly through their effects on gut microbiota and related immune pathways.

An observational study with a large cohort of 34,192 participants from the California Seventh-day Adventist population provided substantial evidence of an inverse relationship between dried fruit consumption and lung cancer risk. This study found that as the intake of dried fruits increased, the likelihood of developing lung cancer decreased significantly, suggesting potential protective effects of dried fruits on lung health.^[28] In a related investigation, a prospective trial analyzed dietary influences on lung cancer by examining 92 dietary factors in a sizable sample of 92,327 participants. This comprehensive analysis highlighted a notable negative association between higher consumption of fruits and vitamin C and the risk of developing squamous cell lung cancer, a specific type of lung cancer. The findings suggested that diets rich in these nutrients might contribute to a lower incidence of this cancer subtype, further supporting the potential role of diet in lung cancer prevention.^[29]

5. Limitations

The present study does have several limitations that should be acknowledged. Firstly, cross-sectional studies inherently lack the ability to establish causality, limiting our capacity to infer direct cause-and-effect relationships from the observed associations. Secondly, although we adjusted for numerous relevant confounding factors, there remains a possibility of residual confounding that could influence the results. Thirdly, the study’s findings cannot be taken as definitive conclusions; additional prospective longitudinal studies are necessary to validate these results and strengthen the evidence base. Fourth, our study included a limited number of lung cancer cases, and the diagnosis lacked confirmation through lung imaging or cytology. Additionally, the reliance on self-reported questionnaires may have introduced recall bias in the diagnosis of lung cancer. We recommend conducting studies with larger cohorts to better investigate this relationship. Finally, the DI-GM assessment was based on self-reported 24-hour dietary records, which are also prone to recall bias, potentially affecting the reliability of the data and exacerbating this bias.

6. Conclusions

The study revealed a significant association between DI-GM levels and the risk of lung cancer, indicating that individuals with higher levels of DI-GM had a lower likelihood of developing this disease. This inverse relationship suggests that DI-GM may act as a protective factor against lung cancer, making it a potentially valuable marker for assessing individual risk. Given these findings, monitoring DI-GM levels could offer a practical approach to identifying those at higher risk of lung cancer. Furthermore, future research should explore dietary interventions that target DI-GM levels in individuals with or at risk for lung cancer, as such strategies may contribute to a reduction in the incidence and progression of this condition. Overall, incorporating DI-GM into lung cancer risk assessments and interventions may enhance early detection and prevention efforts, thereby helping to mitigate the impact of lung cancer on public health.

Author contributions

Formal analysis: Tong Wu, Xiaofei Zhang, Xiaohan Ma, Peiling Zuo, Sheng Chen, Encun Hou.

Funding acquisition: Xiaofei Zhang, Xiaohan Ma, Encun Hou.

Investigation: Tong Wu, Xiaofei Zhang, Peiling Zuo.

Methodology: Tong Wu, Xiaofei Zhang, Xiaohan Ma, Peiling Zuo, Sheng Chen, Encun Hou.

Software: Tong Wu, Xiaofei Zhang, Xiaohan Ma, Peiling Zuo, Sheng Chen, Encun Hou.

Supervision: Xiaohan Ma, Xiaofei Zhang, Peiling Zuo, Sheng Chen, Encun Hou.

Visualization: Xiaohan Ma, Peiling Zuo, Sheng Chen, Encun Hou.

Writing – original draft: Tong Wu, Xiaofei Zhang, Xiaohan Ma, Peiling Zuo, Sheng Chen, Encun Hou.

Writing – review & editing: Tong Wu, Xiaofei Zhang, Xiaohan Ma, Peiling Zuo, Sheng Chen, Encun Hou.

Supplementary Material

medi-104-e46430-s001.docx ^{(13.6KB, docx)}