A prediction model for persistent postoperative frailty in older patients undergoing elective orthopedic surgery

Abstract

Frailty is a medical syndrome with multiple causes and contributors that is characterized by diminished strength, endurance, and reduced physiologic function. It is highly prevalent among older orthopedic patients and significantly increases the risk of surgery and postoperative complications. Therefore, preoperative identification of high-risk groups for postoperative frailty is crucial for optimizing clinical decision making. This study aimed to explore the predictors of persistent postoperative frailty in older patients undergoing elective orthopedic surgery. Patients aged > 65 years who underwent elective orthopedic surgery between January 2020 and January 2022 were included in this single-center retrospective cohort study. Baseline characteristics, clinical data, laboratory findings, and frailty assessments were recorded. Persistent postoperative frailty was defined as a FRAILTY score >2 at the 3-, 6-, and 12-month follow-ups. Patients were randomly divided into training and validation cohorts at a ratio of 7:3 using stratified sampling. Least absolute shrinkage and selection operator and logistic regression were used for variable screening and analysis. A nomogram was constructed to visualize the predicted model. Receiver operating characteristic curves and area under the curve (AUC) values were used to assess the diagnostic accuracy. Calibration curves were performed to evaluate the calibration. A decision curve analysis was used to evaluate the clinical utility. Grip strength, gait speed, systemic immune-inflammation index, and the systemic inflammatory response index were identified as the significant predictors. The corresponding odds ratios were 2.467, 1.214, 1.809, and 1.743, respectively. A nomogram was used to visualize the logistic model, which achieved an AUC value of 0.772 (with a sensitivity of 77.9% and specificity of 64.3%) in the training cohort and an AUC value of 0.788 (with a sensitivity of 86.1% and specificity of 82.2%) in the validation cohort. Calibration curves indicating the acceptable agreements between the nomogram-predicted probability and the actual probability of persistent postoperative frailty. The decision curve analysis curves showed that the prediction model consistently outperformed the “treat-all” and “treat-none” strategies. The prediction model incorporating grip strength, gait speed, systemic immune-inflammation index, and systemic inflammatory response index enables early perioperative identification of older patients undergoing elective orthopedic surgery who are at high risk for developing persistent postoperative frailty.

1. Introduction

Frailty is characterized by a decline in the function of multiple physiological systems, leading to a reduced ability to adapt to environmental changes.^[1,2] The resulting decline in physical function, including the inability to independently perform activities of daily living, is particularly prominent in older patients with orthopedic diseases.^[3] Joint replacement and lumbar fusion surgery, among other common elective orthopedic procedures, are regarded as effective means to combat frailty by improving patients’ mobility and enhancing their quality of life. However, some patients experience persistent postoperative frailty, leading to delayed recovery and suboptimal surgical outcomes. This substantially diminishes the clinical benefits of elective surgery for these patients.^[4,5]

The concept of frailty is similar to that of resilience. Resilience is the human ability to adapt when a traumatic life stressor suddenly occurs. On a purely theoretical basis, the same stressor generates heterogeneous consequences for different individuals.^[6] Therefore, for orthopedic patients undergoing elective surgery, the extent of stress caused by surgical trauma has a significant impact on the progression of frailty status.

Research has confirmed that assessing and predicting the frailty status of surgical patients is of significant value for making surgical decisions, determining surgical indications and selecting surgical and anesthesia plans.^[7] For elective orthopedic surgeries aimed at helping patients reverse their frailty status, the ability to predict postoperative frailty progression would assist doctors in making correct clinical decisions. Regrettably, studies on frailty status and its evolution after elective orthopedic surgery are still relatively scarce at present. We believe that for patients undergoing elective orthopedic surgery, utilizing biomarkers that reflect the body’s internal environmental state,^[8,9] in conjunction with physical examinations that assess functional capacity,^[10–13] to evaluate the patient’s resilience represents an innovative approach to addressing the challenges of frailty prediction in clinical practice.

The objective of this study is to establish a predictive model for persistent postoperative frailty in older patients undergoing elective orthopedic surgery, assisting orthopedic surgeons in identifying high-risk populations for persistent postoperative frailty in the perioperative period. This will have significant clinical implications for guiding surgical plan selection and prognostic evaluation.

2. Methods

2.1. Study design and participants

We conducted a single-center retrospective cohort study. Patients older than 65 years who underwent elective orthopedic surgery (including joint replacement and lumbar spinal fusion) for chronic lumbar or knee diseases at our center between January 2020 and January 2022 were included. This is because the age of 65 has been defined as the threshold age for age-related physiological decline by the European Working Group on Sarcopenia and the American Geriatrics Society.^[14,15] Clinical data and frailty assessments were collected at the time of patient admission and 3, 6, and 12 months after discharge. Patients were excluded if they had a history of recent bleeding, anemia, tumors, immune system diseases, acute trauma, cognitive decline, or postoperative complications (infection, hematoma, delayed union, loosening, or dislocation of the implants). Patients who were unable to complete the grip strength and gait speed tests, as well as those with incomplete clinical or follow-up data, were excluded.

All patients provided informed consent and agreed to participate in follow-up. This study was registered in the Chinese Clinical Trial Registry (ChiCTR, Clinical trial number: ChiCTR1800018840) and approved by the Ethics Committee of our hospital (2018BJYYEC-031-01). The surgeries were performed by senior chief physicians at our center. Patients were randomly divided into training and validation cohorts at a ratio of 7:3 using stratified sampling. A training set of ≥70% can prevent overfitting in small samples,^[16] while a 30% validation set meets the minimum sample size requirement for an independent test set in AI models.^[17]

2.2. Clinical data

Basic patient information was collected, including gender, age, body mass index, type of surgery (joint replacement or lumbar spinal fusion), and medical history (including history of smoking or alcohol use).

Grip strength was measured using a hydraulic hand dynamometer (JAMAR® Hydraulic Hand Dynamometer, Model 5030J1; Performance Health Supply Inc., Warrenville). Participants sat in a chair with elbows bent at right angles on their sides. Each hand was tested 3 times and the results were averaged, with 15 to 30 seconds between each trial. The higher grip strengths of the 2 hands was selected as the result of grip strength. Following the Asian Working Group for Sarcopenia 2019 criteria, we defined grip strength decrease in patients whose grip strength was <28 kg for men and 18 kg for women and treated it as a categorical variable.^[18] We also measured gait speed based on the time required to walk 4.57 m.^[19]

Intraoperative blood loss was visually estimated by calculating the sum of the blood loss in the suction canisters on the surgical drape and gauze. In addition, postoperative drainage, intraoperative transfusion status, and preoperative American Society of Anesthesiologists (ASA) classification data were collected.

2.3. Laboratory examination

Routine preoperative laboratory tests, including red blood cell count (RBC), hemoglobin (HGB), white blood cell count (WBC), lymphocyte count (LYM), monocyte count (MON), neutrophil count (NEU), platelet count (PLT), serum total protein (TP), serum albumin (ALB), blood urea nitrogen, creatinine, and bone metabolism marker tests, including total type I collagen amino acid extension peptide, osteocalcin, β-CrossLaps, and 25-hydroxyvitamin D, were performed.

The systemic immune-inflammation index (SII) and systemic inflammation response index (SIRI) are composite indices calculated from the complete blood count results, reflecting immune cell subpopulations, and platelet aggregation.^[9,20,21] These indices have been widely used to explore the relationship between chronic inflammatory states and various diseases including cancer, metabolic disorders, and inflammatory conditions.^[22,23] The SII was calculated using the following formula: platelet count × NEU count/lymphocyte count. SIRI was calculated using the following formula: monocyte count × NEU count/lymphocyte count.^[24] To account for the positive skewness of this marker, the SII and SIRI values were log-transformed to ln-SII and ln-SIRI, respectively, and analyzed as independent variables.

2.4. Assessment of persistent postoperative frailty

We assessed frailty using the FRAIL scale, which ranges from 0 to 5 at the 3-month, 6-month, and 12-month follow-ups, respectively. Patients with a frailty score of ≥3 at 3, 6, and 12 months were defined as having persistent postoperative frailty.^[25] Conversely, pre-frailty was defined as a postoperative frailty score of <3 at any of these time points. Accordingly, patients were categorized into frailty and pre-frailty groups.

2.5. Variable screening and analysis

Least absolute shrinkage and selection operator (LASSO) regression was used to model and select the most relevant variables using 10-fold cross-validation. The dependent variable was the occurrence of persistent postoperative frailty, while the independent variables were gender, age, body mass index, RBC, HGB, WBC, LYM, MON, NEU, PLT, TP, ALB, blood urea nitrogen, creatinine, peptide, osteocalcin, β-CrossLaps, 25-hydroxyvitamin D, ln-SII, ln-SIRI, ASA classification, grip strength, gait speed, intraoperative blood loss, postoperative drainage, intraoperative transfusion status, type of surgery, smoking, and alcohol use. The LASSO model was optimized by selecting the lambda value that minimized the mean squared error, with one standard error rule applied to ensure model simplicity and efficiency. Factors with nonzero LASSO coefficients were retained for further analysis.

Subsequent selection and analysis were performed using a logistic regression model. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated to assess the association between the variables and persistent postoperative frailty.

2.6. Visualization and validation of the prediction model

A nomogram was constructed to visualize the predicted model according to the selected variables in the training cohort. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were used to assess the diagnostic accuracy of the prediction model in the training and validation cohorts. The optimal cutoff value was determined using the maximum Youden index. Calibration curves and the Hosmer-Lemeshow test were performed to evaluate the calibration of the nomogram in both cohorts. Finally, we conducted decision curve analysis (DCA) to evaluate the clinical utility of our prediction model by quantifying the net benefit at various threshold probabilities. DCA allows for the comparison of the performance of our model and each predictor with 2 extreme strategies: treating all individuals and treating none.

2.7. Statistical analysis

Continuous variables that followed a normal distribution were expressed as mean ± standard deviation (mean ± SD). Non-normally distributed continuous variables are expressed as medians and interquartile ranges (median [Q1, Q3]). Categorical variables are expressed as frequencies and percentages. Variables between different groups were compared using the Mann–Whitney U test and chi-square test, as appropriate. Additionally, multicollinearity analysis was performed for the independent variables included in the prediction model.

Based on the principle of events per variable, each variable had at least 10 outcome events.^[26] This ensured that the sample size in this study was adequate. All analyses were performed using R software (version 4.4.0). A two-sided hypothesis test was used, with P-values <.05 indicating statistical significance.

3. Results

3.1. Patients cohort construction

A total of 579 patients were included in this study. Of these, 28 patients were excluded due to anemia, tumors, immune system diseases, or postoperative complications; 57 patients failed to complete the grip strength or gait speed test; and 20 patients were lost to follow-up (Fig. 1). Consequently, 474 patients (314 women and 160 men) were included in this study. There were 86 patients (18.1%) in the frailty group with persistent postoperative frailty, the mean age was 76.08 ± 7.08, while 388 patients (81.9%) were in the pre-frailty group, the mean age was 74.58 ± 6.25. No statistically significant differences in the baseline characteristics were observed between the training and validation cohorts (Table 1). However, there were significant differences in grip strength, gait speed, RBC, HGB, WBC, LYM, MON, NEU, PLT, TP, ALB, ln-SII, and ln-SIRI (see Table S1, Supplemental Digital Content, https://links.lww.com/MD/P554, which illustrates the differences between the pre-frailty and frailty groups).

Figure 1.

The flowchart of patients cohort construction.

Table 1.

Baseline characteristics.

Group	Training cohort		P	Validation cohort		P
	Pre-frailty	Frailty		Pre-frailty	Frailty
Age (year)	74.49 ± 6.39	76.21 ± 7.14	.092	74.70 ± 6.59	75.80 ± 7.31	.443
Gender (n)			.312			.116
Men	97 (35.7%)	17 (27.9%)		34 (29.3%)	12 (48.0%)
Women	175 (64.3%)	44 (72.1%)		82 (70.7%)	13 (52.0%)
BMI (kg/m2)	25.16 ± 3.31	24.49 ± 3.52	.172	25.40 ± 3.12	25.25 ± 4.12	.833
Operation (n)			.989			.999
Joint	110 (40.4%)	24 (44.3%)		43 (37.1%)	9 (36.0%)
Lumbar	162 (59.6%)	37 (60.7%)		73 (62.9%)	16 (64.0%)

Joint. = joint replacement; Lumbar. = lumbar spinal fusion.

BMI = body mass index.

3.2. Valuable predictors of persistent postoperative frailty

We extracted data on 12 clinical variables and 17 laboratory examinations of all patients. LASSO regression was then performed to identify contributing factors. The results of the LASSO regression indicated that the model was the most simplified and efficient when lambda was selected to minimize the mean squared error, with a one standard error rule applied (Fig. 2). Among the various variables included, grip strength, gait speed, ln-SII, and ln-SIRI were considered the most valuable predictive variables for persistent postoperative frailty, with coefficients of 0.344, 0.107, 0.403, and 0.291, respectively (Table 2). Additionally, these variables exhibited statistically significant differences between the pre-frailty and frailty groups (Fig. 3).

Figure 2.

Variables were selected by LASSO model in the general population. (A) LASSO coefficients of the variables; (B) tuning parameter (λ) selection in the LASSO model used 10-fold cross-validation via minimum criteria. LASSO = least absolute shrinkage and selection operator.

Table 2.

LASSO and logistics regression analysis of selected variables.

Variables	LASSO coefficient	OR (95% CI)	P
Grip strength	0.344		.001
Normal		/
Decrease		2.467 (1.428, 4.262)
Gait speed	0.107	1.214 (1.119, 1.317)	<.001
ln-SII	0.403	1.809 (1.016, 3.221)	.044
ln-SIRI	0.291	1.743 (1.048, 2.899)	.032

LASSO = least absolute shrinkage and selection operator, SII = systemic immune-inflammation index, SIRI = systemic inflammation response index.

Figure 3.

The differences in grip strength (A), gait speed (B), ln-SII (C), and ln-SIRI (D) between pre-frailty group and frailty group in the general population. SII = systemic immune-inflammation index.

We then conducted binary logistic regression analysis using the 4 selected variables. Compared with the normal group, the OR for grip strength decrease was 2.467 (95% CI: 1.428, 4.262; P = .001). For continuous variables, the OR for gait speed, ln-SII and ln-SIRI were 1.214 (95% CI: 1.119, 1.317; P < .001), 1.809 (95% CI: 1.016, 3.221; P = .044) and 1.743 (95% CI: 1.048, 2.899; P = .032), respectively (Table 2). No multicollinearity was detected among the 4 variables in the prediction model because the variance inflation factor values for each variable were <5. Accordingly, grip strength, gait speed, ln-SII, and ln-SIRI were identified as valuable predictors of persistent postoperative frailty and selected for nomogram construction.

3.3. Nomogram construction and validation

Next, we constructed a nomogram for predicting persistent postoperative frailty based on the 4 predictors identified through LASSO and logistic regression (Fig. 4). The risk of persistent postoperative frailty can be determined by drawing a vertical line from the total points on the scale in the last row.

Figure 4.

The prediction model is visualized using a nomogram based on training cohort.

Subsequently, ROC curves were utilized to assess the diagnostic accuracy of the predictors and nomogram (Fig. 5). The AUCs of each predictor and the prediction model were all higher than 0.6 with statistical significance except for the AUC of grip strength in the validation cohort (AUC = 0.599, P = .123). The cutoff values were defined as “normal to decrease” for grip strength (±), 6.75 seconds, 5.89 and 0.10 in the training cohort, and these were similar in the validation cohort. The AUCs of the nomogram prediction model was 0.772 (sensitivity, 77.9%; specificity, 64.3%) in the training cohort and 0.788 (sensitivity, 86.1%; specificity, 82.2%) in the validation cohort (Table 3).

Figure 5.

ROC curves for the predictors and prediction model in different cohorts. The AUCs of the prediction model were 0.772 in the training cohort (A) and 0.788 in the validation cohort (B). AUC = area under the curve, ROC = receiver operating characteristic.

Table 3.

AUCs and cutoff values of the predictors and predictive model for persistent postoperative frailty.

Variables	Cohorts	AUC	Cutoff	Sensitivity	Specificity	P
Grip strength	Training	0.629	±*	70.9%	53.1%	.002
	Validation	0.599	±*	68.0%	51.7%	.123
Gait speed	Training	0.605	6.75s	83.7%	35.3%	.011
	Validation	0.711	7.75s	72.0%	59.5%	.001
ln-SII	Training	0.711	5.89	86.0%	42.0%	<.001
	Validation	0.666	5.91	80.0%	44.8%	.009
ln-SIRI	Training	0.698	0.10	86.0%	49.7%	<.001
	Validation	0.708	0.10	88.0%	55.2%	.001
Prediction model	Training	0.772	0.14	77.9%	63.4%	<.001
	Validation	0.788	0.36	86.1%	82.2%	<.001

SII = systemic immune-inflammation index, SIRI = systemic inflammation response index.

^*Between normal and decrease. It means that patients with grip strength decrease were more likely to have persistent postoperative frailty.

Calibration curves using the Hosmer–Lemeshow test for the training and validation cohorts indicated acceptable agreement between the nomogram-predicted probability and the actual probability of persistent postoperative frailty. The red and black dotted lines represent the good performance of our nomogram, which is close to the blue dotted line representing an ideal diagnosis (Fig. 6). In summary, the nomogram for persistent postoperative frailty showed considerable discriminative and calibration abilities.

Figure 6.

Calibration curves for nomogram in different cohorts. The calibration curves showed good predictive efficiency in the training cohort (A) and the validation cohort (B) with P-value of the Hosmer–Lemeshow test >.05.

The DCA curves showed that the prediction model consistently outperformed the “treat-all” and “treat-none” strategies. When analyzed individually, the 4 predictors showed varying levels of net benefit, but none surpassed the complete prediction model across the clinically relevant threshold range (0.2–0.5). This underscores the value of combining these predictors into a comprehensive model (Fig. 7).

Figure 7.

DCA curves for the predictors and prediction model in the training cohort. DCA = decision curve analysis.

4. Discussion

Frailty was defined as “a medical syndrome with multiple causes and contributors that is characterized by diminished strength, endurance, and reduced physiologic function that increases an individual’s vulnerability for developing increased dependency and/or death” in a consensus statement published in 2013.^[25] The identification of patients with frailty is challenging in clinical practice, and the lack of simple and effective prediction methods is one of the important reasons.^[27] We included 29 variables, which not only reflect the degree of surgical trauma but also provide insights into the patient’s physical function and internal environment. These variables are highly specific for predicting persistent postoperative frailty. A decrease in grip strength, decline in gait speed, and increase in ln-SII and ln-SIRI were associated with an elevated risk of persistent postoperative frailty in older patients undergoing elective orthopedic surgery (Table 2). On the one hand, the cutoff value and nomogram visualization made the results more intuitive (Fig. 4). In contrast, we performed internal validation of the prediction model, and the ROC and calibration curves achieved good agreement in both the training and validation cohorts. DCA highlights the clinical utility of the prediction model in improving decision-making for identifying at-risk individuals within a clinically meaningful threshold range (Fig. 7). Therefore, our study could help physicians easily identify the older population at a high risk of persistent postoperative frailty after elective orthopedic surgery. So that, clinicians can implement preventive measures in advance to reduce the risk of postoperative complications, such as improving the nutritional status, managing chronic diseases, and increasing bone density.^[28,29] This not only benefits the patient but also reduces the increase in total hospitalization costs in elective orthopedic procedures.^[30]

Our study indicates that preoperative frailty is not uncommon among patients undergoing elective orthopedic surgery, with 106 patients (22.4%) identified as frail, surpassing its prevalence in the general older population in our country.^[31] These patients may already be in a latent stage of frailty development before surgery. However, this group of patients is often in relatively good general condition and has the capacity to withstand major surgeries. As a result, doctors tend to overlook the potential for frailty progression. Our study results also confirm this viewpoint. Although their mobility improved following elective orthopedic surgery, trauma from surgery further exacerbated their frailty.^[32] We suggest that this may be attributed to 2 factors: first, patients undergoing orthopedic surgery often experience decreased mobility due to joint or lumbar diseases, increasing their risk of frailty; second, chronic orthopedic conditions may lead to a state of chronic inflammation or immune activation, facilitating the onset of frailty.^[9,33,34] Postoperative complications can influence the development of patients’ frailty status and interfere with the prediction results of perioperative clinical variables. This is also inconsistent with the objective of our study. Moreover, in our study, the number of cases with complications was not large, but the types were diverse. To avoid this uncontrollable bias, we excluded cases with complications. Additionally, persistent postoperative frailty, defined based on 3 follow-ups within 1 year, is a clinically significant outcome of great concern to physicians. One-year post-surgery represents a critical recovery period for patients undergoing elective orthopedic procedures. Persistent frailty not only slows the recovery process and affects surgical outcomes but also increases the risk of complications such as infections, reduced mobility, multi-organ dysfunction, and even mortality.^[35] Therefore, using persistent postoperative frailty as a dependent variable in research is more clinically meaningful than studying frailty at a single preoperative or postoperative time point.

The mechanisms underlying frailty remain unclear; however, they are associated with inflammatory responses and immune aging.^[36,37] We included key indicators from routine blood tests, biochemical parameters, and bone metabolism markers were included to comprehensively reflect the physiological and biochemical status of the patients. Neutrophils serve as key biomarkers of innate immunity, whereas platelets may contribute to immune function; monocytes and lymphocytes provide extensive information on adaptive immunity.^[38,39] Therefore, the SII and SIRI have demonstrated remarkable efficacy as emerging biomarkers for various diseases and are suggested as powerful tools for assessing and managing health in older patients.^[24,40–42] Notably, they are derived from routine blood cell counts and offer advantages of low cost, ease of measurement, and high reproducibility in laboratory settings.^[43] Our study confirmed the value of SII and SIRI in predicting frailty and may support our hypothesis that patients with high preoperative SII and SIRI are more likely to be in a state of chronic inflammation and immune dysfunction.

In addition, frailty is highly correlated with sarcopenia, nutritional status, and age.^[44,45] In particular, grip strength and gait speed tests have certain advantages in reflecting muscle reserves and physiological status and are widely used in the assessment of frailty in older adults.^[10–13] We excluded patients who were unable to complete grip strength and gait speed tests due to trauma (such as hand injuries and femoral neck fractures) or acute exacerbation of chronic diseases (such as lumbosacral pain). This is because these tests cannot reflect the true and objective physiological reserves of these patients. Poor preoperative grip strength and gait speed often indicate a lower physiological reserve capacity, higher sensitivity to adverse events, and an increased likelihood of surgery triggering or exacerbating frailty. Postoperative activity reduction, resulting in functional decline and muscle atrophy, may further contribute to a vicious cycle of frailty.^[46] We defined reduced grip strength was defined according to the Asian Working Group for Sarcopenia 2019 recommendations, enhancing the predictive relevance for Asian populations. Elective orthopedic surgery, as a traumatic intervention, exerts significant stress on the body, leading to the disruption of homeostasis.^[47] We also included surgery-related factors to reflect the stress caused by surgical trauma to the body, such as ASA classification, intraoperative blood loss, postoperative drainage, and transfusion status. However, no significant association was observed with persistent postoperative frailty. We believe that these factors are more likely to influence frailty status in the short term following surgery.

This study has certain limitations. First, our study focused on older patients undergoing joint replacement and lumbar spinal fusion, which limits the generalizability of our conclusions to other populations. Secondly, due to the study design, factors such as postoperative rehabilitation, comorbidities and sociodemographic variables that may affect the outcomes were not included. Future multicenter studies with external validation and incorporating a more comprehensive range of influencing factors will improve the comprehensiveness and generalizability of the conclusions.

5. Conclusions

The prediction model incorporating grip strength, gait speed, SII and SIRI enables early perioperative identification of older patients undergoing elective orthopedic surgery who are at high risk for developing persistent postoperative frailty.

Author contributions

Conceptualization: Guanghan Gao, Zitian Zheng, Fei Wang, Lei Shi, Qingyun Xue.

Data curation: Guanghan Gao, Zitian Zheng, Fei Wang.

Formal analysis: Guanghan Gao, Zitian Zheng.

Funding acquisition: Yaonan Zhang, Qingyun Xue.

Investigation: Yaonan Zhang, Lei Shi, Lin Wang.

Methodology: Guanghan Gao, Zitian Zheng, Zichen Ye, Fei Wang, Yaonan Zhang, Qingyun Xue.

Project administration: Yaonan Zhang, Lin Wang, Qingyun Xue.

Resources: Zitian Zheng, Zichen Ye, Yaonan Zhang, Lei Shi, Lin Wang.

Software: Guanghan Gao, Zitian Zheng, Zichen Ye.

Supervision: Zichen Ye, Fei Wang, Lei Shi, Qingyun Xue.

Validation: Zichen Ye, Lei Shi, Lin Wang.

Visualization: Zichen Ye, Lei Shi, Lin Wang.

Writing – original draft: Guanghan Gao.

Writing – review & editing: Qingyun Xue.