A potential bioelectrical impedance equation for estimating skeletal muscle area using computed tomography in colorectal cancer

Abstract

Bioelectrical impedance analysis (BIA) requires validated equations tailored to specific populations and devices to estimate body composition. In this study, we aimed to develop a predictive equation for BIA to evaluate skeletal muscle area (SMA) using computed tomography (CT) as the reference method. This is bi-center cross-sectional study, involving 211 patients. BIA was conducted using a tetrapolar model, measuring resistance (R), and reactance (Xc) values. The equation was developed using a linear regression model, maintaining variables that best correlate to SMA_CT. Validity was assessed using Bland-Altman plots and bootstrapping resampling method. Lins’ concordance correlation coefficient (CCC), root mean squared error (RMSE), and mean absolute error (MAE) were calculated before and after resampling. The proposed equation included sex, age, weight, height, resistance and reactance. This model accounted for more than 85% of the variability in SMA_CT (R² _adjusted = 0.86), with a RMSE of 10.37 cm² and MAE of 8.28 cm². SMA_BIA was highly correlated with SMA_CT (ρ = 0.93, P < .001). Bland-Altman plots and CCC (0.92) demonstrated a moderate agreement between SMA_BIA and SMA_CT. The newly proposed BIA equation demonstrated potential for predicting SMA_CT as the reference standard. Our hypothesis requires further investigation in both healthy and clinical populations.

Introduction

Catabolic diseases, such as cancer, can severely impact nutritional status^1,2, potentially leading to abnormalities in body composition (e.g., loss of skeletal muscle and fat mass)^3,4. These abnormalities may occur at diseases’ any stage, and may present as important prognostic factors^2,3. The loss of muscle mass in clinical populations has been described as a predictor of adverse outcomes, including poor functional capacity⁵, lower quality of life⁶, treatment interruptions^7,8, higher rates of surgery complications⁹, and shorter survival^10,11. Consequently, it is essential for clinicians to proactively assess and monitor body composition of patients who are predisposed to these abnormalities¹², as a pivotal step in nutritional diagnosis, facilitating the implementation of personalized interventions¹³.

Computed tomography (CT) is an accurate standard technique that can be used in research contexts to assess the body composition of clinical patients, particularly cancer patients¹⁴. CT scans assess skeletal muscle, an important component of the lean soft tissue compartment. The third lumbar vertebra (L3) is the most assessed anatomical landmark given its strong correlation with whole-body muscle area, being widely used in oncology as a surrogate for the gold standard, magnetic resonance imaging (MRI)¹⁵. Body composition variables from CT has demonstrated a strong correlation with MRI and offers higher independent explanatory variance (r²) for both fat and muscle compartments¹⁶. Given its availability in oncology settings and its strong correlation with whole-body values, CT scans provide a reliable and accurate method for assessing body composition. In this context, several studies have utilized CT scans for muscle health assessments in cancer^17–23, as CT provides information on both quantity and quality^24,25. CT is also more accessible and less time-consuming than MRI, and it is not affected by the presence of metal objects, such as prostheses or pacemakers^26–28. Despite a promising and convenient method, CT scans are still not widely available in routine practice, nor are they feasible for long-term monitoring due to concerns such as radiation exposure¹³.

Bioelectrical impedance analysis (BIA) emerges as a potential alternative for estimating body composition in clinical practice, applicable to a range of conditions, beyond cancer¹³. As a doubly-indirect method, BIA distinguishes itself as a validated, portable, cost-effective, and non-invasive bedside approach¹³. From raw values (e.g., resistance and reactance), BIA estimates total body water (TBW) and subsequently body composition using predictive equations, assuming a hydration factor of 0.732 for fat-free mass (FFM), particularly^26,29–32. However, different tissues have diverse electrical conductivity, and hydration variations potentially affect BIA measurements^33,34. In this context, hydration is an important factor to ponder, as it can vary depending on body compartment, race, age, nutritional status, and disease stage^35–38.

The estimation of muscle mass through BIA requires validated predictive equations specific to the population and device, derived from a reference method^39,40. Most studies proposing BIA equations have used dual-energy X-ray absorptiometry (DXA) as the reference method⁴¹. However, DXA does not directly measure skeletal muscle; it measures appendicular lean soft tissue, which includes skeletal muscle. To our knowledge, no equation estimating lean soft tissue has used CT scans as the reference method, no has any been developed for patients with colorectal cancer, although some equations have been validated against DXA in this clinical population^42–44. Due to the significant variability among different BIA devices⁴⁵, using CT scans as a reference method for directly measuring skeletal muscle in certain clinical populations holds great promise, improving muscle health assessments. However, a limitation of CT scans, like most other body composition techniques, is that they typically do not account for variations in hydration status, which poses an additional challenge.

BIA may not be the most reliable technique for estimating body composition in clinical populations such as cancer, due to its sensitivity to changes in body mass, underlying clinical conditions, and hydration status, as previously mentioned. However, in resource-limited settings, it remains a valuable and endorsed technique¹³, particularly for monitoring purposes. In this context, developing a population- and device-specific equation using a robust reference standard, such as CT scans, could enhance the practical utility of BIA. Therefore, this study aimed to develop and propose a BIA equation to predict skeletal muscle cross-sectional area (SMA in cm²), using CT as the reference method, in a population of patients with colorectal cancer.

Methods

Study design and subjects

This study used existing data from two prospective cohort studies conducted at oncology reference centers in Brazil: Onofre Lopes University Hospital (HUOL), Natal, RN, and the National Cancer Institute (INCA), Rio de Janeiro, RJ. Details of these studies were previously described^38,46. As patients with cancer frequently undergo CT scans for diagnostic, staging, and clinical purposes, they were conveniently chosen to have their body composition analyzed as the reference method in this study. Inclusion criteria were consecutive inpatients and outpatients of both sexes aged 18 and above, diagnosed with colorectal tumors, irrespective of stage (I-IV), and treatment performed. Data collection for inpatients extended up to 48 h post- hospital admission. Before undergoing BIA assessment, patients collected at the outpatient clinic were asked about the time of their last meal, considering a fast of at least 2 h. Exclusion criteria were individuals with metal implants and/or pacemakers, amputated limbs, edema/anasarca/ascites/abnormal hydration status, concomitant chronic consumptive diseases (e.g., AIDS, non-oncological liver diseases, tuberculosis, chronic kidney disease), and cognitive conditions hindering the expression of informed consent. This study adhered to the principles of the Declaration of Helsinki and approved by the Research Ethics Committee of the HUOL-RN (CAAE 40045220.7.0000.5292), and INCA-RJ (CAAE 38992014.5.0000.5274). All patients provided written informed consent.

Clinical and covariates

Patient assessments were conducted either at the bedside or in the outpatient clinic. Individuals were queried about their age, ethnicity, and physical functioning. Additional clinical data, including details about current treatment (at the time of collection) and tumor type was collected. Information was double-checked for accuracy and extracted from patient’s electronic medical records.

Nutritional assessments

Body weight (kg) and height (m) were measured to calculate body mass index (BMI); BMI classification followed the criteria outlined by the World Health Organization (WHO)⁴⁷. BIA assessments were conducted using the same Quantum IV Analyzer (RJL Systems^®, Clinton Township, MI, USA), which is a tetrapolar, single-frequency (50 kHz) system. Raw values (resistance and reactance) were assessed from BIA, which can estimate TBW assuming a normal hydration status, as previously mentioned. For BIA assessments, patients fasted for 2 h, wore light clothing, were free of metallic objects or prostheses, and had emptied their bladder. Data collection occurred in Natal (Northeast Brazil) and Rio de Janeiro (Southeast Brazil), tropical cities with consistent high temperatures throughout the year. However, hospital beds and consulting rooms were naturally ventilated and maintained at approximately 24 °C, which aligns with the assumptions for BIA temperature control. Furthermore, the manufacturer of our BIA device (Quantum IV) reported that resistance and reactance values exhibit less than 1% interference at temperatures ranging from − 20 °C to 60 °C⁴⁸.

Patients assumed a supine position on stretchers covered with sheets, with their legs and arms away from their torso⁴⁹. Electrodes were placed following specific criteria: (i) on the dorsal surface of the hand near the third metacarpophalangeal joint, (ii) on the wrist between the styloid processes of the radius and ulna, (iii) on the dorsal region of the foot at the level of the anterior transverse arch near the second toe, (iv) and on the upper surface of the ankle between the medial and lateral malleoli. Three measurements were taken, and the average value was considered for subsequent analysis. Measurement error for BIA can vary between 1% and 2% in resistance values throughout the day, and between 2% and 3.5% over a week. This variation can be even greater in BIA models operating at frequencies below 50 kHz⁵⁰. However, in our study, we used a device operating at 50 kHz, minimizing potential frequency-related inaccuracies.

CT was employed to assess SMA at the level of the third lumbar vertebrae (L3) using the Slice-O-Matic program version 5.0 (Tomovision^®, Montreal, Canada). Hounsfield Unit (HU) thresholds ranging from − 29 to + 150 was used to select and analyze specific tissues for SMA. This landmark includes the following muscles: psoas, erector spinae, quadratus lumborum, transverse abdominis, internal and external obliques, and rectus abdominis⁵¹. Given its convenience, CT images were used if conducted within ± 90 days of the anthropometric assessment. Although debatable, this methodology (timeframe) has been employed in previous studies^23,52.

Statistical analysis

Data were analyzed using R Studio version 4.3.2. Normality of the continuous variables was assessed using the Shapiro-Wilk test. Variables with normal distribution are presented as mean ± standard deviation (SD) and the groups were compared using the t-test for independent samples. Categorical variables are described in absolute and relative frequencies (n, %), and were compared through the Pearson’s χ² or likelihood ratio χ² tests, as appropriate. Correlation analyses were conducted using Pearson’s or Spearman correlation tests. Strength was classified according to the magnitude of the correlation coefficient (r or ρ values) as follows: very high (0.90–1.00), high (0.70–0.90), moderate (0.50–0.70), low (0.30–0.50), or negligible (0.00–0.30)⁵³.

Continuous variables were checked for multicollinearity using Variance Inflation Factor (VIF > 0.1 < 3.0). The normality of residuals was also assessed using the Shapiro-Wilk test. If residuals were non-normally distributed, Cook’s distance was utilized to identify influential data points that could potentially skew the results. Influential points with a Cook’s distance greater than 4/n were carefully examined and, if necessary, excluded from the dataset to ensure the normality of residuals. All assumptions for linear regression were met, including independence of residuals, homoscedasticity, normality of residuals, and no multicollinearity. Thus, linear regression analyses with various approaches were conducted to identify the best predictors of SMA_CT. Regression methods included manual insertion, automated analysis (backward and forward), and stepwise addition and removal. All approaches retained the same variables in the final model used for the equation. Age, sex, resistance, reactance, weight and height were included as their potential relationship with skeletal muscle.

Root mean squared error (RMSE) and mean absolute error (MAE) were calculated using both the optimal and obtained values. Lin’s concordance correlation coefficients (CCC) were estimated to assess the concordance between SMA measurements using both methods. CCC values were classified as follows: (i) poor: < 0.90; (ii) moderate: 0.90–0.95; (iii) substantial: 0.95–0.99; and (iv) almost perfect: > 0.99⁵⁴. Agreement between SMA_BIA and SMA_CT was assessed using Bland-Altman (BA) plots, using SMA from CT as the gold standard. BA findings were expressed as the mean difference (bias) between SMA_BIA and SMA_CT, with limits of agreement (LOA) set at ± 1.96 SD to represent the 95% reference range.

To assess the internal validity of the newly developed equation, a robust methodology was employed. Dataset was subjected to random resampling with replacement, generated hypothetical applications in N diverse samples, utilizing the bootstrapping method, with bias corrected and accelerated (BCa). Results were also validated using a k-fold (10) cross-validation approach, ensuring the model’s predictive performance was consistent across different data subsets. Model’s predictive performance was evaluated by calculating the Prediction Sum of Squares (PRESS). Results were compared to the residual sum of squares (SSE) to assess the model’s fit and ensure the balance between model complexity and predictive accuracy. A ratio of PRESS/SSE close to 1 indicates a good fit, suggesting that the model is not overfitting and has strong predictive performance. Sample size was estimated based on the parameter of 30 individuals per variable included in the equation model. Since 6 variables were retained in the final regression model (as will be detailed later), a minimum of 180 individuals would be required to achieve an R² value of 0.93. A significance level of P < .05 was considered statistically significant for all tests.

Results

Initially, 551 patients were screened for eligibility. After exclusions, a total of 211 patients were included in the final sample (51.7% males, 56.4% older adults, ≥ 60 y). Figure 1 illustrates the study flowchart. Stage IV and colon cancer were the most frequent stage and diagnosis (50.2%, and 42.4%, respectively). Clinical features were not different across sexes. Excess weight by BMI (≥ 25 kg/m²) occurred 58.3% of the sample, followed by normal range BMI (38.4%). Sex-specific differences were observed for SMA_CT, resistance and reactance (BIA-derived). Detailed clinical and nutritional features are shown in Table 1.

Fig. 1.

Study flow chart.

Table 1.

Characteristics of patients with cancer.

Variables	Total (n = 211)	Males (n = 109)	Females (n = 102)	P
Clinical features
Age (years)	61 (53;69)	62 (55;69)	61 (57;71)	.89a
Age (categories)				.33b
< 60 years	92 (43.6)	44 (40.4)	48 (47.1)
≥ 60 years	119 (56.4)	65 (59.6)	54 (52.9)
Tumor site				.49b
Colon	106 (50.2)	52 (47.7)	54 (52.9)
Rectum	93 (44.1)	49 (45.0)	44 (43.1)
Rectosigmoid junction	12 (5.7)	8 (7.3)	4 (3.9)
Tumor, Node, Metastasis classification*				.35c
I	17 (8.1)	6 (5.5)	11 (10.9)
II	45 (21.4)	23 (21.1)	22 (21.8)
III	59 (28.1)	30 (27.5)	29 (28.7)
IV	89 (42.4)	50 (45.9)	39 (38.6)
Type of treatment**				.51b
Adjuvant	33 (55.0)	15 (60.0)	18 (51.4)
Neoadjuvant	27 (45.0)	10 (40.0)	17 (48.6)
Nutritional features
BMI categories				.06c
Underweight	7 (3.3)	1 (0.9)	6 (5.9)
Normal range	81 (38.4)	39 (35.8)	42 (41.2)
Excess weight	123 (58.3)	69 (63.3)	54 (52.9)
CTL3
SMA (cm2)	121.6 (100.9; 144.9)	142.0 (129.0; 156.0)	102.0 (92.1; 111.0)	< 0.001 a
BIA
Resistance (Ω)	529.0 (474.0; 602.0)	484.0 (455.0; 529.0)	582.0 (537.0; 658.0)	< 0.001 a
Reactance (Ω)	53 (46.0; 60.0)	52.0 (45.0; 58.0)	54.5 (50.0; 61.6)	0.019 a

*n = 210; ** n = 60.

^aMann-Whitney U test (Median and IQ ranges).

^bPearson’s χ² test (Frequencies, n %).

^cLikelihood ratio χ² (Frequencies, n %).

BIA bioelectrical impedance analysis, BMI body mass index, CT computed tomography, L3 third lumbar vertebrae, SMA skeletal muscle cross-sectional area.

Table 2 displays the variables and their corresponding coefficients that retained associated with SMA_CT following linear regression analysis. This regression model explained more than 85% of the variability in SMA_CT (_adjusted R² 0.86), with a RMSE of 10.37 cm² and MAE of 8.28 cm². Final regression model included: age (in years), sex (0 for males; 1 for females), weight (kg), height (m), resistance (Ω), and reactance (Ω). Following these results, the proposed equation to estimate SMA_BIA was as follows:

Table 2.

β regression coefficients for each variable to predict SMA_CT.

Variables	Coefficient	Standard error	t	P
Intercept	100.40119	21.27523	4.72	< 0.001
Age (years)	-0.32885	0.07210	-4.56	< 0.001
Sex	-21.24588	2.34499	-9.06	< 0.001
Weight (kg)	0.54541	0.07335	7.44	< 0.001
Height (m)	30.05180	12.94076	2.32	0.050
Resistance (Ω)	-0.12626	0.01380	-9.15	< 0.001
Reactance (Ω)	0.64238	0.07628	8.42	< 0.001

Linear regression analysis.

CT computed tomography, SMA skeletal muscle cross-sectional area.

SMA_BIA demonstrated a very high positive correlation with SMA_CT (ρ = 0.93, P < .001). (Fig. 2). Lin’s coefficient demonstrated a moderate concordance between SMA_BIA and SMA_CT (CCC: 0.92, 95% CI 0.90 to 0.94). Figure 3 depicts BA plots illustrating the mean differences between values derived for SMA. Overall agreement between the two methods was good, with a mean bias of 0.085 (95% CI -1.32 to 1.49; LOA: -20.29 to 20.46). A proportional bias between SMA_BIA and SMA_CT was observed, as indicated by the negative regression slope (coefficient: -0.14, 95% CI -0.18 to -0.08, P < .001), indicating that the difference between the two methods increases as the average SMA value decreases (i.e., BIA tends to overestimate SMA in lower SMA_CT values, and underestimate SMA in higher SMA_CT values).

Fig. 2.

Distribution and Spearman’s correlation coefficient between SMA_BIA and SMA_CT. Abbreviations: CT: computed tomography; BIA: bioelectrical impedance analysis; SMA: skeletal muscle cross-sectional area in cm².

Fig. 3.

Bland-Altman plot illustrating the concordance between SMA_BIA and SMA_CT. Abbreviations: BIA: bioelectrical impedance analysis; CT: computed tomography; LOA: limit of agreement; SMA: skeletal muscle cross-sectional area in cm².

After resampling with bootstrapping analysis (N = 10,000 replications), the RMSE downed to 10.31 cm², while the _adjusted R²_, CCC and MAE remained the same (0.86, 0.92, and 8.28 cm², respectively). Cross-validation test yielded similar results: _adjusted R² 0.87, RMSE 10.47 cm² and MAE 8.42 cm². PRESS value was 23,154.65, and the SSE was 22,709.52. The ratio of PRESS to SSE was 1.02, indicating that the predictive error was only 2% greater than the model’s residual error.

Discussion

This study aimed to propose a predictive equation using raw parameters from BIA to estimate SMA_CT as the reference method. Our main findings demonstrate a high correlation, moderate agreement (CCC, BA), and relatively high errors (RMSE of 10.31 cm² and MAE of 8.28 cm²) between SMA_BIA and SMA_CT. Furthermore, SMA_BIA explained over 85% of the variability in SMA_CT. These results represent an initial step in demonstrating the potential of this new predictive equation.

Despite the RMSE being considered a relatively high error, it is important to acknowledge that this metric can be significantly influenced by high data variations⁵⁵, which is the case for SMA measurements obtained via CT scans among patients with cancer (in our sample – minimum SMA: 61.4 cm², maximum: 212.4 cm²). In this context, an RMSE value of 10.31 cm² might be considered acceptable. Furthermore, the MAE, which is less affected by extreme values, demonstrated a lower mean error of 8.28 cm² for predicting SMA. It is also important to recognize that BIA has several limitations, and all predictive equations are inherently imperfect and estimative rather than providing a direct and precise measurement of body composition/SMA. It is also possible that using CT scans to measure SMA in healthy populations could potentially result in less error due to potentially lower variations in SMA, resistance, and reactance measurements.

BA plots revealed overall good agreement between the newly proposed equation and SMA by CT-scans. Regression line indicated a proportional bias tendency towards both overestimation and underestimation at extreme values. This persistent bias, combined with systematic errors (as indicated by error metrics and analysis), may be partially explained by the presence of non-contractile (i.e., non-muscular) components within muscle tissue, which can influence its electrical properties^56,57. Despite this, our overall findings support the validity of our equation for estimating the SMA_BIA. As earlier stated, previous studies have typically relied on dual-energy X-ray absorptiometry (DXA), which traditionally measures appendicular lean soft tissues (ALST), or magnetic resonance imaging (MRI) for assessing SMA or volume as reference methods for developing BIA equations^58,59. Our innovative approach demonstrates the practicality of using CT scans as a viable/opportunistic alternative reference method for SMA estimation.

In oncology research, muscle cross-sectional area assessed by CT is commonly used to diagnose low muscle mass, often standardized to height squared (skeletal muscle index, cm²/m²), and is a critical predictor of adverse outcomes such as surgical complications, chemotherapy toxicity, and elevated mortality rates^15,60. Therefore, proposing an equation that can estimate this muscle cross-sectional area could facilitate the calculation of SMI, potentially enhancing its clinical significance. Additionally, based on our results, other equations could also be applied to estimate muscle mass volume/weight⁶¹.

In our study, the linear regression model optimized to estimate SMA_CT performed well, accounting for more than 85% of its variability. The model demonstrated a high correlation (ρ = 0.93) and moderate agreement (CCC: 0.92) between SMA_BIA and SMA_CT. These results suggest that our BIA equation can provide good estimates of muscle area, even in individuals with low muscle mass, although some bias may be present. This proportional bias may be inherent to BIA analysis, as individuals at extremes of body size (severe underweight: BMI < 16 kg/m²; or high degrees of obesity: BMI > 34 kg/m²) may exhibit altered electrical properties, potentially impacting the accuracy of SMA_BIA estimates⁶². However, despite these BIA-related limitations, clinicians could still use BIA with our equation for monitoring purposes, helping to mitigate biases and improve practical applicability.

Our regression model incorporated key variables such as sex, age, weight, height, resistance, and reactance, all of which are important determinants of muscle mass. These variables included in our equation are in line with previous studies that have developed BIA equations to estimate ALST in various populations, including healthy individuals^63,64. This consistency in the selection of parameters in different groups, using a device similar to ours, underlines the robustness of our approach. In theory, equations adapted to specific populations, which can differ significantly in body composition, are needed^2–4.

This study also employed the bootstrapping method with 10,000 resampling replications for internal validation, thereby demonstrating the feasibility and reliability of utilizing this newly proposed equation. This statistical approach has been previously employed in other studies as a strategy for equation validation^63,65,66. It consists of a random resampling technique aimed at ensuring replication capabilities⁶⁷. Experts have already underscored the robustness of this method, as bootstrapping exhibits low variance⁶⁷. Furthermore, employing a ten-fold approach (10,000 replications) is described to mitigate bias and facilitate cross-validation for model selection⁶⁷. This method demonstrated that the RMSE decreased, albeit modestly elevated, while the concordance (CCC) and variance explanation (R² _adjusted) remained consistent.

Our study has certain limitations warranting acknowledgment. The relatively small sample size necessitates caution when extrapolating our results to a wider population. Furthermore, we opportunistically utilized a population with cancer, as they are more likely to undergo CT scans in clinical settings. It is important to consider that patients with cancer may experience significant abnormalities in body composition, which could introduce biases into our estimations. Potential differences between the period of anthropometric measurements and CT scans could also influence our findings. Additionally, to the best of our knowledge, there are no prior studies that have developed BIA equations using CT scans as the reference method, limiting the scope of our comparative analyses.

Despite methodological challenges, our research includes patients with a wide age range, improving its potential utilization in both healthy and clinical scenarios. Conducted across relevant healthcare centers in two key regions of Brazil, this broad-based approach enhances the reliability and relevance of our findings, potentially making them more applicable across different settings. However, considering the influence of cancer on body composition, future studies could benefit from investigating and proposing a new BIA equation based on CT scans or other techniques, such as multicompartment models⁶⁸, potentially utilizing a healthy population with fewer abnormalities in body composition. It is important to consider that our proposed equation may be more suitable for clinical populations with similar characteristics.

Although additional research is required to confirm our results, the newly developed equation shows promise in predicting SMA via BIA in Brazilian individuals with similar clinical and demographic characteristics, including diagnoses, sex, and age.

From a clinical perspective, our study contributes to both future research and clinical practice. The application of our developed equation for estimating SMA_BIA may aid in muscle health assessments, although further investigation is warranted. By using a BIA equation tailored to our specific population and device, we can potentially optimize nutritional diagnoses and interventions, ultimately mitigating the negative effects associated with low muscle mass.

Author contributions

ASR and APTF contributed to the conception and design of the research; ASR, RABR, NAB, NCS and SFM acquired the data; JPCP, GOCM, ASD contributed to the data analysis. ASR and JPCP wrote the manuscript. APTF, CMP, and MCG critically revised. All the authors critically reviewed, interpreted, and approved the final version of the manuscript.

Funding

This study was partially funded by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), Brazil (Finance Code 001) and the Brazilian National Council for Scientific and Technological Development (CNPq, Chamada Universal, protocol: 404887/2021-0). APTF, and MCG received a productivity scholarship from CNPq. CMP is partially funded through the Canada Research Chairs Program. The supporting sources have no involvement or restrictions on this publication.

Data availability

Data generated and/or analyzed during the current study are not publicly available due to ethical and privacy restrictions but are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.