Association between fat-to-muscle ratio and secondary osteoporosis in rheumatoid arthritis: a cross-sectional study at a tertiary hospital in China

Abstract

Objectives

To investigate the correlation between fat-to-muscle ratio (FMR) or other body composition and secondary osteoporosis (OP) in patients with rheumatoid arthritis (RA) and to develop a predictive model using FMR and related clinical factors.

Design

Cross-sectional observational study with machine learning-based risk modelling.

Setting

Tertiary hospital in eastern China, secondary care level.

Participants

A total of 670 hospitalised RA patients (135 males and 535 females; aged 58.00 (50.00–67.00) years; disease duration 8.00 (2.00–16.00) years) and 126 healthy controls were recruited between October 2019 and October 2022. There were no differences in basic indicators such as gender, age distribution and body mass index between the two groups. RA diagnosis followed American College of Rheumatology (ACR) 1987 or ACR/European League Against Rheumatism 2010 criteria. Exclusion criteria included major organ dysfunction, endocrine disease, infection or long-term hormone or psychotropic drug use.

Primary and secondary outcome measures

Primary outcomes included total skeletal muscle mass, fat mass, FMR measured by bioelectrical impedance analysis and bone mineral density measured by dual-energy X-ray absorptiometry. Secondary outcomes included RA disease activity scores (clinical disease activity index (CDAI), simplified disease activity index, disease activity score in 28 joints (DAS28)) and glucocorticoid use. Logistic regression and four additional machine learning algorithms were used to build predictive models for OP.

Results

The RA group (age, 58.00; duration, 8.00; DAS28, 5.03; rheumatoid factor, 104.75; C-reactive protein, 25.65; erythrocyte sedimentation rate (ESR), 59.00) exhibited reduced total skeletal muscle mass (19.49 vs 25.38, p<0.001), hip bone mineral density (0.90 vs 1.15, p<0.001) and L1-4 bone mineral density (0.86 vs 1.08, p<0.001), alongside increased total fat mass (18.33 vs 16.37, p=0.020) and FMR (0.98 vs 0.68, p<0.001). Total fat mass was positively correlated with simplified and CDAI (p<0.001). Total skeletal muscle mass was negatively correlated with ESR (p=0.001) and positively correlated with both L1-4 and hip bone mineral density (p<0.001). FMR showed a positive correlation with clinical disease activity index (p<0.001). There were significant differences in total fat mass and FMR among RA patients with varying disease activity levels (p<0.001). RA patients with concomitant OP or using glucocorticoids had a higher total fat mass and FMR than their respective control groups, with only total skeletal muscle mass levels being lower (p<0.01). We developed predictive models using multiple machine learning algorithms, which identified that both age and FMR were key factors associated with secondary OP in RA patients. Subgroup analysis identified an interaction effect between FMR and gender and restricted cubic spline fitted the dose-response relationship between FMR and OP.

Conclusion

FMR may serve as a useful clinical indicator of secondary OP in RA patients. A model based on FMR and associated risk factors can predict the possibility of secondary OP.

STRENGTHS AND LIMITATIONS OF THE STUDY.

• This study used a large, single-centre rheumatoid arthritis (RA) cohort with comprehensive clinical and body composition data.

• Fat and muscle mass were assessed using multi-frequency bioelectrical impedance analysis.

• Bone mineral density was measured using standardised dual-energy X-ray absorptiometry.

• Multiple validated disease activity indices (clinical disease activity index, simplified disease activity index, disease activity score in 28 joints) were applied for RA stratification.

• Several machine learning algorithms, including logistic regression, were used to build and validate predictive models for secondary osteoporosis.

Introduction

Rheumatoid arthritis (RA) is a chronic autoimmune disease of unknown aetiology, characterised by joint synovitis and bone erosion. As the disease progresses, it can lead to the bones and joint capsule destruction, ultimately resulting in joint deformity and bone mass loss.¹ Due to the abnormal immune function in RA patients, their body composition significantly differs from that of healthy individuals,² which may contribute to the exacerbation of RA and the increased risk of osteoporosis (OP).³ Internationally, it is widely recognised that muscle deficiency and excess fat accumulation in RA patients can impact disease progression and the likelihood of OP comorbidity. Recent studies suggest that the prevalence of OP was decreased by degrees with increasing body mass index (BMI).⁴ Yet it may be too one-sided to simply assume that obesity is effective in preventing OP.⁵ In fact, both skeletal muscle and fat mass tend to be higher in people with greater body weight, and the interaction between skeletal muscle and adipose tissue ultimately determines the effect of BMI on bone mineral density (BMD). Skeletal muscle is crucial for movement and postural maintenance through the collaboration of muscle and bone tissue.⁶ Adipose tissue affects bone health by secreting cytokines and reducing the level of lipocalin.⁷ Thus, sarcopenia and body fat percentage have been found to be associated with enhanced odds of OP in normal-weight subjects.⁸ Some scholars have proposed to use the fat-to-muscle ratio (FMR) as a novel indicator to assess its correlation with multisystemic diseases.⁹ However, there is a scarcity of literature on the relationship between FMR and RA, as well as OP. The diagnosis of OP is relatively delayed, and there is a lack of effective bone densitometry tools in primary care. Therefore, the establishment of clinical prediction models is of particular importance.

In this study, we used multifrequency bioelectrical impedance analysis (BIA) equipment to measure the body composition of RA patients and healthy individuals, analysing differences in FMR between the two groups. We further investigated the relationship between FMR and disease status in RA patients, as well as its association with OP risk. This research aims to provide new insights for the assessment and subsequent treatment of RA patients, more crucially in parallel is the development of clinical prediction tools for RA patients with comorbid OP.

Methods

Study participants

The minimum sample size of 380 cases was calculated based on the literature expected effect value of p=0.45, δ=0.05, and α was taken as 0.05 for both sides.¹⁰ A total of 670 inpatients, comprising 535 females (79.9%) and 135 males (20.1%), from the Department of Rheumatology and Immunology, the first affiliated hospital of Anhui Medical University between October 2019 and October 2022 were enrolled in the study. All RA patients fulfilled the 1987 RA classification criteria of the American College of Rheumatology (ACR) or the 2010 ACR and European League Against Rheumatism classification criteria for RA and the disease duration was 8.00 (2.00–16.00) years. Patients with severe liver or kidney dysfunction, endocrine disorders, acute or chronic infections, or those who had been taking sex hormones, anticoagulants or psychotropic drugs for prolonged periods were excluded (online supplemental figure S1). A control group of 126 age- and sex-matched healthy individuals undergoing medical examinations at the hospital during the same period was selected, including 96 females (76.2%) and 30 males (23.8%). There were no statistically significant differences in basic indicators such as gender ratio (x²=0.865, p=0.352), age distribution (z=−1.943, p=0.052) and BMI (z=−0.051, p=0.959) between the RA group and the healthy control group, which were well comparable (table 1). All procedures were performed in compliance with relevant laws and institutional guidelines and were approved by the Ethics Committee of the First Affiliated Hospital of Anhui Medical University (approval number: 2022393), and all participants signed an informed consent form. Additionally, this study was carried out in accordance with the Declaration of Helsinki.

Table 1. Comparison of basic characteristics between RA group and healthy control group.

Indicators	Control group (n=126)	RA (n=670)	x2/z/t	P
Male:female	30:96	135:535	0.865	0.352
Age (years)	56.50 (49.00–63.00)	58.00 (50.00–67.00)	−1.943	0.052
BMI (kg/m2) TFM (kg) TSM (kg) FMR Hip BMD (kg/m2) L1-4 MBMD (kg/m2)	22.19 (19.01–24.78) 16.37 (10.58–21.43) 25.38 (21.51–28.60) 0.68 (0.44–0.83) 1.15 (1.02–1.30) 1.08 (0.90–1.13)	22.01 (19.18–24.07) 18.33 (12.75–22.83) 19.49 (16.82–21.47) 0.98 (0.66–1.21) 0.90 (0.77–1.03) 0.86 (0.74–0.97)	−0.051 −2.334 −10.813 −7.656 −11.929 −9.184	0.959 0.020 <0.001 <0.001 <0.001 <0.001

BMD, bone mineral density; BMI, body mass index; FMR, fat-to-muscle ratio; RA, rheumatoid arthritis; TFM, total fat mass; TSM, total skeletal muscle mass.

Patient and public involvement

Patients or the public were not involved in the design, conduct, reporting or dissemination plans of our research.

Clinical data collection

Detailed records were maintained for the general information of all subjects, including gender, age, height, weight and disease duration. For RA patients enrolled in the study, disease activity indicators were documented by the rheumatology clinical team, which included the count of swollen joints (SJC), the count of tender joints (TJC), the duration of morning stiffness, the overall self-assessment of the disease condition (evaluated using the visual analogue scale (VAS) score (0 to 10 cm) and the patient global assessment (PGA)), the physician global assessment (PhGA), the disease activity score in 28 joints (DAS28), patients’ clinical disease activity index (CDAI), simplified disease activity index (SDAI), glucocorticoid (GC) use and health assessment questionnaire (HAQ) scores. Laboratory indicators extracted from hospital electronic medical records included erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), rheumatoid factor (RF) and antibodies against cyclic citrullinated peptides (CCP), mean arterial pressure (MAP), history of diabetes mellitus, history of smoking, aspartate aminotransferase (AST), alanine aminotransferase (ALT), platelet count (PLT), albumin and globulin ratio (ALB/GLO), blood urea nitrogen (BUN), creatinine (CREA), white blood cell count (WBC), low-density lipoprotein (LDL), high-density lipoprotein (HDL), very low-density lipoprotein (VLDL), prothrombin time (PT) and activated partial thromboplastin time (APTT) as indicators of secondary OP.

Disease activity grouping

DAS28 calculated using the Miwa method (DAS28=0.56 × √(TJC28)+0.28 × √(SJC28)+0.70 × ln(ESR) + 0.014 × PGA) was used to evaluate disease activity in RA patients. The DAS28 grouping criteria in this study were as follows: DAS28 score <3.2 was defined as the low disease activity group, the moderate disease activity group was defined as DAS28 score of 3.2–5.1 and the severe activity group was defined as DAS28 >5.1.^{11 12} SDAI was also calculated using the Miwa method: SDAI=SJC28 + TJC28 + PhGA + PGA + CRP (PGA, PhGA: VAS 0~10 cm). SDAI score ≤11 was defined as the low-disease activity group; it was defined as the moderately active group when the SDAI score was between 11 and 26, and it was defined as the severely active group as SDAI score >26.13,15 CDAI was calculated using the method: CDAI = SJC28 + TJC28 + PhGA + PGA (PGA, PhGA: VAS 0~10 cm). According to the CDAI grouping criteria, CDAI score ≤10 was classified as the low disease activity group, CDAI score of 10–22 was called the moderate activity group and the severe activity group was defined when CDAI score >22.12,14

Bone mineral density (BMD) measurement and diagnostic criteria for osteoporosis (OP)

Bone mineral density (BMD) was assessed using a dual-energy X-ray absorptiometry (DXA) device (Lunar Prodigy DF+3 10 504, GE Healthcare, USA), examining the lumbar spine from L1 to L4 and the total hip, reported in g/cm². OP diagnosis followed the 2022 guidelines by the Osteoporosis and Bone Mineral Research Society of the Chinese Medical Association. BMD values within one SD of peak bone mass in healthy adults were classified as normal; reductions of 1–2.5 SDs indicated reduced bone mass, and OP was diagnosed for reductions of 2.5 SDs or more below peak bone mass.16,18

Measurement of body composition

Body composition was assessed using a bioelectrical impedance analyser (InBody 370, Seoul, South Korea), recording height, weight, total fat mass (TFM) and total skeletal muscle mass (TSM). FMR was defined as the ratio of TFM to TSM (FMR=TFM/ TSM; cut-off value: 0.975, sensitivity=66.3%, specificity=65.3%).¹⁹

Modelling, analysis and enhancement

We adopted five machine learning approaches in building optimal clinical models, including logistic regression (LR), support vector machine, elastic networks, eXtreme Gradient Boosting and Light Gradient Boosting Machine. A total of 28 clinical and laboratory features were used as input variables in the models. These included demographic factors (age, sex), body composition indices (FMR, BMI), vital signs (MAP), inflammatory markers (CRP, ESR), liver and kidney function markers (ALT, AST, BUN, creatinine), lipid profiles (HDL, LDL, VLDL), disease characteristics (RA duration, DAS28, CDAI, SDAI, HAQ, VAS), serological markers (RF, anti-CCP), medication use (GCs, DM (diabetes mellitus) therapy), smoking status and composite indices (albumin/globulin ratio, APTT, PT). To determine the best model, we evaluated the predictive performance of these models using the area under the subject operating characteristic curve (ROC-AUC), accuracy, sensitivity, specificity, precision, F1 scores, precision-recall curves, calibration plots and decision-curve analysis (DCA). In addition, feature significance was assessed using Shapley Additive Explanations (SHAP).

Statistical methods

Statistical analyses were conducted using R Studio and SPSS 26.0. Normally distributed data are presented as mean±SD, while non-normally distributed data are presented as median (25th to 75th percentiles) (M(P25–P75)). T-tests were used for comparisons between two normally distributed groups, with non-parametric tests applied for multiple group comparisons, and the χ² test for categorical variables. Spearman’s correlation was used for association analysis, with results expressed as correlation coefficient r. Machine learning model building and optimisation were implemented by R packages. ROC curves were used to evaluate predictive performance, with the AUC indicating screening ability. Calibration curves and DCA assessed prediction accuracy and clinical utility. Restricted cubic spline (RCS) with three nodes was used to assess the dose-effect relationship between FMR and OP. Significance was defined as p<0.05.

Results

Comparison of total fat mass (TFM), total skeletal muscle mass (TSM), fat-to-muscle ratio (FMR) and bone mineral density (BMD) levels between rheumatoid arthritis (RA) and normal control groups

Table 1 presented the comparison of TFM, TSM, FMR and BMD between RA and healthy individuals. Non-parametric tests indicated that RA patients had significantly lower TSM (p<0.001) and significantly higher TFM (p=0.020) and FMR (p<0.001) than the control group. Additionally, RA patients had significantly lower total hip BMD and L1-4 MBMD (mean value of bone mineral density) (p<0.001) compared with controls.

Correlation analysis of total skeletal muscle mass (TSM), total fat mass (TFM) and fat-to-muscle ratio (FMR) with rheumatoid arthritis (RA) condition indicators

In RA patients, TFM showed a significant positive correlation with CDAI (p<0.001) and SDAI (p<0.001). TSM was positively correlated with hip BMD (p<0.001) and L1-4 BMD (p<0.001) and negatively correlated with ESR (p=0.001). FMR was positively correlated with CDAI (p<0.001) (online supplemental table S1 and figure 1).

Figure 1. Heatmap of correlation between TFM, TSM, FMR and RA indicators. BMD, bone mineral density; CCP, cyclic citrullinated peptides; CDAI, clinical disease activity index; CRP, C-reactive protein; DAS28, disease activity score in 28 joints; ESR, erythrocyte sedimentation rate; FMR, fat-to-muscle ratio; HAQ, health assessment questionnaire; RA, rheumatoid arthritis; RF, rheumatoid factor; SDAI, simplified disease activity index; TFM, total fat mass; TSM, total skeletal muscle mass; VAS, visual analogue scale; MBMD, mean value of bone mineral density.

Comparison of total fat mass (TFM), total skeletal muscle mass (TSM) and fat-to-muscle ratio (FMR) levels among rheumatoid arthritis (RA) patients with different disease activity levels

Non-parametric tests revealed significant differences in TFM and FMR (p<0.001) across different CDAI groups in RA patients, and among different SDAI subgroups, there were statistically significant differences in TFM and FMR (p<0.001). In contrast, when comparing patients across different DAS28 groups, no significant differences were found for TFM, TSM or FMR (p>0.05), as shown in table 2.

Table 2. Comparison of TFM, TSM and FMR in RA patients in DAS28, SDAI and CDAI groups (M(P25-P75)).

Groups	Number of cases	TFM (kg)	TSM (kg)	FMR
DAS28 low, moderate activity	351	18.79 (13.00–23.40)	19.74 (16.96–22.00)	1.01 (0.66–1.26)
DAS28 high activity	319	17.82 (12.60–22.40)	19.21 (16.68–21.20)	0.95 (0.67–1.17)
SDAI low, moderate activity	179	16.05 (11.40–19.60)*	19.77 (16.93–21.83)	0.88 (0.59–1.00)*
SDAI high activity	491	19.16 (13.70–24.00)	19.39 (16.79–21.39)	1.01 (0.72–1.28)
CDAI low, moderate activity	297	16.74 (12.00–20.65)*	19.70 (17.08–22.00)	0.90 (0.60–1.11)*
CDAI high activity	373	19.59 (14.00–24.55)	19.32 (16.68–21.30)	1.04 (0.74–1.29)

^*p<0.001.

CDAI, clinical disease activity index; DAS28, disease activity score in 28 joints; FMR, fat-to-muscle ratio; RA, rheumatoid arthritis; SDAI, simplified disease activity index; TFM, total fat mass; TSM, total skeletal muscle mass.

Comparison of TFM, TSM and FMR between rheumatoid arthritis (RA) patients with and without osteoporosis (OP) or glucocorticoid (GC)

RA patients were divided into OP and non-OP groups. Results showed that TSM in the OP group was significantly lower than in the non-OP group (p<0.001), while TFM and FMR (p<0.001) were significantly higher in the OP group (online supplemental table S2 and online supplemental figure S2A). On the other side, all RA patients in this study were divided into the GC group and the non-GC group based on GC usage. The results showed that the TSM level of patients in the GC group was significantly lower than that of the non-GC group (p<0.001), while the TFM (p<0.05) and FMR (p<0.001) levels of patients in the GC group were significantly higher than that of the non-GC group, as shown in online supplemental table S3 and online supplemental figure S2B.

Development and evaluation of machine learning models

Based on the literature, we collected 28 indicators from all 670 patients with RA, which are considered as potential factors that may influence the complication of OP, and then developed and compared predictive models through five machine learning algorithms after we found no significant covariance among the variables. Each machine learning approach randomly split subjects 3:1 into training and validation sets and used fivefold cross-validation for performance evaluation (online supplemental figure S3). We compared the predictive power of all models in the validation set and found that LR showed excellent performance in predicting secondary OP in RA patients. Comparison of the multimodel ROC curves yielded that LR possessed the largest AUC with the smallest Brier score for its calibration curve, indicating that the model possessed the most significant agreement between predicted and actual outcomes; its DCA curve exhibited the widest threshold space representing the best predictive gain (figure 2).

Figure 2. Construction and comparison of multiple machine learning models in the validation set. (A) Evaluation of the efficacy of five machine learning algorithms for predicting OP in RA patients. (B) ROC curve analysis for multimodel forecasting. (C) Calibration curves for the prediction of secondary OP. (D) Decision curve analysis for each model. OP, osteoporosis; RA, rheumatoid arthritis; ROCAUC, area under the subject operating characteristic curve; SVM, support vector machine.

We followed up with a verification of the accuracy of the LR model and found that the AUC values were 0.816 (95%CI 0.779 to 0.854) and 0.808 (95%CI 0.741 to 0.875) in the training and validation set, respectively. The calibration curves were used to evaluate the agreement between predicted probabilities and realistic outcomes, which manifested a remarkable concordance in the training set as well as the validation set. Moreover, the DCA outcomes illustrated that LR could offer a greater benefit within a threshold probability range of 10%–93% in the training set and 12%–94% in the validation set (figure 3).

Figure 3. Development and evaluation of logistic regression (LR) models. (A–B) ROC curves in the training set and validation set. (C–D) Calibration curves in the training and validation set. (E–F) Decision curve analysis in the training and validation set. ROCAUC, area under the subject operating characteristic curve.

Interpretability of important characteristic variables

SHAP method was used to further analyse the interpretation of LR model results by using Shapley values to calculate the contribution of different features to the outcome. The bar graph showed the effect of each clinical feature on the predicted outcome, with the potency of the effect of the different features assessed using the absolute value of SHAP (figure 4A). Figure 4B and C visualised the importances of continuous and categorical variables on the prediction of outcomes respectively, with the distribution of SHAP values for each feature represented by dots and colour-coded to indicate the magnitude of the feature value. It was easy to see that age and FMR were the two most significant features influencing the outcome among the continuous variables. The trend line revealed a positive correlation between FMR values and SHAP values, indicating that the probability of the model predicting OP increases as the FMR value increases. In addition, histograms at the top and right provided an overview of the distribution of FMR values and SHAP values, further confirming the variability of FMR values and the degree of contribution to the model prediction (figure 4D). A waterfall plot showed the difference between the predicted value of a random patient and the average predicted value (online supplemental figure S4). The contribution to OP outcome was negative when FMR achieved 0.69, whereas the univariate relationship with outcome suggested a significant positive risk when FMR exceeded 1. The conclusions of the before and after graphs were in agreement.

Figure 4. Interpretability of the LR model. (A) Bar chart of feature importance of all variables. (B–C) SHAP summary chart of continuous and categorical variables. (D) Contribution of single factor to the outcome. aptt, activated partial thromboplastin time; alb/glo, albumin and globulin ratio; alt, alanine aminotransferase; anticcp, anti-cyclic citrullinated peptide; ast, aspartate aminotransferase; bmi, body mass index; bun, blood urea nitrogen; cdai, clinical disease activity index; crp, C-reactive protein; das, disease activity score; esr, erythrocyte sedimentation rate; FMR, fat-to-muscle ratio; haq, health assessment questionnaire; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MAP, mean arterial pressure; pt, prothrombin time; rf, rheumatoid factor; sdai, simplified disease activity index; SHAP, Shapley Additive Explanations; vas, visual analogue scale; VLDL, very low-density lipoprotein.

Sensitivity analysis based on variable stratification

We added the subgroup analysis to validate possible group interaction effects for the categorical variables in this model and subsequently plotted RCS curves for FMR vs secondary OP based on the covariates with significant interaction effects. We found marked intergroup differences by various RA activity indices and gender (online supplemental figure S5). Dose-effect analysis revealed significantly different RCS curve profiles of FMR for OP between males and females (P for non-linear<0.05) (online supplemental figure S6).

Discussion

This study demonstrated that RA patients have significantly higher FMRs, lower skeletal muscle mass and reduced BMD compared with healthy controls. FMR was positively associated with RA disease activity and was elevated in patients with secondary OP or those receiving GCs. Machine learning models consistently identified age and FMR as important clinical indicators associated with the presence of OP in RA, with the LR model achieving the highest predictive accuracy. These findings suggest that FMR may be a practical and informative marker for identifying RA patients at higher risk of OP. Subgroup analysis and RCS curves corrected for the effect of confounding factors such as age further support the above conclusions.

Compared with prior studies, our results further validate the clinical relevance of altered body composition in RA and highlight the potential utility of FMR in OP risk stratification. By applying multiple disease activity indices and body composition metrics, we were able to characterise the relationship between inflammation, musculoskeletal health and metabolic status in RA patients with greater detail. Notably, our use of machine learning to model OP risk offers a novel approach that may complement traditional risk assessment tools.

RA is a common autoimmune disease with unknown aetiology, characterised primarily by chronic polyarticular synovitis and extra-articular lesions. It is widely recognised in clinical practice that the progression of RA patients is often accompanied by changes in body composition, such as the loss of skeletal muscle and the increase of fat content of whole body, and changes in height and weight, which may be related to the overproduction of the pro-inflammatory cytokine TNF-α.²⁰ BMI is the classical indicator for evaluating whether the human body is functioning well, but it is not very reliable to use BMI alone to measure changes in body composition caused by some specific chronic diseases.²¹ Given recent advancements in body composition research, we focused on the FMR as a new variable, as it influences the progression and prognosis of various diseases, including cancer,²² Alzheimer’s disease,²³ diabetes mellitus²⁴ and abdominal trauma²⁵ and so forth. Prior studies have reported that RA patients often exhibit increased fat mass and reduced muscle mass due to chronic inflammation and reduced physical activity, contributing to higher rates of sarcopenia and OP. Roos et al studied the body composition of 54 postmenopausal women with RA and 86 healthy women; they found that the thigh muscle area was reduced and the fat area of the thigh and forearm was significantly increased in the RA group compared with the healthy control group, aligning with our observations of elevated FMR in RA patients.²⁶ We collected the data of body fat and skeletal muscle from the RA group and the normal control group, respectively, and found that the RA patient group had significantly higher TFM and FMR than those of the control group, while TSM was significantly lower than that of the control group after statistics. Son et al conducted multivariate regression analysis on 335 RA patients and found a significantly positive correlation between DAS28 score and body fat mass in female patients. The study also concluded that the indicators of RA disease activity were higher in obese patients than in non-obese patients, supporting our finding that FMR correlates with CDAI and SDAI scores.²⁷ Unlike earlier studies, we focused on the relationship between FMR and RA disease activity. The results support the previous hypothesis that FMR is correlated with many RA inflammatory indicators and are in line with the current international mainstream view.

Clinically, RA is often accompanied by bone destruction and skeletal muscle atrophy.²⁸ OP is a chronic metabolic bone disease that results in decreased bone density and quality and disrupts bone microarchitecture, increasing bone fragility and the risk of fractures.²⁹ Theaner et al found that by Cox regression analysis, compared with the normal population, both men and women with RA had a higher risk of OP-related fractures, and male RA patients had a particularly high risk of hip fractures.³⁰ In this study, we found that the total hip BMD and the mean BMD of the lumbar spine 1–4 were significantly lower in the RA group compared with the normal group, while machine learning showed that the duration of RA might be a risk factor for patients to complicate OP.

Identifying factors or indicators associated with concomitant OP in RA patients is important for clinicians to establish new OP prediction models. Recently, the link between FMR and OP has garnered attention. Gong et al found that elevated serum leptin levels were a risk factor for RA-induced OP based on LR analysis.³¹ Studies have demonstrated that adipocytes have a significant endocrine function in addition to their role in energy storage; they can secrete leptin, lipocalin, resistin and other factors, which influence bone metabolism by regulating the balance of OPG (osteoprotegerin) and RANKL (receptor activator of nuclear factor κB ligand).³² Sarcopenia, marked by skeletal muscle loss and decreased strength and function, shares similarities with OP as it represents reduced muscle and bone mass, respectively.³³ The mechanism of sarcopenia inducing OP can be explained by the release of interleukin 6 (IL-6) from skeletal muscle cells, which is responsible for the extensive production of osteoclasts by inducing the release of NF-κB receptor activator (RANK) from osteoblasts and increasing the expression of its ligands located on osteoclasts.³⁴ Conversely, osteoblasts can release osteokines such as osteocalcin and prostaglandin E2 to activate p38MAPK, GPRC6A-ERK1/2 and other signalling pathways to promote the myogenic differentiation of C2C12 cells,³⁵ which is not difficult to explain that decreased bone mass and enhanced bone brittleness are often accompanied by a reduction in muscle mass and muscle strength. Thus, sarcopenia and OP often coexist and are defined by sarcopenia-osteopenia.³⁶ Therefore, the risk of OP and osteochondrosis is theoretically increased in higher FMR populations. To note, as we did not adopt formal diagnostic criteria of sarcopenia outlined in recent guidelines including handgrip strength, the diagnosis of sarcopenia is not included in this study. We included representative disease-related indicators for establishing predictable tools using machine learning. LR stood out among five algorithms, showed satisfactory calibration and discrimination, as well as excellent clinical utility. SHAP plots indicated that FMR was the second most significant factor for secondary OP after age, which was corroborated by subsequent univariate analyses. Results of subgroup analysis suggested a possible interaction effect of FMR with OP among RA patients with different CDAI/SDAI and different gender. Contrary to the indicators that respond to the disease severity, the direction of the effect of FMR on OP between male and female patients seemed to be opposite; FMR showed a positive contribution to OP in females, whereas the risk of FMR to OP in males might be abolished. Indirect corroboration was also provided by RCS curves, which showed an inverse L-shaped trend in the dose-effect relationship between FMR and OP in females, which that had a trend was not significant in males.

Recent works by Weber et al³⁷ and Baker et al³⁸ have emphasised the importance of adjusting appendicular lean mass for fat mass to better capture clinically relevant deficits in muscle mass. They proposed a fat-adjusted lean mass index, derived using regression residuals from sex- and race-specific National Health and Nutrition Examination Survey data, which impacts the definition of lean mass deficits. Their approach suggests that raw FMRs may overlook sarcopenia in obese individuals or misclassify leaner patients. This has important implications for FMR-based risk prediction and supports refinement of composite indicators of body composition. However, we did not use DXA to measure FMR. A strength of our study is the use of multifrequency BIA to assess body composition, which is a non-invasive, rapid and accessible tool suitable for large-scale clinical screening. BIA provides estimates of fat mass, muscle mass and derived indices such as FMR with minimal burden to patients and has been validated in various populations. However, its accuracy may be affected by factors such as hydration status, acute inflammation and disease-related oedema, which are common in RA and may introduce measurement variability. Compared with DXA, the reference standard for body composition, BIA has slightly lower precision and cannot differentiate regional fat or muscle distribution. Despite these limitations, BIA remains a practical and cost-effective method for evaluating body composition in routine rheumatology practice. Other emerging methods such as CT and MRI provide high resolution but are impractical for routine use. Thus, FMR measured via BIA remains a feasible method. Applied to real-world clinical practice, incorporating FMR assessment into routine rheumatologic evaluation could help clinicians detect unfavourable body composition patterns early, particularly in patients on long-term GC therapy or with high disease activity, and thus enhance risk stratification without expensive imaging or laboratory tests.

Inevitably, there are some limitations in the research approach of this article. First, the RA patients we selected as the test group were all from the same hospital, and the data for the control group were also obtained from the physical examination centre of this hospital, which may lead to the biased results. Second, the number of RA patients enrolled was far more than the normal population, which may overestimate the positive rate of the results. Third, there were far more women than men enrolled with RA, although this is consistent with the epidemiology of the disease, the applicability of the results of this paper to male patients remains questionable, and further analysis needs to be conducted with the subsequent inclusion of more male cases. Despite these limitations, this study contributes valuable insights into the relationship between FMR, RA and OP.

Conclusion

FMR is significantly associated with RA disease activity and may serve as a useful indicator of disease burden and altered body composition. FMR is also an important risk factor for RA patients with OP, which has superiority in the predicting of secondary OP. Clinicians should monitor FMR and assess OP risk in RA patients to better manage the disease and prevent comorbidities. This metric also provides a more comprehensive clinical evaluation and supports diagnostic and prognostic adjustments, although its diagnostic performance needs to be examined in prospective cohorts.

Data availability statement

Data are available upon reasonable request.