Investigation of the Trajectory of Muscle and Body Mass as a Prognostic Factor in Patients With Colorectal Cancer: Longitudinal Cohort Study

Background: Skeletal muscle and BMI are essential prognostic factors for survival in colorectal cancer (CRC). However, there is a lack of understanding due to scarce studies on the continuous aspects of these variables. Objective: This study aimed to evaluate the prognostic impact of the initial status and trajectories of muscle and BMI on overall survival (OS) and assess whether these 4 profiles within 1 year can represent the profiles 6 years later. Methods: We analyzed 4056 newly diagnosed patients with CRC between 2010 to 2020. The volume of the muscle with 5-mm thickness at the third lumbar spine level was measured using a pretrained deep learning algorithm. The skeletal muscle volume index (SMVI) was defined as the muscle volume divided by the square of the height. The correlation between BMI status at the first, third, and sixth years of diagnosis was analyzed and assessed similarly for muscle profiles. Prognostic significances of baseline BMI and SMVI and their 1-year trajectories for OS were evaluated by restricted cubic spline analysis and survival analysis. Patients were categorized based on


Introduction
Colorectal cancer (CRC) is the third most common cancer, accounting for approximately 10% of newly diagnosed cancers and cancer-related mortality worldwide [1]. Epidemiologic studies indicate that increasing age and resource-rich countries are associated with CRC development [2]. Modifiable lifestyle factors such as smoking, alcohol intake, physical activity, and obesity possibly impact CRC development [3,4].
Given the rapid global increase in obesity in recent years, understanding the effects of obesity on cancer outcomes is important [5,6]. Previous studies suggest that increased BMI is associated with poor prognosis and resistance to anticancer treatment [7][8][9]. Patients with stage II or III disease with high BMI at the time of diagnosis had decreased overall survival (OS) and disease-free survival [10]. Moreover, studies have described a relationship between muscle and CRC prognosis, suggesting that low muscle mass (ie, sarcopenia) is a prognostic factor for poor survival, increased postoperative complications, and decreased treatment response [11,12]. Despite the well-established prognostic impact of baseline profiles of obesity and muscle mass, personalized risk assessments or intervention decisions regarding these body compositions are limited in clinical practice.
Previously, skeletal muscle was manually measured using computed tomography (CT) images at the level of the third lumbar spine vertebra (L3), which includes an intense localization procedure and inevitable operator errors in the manual identification of the L3 vertebrae [13,14]. Recently, the medical application of artificial intelligence (AI) using machine and deep learning algorithms has increased, enabling the resolution of medical problems with ease of use, robustness, and precision [15]. Moreover, studies have demonstrated that automated CT imaging evaluations using deep learning models are feasible for extracting complicated body composition parameters, and the prognostic effects of these measurements on mortality in CRC have been confirmed [16][17][18]. Although previous automated body profile evaluations have partially fulfilled the gap for clinical application, the evaluation of adiposity and muscularity is limited to one time point, and thus, not considered alternatively during various treatment or interventions.
To comprehend the natural course of adiposity and muscularity of patients with CRC, we continuously tracked skeletal muscle and BMI profiles over 1 year and evaluated their representativeness for longer prospective profiles. Moreover, we evaluated the prognostic effects of the trajectories of changes in skeletal muscle and BMI to identify their longitudinal effects on CRC prognosis and demonstrated predicted mortality risks for each body composition trajectory, specified by initial status and changes in BMI and muscularity. Our study highlights the continuous characteristics of muscle and BMI in OS and supports the clinical applicability of artificial intelligence-powered automated evaluation for CRC risk modification and management.

Study Population and Data Collection
A total of 4056 patients with newly diagnosed CRC were enrolled in the Yonsei Cancer Registry Database between January 1, 2010, and September 30, 2020. Patients with abdominal CT images and BMI information up to 1 year after diagnosis were eligible. CT images within 1 year of diagnosis were obtained from the medical database of Severance Hospital. BMI was measured as the patient's weight in kilograms divided by the square of the patient's height in meters (kg/m 2 ). Skeletal muscle volume index (SMVI) outliers and SMVI change ratios were detected using median absolute deviation and were excluded. Demographic factors, such as age, sex, weight, height, variables related to diagnosis, progression, the treatment of CRC, and the date of death or follow-up loss, were collected.

Automated CT-Derived Skeletal Muscle Mass Measurement
Using the UNet architecture-based automated skeletal muscle measurement algorithm proposed by Islam et al [19], skeletal muscle was analyzed with a series of axial CT images. The overall process of muscle assessment is summarized in Figure  1A. Automated CT-derived skeletal muscle mass measurement involved three steps: (1) axial CT images within 2.5 mm superior and inferior to L3 level were detected, (2) areas of the muscle were calculated, and (3) L3-level muscle volume was calculated. SMVI was defined as L3-level muscle volume divided by the square of the patient's height in meters (cm 3 /m 2 ). The process of skeletal muscle segmentation and patient classification according to SMVI patterns. Axial CT image series were inputted (left). The blue area indicates the segmented muscle area and the volume of the muscle with 5-mm thickness was calculated (middle). Patients were classified into SMVI decreased, steady, and increased groups with the SMVI change ratio (right). (B) The process of patient classification according to 1-year BMI patterns. Lines indicate the 1-year BMI pattern of individual patients. The centers of 3 clusters by k-means clustering are presented. The 1-year BMI trajectories of the patients are demonstrated within each BMI pattern. CT: computed tomography; DICOM: Digital Imaging and Communications in Medicine; L3: third lumbar spine vertebra; SMVI: skeletal muscle volume index.

Patient Classification According to SMVI and BMI Patterns
Baseline SMVI was defined as the mean SMVI within 3 months after diagnosis. SMVI change ratios were calculated as the difference between the baseline and last SMVI values. SMVI patterns were divided into 3 groups according to the SMVI change ratio: below −5% as the decreased group, over 5% as the increased group, and between −5% to 5% as the steady group ( Figure 1A). Patients' 1-year BMI patterns were defined as the trajectory of the 12 monthly mean BMI values within 1 year after diagnosis. Missing monthly mean BMI was imputed assuming a linear change of monthly BMI. Using the k-means clustering method, patients were divided into 3 BMI pattern groups: decreased, steady, and increased ( Figure 1B).

Statistical Analysis
Patient characteristics were compared between BMI pattern groups. The normality of variables was determined using the quantile-quantile plot. Parametric and nonparametric continuous variables were compared using ANOVA and Kruskal-Wallis tests, respectively. Categorical variables were compared using the chi-square test. The effect sizes for ANOVA, Kruskal-Wallis test, and chi-square test were individually calculated using partial eta squared [20], eta squared [21], and Cramer V [22] The correlation between the BMI of the first, third, and sixth years after diagnosis was evaluated with Pearson correlation, among patients with available BMI data for a sufficient period. Identical correlation analysis was conducted with SMVI among 300 random patients with available CT images for a sufficient period. The association of 6-year overall mortality with baseline and change ratio of both SMVI and BMI was analyzed using restricted cubic spline analysis. The nonlinearity was assessed by Wald statistics. Patients' 6-year OS was analyzed for survival analysis, and survival curves were plotted using the Kaplan-Meier method and compared between SMVI pattern groups using the log-rank test. Cox proportional hazard regression was conducted to predict 6-year OS. Similar survival analyses were performed for each BMI pattern group. All statistical analyses were conducted using JupyterLab (version 1.2.6; Project Jupyter) [24], Python (version 3.6.8; Python Software Foundation) [25], and R (version 4.1.3; R Foundation for Statistical Computing) [26]. Two-sided P values of <.05 were considered statistically significant.

Risk Assessment and Heat Map Generation
Three hazard ratio (HR) heat maps were designed with HRs of the survival analysis, with the seaborn package (v.0.11.2) [27]. One heat map demonstrated the HRs of 15 baseline conditions, defined by 5 baseline BMI and 3 baseline SMVI groups. The other heat map demonstrated the HRs of 9 body composition change patterns, defined by 3 BMI patterns and 3 SMVI pattern groups. Another heat map demonstrated the HRs of each baseline condition specified by the body composition change groups.

Ethics Approval
The study design was approved by the institutional review board of Severance Hospital, Seoul, South Korea (IRB 4-2020-1304). The need for informed consent was waived by the ethics committee, as this study used routinely collected log data managed anonymously at all stages, including data cleaning and statistical analyses. The study protocol was performed following the guidelines of the International Conference on Harmonization of Good Clinical Practice, the Declaration of Helsinki, and relevant legislation for observational studies.

Correlations Between First-Year and Subsequent Profiles of Obesity and Muscle
Among the 4056 patients, BMI data were available for 3217 patients 3 years after diagnosis and for 1318 patients 6 years after diagnosis. Among these patients, BMI in the first and third years showed a statistically significant positive correlation (P<.001), whereas the trajectory groups were evenly distributed (Figure 2A). BMI in the first year was also significantly correlated with that in the sixth year with a nonspecific distribution pattern of trajectory groups (P<.001; Figure 2B). Among the randomly selected 300 patients for correlation analysis between first-versus third-year SMVI, data from 278 patients were available for CT analysis, and the correlation was found to be positive (P<.001; Figure 2C). Among the randomly selected 300 patients for correlation analysis between firstversus sixth-year SMVI, 269 patients were analyzed, and the correlation was positive (P<.001; Figure 2D).

Relationship of the Baseline and Change Ratio of Obesity and Muscle Mass With Mortality
According to restricted cubic spline analysis, baseline BMI and mortality risk exhibited a U-shaped relationship (P<.001; Figure  3A). The lowest mortality risk was observed among patients with a normal BMI. Baseline SMVI exhibited an inverse relationship with the risk of death, with an L-shaped pattern, whereby patients with the lowest muscle mass had the highest risk (P<.001; Figure 3B).
Change ratios of BMI demonstrated an inverse correlation with overall mortality (P<.001; Figure 3C), with the lowest risk observed for an increased BMI of >20%. Patients with BMI loss >20% had a 65% increased risk of death. Similarly, an increase in SMVI over time was associated with improved OS (P<.001; Figure 3D). A decrease in SMVI by 20% increased the mortality risk by 78%, suggesting that the restoration of body mass and muscle is critical for improving OS. Red lines represent restricted cubic spline curves, and black dashed lines represent 95% CIs. The reference is the median of each variable. Adjusted variables were age at diagnosis (above or below 65 years); sex; stage; primary cancer location (colon or rectum); histology (adenocarcinoma or others); recurrence or metastasis; the administration of surgery, chemotherapy, or radiotherapy; baseline BMI (underweight, normal, preobese, obesity stage 1, or obesity stages 2-3); baseline SMVI (low, normal, or high); and patterns of BMI and SMVI (decreased, steady, or increased). Baseline BMI, baseline SMVI, BMI pattern, and SMVI pattern were excluded from an adjustment in the analysis against baseline BMI, baseline SMVI, BMI change ratio, and SMVI change ratio, respectively. SMVI: skeletal muscle volume index.

BMI and SMVI as Prognostic Factors for Patient Outcomes in Multivariate Analysis
The trajectory of the decreased SMVI group was associated with shorter OS ( Figure 4A and Multimedia Appendix 2).
Based on the trajectories of BMI changes affecting OS, we performed a subgroup analysis according to 3 BMI trajectories ( Figure 4B-D Figure 4C and Multimedia Appendix 4). In the increased BMI group, an increase in SMVI was associated with better OS (HR 0.73, 95% CI 0.55-0.97; P=.03), suggesting that increased body mass, predominantly due to skeletal muscle mass, positively impacted cancer survival ( Figure 4D and Multimedia Appendix 5). Cyan, magenta, and yellow lines indicate decreased, steady, and increased SMVI groups, respectively. The hazard ratio and 95% CIs are shown on the right. Red and blue marked variables are statistically significant (P<.05) with hazard ratios of >1 and <1, respectively. Variables denoted with stars showed statistically significant P values (<.05). Adjusted variables were age at diagnosis (above or below 65 years); sex; stage; primary cancer location (colon or rectum); histology (adenocarcinoma or others); recurrence or metastasis; the administration of surgery, chemotherapy, or radiotherapy; baseline BMI (underweight, normal, preobese, obesity stage 1, or obesity stages 2-3); baseline SMVI (low, normal, or high); and patterns of BMI and SMVI (decreased, steady, or increased). SMVI: skeletal muscle volume index.

HR Heatmap Representing Associations Between BMI, SMVI, and OS
In consideration of baseline BMI and SMVI at the time of diagnosis, the highest mortality risk was observed in patients with high body mass (BMI >30 kg/m 2 ) and low SMVI (HR 2.09; Figure 5A) trajectories of BMI and SMVI revealed that increased body mass and muscle mass were associated with the lowest mortality risk (HR 0.68; P=.001; Figure 5B), whereas decreased body mass and muscle wasting were associated with the highest mortality risk (HR 1.73; P<.001; Figure 5B and Multimedia Appendices 6 and 7).
Finally, we generated an HR heat map depicting the baseline and trajectories of both BMI and SMVI ( Figure 5C). High baseline BMI (>30 kg/m 2 ) was associated with increased mortality risk regardless of baseline muscle mass or muscle changes ( Figure 5C). High baseline SMVI was associated with improved survival in patients with a BMI range from normal to obesity stage 1. There was a trend toward improved OS in patients with increased muscle without BMI loss. Figure 5. Predicted mortality risk heat map representing associations between BMI, skeletal muscle volume index (SMVI), and overall survival (OS). Patient groups of each heat map were classified by (A) baseline status classified by 5 baseline BMI and 3 baseline SMVI groups, (B) pattern groups classified by 3 BMI and 3 SMVI pattern groups, and (C) all 4 dimensions: baseline BMI and SMVI profiles and trajectories of BMI and SMVI. The predicted mortality risk of each specific patient group is described in each corresponding square, and colors represent the natural logarithm of the predicted mortality risk.

Principal Findings
This study evaluated the characteristics of baseline BMI and SMVI and their 1-year trajectories along with the prognostic impacts of these 4 parameters on OS in patients with CRC. The BMI and SMVI profiles of the first year after diagnosis were highly correlated with the third-and sixth-year profiles, implying that these are surrogate indicators for representing subsequent body composition profiles. Moreover, our survival analysis results indicated that high muscularity positively impacted OS, whereas both depletion and excess adiposity adversely impacted survival. Alterations in adiposity and muscle mass significantly affected 6-year overall mortality. Reciprocal compensation of these 2 factors indicated that changes in BMI had a superior prognostic impact than changes in muscle.

Comparison With Prior Work
According to our restricted cubic spline analysis, baseline SMVI did not exhibit a nonlinear relationship with mortality, whereas baseline BMI demonstrated a U-shaped relationship with mortality. In addition, high muscle mass at baseline exhibited a protective effect against survival, and extreme BMI (underweight or severe obesity) predicted lower survival. These findings suggest that muscularity at the time of diagnosis may have protective effects, whereas both adiposity depletion and excess may have disadvantages for survival. The results of the survival analysis of our study also support the independent prognostic effects of body composition profiles at the time of diagnosis. The mechanisms underscoring the impact of body composition on CRC mortality at the time of diagnosis have yet to be elucidated because completely eliminating methodological biases is challenging [8,9]. A widely accepted explanation is that an adequate amount of adiposity acts as a metabolic reservoir and allows patients to tolerate cachectic situations during cancer progression and treatment [29]. However, excess body fat is correlated with higher mortality via various pathways. Dysregulated adipose tissue can increase inflammatory adipokines, leading to systemic inflammation and a tumor-friendly microenvironment [30], thereby exacerbating catabolic pathways [31]. Patients with low muscularity and high adiposity were associated with higher inflammatory-related serum proteins and cytokines [17]. In addition, the specification of adipose tissue into subcutaneous, visceral, and intra-or intermuscular adipose tissue revealed distinct relationships with CRC mortality: subcutaneous fat exhibits a U-shaped association, whereas visceral and intermuscular fats exhibit a positive relationship, implying context-dependent roles of fat [9].
In this study, an increase in muscle mass or BMI demonstrated higher OS, highlighting the beneficial effects of increasing muscle and body mass. Survival analysis within subgroups of each BMI trajectory group showed a combined prognostic influence of muscularity and adiposity. HR results of the survival analysis conducted within steady and increased BMI groups described that survival improved in the order of increased, steady, and decreased SMVI. However, within the decreased BMI group, an increase in SMVI was associated with poor survival compared with steady SMVI, which contradicts general findings. Although several findings were statistically insignificant, this implies that the protective role of increased muscle differs regarding the characteristics of BMI changes, since the prognostic effects of changes in muscle and adiposity may reciprocally compensate. In addition, according to the survival analysis considering all 9 patterns of BMI and SMVI change, patients in the increased SMVI with decreased BMI group had statistically significant increased mortality risk. The superior decrease of adiposity that leads to decreased BMI despite the improvement of muscularity may override the protective effects of increased muscle mass.
Although BMI is known to have limitations in the direct representation of adiposity [9], BMI profiles within 1 year could be an efficient parameter for predicting the prognosis of patients with CRC in most primary care settings. First, continuously repeated measurements of BMI can be performed, which enables longitudinal evaluation with ease. Second, a 1-year BMI profile highly correlates with future BMI profiles. Third, the trajectory of BMI is a critical independent prognostic factor, where it is a parameter containing longitudinal information of multiple compositions. Adiposity-related [17] and muscle-related indexes such as SMVI, muscle-to-bone ratio, or skeletal muscle radiodensity [16,18] are also essential prognostic factors, yet frequent assessments are limited since these parameters are products of occasional CT imaging studies that depend on cancer treatment progress, patient's medical condition, health insurance, etc. Thus, synergizing CT-driven information and BMI data will promote the future clinical application.
Previous studies support the association between physical activity and cancer prognosis. A prospective cohort study revealed that sedentary behavior is associated with cancer mortality risk and replacing sedentary time with physical activities may improve OS [32]. According to a meta-analysis of nonrandomized trials, higher-intensity physical activity was significantly associated with lower mortality risk [33]. In addition, resistance training in patients with cancer resulted in improvement in muscle strength, body composition, and benefits for cancer survival [34,35]. Exercise is safe and feasible for patients with CRC and results in improvements in various health-related outcomes [36]. Collectively, appropriate interventions through physical activities could improve the outcomes of patients with CRC by improving body composition profiles. Studies evaluating the role of an exercise intervention to improve OS are rare; however, a phase-2 randomized clinical trial proved that high-intensity interval training delayed prostate cancer progression [37]. Thus, further studies evaluating whether exercise intervention improves long-term outcomes should be conducted.

Strengths
Our study has the following strengths. We analyzed a large size of patients and assessed various natural courses of patients with CRC who did not receive additional interventions besides clinical practices. To comprehend the role of adiposity and muscularity in CRC progression, understanding its alternating nature is necessary for evaluating time-dependent prognostic impacts. Therefore, longitudinal patterns of BMI and muscle were analyzed simultaneously. Given that adiposity and muscle commonly change during cancer treatments [9], we longitudinally analyzed BMI and muscle for up to 1 year, and our study shows that the 1-year profile of 2 variables can aptly represent the course over 6 years. As these are proven to be independent prognostic factors, analyzing the 1-year course of adiposity and muscularity will be an efficient and effective method to determine the mortality risk. As a part of continuous body composition analysis, the deep learning-based assessment was implemented to minimize manual processes and enable rapid and accurate analysis of a large number of CT images. The application of automated analysis processes in clinical practice will highly promote the improvement of personalized care and risk management.

Limitations
Our study has several limitations. We evaluated the 6-year OS as a survival outcome using 1-year trajectories of BMI and muscle mass. However, according to the correlation analysis, trajectory analysis up to 1 year was sufficient for representing the courses beyond a year. Since physical examinations and imaging studies are frequent during the first year as a part of the diagnosis or treatment process, we assumed that 1 year was sufficient for classifying body composition profiles. Prolonged longitudinal trajectorial analysis may provide additional insight into the longer-term effects of adiposity or muscle in cancer survival, thereby facilitating personalized risk assessment and management.

Conclusions
In conclusion, our findings demonstrated the natural courses of BMI and muscle and highlight the 1-year trajectory as a surrogate indicator for representing further progress, wherein deep learning-based automated muscle assessment enabled longitudinal analysis of muscle. Thus, this study can be ground evidence for further research regarding prognostic improvements of patients with cancer by interventions that improve body composition profiles. In addition, our results provided crucial insights into the prognostic roles of adiposity and muscle mass in CRC survival. Body composition profiles at the time of diagnosis and alterations of these parameters are independent prognostic factors of CRC survival. The combination of BMI, which is frequently gathered but crude, with muscle-related parameter, which is automatically analyzed but occasionally obtained, could be a precise risk assessment tool. For further clinical application, supporting studies evaluating the natural course of patients with CRC will be necessary and more personalized prognosis prediction methods are required. To fully understand the role of adiposity and muscularity in cancer progression, impacts on other clinical outcomes should also be studied.