Predicted lean body mass trajectories, and cancer risk and cancer‐specific and all‐cause mortality: A prospective cohort study

Abstract Background Although many studies have investigated the association between body composition, cancer risk and mortality, predicting these risks through a single body composition measurement undoubtedly increases the limitations of the study. Few studies have explored the association between the trajectory of changes in body composition and the risk of cancer and death. We aimed to explore the association of predicted lean mass trajectories with cancer risk, cancer‐specific mortality and all‐cause mortality. Methods The participants in this study were all from the Kailuan cohort, a prospective, periodic, resurvey cohort study initiated in 2006. Latent mixture modelling was used to identify predicted lean mass trajectories for 2006–2010. The hazard ratios (HRs) and 95% confidence intervals (95% CIs) of the Cox proportional hazard models were used to describe the association between predicted lean mass trajectories and cancer risk and cancer‐specific and all‐cause mortality during follow‐up (2010–2021). Results A total of 44 374 participants (average age, 53.01 ± 11.41 years, 78.99% men and 21.01% women) were enrolled in this study. Five distinct trajectories were identified: low‐stable (n = 12 060), low‐increasing (n = 8027), moderately stable‐decreasing (n = 4725), moderately stable‐increasing (n = 8053) and high‐stable (n = 11 509). During the 11‐year follow‐up period, 2183 cancer events were recorded. After adjusting for age, predicted fat mass in 2010, sex, BMI, sedentary, physical activity, smoke, alcohol use, salt consumption, high‐fat diet, high‐sensitivity C‐reactive protein, serum creatinine, family history of tumour, hypertension, diabetes mellitus, compared with the low‐stable group, participants in the low‐increasing group (HR = 0.851, 95% CI, 0.748–0.969), moderately stable‐increasing group (HR = 0.803, 95% CI, 0.697–0.925) and high‐stable group (HR = 0.770, 95% CI, 0.659–0.901) had a lower cancer risk, but not in the moderately stable‐decreasing group (HR = 0.864, 95% CI, 0.735–1.015). Compared with the low‐stable group, the risk of cancer‐specific mortality was reduced by 25.4% (8.8–38.9%), 36.5% (20.3–49.4%) and 35.4% (17.9–49.2%), and the risk of all‐cause mortality was reduced by 24.2% (16.9–30.8%), 37.0% (30.0–43.2%) and 47.4% (41.0–53.1%) in the low‐increasing, moderately stable‐increasing group and high‐stable groups, respectively. Conclusions Predicted lean mass trajectories may be closely associated with cancer risk and cancer‐specific and all‐cause mortality. Regular monitoring of body composition is necessary.


Introduction
Cancer remains a major public health threat globally, and it is extremely important to identify the determinants of cancer incidence. 1Numerous studies have investigated the relationship between body mass index (BMI) and cancer risk, and some studies have shown that an increase in BMI is positively correlated with the incidence of colon and liver cancers, whereas it is negatively correlated with lung cancer risk. 2,3owever, there is controversy over whether there are still differences in cancer risk among different races and sexes at the same BMI level.Compared with a BMI of 22 kg/m 2 , the risk ratio of cancer for white adults with a BMI of 35 kg/m 2 is 1.83, whereas that for black adults is 0.89. 4 Even among participants of the same race with the same BMI, there are still significant differences in cancer risk. 5Although BMI is convenient to measure and use, it lacks a detailed assessment of body composition because an increase in either lean mass or fat mass (FM) can cause an increase in BMI.Therefore, relying solely on BMI to judge the risk of tumour occurrence may increase errors.Song et al. showed that increased FM is associated with a higher risk of oesophageal and colon cancer. 6Another prospective study showed that FM versus fat-free mass (FFM) based on bioelectrical impedance analysis influenced the risk of 16 common cancers, of which FFM was a stronger predictor. 7Even more so, studies have illustrated that, based on anthropometric equations, higher lean mass rather than FM is associated with a decreased risk of cancer, 8 indicating the importance of body composition measurements.However, all studies share the common problem of judging the occurrence of tumours years or even decades later from only one body composition measurement.The changes in body composition are highly influenced by various factors, with one of the most significant being aging.As individuals age, mitochondrial function and the synthesis of bodily proteins weaken, while synthetic metabolism resistance increases, consequently, sarcopenia, characterized by the loss of lean mass and a decline in muscle function, has become one of the common chronic diseases associated with aging. 9,10Furthermore, research indicated that individuals with metabolic disorders such as diabetes experience greater muscle loss compared with those without such conditions. 11Therefore, relying on a single measurement of lean mass during youth or a disease-free period to predict outcomes over 5 or even 10 years undoubtedly diminishes the rigour and credibility of the study, as we are aware that aging and metabolic changes are ongoing processes.Therefore, this study aimed to construct trajectories of change in lean mass through a prospective cohort with periodic resurvey and explored the associations between the trajectories and participants' cancer risk, cancer-specific and all-cause mortality.

Study design and participants
All participants were from the Kailuan cohort, an ongoing prospective study in Tangshan, China, as previously described. 12ince 2006, Kailuan General Hospital and its 11 affiliated hospitals have conducted physical examinations and annual follow-ups of Kailuan Group employees.The physical examination included laboratory tests, clinical measurements and questionnaires, and the follow-up information included outpatient follow-up reports, electronic medical records, death certificates from the Provincial Vital Statistics Offices (PVSO), and data entries in the Tangshan Medical Insurance System and Kailuan Social Security Information System.A total of 101 510 participants (81 110 men and 20 400 women) aged 18-98 years underwent their first physical examination and signed a written informed consent in 2006-2007.
The predicted lean mass trajectories were constructed from three sets of physical examination data from 2006, 2008 and 2010 and used to predict the cancer risk of participants from 2010 to 2021.Therefore, patients with the following conditions were excluded: (1) who did not continuously participate in physical examinations from 2006 to 2010; (2) lacked cancer occurrence and survival data or patients with multiple cancers; (3) lacked sex, height, weight, waist circumference and age information; (4) presence or history of cancer; and (5) lacked covariate information.Participant recruitment is shown in Figure 1.

Exposure and covariates
Predicted lean mass was calculated based on a sex-, age-and race-specific anthropometric equations that are widely validated and used and has high predictive power for both fat mass (R 2 = 0.90) and lean mass (R 2 = 0.91). 13Equations are presented in Methods S1.
Covariates recorded in 2010 included age, sex, BMI, sedentary lifestyle, physical activity (PA), smoking, alcohol consumption, salt intake, high-fat diet, high-sensitivity Creactive protein (hs-CRP), serum creatinine (Scr), alanine aminotransferase (ALT), and family history of cancer, hypertension and diabetes mellitus, in addition to adjusting for participants' 2010 predicted fat mass to avoid the effects of fat.All covariate definitions are presented in Methods S2.

Outcome
The primary outcome was cancer occurrence.The diagnosis of cancer was based on pathological, imaging, and clinical diagnoses in electronic medical records.In addition, to obtain cancer incidence information accurately, we integrated data from the PVSO, the Tangshan Medical Insurance System, and the Kailuan Social Security Information System.Diagnoses were coded using the International Classification of Diseases, 10th Revision (ICD-10) (Methods S3).The follow-up time was determined from the date of completion of the investigation in 2010 until the first occurrence of cancer, death, or the last follow-up (31 December 2021), whichever occurred first.
The secondary outcomes were all-cause and cancer-specific mortality, defined as death from any cause, or cancer-related death, or the status at last follow-up (31 December 2021).

Statistical analysis
All statistical analyses were conducted through SAS 9.3 (SAS Institute) or R version 4.2.2.Two-sided P < 0.05 was considered statistically significant.
In our previous study, we found a significant sex-related difference in lean mass. 10Therefore, in this study, we used the sex-specific quintile of the predicted lean mass to build the trajectory model.Otherwise, women would be in the low lean mass trajectory group, while men would account for the main component in the high lean mass trajectory group, resulting in a sex bias.Latent mixture modelling was used to identify shared similar trajectories, which caused participants in the same group to have a similar trajectory of lean mass changes.This model was implemented using the PROC TRAJ program in SAS 9.3. 14The construction of the model was gradual.We first developed a model containing one trajectory, followed by two, three, four and up to five trajectory modes.The Bayesian information criterion (BIC) was used to evaluate the model fitness, with the smallest BIC indicating the most suitable model.Finally, we compared models with different functional forms and optimized them based on the significance of the cubic, quadratic and linear terms.
Variables conforming to a normal distribution are expressed as mean ± standard deviation (SD), and one-way ANOVA was used for inter-group comparison.Continuous variables with skewed distributions are represented as median and quartile range (IQR) and compared using a non-parametric test.Categorical variables are expressed as rates, and the chi-square test was used for comparisons between groups.Incidence rates are presented as per 1000 person-years.First, we compared predicted lean mass trajectories, predicted lean mass in 2006, predicted lean mass in 2008, and predicted lean mass in 2010 through integrated discrimination improvement (IDI) and net reclassification improvement (NRI) to determine which approach is the best for assessing patients' prognosis.After the proportional risk assumption was satisfied, we used COX analysis to describe the association between different trajectories and cancer risk, as well as all cause and cancer-specific mortality, and expressed them using hazard ratios (HR) and 95% confidence intervals (95% CI).Model 1 was adjusted for predicted fat mass only, while Model 2 was further adjusted for other covariates.In terms of digestive system cancer, we also adjusted for cirrhosis and hepatitis B. For gallbladder and extrahepatic cholangiocarcinomas, we additionally adjusted for the presence of gallstones and gallbladder polyps.For liver cancer, we additionally adjusted for cirrhosis, hepatitis B virus infection, gallstones and gallbladder polyps.These confounding factors have been shown to be related to the occurrence of cancer. 12,15,16We also conducted a subgroup analysis of specific cancer sites.Interaction analysis was performed using multiplicative interactions between the factors.We also supplemented the analysis of the association between baseline predicted lean mass in 2006 and cancer risk and used a restricted cubic spline (RCS) function to describe the nonlinear relationship between participants' predicted lean mass and cancer risk.Finally, sensitivity analysis was performed to verify the stability of the results.Sensitivity analyses excluded participants developing cancer within 1 year to avoid the possibility of causal inversion.As this was a rigorous prospective study, we excluded participants infected with hepatitis B virus, which has been proven to be closely related to digestive system cancers, and participants with a family history of cancer or abnormal Scr.At the same time, considering the potential collinearity among age, BMI and predicted lean mass, we included age 2 and BMI 2 as covariates in the model to control for the non-linear relationships with age/BMI and the outcomes of interest.During the follow-up process, covariates may change over time, and death as a competitive event may have affected the accuracy of the results.Therefore, we further adjusted for time-varying variables and repeated our analysis using the Fine and Gray model, which can explain the bias caused by death as a competing risk.

Results
A total of 44 374 adults participated in this study, with an average age of 53.01 (11.41) years, and 35 049 (78.99%) participants were men.During the 11-year follow-up period, 2183 cancer events were recorded.Based on the participants' predicted lean mass from 2006 to 2010, we identified five trajectories: low-stable (n = 12 060), low-increasing (n = 8027), moderately stable-decreasing (n = 4725), moderately stable-increasing (n = 8053) and high-stable (n = 11 509) (Figure 2).The baseline characteristics of the participants in 2006 according to the trajectory determined from 2006 to 2010 are shown in Table 1.Compared with the low-stable group, the participants in the high-stable group were younger, had a higher BMI and waist circumference, and had higher CRP, Scr and ALT levels.In addition, sedentary behaviour, regular PA, smoking, drinking, salt intake, hypertension, diabetes and other factors differed among the five groups.
Table S1 indicated that, compared with predicted lean mass trajectories, the predictive abilities of predicted lean mass in 2006, 2008 and 2010 were all worse for cancer risk, all-cause mortality and cancer-specific mortality.
At the same time, to emphasize the role of body composition, we also divided patients into four groups based on the median values of predicted lean mass and predicted fat mass (Table S6).The results indicated that compared with patients with low fat and low lean mass, participants with low fat and high lean mass had the lowest cancer risk (HR = 0.838, 95% CI, 0.740-0.949),while participants with high fat and low lean mass had the highest cancer risk (HR = 1.207, 95% CI, 1.068-1.365).
In sensitivity analyses (Table S7), similar to the primary results, the participants in the moderately stable-increasing and high-stable groups had significantly reduced cancer risk.In addition, we adjusted for the predicted lean mass in 2010, and the results remained robust (Table S7).The results of the competitive risk analysis indicated that after excluding the impact of competitive events, the risk in moderately stable-increasing and high-stable groups remained low (Table S8).

Discussion
In this large-scale prospective periodic survey study, we described five distinct trajectories of predicted lean mass based on sex-specific quintiles and found that the trajectories of predicted lean mass changes were significantly associated with cancer risk and mortality.Compared with the low-stable group, the moderately stable-decreasing, moderately stable-increasing and high-stable groups all had significantly lower cancer risks and mortality rates, especially for digestive system cancers (gastric cancer and colorectal cancer) and lung cancer.Additional analyses also showed that the baseline level of predicted lean mass was closely associated with cancer risk, and participants in quintile 5 had lower risks of digestive system and lung cancers than those in quintile 1.
Changes in anthropometric parameters, including weight, BMI, waist circumference, waist-to-hip ratio, FM, and FFM, reflect an individual's nutritional status to some extent and can, therefore, greatly predict the risk and prognosis of chronic diseases, including cancer. 17,18Similar to our partial results, some studies based on the UK Biobank have also shown that FFM is strongly associated with the risk of all cancers, including lung and digestive system cancers. 7,19,20Another study on body composition using dual-emission X-ray absorptiometry also demonstrated a significant association between FFM and the malignancy of prostate cancer. 21At the same time, there are also some heterogeneous conclusions.A cross-sectional study by Mathew et al. showed that FFM was associated with non-Hodgkin's lymphoma and melanoma but not with other types of cancers. 22However, these studies are limited, and as far as we know, the lean mass changes significantly with aging, exercise and diet. 23,24Therefore, predicting the risk and prognosis of diseases during a 10-year or longer follow-up period through a single measurement of indicators will undoubtedly increase the bias of predictive effects.Building trajectories of lean mass and exploring their associations with cancer risk and prognosis may be more reliable.
Interestingly, although the HR values of cancer risk for the other four groups were all less than 1 compared with the low-stable group, there were still differences between the groups, especially for the moderately stable-decreasing and moderately stable-increasing groups.In the initial survey (2006), the lean mass of the two groups of participants was basically the same; however, over time and with changes in lean mass, the cancer risk of the increasing group of patients significantly decreased by approximately 20%, whereas the cancer risk of the decreasing group of participants did not show any statistical difference.This trend was observed in overall cancer risk and some specific-site cancers.Even in patients whose lean mass was initially low, their subsequent cancer risk decreased once their lean mass increased over time.It is worth noting that similar trends were observed for cancer-specific mortality.Compared with the low-stable group, the moderately stable-decreasing group did not show a decrease in cancer-specific mortality, whereas the other three groups had varying degrees of risk reduction, which has not been described before.These results further demonstrate the importance of exploring the trajectories of lean mass, cancer risk and mortality.A single measurement of body composition has significance; however, periodic monitoring of lean mass is essential.
In the interaction and subgroup analyses, we found that besides factors such as age and physical activity that directly affect muscle function, there were also significant interactions between the predicted lean mass trajectory and lifestyle habits such as smoking and drinking.In the same lean mass trajectory, smokers' muscles did not seem to significantly reduce cancer risk compared with those of non-smokers.We assumed that this difference was due to differences in muscle quality.Similar to a basic study by Wang et al., the results showed that muscle strength and protein synthesis signalling were reduced in mice after cigarette smoke exposure, but muscle mass was stable. 25Further in vivo and in vitro studies revealed the role of inflammation.Impairment of muscle function may be related to the systemic release of the proinflammatory cytokines TNF-α and adipokines produced in adipose tissue, leading to mitochondrial dysfunction, decreased muscle strength and even muscle wasting. 26,27In addition, increased levels of oxidative stress are one of the underlying mechanisms for this phenomenon. 28This had also been shown in drinkers, where acute and chronic alcohol consumption can cause skeletal muscle myopathy along with impairment of skeletal muscle strength, function and fatigue resistance. 29Alcohol may affect muscle function and performance through the following aspects: First, alcohol disrupts the balance between the anabolic and catabolic pathways by affecting signalling pathways, the imbalance of which can lead to altered muscle morphology or loss of function without timely intervention and repair. 30,31Second, alcohol also reduces the ability of muscle to regenerate after injury, thus causing some irreversible damage. 32,33Finally, alcohol can also cause mitochondrial dysfunction and increased oxidative stress, and these abnormalities can delay the regeneration and recovery of muscle function and contribute to alcohol-related muscle dysfunction and metabolic pattern shifts. 32Altogether, the association of lean mass trajectories with cancer risk was influenced by unhealthy lifestyles.
This study has several limitations.First, the lean mass and fat mass in the present study were calculated using a well-recognized formula, so this may overlook the effect of muscle distribution or intermuscular adipose tissue, 34 but our results can still reflect, to some extent, how lean mass changes in individuals.Second, this equation categorizes the Asian population as 'other races', which could potentially reduce the accuracy of the calculated results.It is necessary to validate and refine the equation using large-scale cohorts.Third, although we adjusted for several confounders, there are still many potential unknowns that we did not consider, such as the type of physical activity.Aerobic and anaerobic exercise are known to be differentially effective for muscle improvement 35 ; meanwhile, special types of diet, especially protein diets, have a significant effect on lean mass and strength in adults. 36Finally, the study was limited by the cohort design, as the Kailuan cohort was dominated by male workers, so the generalizability of the results may require more cohorts and studies to validate.

Conclusions
In summary, this study suggests that the lean mass trajectory is associated with the risk of cancer, cancer-specific and all-cause mortality.Sustained increases in lean mass or maintenance of a high and stable level of lean mass may significantly reduce the risk of cancer incidence, cancer-specific and all-cause mortality.A single measurement of body composition may underestimate these risks, and long-term monitoring of body composition is particularly important for predicting participants' prognosis.Improving and optimizing lean mass and quality have significant implications for reducing the burden on cancer-related public health.

Figure 1
Figure 1 Flowchart of this study.

Table 1
Baseline characteristics of participants based on predicted lean mass trajectory pattern

Table 4
Association of predicted lean mass trajectories with cancer specific mortalityModel was adjusted for age, predicted fat mass in 2010, gender, BMI, sedentary, physical activity, smoke, alcohol use, salt consumption, high-fat diet, hs-CRP, Scr, family history of tumour, hypertension, and diabetes mellitus.
a Mortality rate were presented as per 1000 person-years.b