This study marks the first attempt to use ML models for predicting the risk of CC in stage IV CRC patients. We evaluated four ML prediction methods, utilizing patient clinical features for CC prediction. Among the models assessed, the RF model exhibited superior performance in single model prediction, achieving the highest AUC value. Additionally, we employed the SHAP method to elucidate the RF model, highlighting TC, Hb, Glu, NLR, and N stage as the most crucial features. Importantly, these variables are readily accessible from the medical records of stage IV cancer patients, providing clinicians with valuable routine data to assess these risk factors and identify patients at risk of CC.
One notable strength of our study lies in the use of SHAP values to unveil the black box of ML. While previous risk scoring models identified several risk factors, including cancer site, cancer stage, time from symptom onset to hospitalization, appearance loss, BMI, skeletal muscle index, and NLR[18], our research highlights the significance of dynamic changes in metabolic indicators such as TC, Glu, and Cr, which traditional models often overlook. In this study, we employed metabolic indicators such as TC, Hb, Glu, and Cr to assess the risk of developing cachexia. Cancer cachexia is caused by the activation of energy rich compound mobilization processes [3, 19], such as increased liver gluconeogenesis [20]. Causing impaired glucose and lipid homeostasis[21], Cancer cachexia is also associated with insulin resistance, which enhances liver glycolysis [3, 22, 23] and lipid mobilization in white adipose tissue[24]. Insulin resistance in cancer patients is characterized by decreased insulin sensitivity or impaired glucose tolerance[25], and is suspected to increase during the progression of cachexia[26, 27]. In addition, elevated levels of glucagon[28] promote liver gluconeogenesis, thereby increasing blood sugar levels in cancer cachexia[28, 29]. However, others have also pointed out that there is no change in glucose levels in cancer cachexia[22, 30], which may reflect the complex metabolic dynamics of cancer cachexia[31]. Here, we investigated fasting blood glucose levels, whose damage is the second sign of insulin resistance (unrelated to impaired glucose tolerance)[32]. Our research findings support the view that insulin resistance increases in cancer cachexia. Patients with significantly elevated plasma fasting blood glucose levels also have an increased risk of developing cachexia. However, our research findings suggest that an increase in Glu is associated with the occurrence of cachexia, rather than with cancer staging. Previous studies have also shown that high fasting blood glucose levels during cancer diagnosis are associated with poor prognosis in non-small cell lung cancer (NSCLC) patients[33], supporting the view that elevated fasting blood glucose levels are not only an indicator of metabolic disorders or insulin resistance, but also a marker of poor prognosis.
The increase in lipolysis in CC patients may be mediated by norepinephrine signals involved in the sympathetic nervous system[34, 35]. In addition, systemic inflammation associated with cachexia and/or activation of brown adipose tissue requires a high energy demand[34, 36], and brown adipose tissue can effectively remove energy rich compounds from normal physiological circulation[37]. Adipose tissue is an important regulatory factor for body composition and plays a crucial role in energy balance. White adipose tissue (WAT) and brown adipose tissue (BAT) regulate energy storage and body temperature, respectively. BAT maintains body temperature by generating heat, including increasing the expression of uncoupling protein 1 (UCP1) in adipose tissue and regulating glucose and lipid metabolism[38]. In addition, the thermogenic effect of BAT is believed to enhance resting energy expenditure and lipid mobilization[31]. The browning effect observed in metabolic disorders (such as obesity, diabetes and cancer) is proved to be harmful in CC[2]. Patients with cachexia typically exhibit elevated levels of circulating free fatty acids and glycerol, due to the extensive lipolysis of WAT by activating lipases such as hormone sensitive lipase and triglyceride lipase [39]. Therefore, the interactions between these metabolic processes may manifest as changes in plasma levels of energy rich compounds associated with cachexia[40]. The loss of skeletal muscle tissue is a key characteristic of cancer cachexia and the best studied aspect [1, 11, 41]. Muscle is the source of amino acids, which may be released for energy production during the process of catabolism[42]. Muscle homeostasis is maintained through a balance between the synthesis and degradation of muscle proteins[3]. However, when there is excessive degradation of protein and/or a decrease in protein synthesis in skeletal muscle, this imbalance can lead to muscle atrophy and the occurrence of cachexia [3, 41, 42].The SHAP summary chart indicates that changes in these metabolic characteristics are key predictive factors, indicating that as TC, Glu, and Cr increase, the risk of CC occurrence also increases. In addition, CC is associated with hyperlipidemia and decreased levels of branched chain amino acids[43]. It is currently unclear whether interventions based on these metabolic indicators can regulate CC correlation, and further research is needed.
In our study, we used hemoglobin as a nutritional variable to assess the nutritional status and determine the risk of CC. While some reports suggest that low hemoglobin is a risk factor for CC and is associated with more complications and lower postoperative survival rates[44, 45], others indicate inconclusive results when using nutritional variables, including hemoglobin, to assess nutritional status, including CC[46]. Cancer patients with lower serum hemoglobin levels are more likely to die from this disease[47]. In CC patients, low hemoglobin levels are associated with mortality; therefore, hemoglobin can be used as a prognostic indicator. Our findings, however, indicate a significant association between serum hemoglobin levels and the risk of CC, with lower Hb levels indicating a higher risk of developing cachexia. Additionally, cancer-related indicators, such as cancer staging, were employed to assess the risk of CC in stage IV cancer patients. Given that cancer is the primary cause of cachexia, studying cancer-related factors for assessing cachexia risk in cancer patients holds significance. Our study identified a substantial correlation between cancer staging and CC, aligning with other studies[48]. Advanced-stage cancer patients often exhibit abnormal tumor biomarkers, further emphasizing the importance of detecting these biomarkers in routine cancer management. We questioned whether abnormal tumor biomarkers could serve as predictive indicators of cachexia, and our research indicated their potential relevance. Nevertheless, our study found that the histological type and differentiation of tumors had no significant impact on CC. This observation suggests that histological type may not be related to the patient's metabolic status and, consequently, may not significantly influence WL and cachexia. In conclusion, factors such as cancer stage and location, along with tumor-related indicators, offer valuable insights into assessing the risk of CC in stage IV CRC patients.
One of the primary clinical features of CC is systemic inflammation, a key driver in its development and affects various tissues such as skeletal muscle, fat, brain, and liver[3]. Cancer-related pro-inflammatory cytokines, such as interleukin (IL)-6), IL-1, and tumor necrosis factor-α, contribute to inhibiting albumin production, a factor associated with the occurrence of CC[49]. Notably, NLR has recently been recognized as a new inflammatory evaluation indicator that can be easily obtained from blood routine. Moreover, NLR is a significant negative prognostic biomarker for cachexia patients[50]. In this study, NLR emerged as an independent risk factor for cancer cachexia, emphasizing the convenience of evaluating inflammation in clinical practice, especially in assessing the risk of cachexia. This suggests that anti-inflammatory interventions could potentially regulate systemic inflammation, aiding in muscle protection during treatment. Despite recommendations for nutrition and physical activity to maintain muscle mass, their efficacy might be limited in individuals with increased systemic inflammation[51, 52].
Our research has some limitations. First, due to the retrospective design and data collection from a single institution, there may be selection bias. Second, with evolving treatment strategies over the study period (2018–2022), our research may not entirely represent current medical practices. Third, the study's limited sample size from a single center in China necessitates additional validation in diverse centers to confirm the reliability of the ML model, and its applicability to patients outside China remains uncertain. To address these limitations, prospective trials on a larger and more diverse population with stricter inclusion criteria are essential. Finally, while nutritional management and the use of supplements or therapeutic diets are critical in evaluating the predictive utility of nutritional and inflammatory measures, this study lacks this information. The correlation between nutritional and inflammatory status and the prognosis of stage IV CRC warrants further investigation and validation in prospective cohort studies.