Lung cancer, the most prevalent malignancy globally and a primary contributor to cancer-related fatalities (constituting ~18% of all cancer deaths) (1), was often accompanied by cachexia. Cachexia is a multifactorial and multi-organ syndrome prevalent in the late stages of chronic conditions (2), especially among cancer patients. Cachexia, a metabolic syndrome, is characterized by loss of muscle mass, with or without concurrent loss of fat mass, and diminished bodily functions. This condition is frequently attributed to an inflammatory state or reduced food intake (2). It manifests in nearly 50% of lung cancer patients and indirectly contributes to at least 20% deaths of all cancer patients (3, 4). Cachexia during cancer treatment poses a severe complication as it is linked to increased chemotherapy-related side effects, fewer completed cycles of chemotherapy, and reduced survival rates (5, 6).

Despite the widespread occurrence of cachexia in clinical practice, addressing its prevention, early identification, and intervention remains a challenge. The impact of cancer cachexia on quality of life, treatment-related toxicity (7, 8), physical function impairment, and mortality was well documented. However, focusing on weight loss to establish a clinically meaningful definition for the diagnosis of cachexia had proven challenging because the failure of patients in recalling their weight history can lead to lack of accurate weight loss information. Early identification and management of cachexia is critical for its prevention. Moreover, cachexia is characterized by not just weight loss but also loss of body composition and body function (9). Importantly, emaciation is a late symptom of cachexia and a characteristic of its advanced phase. Although attempts to more comprehensively define cachexia through body composition, body function, and molecular biomarkers were promising, it had not been routinely incorporated into clinical practice (9–11). To date, there are no effective pharmacological interventions that can completely reverse cachexia. Adequate and early nutritional support remains the mainstay of cachexia treatment (12).

In the established diagnostic framework (12), cachexia is defined as an involuntary weight loss of more than 5% or a body mass index (BMI) of <20 kg/m², with sustained weight loss of more than 2% within the past 6 months. Additionally, it included cases of sarcopenia combined with sustained weight loss of more than 2%. Cancer cachexia is categorized into three stages: pre-cachexia, cachexia, and refractory cachexia. Notably, refractory cachexia has been the focus in numerous clinical trials for novel interventional drugs (12). However, demonstrating its therapeutic efficacy in this stage is challenging due to the catabolic state's resistance to anticancer therapy, the low performance state, and a prognosis of survival of <3 months. Currently, the focus of academic research had shifted from advanced cachexia to the etiology of cachexia (13). Evidence has shown that different drivers of inflammation, metabolism, and neuro-modulatory can initiate processes that eventually led to advanced cachexia (14).

This study aimed to use the nutrition-related parameters without weight loss information to establish an ensemble learning system for predicting and diagnosing the occurrence of cachexia in lung cancer patients before weight loss or a lack of weight loss information, which will be beneficial for administrating early nutritional support and improving the overall survival (OS) of lung cancer patients.

This study was a retrospective cohort investigation of 4,712 lung cancer patients at multi-centers in China. The study focused on a real-world clinical challenge, which predicted the onset of cancer cachexia in advance before significant weight loss. To our knowledge, this study represents the initial large-scale investigation tackling this challenge through ensemble learning methods grounded in traditional clinical characteristics. Our discoveries offer valuable insights that may assist clinicians or nutritionists in decision-making for the treatment of high-risk lung cancer patients, guiding effective management strategies to enhance patients' outcomes.

Weight loss was a major factor in the diagnosis of cachexia (12, 36). However, due to the lack of self-awareness in patients' daily life, the delay in recalling weight loss information and the inaccuracy of recalling made the diagnosis of cachexia difficult (37). Previous evidence showed that, once patients lose more than 5% of their body weight, the mortality rate had increased significantly (9). In addition, notable instances of treatment-related toxicity and increased mortality were observed in obese patients with low muscle mass, a condition referred to as sarcopenia obesity (38, 39). Clinical identification of these patients posed a challenge, especially since sarcopenia can be concealed by obesity body mass index (BMI) (40). Consequently, beyond weight loss, increased need for sensitive criteria arises to detect patients in the early stages of cachexia. Such assessments necessitate measurements beyond standard body weight, including instruments for evaluating muscle mass and/or physical activity (38). In the context of cancer, because of the complexity of body composition, some researchers suggested that measurements of specific body compartments using methods such as computed tomography should outweigh the role of weight loss in evaluating cachexia test results (41). We used a convenient and cost-effective GBC ensemble learning model that incorporated demographic, anthropometric, and laboratory parameters, excluding baseline weight loss to construct an effective predictive model for early identification of cachexia. A total of 18 machine learning models were evaluated, and a GBC ensemble learning model with optimal performance was obtained. Compared to the training set, the GBC model showed good performance in the testing set.

In our study, we eliminated weight loss, a dynamically changing feature, and only used currently measurable characteristics to construct a diagnostic model of cachexia. Low BMI, reduced food intake, digestive symptoms, and NLR contributed more to the GBC model. Currently, the GLIM framework suggests the inclusion of at least one phenotypic criterion and one etiological criterion for diagnosing malnutrition. Etiological criteria encompass reduced food intake, digestive or absorption disorders, inflammation, or disease burden in patients (42). BMI was the most important indicator to measure the nutritional status of the human body and is designated to be one of the diagnostic criteria for cachexia (12). The mechanism of anorexia involved tumor-derived humoral factors that can induce cancer anorexia by modulating neuropeptide hormones in the brain associated with eating (43). Furthermore, the elucidation of the process indicates that anorexia preceded tissue wasting in cachexia. In a previous study, severe and moderate reduction in food intake was also found to independently predict OS. When anorexia, vomiting, or other digestive symptoms occurred or food intake decreased, patients were at higher risk for cachexia, which was consistent with findings of previous research (44). In the tumor context, systemic inflammation played an important role in the onset and progression of cancer, existed during cancer-associated cachexia, and served as a diagnostic tool for cancer-associated cachexia (25, 45). Recent meta-analyses demonstrated an association between elevated NLR and reduced progression-free survival and OS both in NSCLC and SCLC (33, 46). A prior study identified an association between elevated NLR and weight loss, as well as an increased prevalence of cachexia in cancer patients with advanced colon, lung, or prostate cancer (22). It had also been found that high levels of NLRs at baseline and a progressive increase in NLRs during treatment were associated with progressive disease, low OS, and weight loss in NSCLC patients (47). In our GBC model, NLR was one of the top five clinical features contributing to the model, indicating the important value of NLR in the diagnosis of lung cancer cachexia.

However, this study had several potential limitations. First, the weight loss relied on patient's reported historical weight, introducing the possibility of recalling bias. To mitigate this, only data from patients who were able to provide past weight information were included. Second, the ASMI in this study was derived from an anthropometric equation validated for the Chinese population. While the equation had demonstrated good agreement with dual-energy X-ray imaging and anthropometric data were readily available (17). It is essential to note that more precise muscle mass measurements for diagnosing cachexia may be obtained through imaging techniques such as dual-energy x-ray imaging or bioimpedance analysis (48). However, our GBC model showed that the classification results of the ASMI were of relatively little importance in the diagnosis of cachexia. Imaging indicators that could more accurately assess muscle mass might improve the performance of our model. For example, transverse CT images, typically at standard markers frequently presented in abdominal CT (such as the third lumbar spine), had established correlations with DXA equivalent total body fat and muscle mass (38, 49). Future studies using more advanced techniques to measure muscle mass are required. Third, as an etiological indicator of cachexia, inflammatory indicators are prognostic indicators of advanced lung cancer, as observed in previous studies. One potential serum biomarker commonly used in clinical practice was C-reactive protein (CRP) (25) that is used to identify patients at risk for cancer cachexia, when combined with other weight loss and nutrient intake factors (50). Due to the substantial amount of missing data on CRP, we could not include CRP and its related systemic inflammation indicators, which would be further improved in the subsequent data collection. However, in our study, we also demonstrated a significant correlation between NLR/ALI/PLR and cachexia through a univariate analysis (P < 0.05). Our model also included NLR, a marker of systemic inflammation, which could compensate for the missing CRP data to some extent. In future studies, we could analyze and clarify the role of systemic inflammatory indicators in the diagnosis of cachexia.

In conclusion, the GBC ensemble learning model using clinical data without weight loss information could identify patients with lung cancer cachexia. Feature importance showed the contribution value of each clinical feature to the diagnosis of cachexia. This study may support patient counseling and targeted interventions to perform nutritional treatment in advance and improve patient prognosis and life quality.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by the Institutional Review Committee of Beijing Shijitan Hospital. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

PJ: Data curation, Funding acquisition, Project administration, Supervision, Validation, Writing—original draft, Writing—review & editing. QZ: Investigation, Methodology, Writing—review & editing, Writing—original draft. XWu: Data curation, Investigation, Methodology, Writing—original draft, Writing—review & editing. FS: Data curation, Methodology, Writing—original draft, Writing—review & editing. KS: Investigation, Methodology, Writing—original draft, Writing—review & editing. XWa: Data curation, Investigation, Methodology, Writing—original draft, Writing—review & editing.