In this secondary analysis of three harmonized datasets of patients with cancer, we identified determinants of aerobic and resistance physical activity using a decision-tree approach. By translating sociodemographic and clinical characteristics into an interpretable classification structure, this approach may help clinicians identify patient profiles associated with lower adherence and guide individualized physical activity support. Overall, we found low adherence to recommended physical activity levels: only 17.5% and 12.7% of patients met the guidelines for aerobic and resistance training, respectively. Just 2.6% of patients achieved both recommendations. A prior systematic review, including 41 studies, found that, on average, 34.2% of patients met the physical activity guidelines [8]. This percentage is higher compared to that found in our study. Nevertheless, it should be noted that in such investigations, meeting the guidelines was defined as 150 min of moderate-intensity or 75 min of vigorous-intensity physical activity per week, without distinguishing between aerobic and resistance activities [8]. However, this criterion may have overestimated the percentage of sufficiently active patients since it also includes the resistance-training component. By contrast, our analysis applied separate thresholds for aerobic and resistance exercise, providing a more nuanced and stringent assessment of guideline adherence. Consequently, the lower prevalence observed in our study may reflect a more accurate estimate of proper compliance, highlighting a substantial gap between current recommendations and patients’ real-world physical activity behaviors. The “dark side” of this approach is that the low proportion of physically active patients found prevented the development of a reliable decision-tree model. On the other hand, it has permitted, for the first time, the analysis of the determinants of aerobic and resistance training separately.

In both models, the cancer site was the primary discriminator of activity levels. In this context, for aerobic activity, patients with lung, hematological, and gynecological cancers were more likely to exhibit lower adherence than patients with other cancer types. By contrast, patients with melanoma, colorectal, and gynecological cancers were more likely to engage in muscle-strengthening activities compared to others. To our knowledge, this is the first study to identify cancer site as a key determinant differentiating adherence patterns between aerobic and muscle-strengthening activities. Although the underlying reasons for these findings remain unclear, they could reflect differences in symptom burden, treatment intensity, functional limitations, and clinical recommendations across cancer types [7, 25, 26]. However, these findings may provide clinically relevant insights for exercise prescription in oncology. For instance, for cancer types commonly associated with substantial weight loss, sarcopenia, and muscle wasting (e.g., pancreatic and head and neck cancers), resistance training may play a more crucial role in preserving or restoring muscle mass and physical function [27]. Conversely, in cancers more frequently associated with excess body weight or adiposity (e.g., breast, prostate, and gynecological cancers), aerobic exercise may warrant greater emphasis to address cardiometabolic risk and weight management [27]. Therefore, knowing that specific cancer types are more likely to engage in either aerobic or resistance exercise could facilitate the development of more targeted, cancer-tailored exercise interventions that focus on the physical activity modality in which patients are least active and could stand to benefit most.

Interestingly, neither cancer stage nor treatment timing was a predictor in the decision trees. Despite clinical assumptions that advanced disease or ongoing treatment may limit physical activity, our models did not identify these variables as determinants of physical activity. This absence suggests that other factors, e.g., functional capacity, symptom profiles, and psychosocial conditions, could mediate the relationship between disease severity and physical activity. Unfortunately, these features were not captured in the present datasets but represent important aspects to be explored in future research. The finding is also consistent with emerging literature that shows high inter-individual variability within the same stage, challenging linear assumptions about disease progression and behavior [28].

Sociodemographic variables demonstrated substantial predictive value in both decision-tree models. Occupational status emerged as a significant determinant of aerobic and resistance activity, with patients who were employed showing higher adherence compared with retirees, homemakers, or those seeking employment. Occupational status may reflect differences in daily routines, functional capacity, social engagement, and socioeconomic resources, all of which may influence physical activity behavior.. In addition, employed individuals may perceive physical activity as more instrumental for sustaining work ability and independence. Interestingly, BMI showed differential associations with physical activity across exercise modalities. In the aerobic activity model, obesity was associated with lower levels of physical activity. It is consistent with prior literature and may reflect barriers such as reduced mobility, heightened fatigue perception, and lower exercise self-efficacy. In the resistance training model, being overweight was associated with higher participation rates than in other BMI categories. To our knowledge, this is the first study reporting this finding, which may be partly explained by physiological/functional factors. In particular, being overweight may reflect a more favorable balance between muscle and fat mass, potentially facilitating engagement in muscle-strengthening activities without the functional limitations often observed at higher BMI levels. Together, these findings suggest that BMI may influence engagement in physical activity in a modality-specific manner rather than acting as a uniform barrier across exercise types. The role of educational level as a determinant of aerobic and resistance activities was heterogeneous and difficult to interpret. While higher education is generally associated with higher physical activity levels, our finding likely reflects indirect effects mediated by broader socioeconomic and contextual factors rather than education alone, and thus warrants further investigation.

Limitations of this study include reliance on self-reported activity, which may be subject to misclassification. In addition, resistance-training and combined physical activity data were unavailable for 448 participants, corresponding to 39.6% of the total sample. This missingness mainly reflects the fact that resistance training was assessed in only two of the three parent studies and may limit the generalizability of findings related to resistance training and combined adherence to physical activity recommendations. Additionally, several potentially relevant predictors, including symptom burden, performance status, objective economic status (e.g., income), and treatment-related toxicity, were not available in the harmonized datasets; their absence may have contributed to the exclusion of stage or treatment characteristics from the decision-tree models. Finally, cancer-site categories were not equally represented in the sample, and this uneven distribution may have influenced the stability of cancer-site splits in the decision-tree models. Future research should validate these findings in prospective cohorts, incorporate objective activity measures (e.g., accelerometers), and include symptom burden, performance status, and treatment toxicity in predictive models. On the contrary, the strengths of this study include the large, heterogeneous sample and the application of an interpretable machine learning approach. Moreover, the separate examination of aerobic and resistance activities represents a step forward in identifying patients at risk of insufficient physical activity and in providing more appropriate and targeted support.

In conclusion, the decision-tree approach revealed that physical activity behaviors among patients with cancer are primarily driven by cancer type and sociodemographic characteristics. The low proportion of patients adhering to aerobic and resistance exercise guidelines underscores an urgent, largely unmet need for targeted physical activity interventions in oncology care. Importantly, recognizing that distinct determinants govern aerobic and resistance activities provides a critical framework for designing more precise, modality-specific exercise strategies that align with contemporary guidelines for patients with cancer.