We have demonstrated, for the first time, that the type of cancer diagnosis, specifically the presence of lung or colorectal cancer versus other cancers, exhibits a causal relationship with global quality of life as shown by 2 separate methods of causal analysis. While some previous studies have indicated an association between cancer type and various quality of life scores, our analysis using machine learning and Bayesian inference techniques revealed that this relationship is not purely correlative but has a causal component [16]. In contrast, some authors have not shown a difference in global quality of life with respect to cancer type [17]. Interestingly, the most influential causal factors for QL2 identified in this study differ from the strongest associates of QL2. Although the type of cancer diagnosis emerged as the most influential causal factor for QL2 in both the manual approach and the LiNGAM algorithm, it did not reach statistical significance in the multivariate analysis conducted using classical regression analysis. Similarly, while social functioning (SF) was the strongest associate of QL2 in the regression analysis, it did not emerge as the most influential causal factor in our causal analyses. This disparity underscores some unique insights that causal analysis could offer, revealing complex causal relationships that might not be captured by traditional associative or correlative approaches, such as regression analysis.

The DAG we manually coded did not include a visible path between the type of cancer diagnosis and QL2, but the total causal effect of the type of cancer diagnosis could be quantified algorithmically. However, looking at the LiNGAM adjacency matrix, we see that there are various possible paths between the type of cancer diagnosis and QL2, and thus, in that fashion, the LiNGAM adjacency matrix captured the complex relationships in our study better than the DAG. A possible explanation for the discrepancy between these 2 separate approaches (Expert-driven construction of the DAG versus LiNGAM) in our study is that the manually coded DAG failed to capture some of the important variables that were visible in the LiNGAM adjacency matrix. In other words, these variables remained latent for some of the relationships in the DAG structure, making LiNGAM, a machine learning model, more explanatory for our purpose.

The use of DAG and LiNGAM approaches in causal analysis offers several strengths and limitations. DAGs provide a clear and intuitive graphical representation of causal relationships, aiding in the identification of confounders, mediators, and potential sources of bias [18]. They facilitate the understanding of complex causal structures and the formulation of testable hypotheses. LiNGAM extends this by enabling the discovery of causal ordering and effects from observational data under the assumptions of linearity and non-Gaussian errors, offering a robust method for causal inference when these conditions are met. However, both approaches have limitations. DAGs rely heavily on domain knowledge for accurate construction, and their effectiveness is constrained by the correctness and completeness of this prior knowledge. LiNGAM’s assumptions of linearity and non-Gaussianity may not always hold in real-world data, potentially limiting its applicability and accuracy. Additionally, LiNGAM can be sensitive to sample size and may struggle with high-dimensional data. Despite these limitations, the combined use of DAGs and LiNGAM provided a powerful toolkit for causal discovery and analysis for our study.

If the type of cancer directly influences global quality of life at the time of presentation, several factors may drive this phenomenon. Previous studies, including ours, have assessed quality of life across different cancer populations and have identified differences in various dimensions of quality of life; those cancer populations included metastatic or nonmetastatic cases, men or women, those with different categories of weight loss, performance status, weight loss, time from last treatment, and emotional and functional well-being [16, 19,20,21]. Specifically, the literature shows differences in the QL2 score between lung cancer patients and patients with other types of cancer [22]. Likewise, the quality-of-life score is also low in patients with colon cancer [23]. These differences are likely attributed to the distinct biological profiles of various cancers, with lung cancer often presenting at more advanced stages and having a poorer prognosis, especially in patients with stage 4 disease. Causal factors alter the magnitude of the affected variable, leading to a difference in its magnitude. This aligns with the basic concept of causality, where one variable (the cause) influences another variable (the effect). Thus, from this perspective, type of cancer causally effects and leads to difference in global quality of life in cancer patients. Additionally, the stigma associated with different types of cancer and perceptions of their curability may impact patients’ experiences [24, 25]. For instance, a lung cancer patient may perceive a more unfavorable outlook than a patient with breast cancer. Thus, not only the biological and symptomatic aspects of the disease but also the patient’s perception and societal view of the disease may influence the QL2 score. Further research focusing on the perception dimension and its relationship with quality-of-life dimensions across various cancers could provide valuable insights in this direction.

Apart from type of cancer, which has a causal effect on global quality of life, stage also emerged as another causal factor. Additionally, functional dimensions of QoL had positive causal effects, whereas symptom scales had negative causal effects on QoL2. Although there is some evidence that there is no association between stage or functional dimensions of QoL and QoL2, some available evidence suggests that some domains of QoL may be marginally different with respect to the disease stage [17, 21].

While our study provides important insights, it is not without limitations. Despite including data from 469 patients, a larger sample size, potentially in the range of thousands of cases, would be preferable for elucidating more intricate causal relationships and accurately assessing their strength. Additionally, obtaining more comprehensive data on social security, financial status, treatment details, and patients’ perceptions of their disease status would enable a more thorough examination of the causal effects of these factors in addition to the other variables considered in this study. Lastly, the validity of LiNGAM has been proven only for continuous variables; thus, our use of LiNGAM, which includes categorical or discrete data, is experimental.

The rise of AI and machine learning techniques is reshaping medical practice [26]. These technologies, including computer vision, natural language processing, and various machine learning algorithms, have been increasingly utilized for predicting outcomes and prognosis in recent years. Causal inference, whether through probabilistic, statistical, or machine learning-oriented algorithms, addresses the question of “Why did this happen?” rather than “What features or factors are associated with this outcome?” [27, 28]. This approach is expected to influence medical policies and interventions by providing deeper insights into causal relationships. Furthermore, causal inference techniques, such as causal survival analysis, are gaining traction in other research areas and are anticipated to play a more prominent role in medical research in the future [29]. The findings from our study lend support to this trend.

In summary, our study highlights the causal effect of cancer type, among other factors, on changes in QL2 scores in cancer patients. Importantly, this causal effect was not evident in classical regression analysis. The causal inference methodologies employed in our study have the potential to inform policies and interventions aimed at improving quality of life in cancer patients and, more broadly, to address numerous causal questions in oncology and medicine.