Machine learning for exploring and predicting cancer symptom clusters.

Abstract

Cancer patients experience multiple symptoms. Leaving these symptoms unrelieved can have deteriorating effects on the patients health related quality of life and functional status. The intensity and load of this symptom burden has implications for possible delay or termination of treatment, increased hospitalizations and medical costs, even for the prognosis and survival of cancer patients. Therefore, understanding the pattern of these symptoms’ interactions and recognising the symptom risk profiles of cancer patients is a major concern for oncology care. This research supports the aforementioned goals by exploring, understanding and identifying symptom clusters as well as their networks. It provides analytical approaches, also, on how to assess and predict the future occurrence of these symptoms. Risk profiling of oncology patients based on their symptom experience assists clinicians to provide more personalized symptom management interventions. Recent findings suggest that oncology patients with distinct symptom profiles can be identified using a variety of analytic methods. The aim for our first experiment was to evaluate the concordance between the number and types of subgroups of patients with distinct symptom profiles using latent class analysis (LCA) and K-modes analysis. Using both analytic methods, four subgroups of patients with distinct symptom profiles were identified (i.e., All Low, Moderate Physical and Lower Psychological, Moderate Physical and Higher Psychological, All High). The percent agreement between the two methods was 75.32 % which suggests a moderate level of agreement. In both analyses, patients in the All High group were significantly younger and had a higher comorbidity profile, worse MSAS subscale scores, and poorer QOL outcomes. Computational tools that predict the course and severity of these symptoms have the potential to assist oncology clinicians to personalize the patient’s treatment regimen more efficiently and provide more aggressive and timely interventions. Three common and inter-related symptoms in cancer patients are depression, anxiety, and sleep disturbance. In our second experiment, we elaborate on the efficiency of Support Vector Regression (SVR) and Non-linear Canonical Correlation Analysis by Neural Networks (n-CCA) to predict the severity of the aforementioned symptoms between two different time points during a cycle of chemotherapy (CTX). Our results demonstrate that these two methods produced equivalent results for all three symptoms. These types of predictive models can be used to identify high risk patients, educate patients about their symptom experience, and improve the timing of pre-emptive and personalized symptom management interventions. Oncology patients undergoing cancer treatment experience an average of fifteen unrelieved symptoms that are highly variable in both their severity and distress. Recent advances in Network Analysis (NA) provide a novel approach to gain insights into the complex nature of co-occurring symptoms and symptom clusters and identify core symptoms. In our third expirement, we present findings from the first study that used NA to examine the relationships among 38 common symptoms in a large sample of oncology patients undergoing chemotherapy. Using two different models of Pairwise Markov Random Fields (PMRF), we examined the nature and structure of interactions for three different dimensions of patients symptom experience (i.e., occurrence, severity, distress). Findings from this study provide the first direct evidence that the connections between and among symptoms differ depending on the symptom dimension used to create the network. Based on an evaluation of the centrality indices, nausea appears to be a structurally important node in all three networks. Our findings can be used to guide the development of symptom management interventions based on the identification of core symptoms and symptom clusters within a network. Finally, we investigate the application of Bayesian Network Analysis (BNA) methods to evaluate the relationships between symptoms in oncology patients receiving chemotherapy (CTX). We compare 3 different types of BNA algorithms and describe the way that BNA methods can be used to explore the interrelationships among cancer symptoms in a large and representative sample of 1328 cancer patients undergoing chemotherapy who reported the occurrence of 38 symptoms. We provide, also, a case study on two different sub-populations of cancer patients, based on gender and age, and explain the appropriate use of BNA for reproducible and reliable findings in future cancer symptom studies . Last, we give an example of how conditional probabilities through BNA can provide insights for the future occurrence of cancer patients’ symptoms, supporting this way cancer care and possibly guiding the design of future intervention studies. Overall, this PhD research has implemented and evaluated different machine learning methods and algorithms to assess cancer patient risk profiles, predict cancer symptoms’ future values and understand the way and order that cancer symptoms interact between them. Though machine learning, we introduce new methods and algorithms for the design and implementation of cancer symptom analytical tools. The results of this research have a significant impact both on the theoretical aspect as well as the clinical aspect of Cancer Symptom Science.