The nested structure of cancer symptoms. Implications for analyzing co-occurrence and managing symptoms.

OBJECTIVE Although many cancer patients experience multiple concurrent symptoms, most studies have either focused on the analysis of single symptoms, or have used methods such as factor analysis that make a priori assumptions about how the data is structured. This article addresses both limitations by first visually exploring the data to identify patterns in the co-occurrence of multiple symptoms, and then using those insights to select and develop quantitative measures to analyze and validate the results. METHODS We used networks to visualize how 665 cancer patients reported 18 symptoms, and then quantitatively analyzed the observed patterns using degree of symptom overlap between patients, degree of symptom clustering using network modularity, clustering of symptoms based on agglomerative hierarchical clustering, and degree of nestedness of the symptoms based on the most frequently co-occurring symptoms for different sizes of symptom sets. These results were validated by assessing the statistical significance of the quantitative measures through comparison with random networks of the same size and distribution. RESULTS The cancer symptoms tended to co-occur in a nested structure, where there was a small set of symptoms that co-occurred in many patients, and progressively larger sets of symptoms that co-occurred among a few patients. CONCLUSIONS These results suggest that cancer symptoms co-occur in a nested pattern as opposed to distinct clusters, thereby demonstrating the value of exploratory network analyses to reveal complex relationships between patients and symptoms. The research also extends methods for exploring symptom co-occurrence, including methods for quantifying the degree of symptom overlap and for examining nested co-occurrence in co-occurrence data. Finally, the analysis also suggested implications for the design of systems that assist in symptom assessment and management. The main limitation of the study was that only one dataset was considered, and future studies should attempt to replicate the results in new data.