Identifying dynamic subtypes of cancer cachexia using longitudinal clinical lab values

Cancer cachexia is a wasting syndrome estimated to affect roughly 50% of all cancer patients and is most commonly identified using a simple weight loss criterion. The heterogeneity in plausible cachexia weight loss mechanisms, along with their potential to evolve over time, suggests the existence of distinct, dynamic subtypes of cachexia. Clinical lab values are frequently measured and, although they do not grant direct mechanistic insight, they provide rich temporal information about patient latent states. In a cohort of lung adenocarcinoma patients from Memorial Sloan Kettering Cancer Center, we clustered 1,630 cachectic episodes using 28 longitudinally measured clinical lab values and body mass index. Clustering was performed using K-means with Dynamic Time Warping Barycenter Averaging to handle trajectories with varying lengths. We identified 4 major clusters with differentiating patterns including a group with relatively higher red cell distribution width and relatively lower mean corpuscular volume and mean corpuscular hemoglobin. Another cluster had lab values that are more consistent with normal ranges, indicating a possible non-cachectic group of weight loss patients. This study demonstrates the potential of longitudinal clinical lab values for stratifying cachexia patients and identifying intervention points in patient care that can improve outcomes.