We used a spectral clustering algorithm to find clusters among medical patients with lower back pain symptoms, and then we assessed the health outcomes within each cluster. First, we mapped all of the variables onto [0,1] intervals. This allowed us to compute a similarity score between every pair of patients, using an adaptation of Pearson correlation. We then calculated the spectral (eigen) decomposition of this similarity matrix, and we used the first few eigenvectors to create a low-dimensional subspace. Finally, we performed k–means clustering in this new subspace to find four clusters. We compared the cluster means and variances for each recovery assessment variable to differentiate the health outcomes for each cluster. Lastly, we highlighted the identifying symptoms of each patient cluster by inspecting any variable whose within–cluster average is extraordinarily low or high, relative to the other clusters.