Establishing Central Sensitization Inventory Cutoff-Scores in Chronic Low Back Pain Population by Deep Learning

BACKGROUND:
Central sensitization (CS) is presented in Chronic low back pain (CLBP). The Central Sensitization Inventory (CSI) is designed to assess CS-related symptoms. A cutoff score of 40 is applied to differentiate between high and low CS, however, recent studies showed that this cutoff score is debatable. It may suggest that the cutoff scores may vary among different diseases. Therefore, this study aims to determine the optimal cutoff score in the CLBP population.

METHODS:
Patients with CLBP were assessed during interdisciplinary rehabilitation (n=92 (n=58 at baseline, n=34 at discharge)). Next to the CSI, a series of questionnaires were used to assess pain-related features: Visual Analogue Scale Pain (pain); Rand36 Physical Functioning subscale Pain Disability Index, WorkAbility Score (physical functioning); Pain Catastrophizing Scale, Injustice Experience Questionnaire, and Brief Symptom Inventory (psychosocial).
The unsupervised deep learning approach autoencoder was used to reduce the questionnaires data to a 4-dimensional vector for every patient, as [pain score, physical score, psychosocial score, CSI score]. A series of state-of-the-art clustering approaches were used to cluster the patients to low or high CS. They were KMeans, Hierarchical clustering, DBSCAN, Self-organizing Map. A receiver operating characteristic was used to analyze the best clustering results and determine the optimal cutoff score.

RESULTS:
KMeans got the best clustering results. The patients with lower CS (n=45) showed statistically significant (p<0.01) less pain, better physical and psychological functioning than higher CS (n=47). The optimal cutoff score is 35, with 74.5% sensitivity, 68.9% specificity, 0.67 of AUC, and 0.73 F1-score.

CONCLUSION:
In patients with CLBP, a CSI cutoff score of 35 differentiated optimally between high and low CS, based on pain, physical, and psychosocial functioning.