Introduction Visceral pain is a complex experience, and is influenced by an array of physiological, psychological and anatomical factors. Previous research from our group has coalesced many of these factors, demonstrating that two major endophenotypic ‘pain clusters’ (PC) exist with the following features: Pain Cluster 1 (PC1), in comparison to Pain Cluster 2 (PC2), had higher neuroticism and anxiety scores, and the serotonin transporter-linked polymorphic region genotype (5-HTTLPR) short allele was over represented. PC1 had greater baseline sympathetic tone and serum cortisol, but during acute pain had a lower stimulus tolerance and increased parasympathetic tone. Meanwhile, PC2 had the converse profile at baseline and during pain. We hypothesised that these PCs could be predicted by whole brain functional connectivity during either rest or oesophageal pain, irrespective of any physiological, psychophysiological and genetic data.
Method We used a previous dataset of 21 healthy subjects (10 male and 11 female; mean age 30 years), all of which were previously allocated to PC1 (n=9) or PC2 (n=12). All had additionally undergone fMRI with an event-related design of acute oesophageal pain and rest periods. Blood oxygen level dependant (BOLD) signal during rest and oesophageal balloon distention to pain tolerance threshold was extracted from a whole brain parcellation map of 346 single-voxel regions of interest (Figure 1a, single node example), which were cross-correlated to produce whole brain correlation matrices of 59 685 r values. Using complex decision trees, an aspect of the ‘classification-learner’ machine learning algorithm within Matlab, we investigated whether whole brain connectivity could accurately predict which cluster subjects were allocated to.
Results Whole brain functional connectivity during oesophageal pain accurately predicted subject PC with 85.7% accuracy: area under curve (AUC) 0.86; true positive rate (TPR) for PC1 and PC2 89% and 83% respectively; false negative rate (FNR) for PC1 and PC2 11% and 17% respectively (Figure 1b, receiver operating characteristic curve). However, PC were only predicted with 58% accuracy during rest: AUC 0.58; TPR for PC1 and PC2 44% and 75% respectively, FNR for PC1 and PC2 56% and 25%.
Conclusion These data suggest that endophenotypic PCs can be accurately predicted from whole brain functional connectivity during acute oesophageal pain, but not by resting connectivity. Future study should investigate for specific brain region connectivity and network differences between PCs to elucidate key differences.
Disclosure of Interest J. Ruffle: None Declared, S. Coen: None Declared, V. Giampietro: None Declared, S. Williams: None Declared, A. Farmer Conflict with: Medical Research Council Grant, Q. Aziz Conflict with: Medical Research Council Grant
- Functional connectivity
- Functional magnetic resonance imaging
- Machine learning
- Pain endophenotypes
- Visceral pain