A replicable and generalizable neuroimaging-based indicator of pain sensitivity across individuals

Abstract

Developing neural indicators of pain sensitivity is crucial for revealing the neural basis of individual differences in pain and advancing individualized treatment of pain. However, it still remains elusive whether pain-evoked neural responses can encode pain sensitivity. To address this issue, we analyzed five large functional magnetic resonance imaging (fMRI) datasets (total N = 1010), where healthy participants received painful and nonpainful tactile, auditory, and visual stimuli, and different pain treatments, including placebo and transcutaneous electrical neural stimulation. We systematically (1) investigated the correlation between pain-evoked fMRI responses and pain sensitivity, (2) evaluated the correlation’s replicability in independent datasets and generalizability across different types of pain, (3) examined whether the correlation between fMRI responses and sensory sensitivity is unique to pain, and (4) how sample sizes affect the relationship between fMRI responses and pain sensitivity. We found that, with a sufficiently large sample size, there were replicable and generalizable correlations between pain-evoked fMRI responses and pain sensitivity across individuals for laser heat, contact heat, and mechanical pains. Despite lacking pain selectivity, fMRI signals exhibited larger correlations with pain sensitivity than with tactile, auditory, and visual sensitivity. Importantly, we developed a machine learning model that could accurately predict not only pain sensitivity to laser heat, contact heat, and mechanical stimuli, but also pain relief from pain treatments. Notably, our findings were influenced considerably by sample sizes, requiring >200 for univariate correlation analysis to reveal the relationship between pain sensitivity and fMRI responses, and >150 for multivariate analysis to decode pain sensitivity with fMRI responses. Altogether, given an enormous sample size, we convincingly showed the validity to decode pain sensitivity and predict analgesic effects using pain-evoked fMRI responses, which holds significant clinical promise in tailoring individualized pain treatments.