Background & Aims

RCTs in neuropathic pain (NeP) are challenged by the high inter-individual variability in treatment response, reflecting the diverse underlying mechanisms of pain. While the Neuropathic Pain Symptom Inventory (NPSI) has previously delineated three pain clusters (Bouhassira et al 2021): pinpointed pain (paresthesia/dysesthesia), evoked pain (stimulus-induced pain), and deep pain (spontaneous pressing pain), this analysis explores whether a similar clustering approach can be applied using the PainDetect questionnaire in patients with acute low back pain and radiculopathy. Although PainDETECT was originally designed to screen for neuropathic pain rather than to subclassify its subtypes, we aimed to determine whether patients with neuropathic features could be stratified into distinct clusters using a pre-defined algorithm transposed from the NPSI methodology.

Methods

This post-hoc analysis used data from the C.L.E.A.R. study, an RCT on SP-102 transforaminal injections for lumbosacral radicular pain. Clustering was based on all screened subjects with possible NeP (PainDETECT score ≥13, n = 1420) from 401 enrolled patients.

A k-means algorithm clustered PainDETECT item scores, optimizing clusters via silhouette statistics. The NPSI-based algorithm was tested by comparing patients to predefined cluster centers. The NPSI-based clustering algorithm was tested on PainDETECT data by assessing each patient’s similarity to predefined cluster centers.

To apply the NPSI-based clustering, PainDETECT items were mapped to four of the five NPSI pain subdomains: burning (item 1), paroxysmal (item 4), evoked (items 3, 5, 7), and paresthesia/dysesthesia (items 2, 6). However, no item corresponded to pressing pain, the defining feature of Cluster 3 (Deep Pain), making it difficult to identify this subgroup. Accordingly, two clusters—pinpointed and evoked pain—were expected to emerge.

We assessed consistency between PainDETECT and NPSI clusters and clustering stability between screening and baseline visits.

Results

Two PainDETECT clusters emerged: Cluster 1 had higher scores in paresthesia and burning pain but lower scores in evoked pain. Cluster 2 showed the opposite pattern. Comparing with the NPSI algorithm, PainDETECT Cluster 1 comprised 60% of the NPSI “Pinpointed Pain” Cluster, while 39.6% belonged to the NPSI “Deep Pain” Cluster. PainDETECT Cluster 2 was primarily composed of 57.3% of NPSI “Evoked Pain” Cluster, with most others classified as NPSI “Deep Pain” subjects.

Cluster stability showed that 56% remained in the same NPSI cluster between screening and baseline. Switching mainly occurred between “Deep Pain” and the other clusters, highlighting the challenge of identifying Deep Pain via PainDETECT.

Conclusions

Our findings support the feasibility of using a pre-defined NPSI-based clustering algorithm to stratify neuropathic phenotypes using the PainDETECT in patients with acute low back pain and radiculopathy. However, the inherent difference of PainDETECT – not assessing pressing pain – limits its ability to fully replicate the three-cluster model of the NPSI. Consequently, for characterization neuropathic pain subtypes, PainDETECT is most appropriately applied in a two-cluster model (Pinpointed Pain and Evoked Pain), while the NPSI remains the more comprehensive tool for phenotyping. Future work will further evaluate the clinical relevance of these clusters.