Used in conjunction with registry data, machine-learning may be a useful tool to predict the occurrence and intensity of persistent pain in patients with rheumatoid arthritis, according to a recent study published in Pain.
Researchers used data gathered from patient questionnaires (n=789) filled from the time of diagnosis for up to 5 years after diagnosis. A total of 21 parameters were assessed, including sociodemographic variables and factors deemed likely to influence persistent pain.
Unsupervised machine learning was used to identify subgroups within the distribution of patient pain levels. Gaussian mixture models were used to identify subgroups, and overfitting was assessed in all models. Supervised machine learning using Random Forest regression was used to identify the parameters that best predicted the subgroup of patients. The investigators also examined whether the evaluation of parameters at 3 months was optimal for predicting the occurrence and intensity of persistent pain.
After filtering the data to include only those patients with ≥4 assessments of pain level from 0 to 5 years after diagnosis, the data of 209 women and 79 men were analyzed (average age, 52.2 years). The use of unsupervised machine learning allowed to separate the patients into 3 subgroups according to their level of pain (ie, low, moderate, and high persistent pain). Patient global assessment and health assessment questionnaires administered at 3 months were found to allow to determine to which pain subgroups an individual would belong, using a supervised method. When the analysis was conducted again, excluding these parameters, tender joint count and swollen joint count assessed 3 months after diagnosis were found to be the most important nonpatient-related parameters for the prediction of pain level. For both patient- and nonpatient-related parameters, assessment at the time of diagnosis was not found to be an accurate predictor of persistent pain, as was the case when evaluations were conducted at 3 months.
Study limitations include the fact that this was a registry study, with no starting hypothesis.
“The results indicate that early functional parameters of rheumatoid arthritis are informative for the development and degree of persistent pain, however, not earlier than 3 months after rheumatoid arthritis diagnosis,” concluded the investigators.
Lötsch, J, Alfredsson L, Lampa J. Machine-learning based knowledge discovery in rheumatoid arthritis related registry data to identify predictors of persistent pain [published online August 30, 2019]. Pain. doi: 10.1097/j-pain.0000000000001693
This article originally appeared on Clinical Pain Advisor