In Frontiers in pharmacology
Background: The selection strategy of non-steroidal anti-inflammatory drugs (NSAIDs) for migraine is hard to judge whether it is effective, leading to unnecessary exposure to insufficient or lengthy treatment trials. The goal of the study was to investigate potential predictors of NSAIDs efficacy in migraine therapy and to explore their influence on efficacy. Methods: 610 migraine patients were recruited and assigned into responders and non-responders. Potential predictors among demographic and clinical characteristics for NSAIDs efficacy were extracted using multivariable logistic regression (LR) analysis, and were applied to construct prediction models via machine learning (ML) algorithms. Finally, Cochran-Mantel-Haenszel tests were used to examine the impact of each predictor on drug efficacy. Results: Multivariate LR analysis revealed migraine-related (disease duration, headache intensity and frequency) and psychiatric (anxiety, depression and sleep disorder) characteristics were predictive of NSAIDs efficacy. The accuracies of ML models using support vector machine, decision tree and multilayer perceptron were 0.712, 0.741, and 0.715, respectively. Cochran-Mantel-Haenszel test showed that, for variables with homogeneity of odds ratio, disease duration, frequency, anxiety, and depression and sleep disorder were associated with decreased likelihood of response to all NSAIDs. However, the variabilities in the efficacy of acetaminophen and celecoxib between patients with mild and severe headache intensity were not confirmed. Conclusion: Migraine-related and psychiatric parameters play a critical role in predicting the outcomes of acute migraine treatment. These models based on predictors could optimize drug selection and improve benefits from the start of treatment.
Lu Zhao-Xuan, Dong Bing-Qing, Wei Heng-Le, Chen Liang
2022
efficacy, machine learning, migraine, multivariable logistic regression, non-steroidal anti-inflammatory drugs
