• Center on Health Equity and Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

Can Machine Learning Predict Medication Overuse in Migraineurs?

Article

A combination of machine learning and random optimization may aid in predicting medication overuse in migraineurs, if clinical/biochemical features, drug exposure, and lifestyle are taken into consideration, according to a study published in Computational and Structural Biotechnology Journal.

A combination of machine learning (ML) and random optimization (RO) may aid in predicting medication overuse (MO) in migraineurs, if clinical/biochemical features, drug exposure, and lifestyle are taken into consideration, according to a study published in Computational and Structural Biotechnology Journal. The combination also holds the potential for improving model precision by weighing the relative importance of specific attributes.

Medication overuse headache (MOH) occurs in migraineurs who frequently use medications like triptans, ergots, barbiturates, or opiates. High-frequency use of these medications may not only increase the frequency and intensity of headaches but can also lead to adverse events such as gastrointestinal issues, renal toxicities, medication dependency, and withdrawal.

Approximately 15% of patients with migraine are affected by MO, and it is the most relevant risk factor for chronic migraine, in addition to development of MOH.

“ML is largely used to develop automatic predictors in migraine classification, but automatic predictors for MO in migraine are still in their infancy,” the authors write.

Researchers analyzed the performance of a customize ML-based decision support system that combined suport vector machines and RO (RO-MO).

The model used in the study was based on Multiple Kernel Learning and combines support vector machine (SVM) algorithms and RO. This way, researchers can inspect the learned model and provide an estimate of the relative weight of demographic, clinical, and biochemical data in predictions.

Data from 777 consecutive migraineurs who presented at the Headache Pain Unit of the Department of Neurological, Motor and Sensorial Sciences and the InterInstitutional Multidisciplinary Biobank in Italy were analyzed.

The dataset was randomly divided into a training set of 543 migraineurs (70%) and test set of 234 migraineurs (30%). The researchers ran baseline SVM and compared its performance with multiple RO-MO classifiers.

In addition, group clustering was performed according to the clinical significance of the attributes included in the patient dataset. Patients were categorized based on demographic characteristics, migraine clinical features, treatment details, presence of co-morbidities, biochemical variables, and lifestyle information.

Data showed 21% (162 of 777) of enrolled patients reported the presence of MO lasting for at least 2 years, while “no substantial differences were observed for clinical and biomolecular variables between patients included in the training and test sets, with the exception of attack frequency, the use of triptans in combination with nonsteroidal anti-inflammatory drugs (NSAID), and the percentage of patients with MO.”

Analyses revealed:

  • Receiver operating characteristic analysis resulted in a c-statistic of 0.83 with a sensitivity and specificity of 0.69 and 0.87, respectively, and an accuracy of 0.87 when MO was predicted by at least 3 RO-MO models
  • Logistic regression analysis confirmed that the derived RO-MO system could effectively predict MO with odds ratios of 5.7 and 21.0 for patients classified as probably (3 predictors positive) or definitely at risk of MO (4 predictors positive), respectively
  • RO-MO-08 was the best performing predictor in the training set, with an area under the curve (AUC) of 0.79, which was significantly higher than that observed for baseline SVM (P = .003) and further increased to 0.81 in the test set
  • 5 RO-MO predictors showed an acceptable predictive performance with AUCs above 0.70 and significant positive and negative likelihood ratios, but only 4 confirmed their predictive value in the test set
  • The 4 models resulted in an overall improvement of MO risk prediction performance, with a 0.765 AUC, which was higher than that observed with SVM or each single predictor

Each of the 4 models weighed demographics and co-morbidities, DBH polymorphism, treatment details and lifestyle-related triggers, and the biochemical and metabolic asset, respectively.

“The clinical soundness of our approach was…supported by the finding that the best scoring models in terms of both f-measure and AUCs were also clinically plausible, as they were all strongly weighted on clinical features, which represent some of the major determinants of MO development identified in epidemiological studies,” authors conclude.

The findings should be validated via multicenter prospective studies prior to making any ML approach into clinical practice available, the researchers note.

Ultimately, the findings strengthen “the theory advocated by precision medicine that data should be considered in a more general association, rather than individually,” they write.

Reference

Ferroni P, Zanzotto FM, Scarpato N, et al. Machine learning approach to predict medication overuse in migraine patients. Comput Struct Biotechnol J. Published online June 12, 2020. doi:10.1016/j.csbj.2020.06.006

Related Videos
David Yeomans, PhD
David Yeomans, PhD
David Yeomans, PhD
David Yeomans, PhD
All of the panelists for this series
An image of all of the participants
An image of all of the participants
An image of all of the participants
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.