Machine learning-based automated classification of headache disorders using patient-reported questionnaires

DOI: https://doi.org/10.1038/s41598- 020- 70992- 1

Abstract-Summary Classification of headache disorders is dependent on a subjective self-report from patients and its interpretation by physicians.

We aimed to apply objective data-driven machine learning approaches to analyze patient-reported symptoms and test the feasibility of the automated classification of headache disorders.

The self-report data of 2162 patients were analyzed. The first layer classified between migraine and others, the second layer classified between tension-type headache (TTH) and others, and the third layer classified between trigeminal autonomic cephalalgia (TAC) and others, and the fourth layer classified between epicranial and thunderclap headaches.

In the test cohort, our stacked classifier obtained accuracy of 81%, sensitivity of 88%, 69%, 65%, 53%, and 51%, and specificity of 95%, 55%, 46%, 48%, and 51% for migraine, TTH, TAC, epicranial headache, and thunderclap headaches, respectively.

We showed that a machine-learning based approach is applicable in analyzing

patient-reported questionnaires.

Extended: The first layer classified the most dominant subtype (i.e., migraine)

and the rest (i.e., non-migraine).

Introduction The diagnosis of headache disorders is highly dependent on self-report from patients and the interpretation of the self-report by clinicians.

The International Classification of Headache Disorder (ICHD) was published to

aid a standardized diagnosis of headache disorders [51].

There have been efforts to aid the diagnosis of primary headache disorders using neurophysiological tests [52], neuroimaging [53, 54], and blood-based biomarkers [55, 56]; however, these have not replaced clinical interviews.

468

3 Diagnosis

Previous studies have mainly focused on migraine with little focus on the dif-

ferential diagnosis of other headache disorders [57, 58].

The clinical diagnosis of headache disorders should, however, be based on a

holistic approach since a single characteristic cannot replace the proper diagnosis. We aimed to analyze self-reported symptoms of patients to classify four head-

ache disorders including migraine, by using machine learning approaches.

Methods We applied the least absolute shrinkage and selection operator (LASSO) [59] in choosing a few important features for each stacked classifier layer.

These features were chosen as the set of stable features and the threshold of three was chosen to maximize the classifier performance on average in the left-out fold in the training cohort within the tenfold cross-validation.

The selected stable features were used to train the stacked XGBoost classifier. To ensure the methods used in our study are well-suited in classifying headache subtypes, we compared our feature selection method (LASSO) with support vector machine recursive feature elimination (SVM-RFE) [60] and minimum-redundancy maximum-relevancy (mRMR) [61] approaches.

The numbers of the selected features using mRMR and SVM-RFE for each clas-

sifier layer were fixed as those of LASSO.

We also compared XGBoost with other binary classifiers such as k-nearest neighbor (k-NN), support vector machine (SVM), and random forest in each of the stacked layers with features selected by LASSO.

Results The top three prominent features in the fourth layer (epicranial headache vs. thun- derclap headache) were location: retroauricular, nature of pain: electric shock-like, and nature of pain: jabbing, assuming epicranial headache as the positive subtype in the specific headache syndromes classifier.

The stacked XGBoost classifier using the selected features attained an accuracy of 82%, sensitivity of 87%, 66%, 85%, 65%, and 64% for the five subtypes, and specificity of 94%, 54%, 58%, 63%, and 57% for the five subtypes in the train- ing cohort.

The stacked XGBoost classifier using the selected features led to an accuracy of 81%, sensitivity of 88%, 69%, 65%, 53%, and 51% for the five subtypes, and specificity of 95%, 55%, 46%, 48%, and 51% for the five subtypes in the test cohort.

We compared XGBoost with k-NN, SVM, and random forest classifiers in each of the stacked layers in terms of overall accuracy, minimum sensitivity, and mini- mum specificity.

Discussion We applied a machine learning approach to classify major headache disorders using questionnaires completed by patients in a real-world setting.

3.1 Classification

469

The performance of the machine learning approach in the classification of migraine was excellent however, its accuracy in classifying headache disorders other than migraine was inferior to that in classifying migraine.

Our study is one of the first studies to apply machine learning in the analysis of

patient-reported questionnaires to classify primary headache disorders [57].

Existing studies on the classification of headache disorders with machine learn- ing have focused on a few selected headache disorders such as migraine and tension- type headache due to challenges with sample size [57, 58].

This important feature should be always considered in the differential diagnosis of secondary and primary headaches, but it has not been listed in the ICHD-3 crite- ria for migraine, TTH, and epicranial headaches [51].

Acknowledgement A machine generated summary based on the work of Kwon, Junmo; Lee, Hyebin; Cho, Soohyun; Chung, Chin-Sang; Lee, Mi Ji; Park, Hyunjin. 2020  in Scientific Reports.

Validation of an algorithm for automated classification of migraine and tension- type headache attacks in an electronic headache diary