Elsevier

Clinical Neurophysiology

Volume 121, Issue 12, December 2010, Pages 1998-2006
Clinical Neurophysiology

A pilot study to determine whether machine learning methodologies using pre-treatment electroencephalography can predict the symptomatic response to clozapine therapy

https://doi.org/10.1016/j.clinph.2010.05.009Get rights and content

Abstract

Objective

To investigate whether applying advanced machine learning (ML) methodologies to pre-treatment electroencephalography (EEG) data can predict the response to clozapine therapy in adult subjects suffering from chronic schizophrenia.

Methods

Pre-treatment EEG data are collected in 23 + 14 schizophrenic adults. Treatment outcome, after at least one year follow-up, is determined using clinical ratings by a trained clinician blind to EEG results. First, a feature selection scheme is employed to select a reduced subset of features extracted from the subjects’ EEG that is most statistically relevant to our treatment-response prediction. These features are then entered into a classifier, which is realized in the form of a kernel partial least squares regression method that performs response prediction. Various scales, including the positive and negative syndrome scale (PANSS) are used as treatment-response indicators.

Results

We determined that a set of discriminating EEG features do exist. A low-dimensional representation of the feature space showed significant clustering into clozapine responder and non-responder groups. The minimum level of performance of the proposed prediction methodology, tested over a range of conditions using the leave-one-out cross-validation method using the original 23 subjects, with further testing in an independent sample of 14 subjects, was 85%.

Conclusions

These findings indicate that analysis of pre-treatment EEG data can predict the clinical response to clozapine in treatment resistant schizophrenia.

Significance

If replicated in a larger population, this novel approach to EEG analysis may assist the clinician in determining treatment-efficacy.

Introduction

Compared with other antipsychotic medications the atypical antipsychotic medication clozapine is recognized to have superior therapeutic effectiveness in the treatment of chronic medication-resistant schizophrenia (e.g., Essali et al., 2009). However, clozapine may produce serious side effects such as seizures, cardiac arrhythmias or bone marrow suppression with neutropenia (Young et al., 1998). According to a recent Cochrane review, about 34% of treatment-resistant patients respond to clozapine while 3.2% develop blood problems (Essali et al., 2009). As the hematological side effects can be life threatening, blood samples to monitor the white blood cell count must be collected as long as the drug is used, at weekly to monthly intervals. The logistic difficulties for the patient and the treatment team are substantial. A method that could reliably determine, before the onset of therapy, whether a given patient will or will not respond to clozapine would greatly assist the clinician in determining whether the risks and logistic complexity of clozapine are outweighed by the potential benefits.

Quantitative electroencephalography (QEEG or EEG) may offer some promise in this regard. EEG abnormalities in schizophrenic subjects and EEG changes due to clozapine therapy have been the focus of a number of clinical studies (see e.g., Gunther et al., 1993, Malow et al., 1994, Freudenreich et al., 1997, Hughes and John, 1999, Knott et al., 2001, Adler et al., 2002, Knott et al., 2002, Birca et al., 2006, Coburn et al., 2006, Dunki and Dressel, 2006, Oikonomou et al., 2006, Sakkalis et al., 2006, Boutros et al., 2008).

Based on findings in 17 schizophrenic subjects, Knott et al. (2000) found that the clozapine-induced improvement of psychopathology symptom ratings using the Positive and Negative Syndrome Scale (PANSS) was correlated with pre-treatment QEEG inter and intra-hemispheric spectral power asymmetry. Greater pre-treatment anterior to posterior asymmetry in the delta frequency range was associated with greater improvement in negative symptoms while greater pre-treatment anterior to posterior theta asymmetry predicted improvement of positive symptoms and global improvement. Larger inter-hemispheric asymmetry in the theta and beta frequencies in the central and anterior temporal regions were, respectively, predictive of greater improvement in positive and negative symptoms. Gross et al. (2004) also found that changes in the theta frequency in QEEG with clozapine treatment, particularly in the midline electrodes over the fronto-central scalp area, were a more sensitive indicator for the evaluation of clozapine treatment efficacy than the serum clozapine level. Though these methods reveal important relationships between QEEG variables and clinical outcome, a series of simple correlational analyses do not readily yield a “responder” or “non-responder” dichotomous categorization for an individual patient.

The above analyses employed standard statistical methods. On the other hand, a more mathematically sophisticated analysis including pattern recognition and dimensionality reduction methods (which together may be categorized as machine learning techniques) can perform a more comprehensive data analysis. Machine learning techniques are finding increasing application in psychiatry, particularly when multi-dimensional, noisy, highly complex data or multi-modal data sets are analyzed together, (see e.g., Gallinat and Heinz, 2006). For example, support vector machine (SVM) techniques that select spectro-temporal patterns from multichannel magnetoencephalogram (MEG) data collected during a verbal working memory task have been used to distinguish schizophrenic from control subjects (Ince et al., 2008). Machine learning algorithms using structural brain magnetic resonance (MRI) images (Fan et al., 2007), functional MRI (fMRI) data (Guo et al., 2008, Kim et al., 2008) and combined genomic and clinical data (Struyf et al., 2008) have been employed to separate schizophrenic, bipolar and healthy control subjects.

Machine learning approaches have also been applied to prediction of clozapine treatment-efficacy. Lin et al. (2008) describes a study in which a feed-forward multilayer perceptron network (with a back-propagation error training technique) is employed using clinical and pharmacogenetic data to predict clozapine response in schizophrenic subjects. Five pharmacogenetic variables and five clinical variables (including gender, age, height, baseline body weight, and baseline body mass index) were collated from 93 schizophrenic subjects taking clozapine, including 26 responders. Using this method, they obtained an overall prediction accuracy rate of 83.3%.

Guo et al. (2008) describes a Bayesian hierarchical model using pre-treatment fMRI and positron emission tomography (PET) information coupled with patient characteristics (e.g. medical or family history and genotype) as training data to predict changes in brain activity in 16 schizophrenic subjects following treatment with two atypical antipsychotics (risperidone or olanzapine). The authors postulated that predicting drug-induced changes in brain activity would assist the clinician in determining optimal drug choice.

However, the clinical utility of these previous approaches is negatively impacted by the expense and unavailability of complex methods such as fMRI, PET, genetic screening and MEG. In contrast, electroencephalography (EEG) is an inexpensive, non-invasive technique widely available in smaller hospitals and in community laboratories. Therefore, predictive algorithms dependent on EEG measurements are more practical. Furthermore, since the required EEG data is acquired during the resting state, only minimal cooperation is required from the patient. Thus, an EEG based method of predicting treatment response would have many advantages over imaging methods such as MRI, PET or MEG.

The goal of the present pilot study is to examine the utility of machine learning (ML) methods for processing EEG signals to predict the response of schizophrenic subjects to clozapine.

Section snippets

Quantitative EEG recordings

We collected pre-clozapine resting EEG data from chronically ill, treatment-resistant schizophrenic subjects prior to beginning clozapine therapy. The data were collected without change to the patient’s current medication regimen. EEG was recorded with the patient in a semi-recumbent position in a sound attenuated, electrically shielded room by an experienced technician who prompted patients on signs of drowsiness. Sessions were arranged in the mornings and patients were requested to avoid

Treatment-efficacy prediction performance

The first set of results uses data from Group A which consists of 23 subjects. The set of candidate features were extracted from the pre-treatment EEG data and then reduced into a set of Nr = 8 most-relevant features using the available training set data, as discussed in Section 2.3. The prediction performance was then evaluated using the leave-one-out cross-validation procedure discussed previously. The performance evaluation results using the combined EO and EC EEG data sets together for the 23

Discussion

Our findings support the potential utility of machine learning methods in clinical psychiatry. In the current example we have been able to predict, in advance of the first dose, whether a treatment–resistant patient will or will not respond to a powerful but potentially toxic medication. In various experiments, we evaluated the performance of advanced prediction models in conjunction with kernelization methods to analyze pre-treatment EEG to predict the responsiveness to clozapine. These

Appendix A. The QCA clinical rating procedure

The QCA clinical rating procedure was devised in the context of an un-related earlier naturalistic retrospective un-published clinical study of treatment-resistant schizophrenic patients being considered for clozapine treatment. The subjects in the present study were included in this previous study. An experienced clinician reviewed all the available clinical descriptive information of the patient’s symptomatology prior to beginning a course of clozapine. Reported symptoms, corresponding to

Acknowledgements

The authors would like to thank Margarita Criollo, Joy Fournier, and Eleanor Bard for their help in clinical experiments. This work was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC).

References (35)

  • K.L. Coburn et al.

    The Value of Quantitative Electroencephalography in Clinical Psychiatry: A Report by the Committee on Research of the American Neuropsychiatric Association

    J. Neuropsychiatry Clin. Neurosci.

    (2006)
  • Cover TM, Thomas, JA. Elements of Information Theory, 2nd Ed. John Wiley & Sons,...
  • Essali A, Haj-Hasan NA, Li C, Rathbone J. Clozapine versus typical neuroleptic medication for schizophrenia. Cochrane...
  • Y. Fan et al.

    COMPARE: classification of morphological patterns using adaptive regional elements

    IEEE Trans. Medi. Imaging

    (2007)
  • S.F. Faux et al.

    P200 topographic alterations in schizophrenia: evidence for left temporal-centroparietal region deficits

    Electroencephalogr. Clin. Neurophysiol. Suppl.

    (1987)
  • J. Gallinat et al.

    Combination of multimodal imaging and molecular genetic information to investigate complex psychiatric disorders

    Pharmacopsychiatry

    (2006)
  • A. Gross et al.

    Clozapine-induced QEEG changes correlate with clinical response in schizophrenic patients: a prospective, longitudinal study

    Pharmacopsychiatry

    (2004)
  • Cited by (73)

    • Pattern classification as decision support tool in antipsychotic treatment algorithms

      2021, Experimental Neurology
      Citation Excerpt :

      Although it is not clear how antipsychotics contribute to these abnormalities, EEG connectivity measures as well as other quantitative indices from resting state EEG or evoked potentials have been used to predict antipsychotic drug response: For example, Masychev et al. (Masychev et al., 2020) applied symbolic transfer entropy to resting state EEG in combination with machine learning algorithm to discriminate responders vs. non-responders to clozapine treatment with 89.90% accuracy. Moreover, a machine-learning algorithm applied on pre-treatment EEG features for classification of clozapine responders and non-responders, using first the leave one-out cross-validation procedure and then external validation datasets, resulted in more than 85% accuracy, with EEG features from the left temporal areas having the greatest predictive value (Khodayari-Rostamabad et al., 2010). Finally, Ravan et al. assessed EEG predictors to clozapine response (Ravan et al., 2015) based on evoked responses to an auditory oddball paradigm recorded before and after one year of clozapine treatment.

    • Quantitative biomarkers to predict response to clozapine treatment using resting EEG data

      2020, Schizophrenia Research
      Citation Excerpt :

      In comparison to other neuroimaging techniques, EEG is an inexpensive and non-invasive technique that is available even in small hospitals and private clinics. Therefore, predictive algorithms based on EEG signals are more practical and thus have been the focus of a number of clinical studies (e.g. Gross et al., 2004, Kang et al., 2001, Khodayari-Rostamabad et al., 2010, Knott et al., 2000, Pillay et al., 1996, Ravan et al., 2015). Pillay et al. (1996) investigated whether clozapine response can be predicted by minor EEG abnormalities such as generalized slowing or sharp waves, spikes, spike-wave patterns, and focal dysrhythmias using 86 patients.

    • Support vector machine

      2019, Machine Learning: Methods and Applications to Brain Disorders
    View all citing articles on Scopus
    View full text