Multivariate eye movement psychophysiology accurately differentiates schizophrenia cases from unaffected controls

Philip J. Benson, Sara A. Beedie, Ina Geigling, Dan Rujescu, David St. Clair

Research output: Contribution to journalArticlepeer-review


Background: The ease with which eye movements can be measured initiated a century of research into psychophysiological dysfunction in psychiatric populations. Initial findings were promising, suggesting performance on oculomotor tasks was influenced by the status of the observer's mental health. Research using family pedigrees subsequently provided evidence for a link between eye movement dysfunction and inherited risk of serious mental illness. Most recently, new and unexpected genetic and clinical overlaps have been discovered between debilitating neurodevelopmental illnesses such as schizophrenia and affective disorders. This has naturally cast doubt upon the likelihood of finding an endophenotype for schizophrenia per se, and may also explain why a trait marker for the illness has not been found. While eye movements can be assumed to act as an index of various on-going neural and mental processes, their prevalence as an abnormal feature of schizophrenia is as varied as the illness is heterogeneous. Equally important is the fact that 'abnormal' eye tracking is also found in the normal population. The use of visual scanpaths as method for examining more general scene perception in schizophrenia has prompted a resurgence of interest in using eye movements in the test repertoire. The relative importance of scanpaths, smooth pursuit and saccadic control as atypical processes in schizophrenia is unknown as is their combined efficacy in delineating major psychiatric illnesses.

Methods: Eye tracking performance was measured in 95 out-patients meeting clinical diagnostic criteria for schizophrenia, and 93 age-matched controls. Tasks included smooth pursuit of horizontal and Lissajous sinusoidal targets, fixation stability (as a proxy for saccadic control), and scanpath formation during free-viewing of natural scenes. Statistical models were then used to evaluate composite fixation and saccade activity in each participant group. Cross-validation was used to prevent over-fitting of data to the model. Individuals with incomplete tests (missing values) were included in model training and testing.

Results: Perfect separation of cases and controls was obtained using a supervised gradient boosting model to build a series of decision trees using the multivariate data. Classification was determined by the weighted sum of each tree's fit for a given observer. Eye movement data from 30 new cases and controls and re-test sessions confirmed the high predictive validity of the model.

Discussion: Restricted scanpaths emerged as the most influential aspect of eye movement dysfunction in schizophrenia. Particular elements of pursuit accuracy and saccade inhibition also featured prominently in the model. Each of these discriminators confirmed findings from previous studies, but only by combining measures from several tasks could their utility be exploited to discriminate between individuals with schizophrenia and unaffected controls. The multivariate model succeeded because it was able to capture phenotypic variation on a continuum within a category. No consistent patterns of association were found with illness duration or onset, nicotine and caffeine intake, age, sex or medication. These are extremely exciting findings. We can now examine the specificity of these results and delineation of the natural boundaries of illnesses sharing common genetic pathways with schizophrenia.
Original languageEnglish
Pages (from-to)248-249
Number of pages2
JournalSchizophrenia Research
Issue number2-3
Early online date20 Mar 2010
Publication statusPublished - Apr 2010


Dive into the research topics of 'Multivariate eye movement psychophysiology accurately differentiates schizophrenia cases from unaffected controls'. Together they form a unique fingerprint.

Cite this