Background: At present there are no objective tests or biomarkers in routine use that can help with the diagnosis of major mental disorders. This study aimed to determine the potential value of eye movement behaviour patterns for distinguishing individuals with schizophrenia compared to those with major affective disorders and control groups. Methods: Using a boosted decision tree multiclass classifier in a machine learning framework, we compared eye movement viewing patterns in a training set of UK subjects with Schizophrenia (SCZ) n=120, Bipolar affective disorder (BPAD) n=141, Major depressive disorder (MDD) n=136 and control (CON) n=142 group. A total of 133 individuals with group ratios similar to the training set acted as a hold-out test set for internal validation and a German cohort of SCZ (n=60) and a Scottish cohort of CON subjects (n=184) were used as a semi-independent test set for external validation. Findings: Based on internal validation, estimates of the area under the curve (AUC) for the one-versus-all comparisons for SCZ (0.85), BPAD (0.78), MDD (0.76) and CON (0.85) (with an overall estimate of 0.81) suggested high levels of discriminability between groups. The estimates of AUCs for the one-versus-one comparisons were also reasonably high; SCZ v CON 0.84, SCZ v BPAD 0.82, SCZ v MDD 0.76. The classifier showed good discrimination of BPAD and MDD from CON with AUCs 0.85 and 0.82, respectively; however, the performance was relatively lower between BPAD versus MDD with AUC 0.69. The estimates of AUC based on external validation were comparable: 0.89 for German SCZ and 0.65 for Scottish CON. The best individual discriminators included free viewing, fixation duration and smooth pursuit tasks. Our findings appear robust in the presence of potential confounders such as age, sex, medication or mental state at the time of testing. Interpretation: Eye movement behaviour patterns have good power to discriminate schizophrenia from both the major mood disorders and control subjects. The performance characteristics are within the range required of diagnostic biomarkers. Our findings are consistent with the Kraepelinian dichotomy of the functional psychoses. Funding: Royal Society of London, Chief Scientist Office Scotland (CZB/4/734), NHS Grampian,Tenovus Scotland (G12/31), Miller MacKenzie Trust, EU-FP6 (SGENE) and Health Innovation Challenge Fund, Wellcome Trust and Department of Health (WT-103911/Z/14/Z Declaration of Interest: Philip Benson and David St Clair are co-founders of SACCADE Diagnostics Ltd a spin out company tasked to develop eye movement technology to assist diagnosis of major mental health disorders. The University of Aberdeen has patents pending in Europe (PCT/GB2013/050016) and USA (14/370,611). The data reported in this paper arose solely from the acknowledged UK research bodies and charities none of whom have vested interests in the company. There are otherwise no competing interests. Ethical Approval: The Scottish and German studies obtained full multiregional ethics committee (MREC) and institutional review board (IRB) approvals respectively and were conducted in accordance with the Declaration of Helsinki.
Read full abstract