Issues in the Aggregation of Data on Retinal Structure and Function in Schizophrenia

It may seem to some readers that the retina is far removed from what is important about schizophrenia. However, the retina is part of the central nervous system (CNS), and published reports continue to indicate that schizophrenia is associated with thinning of retinal layers (using optical coherence tomography; OCT), reduced retinal cell responses (using electroretinography; ERG), and, in some cases, widened retinal venules as measured with fundus photography.^1,2 These data may both contribute to a better understanding of the multiple forms of subjective and laboratory-based visual processing changes observed in people with schizophrenia,³ and serve as a window into brain function in the syndrome. Kazakos and Karageorgiou⁴ have provided a valuable contribution to this growing field by providing an up-to-date meta-analysis of this literature, and the first meta-analysis based on individual participant data.

One of the main conclusions from the paper is that, despite nearly every published study indicating significant between-group differences on multiple variables, the ability of retinal indices to discriminate schizophrenia and control groups was generally fair to poor, with only OCT indices related to macula volume and thickness performing better. Such analyses are critical for evaluating cumulative progress in this growing field of research. However, considering the heterogeneity within the diagnostic category of schizophrenia and the substantial overlap typically observed between patient and control groups (e.g., 45%, 50%, and 65% for biological, psychophysiological, and cognitive markers, respectively⁵), mapping onto psychiatrist- or structured interview-generated diagnoses may not be the most important outcome to assess in this field (or for the cognitive neuroscience of schizophrenia in general). (Even with a [typically not achieved] between-groups effect size of d = .80 (large), there will be 52.6% overlap between group scores when values in both groups are normally distributed.) Instead, a more productive approach may be to determine the concurrent and predictive implications of abnormal retinal findings in subgroups within the category of schizophrenia and/or psychotic disorders. Related to this, one interesting feature of OCT findings in other neuropsychiatric conditions (e.g., multiple sclerosis, Parkinson’s disease) and in normal aging is that retinal thinning significantly predicts features such as cognitive impairment, cognitive decline, brain volume loss, and overall disease progression.^6–9 Therefore, the primary clinical utility of techniques such as OCT may be in rapid and reliable screening for overall CNS difficulties and their progression over time. This could help identify people who could benefit from more intensive evaluations (e.g., MRI, neuropsychological assessment) to clarify level of CNS impairment and plan for rehabilitative interventions.

It must also be kept in mind that meta-analysis has its limitations and is not free from bias.¹⁰ For example, in the case of ERG, we and others have shown that participants with schizophrenia show no anomalies in their retinal response under certain conditions (e.g., low intensity light), and the differences between the parameters under which patient and control groups do and do not differ can be very informative.¹¹ However, when data from all ERG tests are pooled to get a composite score for each waveform (as in ref. 4), the high sensitivity of some variables is lost by being combined with data from tests for which there are no between-group differences. Also, when state or treatment effects exist,¹² pooling across studies that include samples at different phases of illness washes out these effects in the service of looking for an overall effect of diagnosis. This issue is also relevant to OCT studies because individuals in a current or recent psychotic episode, or within the first few years of illness, are more likely to be characterized by neuroinflammation, which can lead to edema and temporally inflated retinal thickness values that mask the presence of underlying tissue loss.¹³ In addition to contributing to an underestimate of effect sizes for OCT metrics in meta-analyses, neuroinflammation could contribute to a more pronounced inverse relationship between age and OCT values in schizophrenia samples, similar to the one reported in the current paper. Explicit modeling of phase effects in future meta-analyses would therefore be useful.

In summary, the report by Kazakos and Karageorgiou should serve as a wake-up call to many readers that pathology of neural structure and function extends beyond the brain in schizophrenia. The findings emerging from this growing field should also remind us that characterizing schizophrenia by a single feature (e.g., as a cognitive disorder; as a disorder of prefrontal cortex function) is incomplete and misleading. At the same time, while the meta-analysis indicates that patient-control group separation is not optimal at present, this should motivate studies that move beyond trying to map retinal metrics onto often unreliable and questionably valid heterogeneous diagnostic categories, and onto factors such as neurodevelopmental profiles (e.g., presence vs. absence of neurological soft signs and sensory impairments in childhood¹⁴; good vs. poor premorbid social and academic functioning¹⁵), short- and long-term trajectories of CNS change (e.g., extent and patterns of gray and white matter loss¹⁶), and levels of functioning (e.g., courses characterized by stability and recovery vs. those involving chronic disability, frequent hospitalization, and cognitive decline^17,18).