Schizophrenia Research Forum - A Catalyst for Creative Thinking

Study Finds Poor Eye Movement and Attention Predict Schizophrenia Risk

20 March 2007. Deficits in attention and visual performance can identify individuals at risk for developing schizophrenia, according to a study by Mark F. Lenzenweger of the State University of New York at Binghamton and colleagues. The findings were published in the February issue of the Journal of Abnormal Psychology. Despite reports that people with schizophrenia and their biological relatives do poorly at paying attention and visually following moving objects, researchers had not previously taken the leap to using these dysfunctions to predict who is or is not at risk for the disease.

Attention and Eye-tracking Deficits as Endophenotypes
To speed the search for the genetic underpinnings of disease, some researchers look to endophenotypes, heritable disease markers that, although invisible to the naked eye, can be measured (see SRF Live Discussion led by Irving Gottesman). Endophenotypes may have simpler genetic determinants than diagnostic groups because they lie earlier in the gene-to-behavior pathway.

Lenzenweger and collaborators Geoff McLachlan at the University of Queensland, Australia, and Donald B. Rubin at Harvard University write that research supports regarding deficits in sustained attention (Wang et al., 2007) and visual tracking (Calkins and Iacono, 2000; Levy et al., 1994) as schizophrenia endophenotypes. Although most genetic models assume that people either have or lack schizophrenia risk, studies had not tested whether the structure underlying these abnormalities echoes this assumption.

To probe these markers’ latent structure, the investigators recruited community residents 18 to 45 years old for a study. They excluded those with a history of psychosis or of using antipsychotic medications. No participant tested psychotic.

The study measured sustained attention using the Continuous Performance Test—Identical Pairs Version, which asks participants to respond when two identical numbers appeared consecutively. In the eye-tracking task, people watched a red square move across a computer screen and pressed a button whenever they saw an “X” in the square change to an “O” or vice versa. The test gauged smooth pursuit eye movement, or the ability to follow a target, as well as saccades—small, jerky movements—to catch up with the target.

A statistical approach called finite mixture modeling showed that a two-group solution best fit the attention and eye-tracking data from 294 study participants. It estimated that 27 percent of subjects belonged in the schizotypic, or at-risk group, versus 73 percent not at risk. In contrast, the authors note, studies using psychometric measures have typically deemed about 10 to 15 percent of participants as at risk.

The analysis also estimated each individual’s probability of group membership. Splitting the sample based on those probabilities put 62 people in the presumed schizotypic group.

A Robust, Specific Finding
If the components truly reflect schizotypy, the investigators reasoned, those in the presumed schizotypic group should have more schizophrenia-related symptoms and relatives with the disorder than the other group. As expected, at-risk participants scored higher on reality distortion, negative symptoms, and disorganization, as well as overall symptoms, on the Schizotypal Personality Questionnaire. In addition, the participants reported that more of their biological relatives received treatment for schizophrenia.

Next, the researchers wondered whether the at-risk group might simply be more impaired in ways not specific to schizophrenia. Analyses ruled out between-group differences in education, intelligence, and family history of psychiatric disorders, including depression, bipolar disorder, anxiety disorders, alcohol or other drug abuse, and obsessive-compulsive disorder. “Thus, it appears that on the basis of the family history data, the schizotypic component was not merely tapping general psychosis-related liability in the subjects,” the authors note.

Lenzenweger and associates even tested whether a statistical technique called taxometric analysis would uphold the mixture modeling results. It not only confirmed the two groups, but also the estimate that 27 percent of the population belonged to the at-risk group, findings they called “relatively robust.”

“We stress that the 27 percent figure should not be taken to mean that 27 percent of the population is going to develop schizophrenia, as epidemiological data clearly do not support this,” the investigators write. More people may harbor liability for the illness than show schizophrenia-related psychopathology.

The researchers concluded that their findings support the all-or-none and tipping-point models of schizophrenia risk. They bring hope that clinicians might someday be able to spot schizophrenia-prone patients by checking their eye tracking and attention. Meanwhile, the approach used by Lenzenweger and colleagues may offer a more objective way to choose people for genomic study, using laboratory measures rather than clinical judgments.—Victoria L. Wilcox.

Lenzenweger MF, McLachlan G, Rubin DB. Resolving the latent structure of schizophrenia endophenotypes using expectation-maximum-based finite mixture modeling. J Abnormal Psychol. 2007;116(1):16-29. Abstract

Comments on News and Primary Papers

Primary Papers: Resolving the latent structure of schizophrenia endophenotypes using expectation-maximization-based finite mixture modeling.

Comment by:  Deborah Levy
Submitted 20 March 2007
Posted 20 March 2007

The recent paper by Lenzenweger, McLachlan, and Rubin unifies several important themes in schizophrenia research in a manner that provides a powerful methodological and statistical approach for uncovering the latent structure of empirical data. In particular, it provides a quantitatively based method for assigning risk for schizophrenia while also helping to resolve heterogeneity. Specifically, what are the different meanings and significances of different conceptualizations about the relationship between schizophrenia liability and endophenotypes? This study represents an ambitious effort to grapple with critical and complex issues facing schizophrenia researchers.

Endophenotypes for schizophrenia liability have received considerable attention in recent years, beginning with the pioneering work of Philip S. Holzman, who first recognized the potential significance of the unusually high rates of certain traits in clinically unaffected relatives. The general appeal of endophenotypes is that they are strongly associated with schizophrenia and occur in first-degree relatives of schizophrenics at a much higher rate than schizophrenia itself. In the typical linkage and family study design, ascertainment of families occurs through a schizophrenic proband. Conditioning on schizophrenia in the proband capitalizes on the strong coupling between the disease phenotype, which has relatively low penetrance, and endophenotypes, which have much higher penetrance. If a major gene for an endophenotype is one of several genes for schizophrenia, conditioning on schizophrenia in the proband maximizes selection for a gene that has a higher probability of being detected through its effect on an endophenotype than through its effect on the disease phenotype. The presence of schizophrenia in the proband selects for genes for the endophenotype in relatives, and allele sharing for a closely linked marker would increase the rate of the endophenotype in first-degree relatives. These considerations are the key rationale for using endophenotypes as surrogates to detect non-penetrant gene carriers in schizophrenia families.

One of the underlying assumptions of the endophenotype strategy is that the presence of the endophenotype has some likelihood of predicting schizophrenia liability, at least in certain populations (schizophrenic patients, first-degree relatives, and possibly in clinical and/or psychometrically defined schizotypes). As Lenzenweger and colleagues point out, this assumption can be stated in the following form: what is the probability of showing the endophenotype given that one has a liability for schizophrenia (P[Endophenotype|Schizophrenia Liability])? As useful as this question is in samples conditioned on the presence of schizophrenia spectrum conditions, it does not address the relation between an endophenotype and schizophrenia liability in the general population. This question asks what is the probability of having schizophrenia liability given that one has the endophenotype (P[Schizophrenia Liability|Endophenotype]). Using a general population sample, Lenzenweger and colleagues do just that.

The Lenzenweger et al. study also tests a fundamental assumption of most major theories of the underlying nature of schizophrenia liability. Specifically, it tests whether or not there is evidence of a discontinuity in the underlying structure of two widely used candidate endophenotypes, eye tracking dysfunction (ETD) and sustained attention. Meehl’s model, the Holzman-Matthysse latent trait model, and Gottesman’s multifactorial polygenic threshold model all assume the presence of a marked discontinuity in the latent structure of schizophrenia liability. Using finite mixture modeling, Lenzenweger et al. found compelling evidence for a discontinuity in the latent structure of the liability space mapped by eye tracking dysfunction and sustained attention. Their findings were essentially the same when a different mathematical formalism was applied, namely maximum-covariance taxometric analysis. These findings are especially salient, because the endophenotypes the authors studied (ETD, sustained attention deficits) have a ratio-scale measurement structure in that they were derived from laboratory measures. Thus, the findings cannot be a function of psychometric artifact, an issue that arose in discussions of earlier taxometric studies in the area.

What was especially compelling about the two-class solution obtained in this study, beyond its theoretical implications, was the apparent validity of the parsing. The smaller component of the two-class solution contained individuals that were highly schizotypic (although they themselves had no prior history of psychosis). Importantly, all cases of expressed schizophrenia were found only among the first-degree biological relatives of those subjects in the smaller of the two components—the schizotypy component. Moreover, other forms of psychopathology were found among first-degree biological relatives in the larger, non-schizotypy component (e.g., bipolar disorder, autism).

The precision and power that accrues from the statistical approach used by Lenzenweger et al.—finite mixture modeling—warrants special notice. This approach allows one to establish with quantifiable precision the actual probability that a given subject is a member of one or another component. This approach is far superior to cruder methods used to resolve heterogeneity with respect to deviance, such as decile cuts, median splits, and so on. The Lenzenweger et al. study illustrates the potential utility of finite mixture modeling for psychopathology research, and would be especially probative in research that has a genomic focus. For example, one could use this approach to select individuals for which genome-wide scans or association studies of candidate genes would be most informative.

The NIMH is currently seeking the unification of advanced statistical methodologies and enhancements in measurement that will facilitate the reduction of heterogeneity and further the search for the genetic basis of psychopathology: this study addresses that aim and more.

View all comments by Deborah Levy

Comments on Related Papers

Related Paper: The endophenotype concept in psychiatry: etymology and strategic intentions.

Comment by:  Patricia Estani
Submitted 3 January 2006
Posted 3 January 2006
  I recommend this paper

Related Paper: The endophenotype concept in psychiatry: etymology and strategic intentions.

Comment by:  Perla Villamor
Submitted 22 November 2007
Posted 25 November 2007
  I recommend this paper