Antisaccade Eye Movement Task
The saccadic distractibility score represents the frequency of the disinhibited reflexive saccades directed toward the target in the antisaccade task. The mean error rates for each of the three groups of subjects are shown in
table 1.
Table 1 also shows the effect of the explanatory variables—group and a combined variable (age and gender)—on the reflexive saccade and antisaccade tasks by showing the deviance (the amount of variance remaining unexplained after a model has been fitted) of the null model and the deviances when each explanatory variable was included in the model separately. Four models were fitted to each eye movement variable: 1) model A, the null model, which included the intercept/constant term, or “grand mean,” but none of the explanatory variables; 2) model B, which included only the explanatory variable group as a factor at three discrete levels (proband, relative, healthy comparison subject); 3) model D, which included age as a continuous covariate and gender as a factor at two discrete levels; and 4) model E, which included age, gender, and group simultaneously. The deviances for models A, B, D, and E are shown in
table 1. To test whether there was a significant group effect irrespective of age and gender, for each saccadic parameter we compared the difference in deviance between models A and B; these differences are also given in
table 1. To test for a significant difference between the three groups after adjustment for age and gender, we examined the change in deviance when moving from model D to model E. The level of significance for differences was determined with reference to the chi-square distribution with two degrees of freedom.
Table 1 shows that for saccadic distractibility, the deviance for the model that included only group was 1067.45. When subtracted from the deviance for the null model, this figure produced a significant chi-square value of 15.69. Pairwise comparisons (with the Wald test in Mln) showed that the schizophrenic probands displayed a higher saccadic distractibility error rate on the antisaccade task (mean=53%) than both the healthy comparison subjects (mean=27%, estimate=25.6, standard error=6.396, df=1, p<0.001) and the first-degree relatives (mean=33%, estimate=22.28, standard error=5.553, df=1, p<0.001). (Statistical significance was determined by dividing the parameter estimate by the standard error for the degrees of freedom.) There was no significant difference in saccadic distractibility between the relatives and the healthy comparison subjects. However, all groups showed a high proportion of corrective antisaccades following the errors (probands, 86%; relatives, 91%; healthy subjects, 94%). There was a significant main effect of gender (χ
2=5.49, df=1, p<0.05): female subjects had consistently higher saccadic distractibility error rates across all groups (
table 2). The effect of age across all groups was not reliable (χ
2=3.55, df=1, n.s.). (These data yielded similar results with conventional analysis of variance [ANOVA] and nonparametric analyses.)
The findings in patients and unrelated healthy volunteers resemble those of previous studies in our laboratory (
15,
25). Overall, the relatives, when considered as a homogeneous group, did not show a significantly elevated rate of saccadic distractibility errors compared with the healthy subjects. However, further analyses revealed that the relatives of the schizophrenic probands with normal saccadic distractibility error rates (N=19) had a mean saccadic distractibility error rate of 17%, which did not differ significantly from the mean of the healthy subjects (27%), whereas the relatives of the probands with high saccadic distractibility error rates (N=25) had a mean saccadic distractibility error rate of 39%. A one-way between-groups ANOVA to compare these independent groups (the relatively high ratio of families to relatives rendered these data unsuitable for Mln modeling) showed that the error rate of the relatives of the probands with high errors was significantly greater than that of the relatives of the probands with normal errors (F=7.24, df=1, 43, p<0.02) and that of the healthy comparison subjects (F=4.91, df=1, 56, p<0.05). There were no significant differences on any other saccadic parameter between the relatives of the probands with high errors and the relatives of the probands with normal errors.
The demographic profile of the group of relatives differed from that of both the probands and the healthy subjects on mean age and male/female ratio. We therefore conducted further analyses to examine the possible effect of these factors on saccadic distractibility errors. There was a highly significant group effect on antisaccadic distractibility even after we took into account the variance attributable to age and gender (χ
2=19.79) (
table 1).
A number of considerations undermine the view that a specific age-related parameter could account for these results. 1) Age correlated only poorly with saccadic distractibility errors across all groups, accounting for only 2.9% of the overall variance (r=0.17, N=117, R2=0.03). The correlation was also poor in the group of relatives (r=–0.05, N=50, n.s.) and in the group of healthy subjects (r=0.24, N=38, n.s.), although there was a positive correlation for the probands (r=0.45, N=29, p<0.05). 2) The schizophrenic probands and the healthy subjects, who showed highly significant differences on saccadic distractibility errors, were closely matched on mean age (see above). 3) The relatives of the probands with high saccadic distractibility errors (mean age=47.9 years, SD=15.6) and probands with normal saccadic distractibility errors (mean age=45.5 years, SD=14.2) were also matched on mean age.
Primary saccade gain is defined as primary saccade amplitude divided by target amplitude. There was no main effect of group on primary antisaccade gain (χ
2=4.25) (
table 1). The mean final eye position of antisaccades was mildly hypermetric for the relatives compared with both the probands, who had mildly hypometric eye positions, and the healthy comparison subjects (
table 1). The group factor accounted for a significant proportion of the final eye position variance (χ
2=7.33); however, the effect was no longer evident once age and gender were introduced in the model (χ
2=4.07). No group effect on antisaccade latency was found (χ
2=5.04), although a weak group effect was obtained in the covariance analysis that included age and gender (χ
2=6.51) (
table 1).