Exploring the Temporal Relationship Between Cognitive Insight and Neurocognition in Schizophrenia: A Prospective Analysis

Participants

All study participants met full criteria for DSM–IV diagnoses of schizophrenia or schizoaffective disorder according to a consensus best estimate by psychiatrists (MD) and psychologists (PhD). Information generated from semistructured clinical symptom interviews (see the later discussion) served as the basis for diagnostic determinations. Data about socioeconomic status (SES) were not systematically collected in these studies. Nevertheless, because both samples consisted of individuals suffering from severe and persistent mental illness and were recruited from the local (Philadelphia) community, the majority of participants were likely low SES.

Measures

Cognitive insight was measured using the BCIS (Beck & Warman, 2004). This scale consists of 15 items (see Table 1), 9 assessing self-reflectiveness (e.g., “I have jumped to conclusions too fast”) and 6 assessing self-certainty (e.g., “My interpretations of my experiences are definitely right”). Participants are asked to rate how much they agree with each statement on a scale from do not agree at all (0) to completely agree (4). Scaled responses for the self-reflection and self-certainty items are summed separately, and then a composite index score is created by subtracting the latter from the former (so higher index scores reflect greater cognitive insight). This scale has been shown to be reliable, as well as to have good convergent and discriminant validity in schizophrenia research (Beck & Warman, 2004; Martin et al., 2010; Pedrelli et al., 2004). Internal consistency of the BCIS index in the present research was acceptable (Cronbach’s α = .71 at baseline in Study 1).

OPEN IN VIEWER

Table 1.

Beck Cognitive Insight Scale Items

Item	Do not agree at all	Agree slightly	Agree a lot	Agree completely
1. At times, I have misunderstood other people’s attitudes toward me.
2. My interpretations of my experiences are definitely right.
3. Other people can understand the cause of my unusual experiences better than I can.
4. I have jumped to conclusions too fast.
5. Some of my experiences that have seemed very real may have been due to my imagination.
6. Some of the ideas I was certain were true turned out to be false.
7. If something feels right, it means that it is right.
8. Even though I feel strongly that I am right, I could be wrong.
9. I know better than anyone else what my problems are.
10. When people disagree with me, they are generally wrong.
11. I cannot trust other people’s opinion about my experiences.
12. If somebody points out that my beliefs are wrong, I am willing to consider it.
13. I can trust my own judgment at all times.
14. There is often more than one possible explanation for why people act the way they do.
15. My unusual experiences may be due to my being extremely upset or stressed.

Neurocognitive performance was assessed with the University of Pennsylvania Computerized Neurocognitive Battery (Gur, Ragland, Moberg, Turner, et al., 2001; Gur et al., 2010). The battery takes 1 to 1.5 hours to complete and consists of behavioral tasks from established paradigms shown to tap domains of neurocognition commonly impaired in individuals with schizophrenia (Gur et al., 2014; Gur, Ragland, Moberg, Bilker, et al., 2001). Previous research has shown that this battery has strong reliability (Gur et al., 2010) as well as convergent validity (e.g., with traditional, nonautomated neurocognitive tasks) and sensitivity to individual and developmental differences in unselected samples (Gur, Ragland, Moberg, Turner, et al., 2001; T. M. Moore, Reise, Gur, Hakonarson, & Gur, 2015) and individuals with schizophrenia (Gur et al., 2014; Gur, Ragland, Moberg, Bilker, et al., 2001). Scores from this test battery have generally been examined separately (for each neurocognitive domain). Because we did not formulate hypotheses regarding specific domains, we calculated composite neurocognitive performance scores to avoid inflated risk of Type I error due to multiple comparisons, as well as to maximize reliability and validity. Specifically, standardized (and norm-referenced) accuracy scores were averaged from domains with the strongest evidence for associations with both cognitive insight (Nair et al., 2014) and functional outcomes (Green et al., 2000), as well as specificity in schizophrenia (Heinrichs & Zakzanis, 1998; Lencz et al., 2006; Reichenberg & Harvey, 2007): abstraction/mental flexibility, attention/vigilance, and verbal learning/memory. Correlations between the three domain scores were strong (ranging from r = .29 to r = .56 at baseline in Study 1), in line with previous factor analytic work on this test battery (T. M. Moore et al., 2015). Higher scores reflect better neurocognitive performance.

Levels of positive and negative symptoms were also determined, using global rating scores from the Scale for the Assessment of Positive Symptoms (SAPS; Andreasen, 1984b) and the Scale for the Assessment of Negative Symptoms (SANS; Andreasen, 1984a). Previous research on schizophrenia supports the reliability and validity of these scales (Andreasen, Flaum, Arndt, Alliger, & Swayze, 1991; Norman, Malla, Cortese, & Diaz, 1996). Reliability of ratings in the present research is supported by both internal consistency of the individual items (at baseline in Study 1, α = .84 for the SAPS and α = .85 for the SANS) as well as test-retest reliability of global ratings (in Study 1, r = .62 for the SAPS and r = .64 for the SANS). Higher scores indicate more severe symptomatology.

Procedures

Institutional review boards from the University of Pennsylvania and the City of Philadelphia approved study procedures. Clinical interviews and ratings were administered by master’s- or PhD-level research personnel trained to criterion (intraclass correlations > .80). Collateral information obtained from family members, treatment providers, and chart review assisted in the determination of interviewer ratings. Participants received financial compensation for completing study assessments.

Statistical analyses

Research questions/hypotheses were tested using hierarchal linear regression. To account for age variability in our samples and potential covariation with neurocognitive performance, age was entered as a predictor in the first step of all regression analyses.

To test cross-sectional associations between cognitive insight and neurocognition, regression analyses were conducted examining whether BCIS index scores predict neurocognitive performance within the same time point. ² Central to the present research, prospective associations were tested by examining whether (a) BCIS scores predict neurocognitive performance at the following time point, beyond variance accounted for by initial neurocognitive performance (i.e., with neurocognitive performance scores from the previous assessment entered in step one); or conversely (b) neurocognitive performance scores predict later BCIS scores, beyond variance accounted for by initial BCIS scores. This analytic strategy is designed to model changes over the course of the study (here, in neurocognitive performance and cognitive insight) and is widely used to test temporal questions/hypotheses in longitudinal research (see, e.g., Arseneault et al., 2002; Mohamed et al., 2009). In the present research, if BCIS scores predict later neurocognitive performance (controlling for previous neurocognitive performance), this would be in line with the proposal that reduced cognitive insight precedes decrements in neurocognition. Conversely, if neurocognitive performance predicts later scores on the BCIS (controlling for previous BCIS scores), this suggests that decrements in neurocognition precede reductions in cognitive insight.

Test-retest (Pearson) correlations were also examined for both BCIS scores and the neurocognitive composites, to evaluate whether any differences in prospective associations might be due to differential stability of the measures. To evaluate whether there was adequate variability in cognitive insight and neurocognitive performance over time in our samples, unstandardized residualized change scores were computed (using regression) and examined. Finally, follow-up regression analyses (including residual scores from the symptom measures as predictors) were conducted to test whether any observed prospective associations between cognitive insight and neurocognition were driven by changes in other symptoms.

Participants

The sample consisted of 168 outpatients (64% male; ages 18–66: M = 44.1, SD = 11.6) with schizophrenia (n = 135) or schizoaffective disorder (n = 33). Most participants identified as African American (76%), Caucasian (17%), or biracial (4%). Within the sample, there was considerable variability in length of illness, ranging from 1 to 52 years (M = 22.7, SD = 12.7). Participants were recruited from the local community through referrals and a subject pool maintained at the University of Pennsylvania. Exclusion criteria for the study were as follows: (a) experiencing a head injury with documented loss of consciousness and (b) evidence of a condition that would compromise neurocognitive functioning. Throughout the study period, participants received standard community treatment (e.g., psychiatric medication, case management); 95% were currently taking antipsychotic medications at the time of study participation.

Additional measures

Severity of depression is also (negatively) correlated with performance on neurocognitive tasks (Brebion et al., 2000; Halari, Mehrotra, Sharma, & Kumari, 2006; McClintock, Husain, Greer, & Cullum, 2010; Moser, Krieg, Zihl, & Lautenbacher, 2006). To examine symptoms of depression in this study, we used scores from the Beck Depression Inventory–II (BDI; Beck, Steer, & Brown, 1996). The BDI is widely used and well-validated in research on depression in schizophrenia (Barnes, Curson, Liddle, & Patel, 1989; Berenbaum & Oltmanns, 1992; O. Moore, Cassidy, Carr, & O’Callaghan, 1999). Reliability in this sample was good, as evidenced by internal consistency (α = .88) and test-retest reliability (r = .58). Higher scores reflect more severe depressive symptoms.

Similarly, some studies have shown that clinical insight is linked with neurocognitive performance (see Nair et al., 2014). To test the specificity of our findings to cognitive insight, we examined ratings of clinical insight from the Positive and Negative Syndrome Scale (PANSS; Kay, Fiszbein, & Opler, 1987). These ratings are based on awareness of symptoms, need for treatment, and consequences of illness, with higher scores indicating greater observed impairments in clinical insight. Previous research supports the validity of these ratings, based on convergence with other established measures of clinical insight in schizophrenia (e.g., the Scale to Assess Unawareness of Mental Disorder; see Lysaker et al., 1997). Reliability in this sample is supported by a test-retest correlation of r = .42. It is notable that correlations between the PANSS insight ratings and BCIS index scores were modest (r = −.29 at baseline, r = −.23 at follow-up), supporting the distinction between clinical and cognitive insight (see also Greenberger & Serper, 2010; Tranulis et al., 2008).

Procedures

After providing informed consent, participants completed self-report questionnaires, clinical interviews, and the neurocognitive performance battery. Approximately 6 months later, participants were invited to return to the laboratory to complete the same measures again. ³ Data from this study were included in other manuscripts examining separate questions and constructs (Bredemeier, McCole, Luther, Beck, & Grant, 2017; Thomas et al., 2017).

Prospective associations between cognitive insight and neurocognitive performance were tested by computing hierarchal regression analyses to determine whether (a) baseline BCIS scores predict neurocognitive performance at 6 months (beyond variance accounted for by neurocognitive performance at baseline) or (b) baseline neurocognitive performance predict BCIS scores at 6 months (beyond variance accounted for by baseline BCIS scores). Follow-up regression analyses were conducted to determine if any prospective associations that emerged would be robust when accounting for (a) changes on the SAPS, SANS, and BDI and (b) baseline PANSS clinical insight ratings.

Results

Descriptive statistics from Study 1 are presented in the left panel of Table 2. It is notable that there was substantial variance in the residual scores (reflecting within participant change from baseline to follow-up) for the neurocognitive composite (SD = 0.79, range = −3.42 to 2.30), consistent with previous studies showing variability in neurocognitive performance over relatively short time frames (Nakagami et al., 2010; Szoke et al., 2008). Likewise, there was considerable variance on residual BCIS index scores (SD = 4.51, range = −11.10 to 9.96). This variability supports the validity of conducting prospective analyses.

OPEN IN VIEWER

Table 2.

Descriptive Statistics for All Study Measures

	Study 1				Study 2
	Baseline		6 months		Baseline		6 months		12 months		18 months		24 months
Measure	M	SD	M	SD	M	SD	M	SD	M	SD	M	SD	M	SD
BCIS index	2.97	6.82	3.57	5.94	6.15	6.27	5.98	7.04	6.24	6.34	6.32	6.01	5.44	5.41
NP battery	−1.19	1.17	−1.03	1.03	−0.97	1.28	−0.73	1.12	−0.68	0.93	−0.63	1.03	−0.62	0.97
SANS global	8.91	3.18	8.50	3.31	10.48	2.74	10.27	3.01	10.12	0.08	8.88	3.70	8.89	3.14
SAPS global	5.49	3.55	4.66	3.57	6.33	4.16	5.15	3.78	6.19	3.90	5.55	3.21	5.07	3.39
BDI	13.17	10.24	12.56	11.38	—	—	—	—	—	—	—	—	—	—
PANSS insight	3.45	1.28	3.76	1.32	—	—	—	—	—	—	—	—	—	—

Note: BCIS = Beck Cognitive Insight Scale; NP = neurocognitive performance; SANS = Scale for the Assessment of Negative Symptoms; SAPS = Scale for the Assessment of Positive Symptoms; BDI = Beck Depression Inventory; PANSS = Positive and Negative Syndrome Scale.

At baseline, the (cross-sectional) association between BCIS scores and neurocognitive performance was in the expected direction, but did not reach statistical significance, β = .12, p = .13 (n = 156). However, follow-up BCIS scores and follow-up neurocognitive performance were significantly associated, β = .19, p = .02 (n = 140). More important, baseline BCIS scores prospectively predicted neurocognitive performance at the 6-month follow-up, beyond variance accounted for by baseline neurocognitive performance (i.e., baseline BCIS scores predicted changes in neurocognitive performance from baseline to follow-up), β = .19, p < .01 (n = 130). Even when accounting for (residualized) changes in SAPS global, SANS global, and BDI scores (in addition to age and baseline neurocognitive performance), BCIS scores from baseline continued to predict neurocognitive performance at follow-up, β = .23, p < .01. ⁴ Conversely, baseline neurocognitive performance was not significantly associated with BCIS scores at follow-up when accounting for baseline BCIS scores, β = −.02, p = .79 (n = 131). Also, the BCIS and neurocognitive performance residual scores were not significant correlated with one another, r = .04, p = .69 (n = 129), and the test-retest correlations were quite comparable (r = .63 for the BCIS, r = .65 for neurocognitive performance). Finally, baseline PANSS clinical insight ratings were not significantly associated with neurocognitive performance at follow-up when accounting for baseline neurocognitive performance, β = −.12, p = .12 (n = 121), and the association between baseline BCIS scores and later neurocognitive performance remained robust when accounting for baseline levels of clinical insight, β = .23, p < .01.

Discussion

In summary, the findings from Study 1 provide further evidence for a link between cognitive insight and neurocognition in schizophrenia (Nair et al., 2014). Study 1 is the first study to show that cognitive insight prospectively predicts neurocognitive performance. These findings suggest two possibilities. The first is that declines in cognitive insight temporally precede diminished neurocognitive performance (or conversely, increases in cognitive insight precede improved neurocognition performance). The second is that persistently low cognitive insight is associated with diminishing neurocognitive performance (or conversely stable high levels of cognitive insight are associated with improving neurocognitive performance). Both accounts are consistent with the broader theory that diminished cognitive insight may contribute to neurocognitive impairment in schizophrenia.

We did not find evidence that neurocognitive performance prospectively predicts cognitive insight, which is inconsistent with the (more common) theory that neurocognitive deficits cause diminished cognitive insight. Similarly, we did not find evidence that scores changed concurrently (given that the two change scores were not correlated with each other). The prospective link between cognitive insight at baseline and neurocognitive performance at follow-up does not appear to be explained by the differential stability of the measures or changes in other symptoms/distress. Finally, this relationship did not emerge for ratings of clinical insight from the PANSS, and the prospective link between BCIS and neurocognition held when accounting for levels of clinical insight, suggesting that this prospective link is unique to cognitive insight.

Participants

The sample consisted of 60 outpatients (67% male, ages 19–60, M = 38.4, SD = 11.6) with schizophrenia (n = 48) or schizoaffective disorder (n = 12) recruited to participate in a randomized treatment trial comparing standard (community) treatment with standard treatment plus recovery-oriented cognitive therapy. Like Study 1, most participants identified as African American (65%) or Caucasian (32%). All individuals had prominent negative symptoms (at least moderate severity on two SANS subscales, or marked severity on one subscale), and thus neurocognitive impairments were also expected to be highly prevalent (given the established linked between neurocognition and negative symptoms; (see O’Leary et al., 2000, and Voruganti, Heslegrave, & Awad, 1997). For additional details about the sample and recruitment, as well as the study design and treatment conditions, see Grant et al. (2012).

Procedures

After providing informed consent, participants completed baseline assessments and then were randomized to a treatment condition. Blind follow-up assessments were conducted at 6 months, 12 months, 18 months (end of treatment), and 24 months (follow-up). ⁵ Although this sample was smaller than that of Study 1, additional variability in both constructs was anticipated (based on the longer follow-up and measurement in the context of treatments that may improve neurocognition or cognitive insight), mitigating concerns about reduced statistical power.

The same statistical approach (hierarchal regression) was used as in Study 1. Specifically, we tested whether cognitive insight predicts neurocognitive performance prospectively (at the next assessment, beyond variance accounted for by previous neurocognitive performance), or the reverse. To ensure that prospective associations were not driven by differences across the experimental groups, treatment condition was also entered as a predictor in these analyses (along with age). For data on improvements in positive and negative symptoms (as well as global functioning) over the course of the trial, see Grant, Bredemeier, and Beck (2017).

Because both variables were measured at more than two time points, we were able to examine the time course of the association between them more extensively. Specifically, this permitted us to examine changes on these variables in a time-lagged fashion (and thus, more directly confirm if changes on one precede changes on the other). In analyses predicting scores for later time frames (6 months predicting 12, 12 months predicting 18, and 18 months predicting 24), scores from the previous assessment point were also entered in the first step of the regression model, to examine unique variance at that assessment point. Based on the results from Study 1, we hypothesized that lower cognitive insight would prospectively predict diminished neurocognitive performance—thus, one-tailed p values are reported. Nevertheless, because we did not have a priori hypotheses about the four separate time frames tested in this study, a Bonferroni correction was used (resulting in an alpha level of .0125).

Results

Descriptive statistics from Study 2 are presented in the right panel of Table 2. As in Study 1, substantial variance was observed in residual scores for both the BCIS index (SDs from 3.38 to 5.25) and neurocognitive composite (SDs from 0.53 to 0.75). ⁶

Cross-sectional associations between BCIS and neurocognitive performance scores within each assessment point were consistently positive but did not reach statistical significance, β = .00 to .28. Consistent with our hypotheses and the central finding from Study 1, BCIS scores at the end of active treatment (18 months) prospectively predicted neurocognitive performance at the end of the follow-up period (24 months), after accounting for BCIS scores at 12 months and neurocognitive performance at 18 months (i.e., BCIS changes from 12 to 18 months predicts neurocognitive performance changes from 18 to 24 months), β = .31, p < .01 (n = 35). A comparable effect was also found earlier in the trial, with changes on the BCIS from baseline to 6 months predicting neurocognitive changes from 6 to 12 months, β = .43, p = .02 (n = 37); however, this association did not reach statistical significance with correction for multiple comparisons. ⁷ Baseline BCIS scores were not significantly associated with neurocognitive performance at 6 months (accounting for baseline neurocognitive performance), β = −.01 (n = 49), nor were BCIS scores at 12 months significantly associated with neurocognitive performance at 18 months (accounting for BCIS scores at 6 months and neurocognitive performance at 12 months), β = .05 (n = 35).

Also consistent with Study 1, reliable evidence for the reverse order (neurocognitive performance predicting later BCIS scores) was not found, β = −.20 to .30, ps > .05. In addition, correlations between BCIS scores at adjacent assessments (rs = .52 to .78) were comparable to those between neurocognitive performance at adjacent assessments (rs = .65 to .83). Finally, the prospective association between BCIS scores at 18 months and neurocognitive performance at 24 months remained significant when accounting for changes in SAPS and SANS global scores over this time frame, β = .47, p < .01.

Discussion

In summary, the hypothesis that changes in cognitive insight would predict later changes in neurocognitive performance was supported, replicating the central finding from Study 1. It is important that here in Study 2 we were able to find more direct evidence that changes in cognitive insight temporally precede changes in neurocognitive performance (because both were measured at more than two time points). Again, this effect did not appear to be accounted for by changes in other symptoms or differential stability of the measures. Furthermore, there was not clear evidence for the reverse temporal pattern (neurocognitive performance predicting later changes in cognitive insight).