In this study, we evaluated the interaction among 3 neurocognitive domains (i.e., working memory, verbal memory, and processing speed) and 3 discrete domains of functioning (i.e., QLS Instrumental, QLS Intrapsychic, and QLS Interpersonal) at baseline and following 24 weeks of antipsychotic drug therapy in patients with schizophrenia. In our path-analytic models, we also incorporated positive and negative symptoms to assess whether cognitive variables may directly or indirectly affect functioning; in the latter case, via changes in symptoms. We found that only processing speed was significantly associated with functioning at both baseline and at 24 weeks. Processing speed affected functioning both directly and indirectly via negative symptoms, although the partitioning of the total effect into direct and indirect parts varied somewhat for each domain of functioning assessed, including QLS Instrumental (work), QLS Intrapsychic, and QLS Interpersonal (psychosocial) domains. In addition, we found that positive symptoms also affected functioning at baseline.
As reviewed by Bowie et al. , processing speed impairments may reflect a core cognitive deficit in schizophrenia. In a recent meta-analytic study, Dickinson and colleagues  reported that the digit-symbol coding task, a measure of processing speed, represented the greatest deficit among cognitive abilities in patients with schizophrenia. Processing speed was also identified as the most sensitive index in patients with schizophrenia for the WAIS-III technical manual . Processing speed, as measured by the digit-symbol coding task, accounted for 65% of the variance in overall cognitive performance and was the best single predictor of total score in the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) analyses . Bowie and colleagues  found that the processing speed factor, which included the digit-symbol coding task, consistently predicted social competence and living skills and was the only factor to have a direct effect on all 3 real-world behaviors. It has been suggested that slowed information processing (including slowed encoding) can account for deficits in a range of higher level cognitive functions in schizophrenia including deficits in working memory, executive function and memory [23, 24].
In the current study, we focused on the digit-symbol coding task and verbal fluency as two measures of processing speed. We were unable to utilize data from the Grooved Pegboard Tests because there were a large number of unusually high values. However, a recent review on psychomotor slowing in schizophrenia has suggested that psychomotor slowing may be distinguishable from a reduction in the speed of information processing (reviewed in ). While a number of neurocognitive processes that support motor control are likely involved in psychomotor functioning, in fact some tasks may be more sensitive to psychomotor speed while others are more sensitive to the speed of information processing. In this regard, the digit-symbol coding task would be more sensitive to the speed of information processing while the pegboard task would have been more sensitive to psychomotor speed.
Based on our proposed model, processing speed worked, in part, through negative symptoms to impact functioning. As proposed by Green and Nuechterlein , it is possible that the relationship between negative symptoms and functioning may reflect the shared variance between negative symptoms and neurocognition, or the stronger association between neurocognition and functional outcomes. There is some data to suggest that while cognitive deficits appear to develop at an earlier age than negative symptoms, some neurocognitive deficits may act as vulnerability factors for development of negative symptoms (reviewed in ). However, negative symptoms typically have shown relatively weak correlations to cognitive deficits , a finding that suggests that while there is some shared variance in predicting outcomes, negative symptoms and neurocognitive deficits are relatively distinct constructs. Thus, negative symptoms may also have a direct impact on a patient's overall level of functioning.
It has also been proposed that the causal pathways from negative symptoms to functional outcomes may be due to a common neurocognitive intersection . Does processing speed possibly serve as a neurocognitive intersection between negative symptoms and functioning? Hughes and colleagues  reported that significant improvements in symptom ratings over time in patients with chronic schizophrenia did not predict improvements in any aspect of cognitive functioning measured, except motor speed. Rodriguez-Sanchez and colleagues  analyzed the relationship between different cognitive tasks and clinical symptoms in patients with first-episode schizophrenia and found that negative symptoms were significantly associated with performance on executive functions and motor coordination tasks, with a significant association of negative symptoms seen only for those executive functions requiring speeded performance. They concluded that the widely described relationship between negative symptoms and executive impairments in schizophrenia appears to be mediated by dysfunction in processing speed.
Some have argued that the Heinrichs-Carpenter QLS subtitled as "an instrument for rating the schizophrenia deficit syndrome" provides a clinical assessment of negative symptoms more than a subjective measure of a patient's quality of life . Although not without controversy , we cannot exclude the possibility that the strong support for negative symptoms as a mediator between cognition and functioning may be somewhat unique to the use of the QLS.
In contrast to the Bowie et al. study , our path-analytic models did not suggest the presence of a significant relationship between working memory or verbal memory and functioning. However, previous work has incorporated additional constructs, such as social competence and functional competence, which may mediate the effect of other cognitive parameters, such as memory, on functioning [15, 31].
The observation that positive symptoms affected functioning at baseline, but not after 24 weeks of treatment, was somewhat surprising. However, positive symptoms can improve quite rapidly with antipsychotic medication, and it is possible that the timing of improvement in positive symptoms may have occurred more quickly than improvement in functioning. Negative symptoms at baseline (and with change) appeared to play a more dominant role in functional outcomes.
Our inability to detect a stronger relationship between cognition and functioning may have been hindered by the lack of substantial improvement in both functioning and cognition observed during the 24-week time period that was analyzed. Long-term data on cognition and functioning are needed to obtain better estimation of potential direct and indirect effects of cognition on functioning of patients with schizophrenia.
A large number of dropouts (especially when caused by lack of symptom improvement) may have introduced bias in evaluating associations among changes in symptoms, cognition, and functioning at the 24-week endpoint. In the original report, 95 (59.7%) olanzapine-, 104 (65.8%) risperidone-, and 70 (72.2%) haloperidol-treated patients had discontinued for any reason; out of these, 20 (12.6%) olanzapine-, 18 (11.4%) risperidone-, and 16 (16.5%) haloperidol-treated patients had discontinued the study for lack of efficacy .
To evaluate whether dropouts could have biased the path analytic models estimated from observed changes at 24-week endpoint we performed a sensitivity analysis by imputing missing outcomes for subjects who discontinued prior to week 24 using a single imputation from a posterior predictive distribution of missing values given all the subjects' outcome data observed up to their dropout (with Bayesian regression for monotone missing data in SAS PROC MI). The results (not reported here) were qualitatively similar to those based on observed data at week 24.
While the relationship among variables could vary by treatment, analyzing the data separately by treatment groups would result in small subgroups with limited power to detect significant differences. In the original study , at the 52-week endpoint, neurocognition had significantly improved in each group, with no significant differences observed between groups. Although differences were observed on what specific domains improved in the three treatment arms, the lack of differences observed across the treatment groups overall and discontinuation of the haloperidol arm per protocol limit further analysis of this important issue in the current study.
Lastly, some studies have suggested that the effects of neurocognition are mediated primarily through social cognition and subsequently through social competence and social support [15, 31]. We did not have data in which to assess the role of social cognition or other social variables in functioning.