Generalized neurocognitive impairment in individuals at ultra‐high risk for psychosis: The possible key role of slowed processing speed

Abstract Objective Widespread neurocognitive impairment is well‐established in individuals at ultra‐high risk (UHR) for developing psychoses, but it is unknown whether slowed processing speed may underlie impairment in other neurocognitive domains, as found in schizophrenia. The study delineated domain functioning in a UHR sample and examined if neurocognitive slowing might account for deficits across domains. Methods The cross‐sectional study included 50 UHR individuals with no (n = 38) or minimal antipsychotic exposure (n = 12; mean lifetime dose of haloperidol equivalent = 17.56 mg; SD = 13.04) and 50 matched healthy controls. Primary analyses compared group performance across neurocognitive domains before and after covarying for processing speed. To examine the specificity of processing speed effects, post hoc analyses examined the impact of the other neurocognitive domains and intelligence as covariates. Results UHR individuals exhibited significant impairment across all neurocognitive domains (all ps ≤ .010), with medium to large effect sizes (Cohen's ds = −0.53 to −1.12). Only processing speed used as covariate eliminated significant between‐group differences in all other domains, reducing unadjusted Cohen's d values with 68% on average, whereas the other domains used as covariates averagely reduced unadjusted Cohen's d values with 20% to 48%. When covarying each of the other domains after their shared variance with speed of processing was removed, all significant between‐group domain differences remained (all ps ≤ .024). Conclusion Slowed processing speed may underlie generalized neurocognitive impairment in UHR individuals and represent a potential intervention target.


| INTRODUC TI ON
Individuals at ultra-high risk (UHR) for schizophrenia and other psychoses exhibit neurocognitive impairment across more or less all domains, with small to medium effect sizes typically reported in meta-analyses (e.g., Bora et al., 2014;Giuliano et al., 2012). Slowed speed of processing has yielded the highest effect size (Hedges' g = −0.43) across neurocognitive domains in the largest meta-analysis (Hauser et al., 2017) to date, and several UHR studies have suggested prominence of this domain (e.g., Carrión et al., 2011;Keefe et al., 2006;Kelleher et al., 2013;Metzler et al., 2014), just as impairment on processing speed may be predictive of transition to psychosis (Riecher-Rössler et al., 2009) and specifically the development of schizophrenia (Velthorst et al., 2019). However, despite this interest in neurocognitive slowing, its potential key role in UHR individuals' broader neurocognitive impairment remains understudied and unclear.
Processing speed refers to how fast neurocognitive operations are executed and represents a wide-reaching neurocognitive domain that may underlie functioning in many other domains (Dickinson & Gold, 2008b;Salthouse, 1996). As a general processing constraint, it may impose limits on an array of processing operations, for instance by reducing the number of times an item is rehearsed during memory encoding (Hartman et al., 2003). Processing speed may be described as a multidimensional domain that includes several basic, relatively simple neurocognitive components, including perceptual and motor functions, and it invariably emphasizes the velocity of information processing (Nuechterlein et al., 2004). In terms of psychometrics, this domain is generally measured by quantifying the number of correct responses made while performing a task in a finite amount of time. Among the most common types of instruments used to measure processing speed are digit symbol coding tasks (Dickinson et al., 2007), but there is no agreed consensus as to what the specific components of this domain may be (Low et al., 2017).
A specific component often considered is motor speed, as it may be measured with the Token Motor Task of the Brief Assessment of Cognition in Schizophrenia (BACS) that requires the coordinated use of both hands (Keefe et al., 2004). As another example, finger tapping tests have been applied to measure the tapping speed of the index fingers (Reitan & Wolfson, 1993). Another specific component of processing speed often considered is cognitive speed that has been measured using nonmotor inspection time tasks. Such tasks measure the shortest target exposure duration needed to make a reliable perceptual discrimination without having to produce a psychomotor response (Low et al., 2017) and may thus index speed of apprehension (Kranzler & Jensen, 1989). Yet another component of processing speed having been considered is response selection that refers to the process of mapping stimuli to specific motor responses and decision making (Woodward et al., 2013). In addition, factor analytical studies indicate that verbal fluency tasks which focus on rapid spontaneous generation and articulation of words under restricted conditions typically load on processing speed (Nuechterlein et al., 2004). Verbal fluency tasks include, for example, controlled oral word association tests (Keefe et al., 2004). In schizophrenia, a substantial generalized neurocognitive impairment is well-established, with slowed speed of processing likely being a core deficit (Dickinson et al., 2007). Processing speed has yielded the highest meta-analytic effect size (g = −1.25) across all neurocognitive domains in this disorder (Schaefer et al., 2013), and several studies indicate that impairment in an array of neurocognitive domains may reflect reduced speed of processing to a significant degree in individuals with schizophrenia (e.g., Andersen et al., 2013;Rodríguez-Sánchez et al., 2007). To our knowledge, the potential influence of slowed processing speed on deficits in a broad range of other neurocognitive domains has not previously been examined in the UHR population (Frommann et al., 2011;Koutsouleris et al., 2010).
Thus, it remains to be investigated whether two cardinal features of neurocognition in schizophrenia, that is, a generalized deficit profile with slowed neurocognitive speed at its core, also characterize neurocognitive functioning of UHR individuals. This examination may provide insight into underlying mechanisms of broader neurocognitive impairment in UHR individuals and therefore be important for designing appropriate neurocognitive assessment and treatment (Dickinson & Harvey, 2009).

| Aims of the study
This cross-sectional study was designed to delineate neurocognitive domain functioning in UHR individuals compared to demographically matched healthy controls. The current aim was to clarify the potential key role of slowed processing speed for other neurocognitive domains. First, we examined the hypothesis that UHR individuals were characterized by a generalized deficit profile, with small to medium effect sizes across domains. Second, we explored the domains after their shared variance with speed of processing was removed, all significant between-group domain differences remained (all ps ≤ .024).
Conclusion: Slowed processing speed may underlie generalized neurocognitive impairment in UHR individuals and represent a potential intervention target.

K E Y W O R D S
at-risk mental state, clinical high risk, cognition, neuropsychology, schizophrenia hypothesis that decreased processing speed accounted for significant between-group differences in other domains.

| Participants and recruitment
The sample consisted of 50 UHR individuals and 50 healthy controls who all participated in the Prodromal Project, a Danish case-control research project on individuals at UHR of first-episode psychosis.
Inclusion period was September 2009 to August 2014. UHR individuals were referred to the Research Unit, Mental Health Center Copenhagen, from psychiatric in-and outpatient facilities in the Copenhagen catchment area. Healthy controls living in the same geographical area as the UHR individuals were recruited via a website for study participants and received payment for participation.
They were matched one-to-one with UHR individuals on sex, age (within two years), parental socioeconomic status (total household income and highest parental education combined), and race/ethnicity (White/Asian/Mixed White-Asian). Inclusion and exclusion criteria are listed in Table 1. Clinical, functional, and cognitive data on part of the sample have previously been reported (Dannevang et al., 2018;Krakauer et al., 2017Krakauer et al., , 2018Madsen et al., 2018;Nordholm et al., 2016Nordholm et al., , 2018. The Prodromal Project was approved by the Regional Ethics Committee of the Capital Region of Denmark (H-D-2009-013) as well the Danish Data Protection Agency (2014-41-2861). It was carried out in accordance with the Declaration of Helsinki II, and participants signed informed consent.

| Clinical and functional measures
To assess if participants met UHR criteria, the Comprehensive Assessment of At-Risk Mental States (CAARMS) was applied (Yung et al., 2005) Psychiatric disorders were evaluated employing the Structured Clinical Interview for DSM-IV-TR Axis I Disorders (SCID-I) (First et al., 2002) and Structured Clinical Interview for DSM-IV Axis II Personality Disorders (SCID-II) (First et al., 1997). Additional psychopathological and functional measures are listed in Table 3.

| Data analyses
Neurocognitive scores were converted to standard equivalents (z-scores) based on means and standard deviations (SDs) of the healthy controls. Skewed and/or kurtotic distributions were approximated to normality by appropriate transformations (as specified in Table 2). In case of negatively skewed distributions, scores were first reflected, and reverse scoring was used when necessary to ensure that higher z-score indicated better performance. If neurocognitive domains included more than one variable, contributing z-scores were averaged by number of tests included, using equal weights. The domain z-scores and the overall neurocognitive composite z-score were standardized to obtain a mean of 0 and a SD of 1 in the healthy control group. Two UHR individuals did not complete the four WAIS-III subtests, and extrapolation was therefore performed by replacing these missing data with the UHR group's mean estimated full scale intelligence z-score.  (Cohen, 1988). In ANCOVAs, we also report percentage reduction in d values after adjustment for each neurocognitive covariate.
IBM SPSS Version 22.00 (Armonk, NY: IBM Corp) was used for all analyses.
F I G U R E 1 Neurocognitive functioning of the ultra-high risk group (n = 50) presented as z-score deficits relative to the healthy control group (n = 50) (with a mean of 0

| Between-group comparisons of demographic and clinical characteristics
Information on demographic and clinical variables in the UHR and healthy control group is summarized in Table 3. Groups were highly similar on basic demographic parameters, except controls having significantly more years of education. The UHR group had markedly elevated scores on all psychopathological measures.

| Unadjusted between-group differences in neurocognitive functioning
Mean or median neurocognitive performance for each measure across groups is summarized in Table S1. Results of primary unad-

| Confounder analyses
No significant associations between psychopathological measures and neurocognitive domain or overall neurocognitive composite functioning were observed within the UHR group, as shown in Table S4. There were also no significant differences in neurocognitive functioning between subsamples of UHR individuals that (had) received medication vs. those that did not.

| D ISCUSS I ON
We examined neurocognitive domain functioning in a sample of UHR individuals compared with demographically well-matched healthy controls, focusing on two cardinal features characterizing neurocognition in schizophrenia, that is, a generalized deficit profile with slowed neurocognitive speed at its core. Our study essentially confirmed our two hypotheses; the UHR group was globally impaired across all neurocognitive domains, and reduced speed of processing appeared to account for all significant domain group differences.
The finding that the UHR group was globally neurocognitively impaired is in agreement with meta-analyses (e.g., Giuliano Our UHR sample, however, appears to be more neurocognitively impaired than many other UHR samples. Generalized impairment does not characterize all UHR samples (Pukrop & Klosterkötter, 2010); some studies have reported no significant deficits (e.g., Thompson et al., 2012) or only deficits in one or some measured domains (e.g., Niendam et al., 2006;Woodberry et al., 2010). Impairment in estimated current intelligence (d = −0.71) was also more pronounced in our UHR sample than in meta-analyses, with effect sizes ranging from g = −0.21 (Hauser et al., 2017) to d = −0.53 (Giuliano et al., 2012). Furthermore, our sample demonstrated neurocognitive deficits of larger magnitudes than hypothesized, with a substantial composite effect size (d = −1.07). Even though many UHR studies have detected comparable or even larger effect sizes across domains (e.g., Frommann et al., 2011;Ohmuro et al., 2015;Simon et al., 2007) and a UHR meta-analysis (Zheng et al., 2018) based only on studies using the MATRICS Consensus Cognitive Battery has reported effect sizes similar to the ones found in our study, meta-analyses have typically reported effects sizes in the small to medium range (e.g., Bora et al., 2014;Giuliano et al., 2012;Hauser et al., 2017). We suspect that our findings may reflect that only UHR individuals with sustained low or significant drop in functioning were included in the study, as per inclusion criterion.
In most other UHR studies, the functional impairment criterion is only part of the Vulnerability (Trait and State Risk) Group, but not of the Attenuated Psychotic Symptoms (APS) Group nor of the Brief Limited Intermittent Psychotic Symptoms (BLIPS) Group. A link between neurocognitive deficits and functional impairment has been documented in a UHR meta-analysis (Bora et al., 2014) as well as systematic review (Cotter et al., 2014) and is well-established in schizophrenia (Fett et al., 2011;Green et al., 2000). The difference in mean SOFAS scores between the UHR and healthy control group corresponded to g = −6.78, which is substantially larger than the meta-analytic mean effect size of g = −3.01 characterizing UHR individuals' low functioning (Fusar-Poli et al., 2015). Functional impairment in our UHR sample is more comparable to that of first-episode psychosis individuals in most studies comparing these individuals to UHR individuals (e.g., Eastvold et al., 2007).
Both primary multivariate and univariate covariate models sug- It is well-documented that performances on a broad variety of neurocognitive tasks and composite domains share considerable common variance, both in healthy populations (Carroll, 1993) and schizophrenia (Dickinson & Gold, 2008a). Multiple studies have considered the influence of intelligence on neurocognitive impairment in UHR individuals and found that it may account for or eliminate some, but not all, significantly impaired performances across specific neurocognitive tasks or domains (e.g., Seidman et al., 2016), in accordance with our results. However, to our knowledge, only few UHR studies (Frommann et al., 2011;Koutsouleris et al., 2010;Woodberry et al., 2010) have considered the potential influence of specific neurocognitive domain or task performances on between-group neurocognitive functioning, and it remained to be determined whether a distinct processing speed domain may significantly contribute to deficits across a broad spectrum of neurocognitive domains. Our findings are in general agreement with a growing body of schizophrenia studies showing that decreased processing speed, to a significant degree, may underlie impairment in an array of neurocognitive domains (e.g., Andersen et al., 2013;Brébion et al., 2014;Fuller et al., 2005;Hartman et al., 2003;Holthausen et al., 2003;Kochunov et al., 2017;Ojeda et al., 2012;Rodríguez-Sánchez et al., 2007;Salamé et al., 1998;Sanfilipo et al., 2002;Schatz, 1998). Our findings likewise coincide with similar research on aging and other neurocognitively impaired populations (e.g., Butters & et al., 2004;Lee et al., 2012;Liebel et al., 2017;McGrath et al., 2011;Salthouse, 1996;Su et al., 2015).
The influence of speed of processing on other domains may be obvious in the case of attention/vigilance that requires speeded response to a considerable degree. Still, most neurocognitive variables included in the other domains are untimed, indicating that the processing speed influence on these domains is not just secondary to a common procedural factor. Processing speed, as a nontask-specific mental capacity, may impose limits on a broad diversity of neurocognitive processing operations and therefore constitute a rate-limiting factor for performances (Kail & Salthouse, 1994). Slower speed of executing a variety of neurocognitive processing operations results in less completion of processing in a given amount of time and reduces the amount of simultaneously available information when needed (Salthouse, 1996). Therefore, although this remains speculations, UHR individuals may not be able to complete and coordinate all the information processing needed for adequate performances within a given amount of time.
Supplementary analyses showed that a revised processing speed composite only including the motor-based tasks, i.e. excluding verbal fluency, when used as the covariate, eliminated significant be- to perform such tasks adequately may therefore reflect insufficient coordination or inability to efficiently and rapidly connect spatially distributed and interconnected brain regions, that is, deficits in connectivity (Dickinson et al., 2007). Our study may suggest a systemic perspective on neurocognitive impairment in the UHR state (Dickinson & Harvey, 2009;Kelleher et al., 2013), including reduced white matter integrity (Krakauer et al., 2017;Kristensen et al., 2019).
Our findings have other important research and clinical implications. Processing speed appears to be particularly important to measure in UHR individuals, also for screening purposes (González-Blanch et al., 2011), because it may capture the generalized impairment. From a treatment perspective, our study may provide a rationale for targeting this apparently domain-general neurocognitive mechanism, using behavioral and/or pharmacological interventions to boost neurocognitive processing efficiency (Brébion et al., 2014;Cassetta & Goghari, 2016;Cassetta et al., 2019;Takeuchi & Kawashima, 2012). Thus, a double-blind randomized clinical trial has shown that UHR individuals receiving processing speed training exhibit improvement not only in processing speed but also in social functioning (Choi et al., 2017).

| LI M ITATI O N S
Our findings should be interpreted cautiously in the context of several well-acknowledged limitations. First, construct validity of the speed of processing domain may be questioned (Carter & Barch, 2007). Given its multi-componential nature, it is likely simultaneously sensitive to and taps multiple neurocognitive functions (Dickinson et al., 2007).
It may therefore be difficult to classify and isolate a distinct, unitary speed domain, and future studies need to examine the neurocognitive underpinnings of speed of processing in UHR individuals in more detail (Chiaravalloti et al., 2003;Knowles et al., 2012). Nevertheless, the domain-general and broad-ranging nature of processing speed is likely quintessential to understanding its key role in the neurocognitive architecture of UHR individuals. Our categorization of this domain also followed the MATRICS recommendations (Nuechterlein et al., 2004), and five outcome variables were included in the speed composite to enhance both validity and reliability. Concerning the multi-componential nature of processing speed, it should also be noted that neurocognitive tests of processing speed typically lack the precision of determining the more specific neurocognitive component operations involved (Dickinson et al., 2007). Furthermore, there is no agreed upon consensus as to what components may constitute processing speed (Low et al., 2017), and an analysis of the potential components of this domain was not performed in the present study. Thus, for instance, response selection was not specifically identified as part of the processing speed, even though this may be an important component to take into consideration (Woodward et al., 2013(Woodward et al., , 2014. Future UHR studies should include measures that make it possible to fractionate processing speed into its distinct components, just as using cognitive neuroscience-based approaches would allow for examining relevant components, including the response selection stage of information processing (Woodward et al., 2013(Woodward et al., , 2014. Second, the two neurocognitive domains demonstrating the numerically smallest effect sizes, verbal as well as visual learning and memory, each consisted of only one outcome variable, and this may have caused these domains to be less sensitive to group differences. It has been suggested that at least three outcome variables should be included in a domain composite to ensure adequate psychometric quality (Kenny et al., 1998), and UHR

meta-analytic results indicate that the verbal learning and memory
domain is impaired at the same level as processing speed (Hauser et al., 2017). To add to this limitation, neurocognitive tests were generally not psychometrically matched in our study, and it is uncertain if discriminating power was comparable across tests (Chapman & Chapman, 1973). Outcome variables associated with the processing speed domain may have been somewhat better at capturing an underlying generalized performance deficit than were outcome variables associated with other domains (Dickinson & Harvey, 2009). Still, we used well-validated tests and outcome variables classified according to the MATRICS recommendations. Effect sizes across neurocognitive domains were also comparable in the UHR group given that an essentially flat deficit profile was detected, at least indicating that out study did not artifactually produce differential domain deficits. Third, slowed neurocognitive speed may be a robust correlate of psychopathology in general (Nigg et al., 2017) and thus a nonspecific marker of mental illness (Pukrop et al., 2007); a digit symbol substitution test has failed to detect significant differences between UHR individuals and psychiatric controls (Ilonen et al., 2010;Lindgren et al., 2010). Thus, it is a limitation that the comorbidity in the UHR group was not addressed in the present study, which makes it difficult to address the specificity of the observed impairment in processing speed and of its effects on the between-group differences in other neurocognitive domains. Fourth, we only used cross-sectional data, and even if tests covering speed of processing may be relevant for psychosis prediction (Hauser et al., 2017;Studerus et al., 2016;Velthorst et al., 2019), the potential key role of slowed neurocognitive speed in UHR individuals with later transition to psychosis needs to be examined in depth in future longitudinal studies. Fifth, we used covariate analyses that have been extensively used in studies similar to ours, including UHR studies controlling neurocognitive functioning for specific domains (e.g., Frommann et al., 2011) and/or intelligence (e.g., Seidman et al., 2016). Such analyses may be reasonable for descriptive model building (Tabachnick & Fidell, 2007), and only in this way was it possible to examine the extent to which impairment in one domain might reflect impairment in another domain (Esbjørn et al., 2006). Still, hypothetical group matching does not allow for inferring causality and remains debatable (e.g., Dennis et al., 2011 (Broyd et al., 2016;Potvin et al., 2018;Scott et al., 2018;Stavro et al., 2013). Unfortunately, such potential effects have often not been considered in UHR studies. It may be informative for subsequent UHR studies to examine and account for the potential effects of alcohol and drug use behaviors on neurocognition, including processing speed.

| CON CLUS ION
Our study found evidence that decreased speed of processing may account for the global impairment across other neurocognitive domains in UHR individuals. Future studies are needed to examine if findings can be replicated in UHR samples with varying characteristics. We hope the study will stimulate further UHR research designed to understand the possible contribution of general neurocognitive slowing to broadly impaired neurocognitive functioning.
Future studies should also further assess the associations between processing speed and social functioning (Carrión et al., 2011) as well as social cognitive domain functioning  in the UHR population.

ACK N OWLED G M ENTS
We are grateful to the patients and healthy controls who participated

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/brb3.1962.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.