Cognitive deficits in youth with familial and clinical high risk to psychosis: a systematic review and meta-analysis


  • E. Bora,

    Corresponding author
    1. Department of Psychiatry, Melbourne Neuropsychiatry Centre, The University of Melbourne and Melbourne Health, Melbourne, Vic., Australia
    • Emre Bora, Department of Psychiatry, Melbourne Neuropsychiatry Centre, The University of Melbourne and Melbourne Health, Alan Gilbert Building NNF level 3, Carlton, Vic. 3053, Australia.


    Search for more papers by this author
  • A. Lin,

    1. School of Psychology, University of Birmingham, Birmingham, UK
    Search for more papers by this author
  • S. J. Wood,

    1. Department of Psychiatry, Melbourne Neuropsychiatry Centre, The University of Melbourne and Melbourne Health, Melbourne, Vic., Australia
    2. School of Psychology, University of Birmingham, Birmingham, UK
    Search for more papers by this author
  • A. R. Yung,

    1. Institute of Behaviour and Mental Health, University of Manchester, Manchester, UK
    2. Orygen Youth Health Research Centre and Centre for Youth Mental Health, University of Melbourne, Melbourne, Vic., Australia
    Search for more papers by this author
  • P. D. McGorry,

    1. Orygen Youth Health Research Centre and Centre for Youth Mental Health, University of Melbourne, Melbourne, Vic., Australia
    Search for more papers by this author
  • C. Pantelis

    1. Department of Psychiatry, Melbourne Neuropsychiatry Centre, The University of Melbourne and Melbourne Health, Melbourne, Vic., Australia
    Search for more papers by this author



It is likely that cognitive deficits are vulnerability markers for developing schizophrenia, as these deficits are already well-established findings in first-episode psychosis. Studies at-risk adolescents and young adults are likely to provide information about cognitive deficits that predate the onset of the illness.


We conducted meta-analyses of studies comparing familial-high risk (FHR) or ultra-high risk (UHR; n = 2113) and healthy controls (= 1748) in youth studies in which the mean age was between 15 and 29.


Compared with controls, high risk subjects were impaired in each domain in both UHR (d = 0.34–0.71) and FHR (d = 0.24–0.81). Heterogeneity of effect sizes across studies was modest, increasing confidence to the findings of the current meta-analysis (I2 = 0–0.18%). In both risk paradigms, co-occurrence of genetic risk with attenuated symptoms was associated with more severe cognitive dysfunction. In UHR, later transition to psychosis was associated with more severe cognitive deficits in all domains (d = 0.31–0.49) except sustained attention. However, cognitive impairment has a limited capacity to predict the outcome of high-risk patients.


Cognitive deficits are already evident in adolescents and young adults who have familial or clinical risk for psychosis. Longitudinal developmental studies are important to reveal timing and trajectory of emergence of such deficits.


  • Both familial and clinical high-risk for psychosis is associated with cognitive deficits.
  • Co-occurrence of both risk factors is associated with more severe cognitive dysfunction.
  • Youth who develops psychotic disorders at follow-up have more severe cognitive dysfunction at baseline assessment.


  • More studies are needed to directly compare cognitive functioning in genetic and clinical high-risk risk subjects.
  • Effect of the diagnosis on baseline cognitive deficits at individuals who developed psychotic disorders at follow-up needs further clarification.
  • There is a significant overlap of baseline cognitive performances of clinical high risk individuals who did or did not develop psychosis, despite the fact that first group is significantly more impaired. There is no sufficient data regarding the effect of baseline cognition on transition to psychosis in familial high risk individuals.


Cognitive dysfunction is one of the core features of schizophrenia and is an important predictor of occupational and functional impairments observed in people with this disorder [1-5]. Reduced cognitive performance is already evident at the first-episode of psychosis and is a stable characteristic of the illness [6-8]. This implies that cognitive dysfunction may be a likely neurobiological marker of psychosis before the onset of the illness.

The most important paradigms to investigate cognitive deficits before the onset of the illness are the genetic and clinical high-risk approaches. The genetic high risk paradigm has traditionally been used to study neurobiological alterations prior to onset of schizophrenia [9, 10]. There are a substantial number of studies that examined cognitive functions in the offspring and siblings of patients with schizophrenia in childhood, adolescence and adulthood [11-14]. Over the last 15 years, a clinical-high risk paradigm targeting identification in the prodrome of psychosis, known as ultra-high-risk (UHR), clinical high risk, at-risk mental state or psychosis risk syndrome, has successfully been used to study individuals who are at substantial risk for developing a psychotic disorder in the near future [15-17].

There is evidence of cognitive impairment of both individuals with UHR and young people with family history of schizophrenia (FHR); however, there are inconsistent findings regarding specific abnormalities associated with clinical and familial high risk to psychosis. Two recent meta-analyses examining cognitive deficits in UHR subjects in comparison to healthy controls found evidence of cognitive impairment in adolescents and youth with UHR [18, 19], but no meta-analysis has examined cognitive deficits in FHR at this age period, or compared this group to UHR populations.

One important reason for inconsistent results is the small sample size of most studies. Another is the fact that both of these risk paradigms will recruit a substantial proportion of ‘false positives’ that have no real susceptibility to develop psychotic disorders. These young people may never be destined to develop frank psychotic disorder. Combining familial and clinical risk could be important, but to date it is not clear whether such an approach is associated with more severe cognitive deficits. It is also unclear whether cognitive deficits can contribute to prediction of individuals who will develop frank psychosis among high-risk subjects. Meta-analytic methods can be beneficial to reveal the most consistent cognitive deficits in FHR and UHR subjects and potential predictors of transition to psychosis.

Aims of the study

Our goals were to conduct a meta-analysis of cognitive deficits in familial-high risk and ultra-high risk to i) examine consistent cognitive deficits in youth with familial-high risk and ultra-high risk; ii) examine effect of co-occurrence of genetic and clinical risk on cognitive deficits; iii) examine whether baseline cognitive deficits in both high risk groups are associated with transition to psychosis.

Material and methods

Study selection

We followed PRISMA 2009 guidelines in conducting this meta-analysis [20]. A literature search was conducted using the databases Pubmed, PsycINFO and Scopus to identify the relevant studies (January 1990 to April 2013). We used the combination of the following keywords: high risk, familial high risk, relatives, clinical high risk, ultra-high risk, prodrome, at-risk mental state, psychosis, schizophrenia, cog*, neurops*, memory, attention and executive function. We also reviewed the reference lists of published studies. Inclusion criteria were studies that i) reported neurocognitive data or statistics sufficient to calculate effect sizes (where necessary, authors were contacted to provide such data); ii) written in English; iii) compared performance of a help-seeking clinical-high risk group (UHR) and/or unaffected relatives of people with schizophrenia (FHR) with healthy controls; or compared transitioned and non-transitioned at-risk subjects, and each group included at least five subjects; iv) had a mean high-risk sample age between 15 and 29 to examine FHR and UHR subjects in the same developmental stage. All available UHR studies met this fourth criterion, and FHR studies included were selected from larger number of studies with a wider age range. FHR was defined as having a parent or sibling with schizophrenia, or at least two relatives with schizophrenia. UHR was defined as having one or more of three psychosis risk syndromes in help-seeking youth or young adults. Although there are various tools used to assess these, (e.g. CAARMS, SIPS), essentially these three risk syndromes included i) recent onset or worsening of attenuated positive symptoms (APS); ii) recent onset of frank level psychotic symptoms which were significant but not sufficiently sustained to meet the criteria for DSM-IV psychotic disorder (brief limited intermittent psychotic syndrome: BLIPS); iii) genetic/familial risk to psychosis plus deterioration in functioning (recent onset or worsening functional decline, or chronic poor functioning) syndrome (GRDS). Some studies used an alternative basic symptom approach by combining three syndrome criteria above (late prodromal state) with other clinical features that was hypothesized to predate them (early prodromal state). In these studies, only late prodromal state data was included if it was reported separately. Studies comparing high risk samples with help-seeking controls, recruited non-help-seeking clinical high risk groups based on screening, and those which defined risk syndrome based on psychometric risk or schizotypal personality were excluded. For each cognitive measure, only the study with largest sample size was included from each group. Studies that were included in reports of multisite studies were not included in the meta-analysis unless they reported a different cognitive measure. A flow chart of study selection is shown in Fig. 1. Forty-four Studies [21-64] including two multisite studies included in the study (Tables 1 and 2).

Table 1. Study characteristics of FHR studies included
StudiesSampleAgeCharacteristicsCognitive tasks
  1. HC, Healthy controls; FHR, Familial high risk to psychosis; WMS, Wechsler memory scale; CPT, continuous Performance test; CVLT, California verbal learning test; LNS, Letter number sequencing; TMT, trail making test; RCFT, Rey complex figure test; WCST, Wisconsin cart sorting test; HSCT, Hayling sentence completion test; RAVLT, Rey Auditory verbal learning test; IQ, intelligence coefficient; EHRS, Edinburgh High Risk Study; NAPLS, North American Prodrome Longitudinal Study.

Barrantes-Vidal et al. 2007 [20]

38 FHR

63 HC

FHR = 27.3


No personality disorder

clusters A or B

WAIS-III, CPT, WMS-R, WCST, letter and category fluency
Bryne et al. [27]

157 FHR

34 HC

FHR = 21.2With at least 2 family member schWAIS-R, NART, Stroop, letters/Category fluency, HSCT, digit symbol, Digit span, RAVLT, CPT, WMS-R visual
Seidman et al. [54]

73 FHR

84 HC

FHR = 18.1 WRAT reading, vocabulary, BD, TMT, CPT, Digit symbol, Digit span, Story recall, letter fluency
Hans et al. [35]

39 FR

36 C

FHR = 16.9




TMT, CPT, Stroop, WCST, digit


Klemm et al. [40]

32 FHR

32 HC

FHR = 16.0



WCST, TMT, Stroop, d2

Concentration test, SPM

Hughes et al. [36]

25 FHR

25 HC

FHR = 28


SPD excluded

IQ, CPT, LNS, WMS-R, Visual memory, RAVLT, WCST, TMT, Stroop, letter, category fluency
Seidman et al. [55]

49 FHR

109 HC

FHR = 18.7

Offsprings, siblings

No prodrome

Groom et al. [34]

36 FHR

72 HC

FHR = 17.5Siblings No prodromal symWASI, RAVLT,CPT, HSCT, Letter fluency
Wolf et al. [61]

73 FHR

120 HC

FHR = 25.6


Myles Worsley et al. [45]

99 FHR

99 HC

FHR = 16.9

Offspring or 2 siblings +

55/99 mild symptoms

WMS, digit symbol, digit span, spatial

Span, LNS, CPT

Nam et al. [46]

44 FHR

100 HC

FHR = 28.8SiblingsDSST, letter/category fluency, Digit span, N-back, TMT, CPT, RAVLT, RCFT
Prasad et al. [48]

74 FHR

86 HC





Maziade et al. [44]

22 FHR

45 HC

FHR = 17.3OffspringsIQ, CPT, CVLT
Seoul Youth Clinic
Choi et al. [29]

23 FHR

34 HC

FHR = 23.42 relativesIQ
St Louis
Delawalla et al. [30]

31 FHR

42 HC

FHR = 21.3SiblingsVocabulary, CPT
Schubert and Mc Neil [52, 53]

38 FHR

88 HC

FHR = 22.4


8/74 has cluster A personality

WCST, Word-pair, verbal fluency, TMT, Digit span, reaction time, BD, Selective attention
Washington University
Bonner-Jackson et al. [24]

24 FHR

40 HC

FHR = 21.1SiblingsLogical memory, family pictures, CVLT
New York
Bertisch et al. [22]

16 FHR

31 HC

 One first-degree or Multiple second degree 6/16 prodromal symptomsVocabulary, CVLT, digit span, LNS, Stroop, Visual memory, word pairs, Matrix reasoning
Table 2. Study characteristics of UHR studies included
StudiesSampleAgeCharacteristicsCognitive tasks
  1. HC, Healthy controls; UHR, Ultra-high risk to psychosis; COPE, Center of prevention and evaluation; NAPLS, North American Prodrome Longitudinal Study; RAP, Recognition and prevention programme; PRIME, Prevention through risk management and education; OASIS, Outreach and Support in South London service; WMS, Wechsler memory scale; CPT, continous Performance test; CVLT, California verbal learning test; LNS, Letter number sequencing; TMT, trail making test; RCFT, Rey complex figure test; WCST, Wisconsin cart sorting test; HSCT, Hayling sentence completion test; RAVLT, Rey Auditory verbal learning test; HVLT, Hopkins verbal learning test; CANTAB, Cambridge Neuropsychological Test Automated Battery; WM, Working memory; SWM, Spatial working memory; SOPT, Self-ordered pointing test; ToH, Tower of London test; D-KEFS, Delis-Kaplan Executive Function System; DRT, Delayed response task; IQ, intelligence coefficient.



Lin et al. 2013 [42]

325 UHR

66 HC

UHR = 19.1

244 UHR-NP

81 UHR-P

IQ, premorbid IQ, WMS-R, RAVLT

Trails A &B, Stroop

Letter fluency

Additional data below

Francey et al. [31]

70 UHR

51 HC

UHR = 20.1

20 UHR-P


Wood et al. 2003 [62]

16 UHR

17 HC

UHR = 18.8

1–2 year follow-up






Stanford et al. 2011 [58]

63 CHR

24 HC

UHR = 19.6Cross-sectionalWMS, WAIS


Multisite Emory, Harvard, UCSD UCL, Toronto, Yale, North Carolina, Zucker-Hillside

304 UHR

193 HC

UHR = 18.2 WCST, CPT, Premorbid IQ, Symbol coding, letter fluency, Verbal memory, symbol coding, IQ
Seidman et al. [55]

167 UHR

109 HC

UHR = 18.2

2.5 year follow-up

54 UHR-P

113 UHR-NP

WCST, CPT, Premorbid IQ, Symbol coding, letter fluency, Verbal memory, BD, IQ




Data included in NAPLS

Not overlapping data (below)

Smith et al. 2006 [57]


10 HC

UHR = 16.3BaselineSWM
Carrion et al. 2011 [28]

127 UHR

80 HC

UHR = 16.0BaselineVerbal WM (Digit span, LNS)


UNC, Toronto

NAPLS LongitudinalData included in NAPLS
UCL and Emory NAPLS  Data included in NAPLS

Data included in NAPLS

Not overlapping data (below)

Jahshan et al. 2010 [37]

48 UHR

29 HC

UHR = 18.7LongitudinalSpatial Span, Stroop, HVLT, LNS


Bonn, Cologne Dusseldorf, Munich

Frommann et al. [32]

89 LPS

87 HC

LPS = 25.3Cross-sectionalAVLT, SOPT, TMT A &B, CPT, symbol coding, LNS, Letter fluency




Data included in GRNS

Not overlapping data (below)

Pukrop et al. 2006 [49]

90 UHR


Pukrop et al. 2007 [50]

83 CHR

44 HC

UHR = 24

44 UHR-P




Letter/Number, WCST, SC

Trail A B, Fluency, DRT

Munich GRNS  

Data included in GRNS

Not overlapping data (below)

Koutselaris et al. [41]

48 UHR

30 HC

UHR = 24.7

20 EPS/28 LPS

15 UHR-P/



Digit span, SOPT, AVLT,

Letter/Number, symbol coding

Trail A B, Fluency, DRT

UHR vs. HC

Digit span

Pflueger et al. 2007 [47]

54 CHR

51 HC

UHR = 2754 HR, 6 low riskIQ, ToH, WCST, Go/No-Go, WM-visual 2 back, CPT
Becker et al. [21]

47 UHR

42 UHR

UHR = 20.7

18 UHR-P, 2 year

69 FE

Letter and category fluency
Becker et al. [22]

41 UHR

17 HC

UHR = 19.2

17 UHR-P


CVLT, Letter and category fluency, CPT, SWMT, ROCF,Premorbid IQ



Woodberry et al. 2010 [63]

73 HR

34 HC

UHR = 16.5

13 UHR-P


WCST, LM, CVLT, WAIS, Fluency, Trail making test, WRAT, CPT-IP

Seoul 1

Seoul Youth Clinic

Kim et al. [38]

49 UHR

45 HC

UHR = 21.1

2.8 year mean follow-up

13 UHR-P

Trail A, Stroop, CVLT, ROCF

WCST, letter fluency, DS, Spatial location,

Seoul 2

Yonsei University

Kim et al. [39]

33 UHR

27 HC

 17% transit over 5–17 monthsVerbal and visual memory
Lingdren et al. [43]

62 UHR

72 HC

UHR = 16.6Cross-sectionalLetter, semantic fluency, TMT-A/B, list learning, Vis and digit span, reaction time, Dot cancellation



Broome 2007 [25]

35 UHR

23 HR

UHR = 24.9Cross-sectionalPremorbid IQ, Beads task, IQ?
Broome 2012 [26]28 UHRUHR = 24.4

2 years follow-up



WM, IQ, premorbid IQ
Buones Aires
Serrani 2011 [56]

27 UHR

38 HC

UHR = 17.4Cross-sectionalIQ, Premorbid IQ, category fluency, CPT, symbol coding, HVLT, LNS, TMT A, spatial span. Mazes, Visual memory
Van Rijn et al. [60]

36 UHR

21 HC

UHR = 15.2Cross-sectionalIQ, shifting set task
Szily et al. 2009 [59]

26 UHR

50 HC

UHR = 22.0Cross-sectionalIQ
Figure 1.

Flow diagram for meta-analysis of cognitive deficits in UHR and FHR. From: Moher D, Liberati A, Tetzlaff J, Altman DG, The PRISMA Group (2009). Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLoS Med 6(6): e1000097. doi:10.1371/journal.pmed1000097.

FHR or UHR vs. Controls

The FHR and healthy control comparison included 18 main studies (20 samples) including one multi-site study. About 929 FHR (178–757 FHR for each variable) subjects (48.3% males) and 1140 (126–881 for each variable) controls (46.6%) males were included. There was no significant between-group difference for age (d = 0.01, CI = −0.11 to 0.14, Z = 0.23, = 0.82).

Sixteen main UHR studies including two multisite studies consisting of 1184 UHR (351–907 UHR for each variable) subjects (53.8% males) and 717 (311–557 for each variable) controls (50.1% males) were included. There was no significant difference for age (d = 0.05, CI = −0.12 to 0.21, Z = 0.54, = 0.59).

Co-occurrence of genetic and clinical risk

Meta-regression analyses were used to examine effect of the percentage of cases with genetic risk (11 studies) in UHR samples. 20.3% of UHR subjects met the genetic risk criteria. A vast majority of UHR subjects with genetic risk met criteria for GRDS+APS rather than pure GRDS. In FHR, a preliminary meta-analysis conducted for three studies which provided separate data for genetic at-risk subjects with (= 116) and without (= 159) symptoms. This analysis in FHR was only conducted for verbal memory and processing speed as they were the only cognitive domains reported in all three studies.

Baseline cognition and prediction of psychosis

Comparison of UHR subjects who developed psychotic disorder (UHR-P) and did not (UHR-NP) were based on data from nine samples (263 UHR-P and 586 UHR-NP, 11 reports) including up to 229 UHR-P and 518 UHR-NP for each of the variable. There was no significant difference for gender (60.9% vs. 52.1% male, RR = 1.10, CI = 0.98–1.24, = 0.12) and age (= 0.06, CI = −0.11 to 0.22, = 0.50) between UHR-P and UHR-NP. Only one study [the Edinburgh high risk study (EHRS)] compared FHR subjects who developed illness with others, and therefore meta-analysis was not possible for this group.

Cognitive measures

We combined individual tasks into broader cognitive domains of current and premorbid IQ, verbal memory, visual memory, executive functions, fluency, sustained attention, verbal and visuospatial working memory. This step was undertaken because there were not sufficient studies to perform meta-analyses for all individual tasks (See Appendix). In addition to cognitive domain analyses, task-specific analyses were conducted when at least three independent studies had employed a given task. Individual tasks that were analyzed separately included list learning and delayed recall, the Wisconsin cart sorting test (WCST) perseveration errors, Continuous performance test (CPT) sensitivity (d) score, Trail Making Task A and B, digit span, symbol coding, letter number sequencing and letter (phonemic) fluency.

Statistical analyses

Meta-analyses were performed using mix software version 1.7 on a Windows platform [65]. For each cognitive task, an effect size and standard error was estimated. Effect sizes were weighted using the inverse variance method, and a random effects model (DerSimonian–Laird estimate) was used because the distributions of effect sizes were heterogeneous for a number of variables. For studies that reported more than one cognitive task for each domain, a pooled effect size was calculated. The Q test was used to measure the ‘heterogeneity of the distribution of effect sizes’ (the dispersion of the effect sizes around their mean is greater than expected from sampling error alone). When the Q test was significant, ‘I2′ was used to quantify heterogeneity [66]. I2 estimates the percentage of total variation across studies due to heterogeneity rather than chance. I2 values between 0 and 0.25 indicate small magnitudes of heterogeneity, I2 values in the range 0.25–0.50 indicate medium magnitudes and I2 > 0.50 indicate large magnitudes. Publication bias was assessed by Egger's test that rely on the theory that small studies with significant rather than negative findings in studies with small sample size would be likely to be reported, while large-scale studies would be more likely to be published regardless of significance of findings. We also calculated homogeneity statistics using Qbet to test the differences between FHR and UHR.

Meta-regression analyses in UHR were conducted for age, gender, transition rate, duration of follow-up, the ratio of subjects with genetic risk and Global assessment of functioning (GAF) score. Meta-regression analyses in FHR were conducted for age and gender. Meta-regression analyses (weighted generalized least squares regressions) were conducted using SPSS version 11.0 (SPSS Inc., Chicago, IL, USA). Meta-regression analyses were performed with a random effects model and conducted using the restricted-information maximum likelihood method with a significance level set at < 0.05.


Cognitive deficits in FHR and UHR

FHR vs. HC

Compared with controls, FHR subjects were impaired in every domain (= 0.24–0.81; Fig. 2) and on all individual tasks (= 0.25–0.51; Table 3). There was significant heterogeneity in some of the cognitive variables, but the magnitude of heterogeneity was small for each of these variables (range of I2 = 0–0.09). There was no evidence of publication bias in any of the analyses. In meta-regression analyses, age and gender had no effect on observed between-group differences.

Table 3. Mean weighted effect sizes for cognitive differences between UHR, FHR and healthy controls
TestSampleHRHC D 95% CI Z P Q-test P I 2 Bias
  1. D, Effect size of between-group difference; Bias, P value of the Egger's test; UHR, Ultra-high risk to psychosis; FHR, Familial high risk to psychossi; TMT, Trail making test; WM, Working memory; EF, executive functions; CPT, Continous performance test; LNS, Letter Number sequencing; CI, Confidence interval; HC, Healthy controls; HR, High risk; WCST per, Wisconsin cart sorting test perseverative.

Premorbid IQ
UHR9 (2)8734970.300.13–0.483.35<0.0010.020.040.13
Current IQ
UHR12 (2)9075330.400.25–0.545.48<0.0010.150.020.36
Processing speed
FHR13 (11)6868080.350.22–0.495.07<0.0010.130.02
UHR5 (1)3442790.410.17–0.663.30.00090.080.040.88
Symbol coding
UHR4 (2)4172710.700.41–0.994.76<0.0010.050.050.41
FHR6 (5)4204260.370.23–0.515.20<0.0010.460
UHR6 (2)5563840.410.16–0.673.150.0020.0050.070.63
FHR8 (7)5665030.380.17–0.603.46<0.0010.030.05
Verbal memory
FHR12 (11)7577900.450.29–0.615.65<0.0010.060.03
UHR6 (1)3962420.680.34–1.013.97<0.0010.0030.120.21
Visual memory
FHR8 (7)5824030.510.30–0.724.79<0.0010.080.04
FHR11 (10)6056720.240.06–0.422.630.0090.020.04
WCST per
Verbal WM
FHR10 (9)6056010.320.12–0.513.230.0010.020.05
FHR5 (4)1782180.500.22–0.773.53<0.0010.100.05
Digit span
FHR7 (6)3023970.250.04–0.452.380.020.120.03
VisSpa WM
UHR8 (2)5834590.370.25–0.505.78<0.0010.5900.35
UHR7 (2)5213870.400.26–0.535.60<0.0010.5300.28
FHR12 (11)5307610.300.15–0.463.84<0.0010.040.04
Letter fluency
UHR5 (2)4642950.450.20–0.703.550.00040.060.040.91
Figure 2.

Cognitive deficits in UHR and FHR. PS, Processing speed; VerMem, Verbal memory; VisMem, Visual memory; Ver WM, Verbal working memory; EF, executive functions; FHR, Familial high risk; UHR, Ultra-high-risk to psychosis.

UHR vs. HC

Compared with controls, UHR subjects were impaired in every domain (= 0.34–0.71; Fig. 2) and on all individual tasks (= 0.21–0.70), except the Stroop interference score (= 0.38, = 0.11; Table 3). There was significant heterogeneity in some of the cognitive variables, but the magnitude of heterogeneity was small for each of these variables (range of I2 = 0–0.23). There was no evidence of publication bias in any of the analyses. The most robust impairments in the UHR subjects compared with controls were observed for symbol coding (= 0.70), visuospatial working memory (= 0.71) and smell identification (= 0.77).

In meta-regression analyses, there was a significant negative association between GAF score (level of functioning) and executive dysfunction (2.75, = 0.006), verbal working memory (= 3.95, = 0.0001) and premorbid IQ (= 2.87, = 0.004). There were no significant associations between any cognitive impairment and any other variables (with the exception of the association between genetic risk ratio and cognitive impairment in UHR, which is reported in the section 'Effect of combined genetic and clinical risk on cognition').

Comparison of cognitive deficits in FHR and UHR

In FHR subjects, impairment in IQ (= 0.81 vs. = 0.40, Qbet = 20.0, < 0.001) and premorbid IQ (= 0.64 vs. = 0.30, Qbet = 13.1, < 0.001) were more robust than deficits in UHR. In other cognitive domains, impairments in UHR and FHR groups were comparable, except more impaired visuospatial working memory in UHR than FHR (= 0.71 vs. = 0.35, Qbet = 4.6, = 0.03; Fig. 2). Examination of individual tasks showed more impairment in the UHR group on symbol coding (= 0.70 vs. = 0.37, Qbet = 6.40, = 0.01).

Effect of combined genetic and clinical risk on cognition

Meta-regression in the UHR sample showed that a higher ratio of GRDS was associated with more severe deficits in verbal memory (= 2.0, = 0.04), premorbid IQ (= 1.95, = 0.05) and CPT (= 2.06, = 0.04). There was also a trend association for verbal working memory (= 0.07).

In FHR studies, symptomatic subjects were significantly more impaired than asymptomatic subjects in the two domains examined: verbal memory (= 0.28, CI = 0.0–0.56, = 0.05, I2 = 0) and processing speed (= 0.51, CI = 0.0–1.02, = 0.05, I2 = 0.14).

Baseline cognitive functioning and transition to frank psychosis


At baseline, UHR-P subjects were cognitively more impaired than UHR-NP in all domains (= 0.31–0.49), except sustained attention (= 0.11; Table 4). UHR-P were also more impaired in most individual task analyses (= 0.26–0.52), except Trail Making Task A and letter number sequencing. Non-overlap ratios for these effect sizes were only between 9 and 34%, suggesting that the vast majority of UHR-P and UHR-NP subjects have similar performances. Heterogeneity of distribution of effect sizes was small (range of I2 = 0–0.22) and significant only for verbal fluency tasks. There was no evidence of publication bias.

Table 4. Mean weighted effect sizes for cognitive differences between UHR-P and UHR-NP
TestSampleUHR-PUHR-NP D 95% CI Z P Q-test P I 2 Bias
  1. UHR, Ultra-high risk to psychosis; UHR-P, UHR who converted to psychotic disorder at follow-up; UHR-P, UHR who has not converted to psychotic disorder at follow-up; TMT, Trail making test; WM, Working memory; EF, executive functions; CPT, Continous performance test; LNS, Letter, Number sequencing.

IQ 61724720.390.13–0.652.960.0030.140.040.39
Processing speed61913450.330.14–0.523.390.00070.470.060.94
Symbol coding41652540.390.06–0.722.300.
TMT B61863320.370.18–0.563.810.00010.8200.37
Verbal memory72033520.430.24–0.624.43<0.0010.3900.60
Visual memory 41251660.490.24–0.753.810.00010.8700.45
EF 61624410.340.15–0.533.530.00040.4800.37
Verbal WM 61672760.490.20–0.783.340.00080.120.050.45
Visspa WM 71161940.300.06–0.542.450.010.8400.99
Attention 61693130.11–0.08–0.311.150.250.8500.95
Fluency 72043570.490.11–0.862.520.010.00080.180.37
Letter fluency61913020.520.09–0.952.350.020.00040.220.36

In meta-regression analyses, lower transition rate was significantly associated with larger IQ (= 3.9, SE = 1.8, = 0.03) differences between UHR-P and UHR-NP.

The effect of transition to psychosis in FHR

In the current review, only one study examined the relationship between baseline cognitive deficits and the development of psychosis by follow-up. Verbal memory was the only measure that was significantly impaired in FHR subjects who developed psychosis by follow-up compared with those who did not [67]. There is some additional evidence of for cognitive deficits in childhood as predictors of psychosis in the follow-up in FHR samples, but these studies are excluded based on inclusion criteria.


Findings of this meta-analysis suggest that cognitive functions are significantly impaired in both FHR and UHR. The severity of most of these deficits is modest, and cognitive deficits appear to be more severe in people who have attenuated psychotic symptoms together with genetic risk for psychosis. General intellectual deficits were more strongly associated with genetic risk and poor performance on visuospatial working memory and digit symbol coding was more strongly associated with clinical high risk. This meta-analysis also provided evidence that cognitive functions were significantly more impaired in UHR-P compared with UHR-NP, although non-overlap ratios of both groups were small. These differences were most pronounced for general intellectual capacity in samples with lower transition rate, suggesting that UHR subjects in enriched samples are more similar to each other, independent of outcome.

Cognitive deficits in FHR

Findings of this meta-analysis suggest that FHR is associated with modest sized cognitive deficits, and distribution of effect sizes were quite homogenous. Cognitive deficits in FHR are comparable to abnormalities reported in meta-analysis of older unaffected relatives of schizophrenia [11]. However, older relatives have less severe deficits than our findings in general intelligence [12]. This difference might be related to the fact that older relatives are past the age of peak risk for psychosis onset, while young relatives may still develop the disorder. In older samples, better intelligence might be a protective factor in this population, as intellectual impairment is associated with higher risk for psychosis [68]. The relative strength of IQ finding in FHR suggests that general rather than specific cognitive deficits might be the genetically transmitted neurocognitive correlate of schizophrenia. This is consistent with evidence suggesting that genetics of intellectual disability and schizophrenia might be related [69, 70], and the fact that psychosis is more common among people with intellectual disabilities than the general population [68]. It could be argued that deficits observed in range of cognitive domains in FHR might be secondary to general cognitive dysfunction, and some authors have already suggested that this might be also true in established schizophrenia [71]. However, it is not possible to exclude the likely contributions of more specific cognitive deficits to psychosis risks. Future FHR studies controlling IQ might likely reveal such deficits.

Another important question is whether neurocognitive deficits in FHR are specific to genetic risk to schizophrenia or they are shared risk factor for psychosis in general. Few studies have compared neurocognitive deficits in young relatives of schizophrenia and affective psychoses and bipolar disorder. Schubert et al. [53] reported that offspring of patients with schizophrenia have more severe cognitive deficits than those of parents with affective psychosis, but offspring of patients with affective psychosis also underperformed in some tasks including selective attention and grammatical reasoning. Maziade et al. [72] found shared neurocognitive dysfunction in young offspring of patients with schizophrenia and bipolar disorders. Available evidence does not support the specificity of cognitive deficits to genetic risk to schizophrenia. Current data suggest that cognitive deficits are evident in young relatives of patients with psychosis (affective and non-affective), but might be modestly more severe in relatives of schizophrenia.

Cognitive deficits in UHR in comparison to FHR

Similarly, UHR status was associated with modest sized cognitive deficits and distribution of effect sizes across studies were more homogenous than reported in Giuliano et al. [19] (I2 = 30–70% for most measures) and Fusar-Poli et al. [18] (I2 = 50–64% for 4/9 of measures) who reported high I2 scores. The differences in the studies included in the current, meta-analysis and previous studies (i.e. new PACE data) are likely to explain these differences. Also, Fusar-Poli et al. [18] combined the data of the studies including healthy controls and help-seeking patient controls, which is problematic and would contribute to heterogeneity. Our findings of consistent cognitive deficits in UHR suggest that cognitive dysfunction predates the onset of first-episode psychosis, and that these cognitive deficits contribute to functional impairment. However, magnitudes of these deficits are modest compared with FEP and established illness. Indeed, the size of effects is more comparable to findings in FHR. For example, the magnitude of impairment in UHR in digit symbol task is half of that observed in schizophrenia [73]. A number of factors are likely to contribute to the observed cognitive differences between UHR and first-episode psychosis. Most importantly, like any other construct defining milder severity cases within a category of a mental illness, UHR is expected to recruit higher percentage of patients with good prognosis and false-positives [74]. Second, UHR is a more heterogeneous concept than established schizophrenia and FEP: it is likely to include a mixture of true prodromal schizophrenia, affective psychosis and other psychotic disorders; subjects who are in psychotic disorders spectrum but have a favourable outcome or present predominantly with negative symptoms and subjects with non-psychotic conditions (false positives). Similarly, many relatives in FHR group might not carry susceptibility genes that their parents or siblings carry. Also, use of antipsychotics is relatively less common in at-risk subjects than FEP and chronic samples which might be related to poor performance in some cognitive tasks. Therefore, it is not unexpected that lesser percentage of UHR and FHR subjects would have cognitive deficits of modest effect sizes. These findings has also been interpreted as an indirect evidence of neuroprogression; however, most of the available longitudinal studies did not show progressive cognitive deficits in UHR and FE patients [23, 75, 76], and a recent meta-analysis of follow-up studies of neurocognition in UHR and FEP studies did not find any evidence of cognitive decline [77].

The preliminary comparison between FHR and UHR suggests that general intellectual deficit is more likely associated with FHR. One explanation is that genetic risk paradigms might be more likely to recruit subjects with vulnerability to certain subgroups of schizophrenia (chronic cases that remain in care and may be associated with lower IQ) than clinical risk. The UHR criteria rely on the presence of positive, but not negative, symptoms and referral system through the school system is often used in recruitment of UHR. In contrast, visuospatial working memory and digit symbol tasks were more impaired in UHR compared with FHR. However, more studies directly comparing FHR and UHR are necessary to be conclusive in the subject as statistical comparison of findings of independent studies has limitations.

Co-occurrence of genetic and clinical risk

Our meta-analysis suggests that co-occurring attenuated symptoms and genetic risk was associated with more severe cognitive deficits. Seidman and colleagues found similar results in a clinical high risk sample [52]. Our meta-analysis provides additional evidence for the effect of co-occurrence on FHR studies. One explanation might be cumulative effect of different risk factors on cognition. However, association between GRDS+APS and cognitive deficits might be also a severity and reliability issue: clinical high risk together with family history is associated with higher risk of transition in follow-up, and attenuated symptoms on the background of genetic risk are expected to be a better indicator of to true prodrome or vulnerability. Cognitive deficits in pure UHR or FHR are likely to be less than reported in studies included. These findings might have implications for the discussions and reservations about introducing attenuated psychosis syndrome in psychiatric classification systems due to concern regarding validity and false-positives. Defining a risk syndrome based on attenuated psychotic symptoms only in relatives of people with schizophrenia can significantly increase the validity of the syndrome and decrease false positives. However, such a risk syndrome would have low sensitivity as only one third of the patients with schizophrenia have a family history of the illness.

Prediction of psychosis in high-risk samples

One important goal of the high risk research is to identify markers which can be use to predict illness onset [78]. In our meta-analysis, cognitive deficits are significantly more severe in subjects who developed psychosis by follow-up, suggesting that some of cognitive impairments in UHR samples are associated with risk to transition to full-blown psychotic disorder. This may also be true for FHR, but more research examining the transition to psychosis in FHR samples is needed. Interestingly, unlike other domains, sustained attention performance was not different between UHR-P and UHR-NP. One explanation might be that sustained attention is a vulnerability marker for being at risk for psychosis but not related to onset of the illness. Alternatively, sustained attention deficits might be observed not only in conditions within psychotic spectrum but also in other conditions which might be overrepresented in UHR-NP group such as mood disorders in which sustained attention deficit suggested to be a trait-marker [79-82] and has been shown to be not related to history of psychosis [83].

While cognitive dysfunction was more severe in UHR-P compared with UHR-NP, effects sizes for between-group differences were modest, approximately Cohen = 0.5 at most (for verbal fluency, verbal and visual memory and working memory), suggesting that there is significant overlap even for these measures (67%) between cognitive performances of UHR-P and UHR-NP. Such a high overlap rate suggests that potential of neurocognition to predict illness prognosis is very limited. In addition, not all UHR-P subjects will be diagnosed with schizophrenia, and it is possible that cognitive deficits might have a better predictive value for schizophrenia than other psychotic illnesses. A recent study failed to show significant differences between UHR subjects who developed schizophrenia and affective psychosis [84]. Still, more research is needed in this subject. On the other hand, Lin et al. [85] showed larger and broader differences in cognitive performance when UHR individuals with a poor functional outcome compared with good outcome (regardless of transition status), than when UHR-P and UHR-NP were compared. In this meta-analysis, we also found that decreased GAF scores are related to more severe cognitive deficits. These findings suggest that reducing the heterogeneity of the UHR sample by examining functional outcome may be a useful approach. Of note, our confounding factor analyses suggested that these differences in general intellectual abilities might be more severe in samples with lower transition rates and that UHR-P and UHR-NP groups perform more similarly in samples with a high transition rate. This finding supports the possibility that low intellectual capacity might be a vulnerability factor for risk of developing psychosis (whether they develop psychosis or not) as samples with high transition rates are likely to be enriched UHR samples. In contrast, many UHR-NP subjects in samples with low transition rate would be false-positives who have preserved intellectual abilities and would drive the between-group differences. In accordance with these findings, studies examining non-help-seeking UHR samples reported less cognitive deficits [86].

The advantages of this meta-analysis include homogenous distribution of effect sizes, co-examination of both familial and genetic risk and large sample size. Limitations of this meta-analysis include differences in methodology, such as follow-up duration and the various risk criteria and exclusion criteria used in the included studies. Many studies did not report variables such as positive/negative symptoms, functioning levels, antipsychotic use, and cannabis use and other substance use.

In conclusion, findings of this meta-analysis confirm that UHR and FHR subjects have cognitive deficits that are modest compared with established schizophrenia. Cognitive impairment is larger in subjects who have both genetic and clinical risk and those who developed psychotic disorder at follow-up. However, it is not possible to know timing of the development of cognitive deficits with cross-sectional methods. There is a need for large-scale longitudinal studies that follow at-risk subjects many years before developing UHR to reveal timing and development of cognitive deficits in psychotic disorder.


Prof Christos Pantelis and Prof Patrick McGorry were supported by a National Health and Medical Research Council of Australia (NHMRC) Program Grant (ID: 566529). Prof Christos Pantelis was also supported by a NHMRC Senior Principal Research Fellowship (ID: 628386).

Declaration of interest

Over the last two years, Christos Pantelis has participated on Advisory Boards for Janssen-Cilag and Lundbeck. He has received honoraria for talks presented at educational meetings organized by Janssen-Cilag and Lundbeck. Patrick McGorry received research grant support from Janssen Cilag and Astra Zeneca and honoraria for educational lectures from Janssen Cilag, Eli Lilly, Lundbeck, Pfizer and Roche. Other authors have no conflict of interests to be reported. None of the authors have financial or personal relationships, interests and affiliations relevant to the subject matter of the manuscript.


Individual cognitive tests under each domain

  1. TMT, Trail making test; HSCT, Hayling sentence completion task; RAVLT, Rey auditory verbal learning test; HVLT, Hopkins verbal learning test; WMS, Wechsler memory scale; WMS, Wechsler memory scale; RCFT, Rey complex figure test; CVLT, California Hopkins verbal learning test; SOPT, Self-ordered pointing test.

Processing speedTMT-A, digit symbol, letter fluency, category fluency, reaction time, TMT-B, Stroop interference, HSCT, set shifting
Verbal memoryRAVLT, CVLT, HVLT, logical memory, word pair, paired associate learning
Visual memoryVisual reproduction/ WMS visual memory, RCFT
Verbal working memoryDigit span, letter number sequencing
Visual working memorySpatial span, SWM, DMTS, SOPT, DRT, visual 2 back, beads task
Executive functionsWCST, mazes, matrix reasoning
AttentionContinuous performance test, d2 concentration test, cancellation test, selective attention
FluencyLetter fluency, category fluency