A Longitudinal Study of Premorbid IQ Score and Risk of Developing Schizophrenia,Bipolar Disorder, Severe Depression, and Other Nonaffective Psychoses FREE

There is strong evidence that patients with schizophrenia show impairmentsin a wide variety of neuropsychological tasks, including attention, executivefunction, language, memory, and general intellectual ability.¹ Thus,when investigating risk factors for schizophrenia, case-control studies ofpremorbid cognitive ability may be particularly biased if this is assessedafter onset of the disorder. Overall results from such studies for schizophreniaindicate that impairment in intellectual ability may exist from early in lifeand is not just a consequence of the pathological process of disease onset.²

Findings from population-based studies support this view. Jones et al,³ using the 1946 birth cohort, found that subjects whodeveloped schizophrenia were more likely to have had impairments in childhoodeducational scores that persisted after adjusting for sex and social class.David et al⁴ found a strong association betweenIQ score at the age of 18 years and subsequent risk of developing schizophreniain a cohort of male Swedish conscripts. In a separate Swedish conscript cohort,Gunnell et al⁵ found that IQ was associatedwith risk of early-onset schizophrenia after adjusting for markers of prenataladversity and obstetric complications. Lower verbal and nonverbal IQ scoresat the age of 11 years were reported for subjects who developed schizophreniain the National Child Development Study.⁶ Inthe Dunedin birth cohort, children who developed a schizophreniform disorderin adulthood were also more likely to have worse childhood IQ scores and receptivelanguage impairments.⁷

A recent nested case-control study of subjects drafted into the Israelimilitary service found that subjects who later developed schizophrenia showedhighly significant impairments in premorbid adolescent intellectual functionand reading tasks compared with controls.⁸ Controlswere matched for sex, high school attendance, and age at assessment. Subjectswith schizoaffective disorder also showed significant impairments in 1 ofthe 4 subtests of intellectual performance, although the small numbers inthis group may have limited statistical power to observe significant differencesin the other subtests measured. Another population-based, nested case-controlstudy,⁹ however, found no association betweenschool academic performance and risk of schizophrenia. Although this studywas well powered, school education results may reflect intellectual abilityless adequately than IQ score does.

Cognitive ability in bipolar disorder has been less well studied, butthere are reasonably consistent findings that subjects with this disorderalso show evidence of cognitive impairments after onset of the illness.^10- 11 These deficits are observed evenwhen subjects are euthymic, although impairments are more marked when affectivesymptoms are present. The few studies that adjust for residual affective symptomsshow evidence for impairments, mainly in verbal learning and memory and sustainedattention during remission.^10- 11

Studies of premorbid intellectual functioning in subjects who developbipolar disorder show some variation in results. In the 1946 birth cohort,van Os et al¹² found an association betweenchildhood cognitive ability and risk of both childhood and adult-onset affectivedisorders. In a case-control design, Gilvarry et al¹³ foundthat affective psychosis subjects had significantly lower premorbid IQs, asestimated by the National Adult Reading Test (NART), than their first-degreerelatives but premorbid IQs that were not different than those of controls.Selection of controls from job centers may have biased the comparison withthe control group. Neither of these studies distinguished between unipolarand bipolar disorders, making the association between premorbid intellectand bipolar disorder difficult to interpret. A comparison of NART scores amongsubjects with bipolar disorder found that subjects were no different fromcontrols but performed significantly better than subjects with schizophrenia.¹⁴ In an early cohort study of US Army recruits, Mason¹⁵ found that subjects who developed a manic-depressivedisorder had a significantly raised IQ score at induction vs noncases. Itis unclear what the criterion for diagnosis of manic depression was in thisstudy, but it was probably a clinical diagnosis. No other study since hasreplicated this observation of a raised premorbid IQ in this patient group.

Two additional case-control studies^16- 17 foundno difference in NART scores between euthymic bipolar patients and controls.Goldberg et al¹⁸ found no difference in premorbidIQ as measured by the Wide Range Achievement Test–Revised reading testbetween subjects with bipolar disorder and those with either unipolar depressionor schizophrenia. Sample sizes were small for all these studies (approximately20 subjects), and statistical power is likely to have been limited. Similarly,in the Dunedin birth cohort, there was no association observed between childhoodIQ or language measures and risk of developing bipolar disorder (N = 20).⁷ However, the confidence intervals (CIs) narrowly overlapwith those for schizophrenia, suggesting that a difference between the 2 disordersmay exist for measures of premorbid intellect. Furthermore, Reichenberg etal⁸ found no association with premorbid intellectualor language measures for 68 subjects with nonpsychotic bipolar disorder.

Comparisons between subjects with schizophrenia and bipolar disorderon postmorbid tests of cognitive function show that those with schizophreniaperform consistently worse than those with bipolar disorder.^11,19 Inboth disorders, though there is a wide variation in study results, subjectsseem to perform more poorly on performance-based rather than verbal-basedneurocognitive tests.^1,11

A limitation of many studies to date has been the difficulty of untanglingeffects of IQ score on risk of bipolar disorder as opposed to affective disordersin general. This is important given the theoretical implications of a separatepathogenesis for the 2 disorders. It is also unclear whether intellectualfunction is associated with risk of psychosis per se, regardless of the diagnosis.For example, Gunnell et al⁵ found an associationbetween IQ score and nonaffective psychoses, whereas David et al⁴ foundan association between IQ and nonschizophrenia psychotic disorders that includedsubjects with affective psychoses.

Furthermore, few studies have examined the possible role of confoundingas a possible explanation for associations between IQ and psychiatric diagnoses.The aim of this study was to further examine the association between premorbidIQ score and risk of developing bipolar disorder, severe depression, schizophrenia,and other psychotic disorders in men. Specifically, we wanted to (1) examinethe relationship between premorbid IQ score and risk of developing bipolardisorder by studying a larger sample of cases than previous studies and byincluding subjects both with and without psychotic symptoms; (2) examine whetherit is specifically low IQ score that increases risk of schizophrenia or whethereven average IQ is associated with increased risk compared with high-scoringsubjects; (3) investigate the effect of premorbid IQ score on risk of developingother nonaffective psychoses and depression severe enough to warrant hospitaladmission; (4) investigate the effect of potential confounders on the relationshipof these disorders with IQ score, including drug use, paternal age, familyhistory, social class, and nonpsychotic psychiatric disorders at the age of18 years; (5) investigate possible causal pathways by which IQ score may operateto alter risk of developing these disorders; and (6) investigate the relationshipbetween premorbid IQ score and age at onset, number of admissions, and totaldays of admission as markers of illness severity.

The cohort consisted of 50 087 Swedish men conscripted for compulsorymilitary training during 1969-1970. More than 98% were 18 to 20 years of age.Only 2% to 3% of the male population were excused conscription on accountof severe mental or physical disability. This is the same cohort used by Davidet al,⁴ who examined the relationship betweenIQ score and risk of both schizophrenia and other psychoses. However, we nowhave additional cases identified due to a longer follow-up period, and wealso investigate other psychotic diagnoses separately.

The conscription procedure included tests of intelligence and nonanonymous,self-reported questionnaires on family, social background, behavior duringadolescence, and substance use. All subjects underwent a structured interviewconducted by a psychologist, and those reporting any psychiatric symptomswere interviewed by a psychiatrist and given a diagnosis according to the International Classification of Diseases, Eighth Revision (ICD-8)²⁰ when applicable.Thirty-four cases of psychosis diagnosed at conscription were excluded fromthe study, resulting in a sample size of 50 053. Permission to use theanonymized database was granted by the Karolinska Institute Research EthicsCommittee and the Swedish Data Inspection Board.

There were 4 main subtests to the assessment of intelligence, each yieldinga 9-point summary score. These subtests assessed verbal IQ, visuospatial ability,general knowledge and intelligence, and mechanical knowledge. Further detailsof these subtests are given in a previous study using this cohort.⁴ These 4 subtests were aggregated to give an overallstandardized intelligence score, ranging from 1 to 9 and corresponding toapproximate IQ bands of less than 74, 74 through 81, 82 through 89, 90 through95, 96 through 104, 105 through 110, 111 through 118, 119 through 126, andmore than 126.

The Swedish National Hospital Discharge Register recorded approximately70% of all psychiatric admissions in 1970, increasing to 83% in 1973. Coveragewas 97% in 1974-1983 and 80% to 95% in 1984-1986 and has been virtually completesince 1987. The linkage reported herein was from 1970-1996. The incompleteregistration during some periods is unlikely to have affected the results.

Patients were given clinical diagnoses according to the Nordic versionof ICD-8 (ICD-9 from 1987).Satisfactory validity of schizophrenia diagnoses in Sweden have been previouslyreported.^21- 22 Outcomes investigatedwere (1) schizophrenia (schizophrenia/schizoaffective disorder, ICD-8 and ICD-9: 295.00-99), (2) bipolar disorder(manic-depressive psychosis, manic/circular type, ICD-8: 296.1, 296.3, 298.1; ICD-9: 296.0, 296.2-5,298.1), (3) severe depression (manic-depressive psychosis, depressed type, ICD-8: 296.0, 296.2, 298.0; ICD-9: 296.1,298.0), and (4) other nonaffective psychoses (paranoid states, other psychoses,substance-induced psychoses, ICD-8: 297.0-9, 298.2-3,298.9, 291.2-3; ICD-9: 297.0-9, 298.2-4, 298.8-9,291.3, 291.5, 292.1). Where subjects had 2 or more different psychotic diagnosesduring the follow-up period, the following approach was used: if 1 diagnosisclearly dominated the admission diagnoses, this was the one used. However,if there was a roughly equal distribution of 2 or more diagnoses, a hierarchicalapproach was used. The diagnostic hierarchy was as follows: schizophrenia,schizoaffective disorder, bipolar disorder, other psychoses, and severe depression.Age at first admission was investigated as a proxy measure of age at illnessonset. The number of psychiatric admissions and total duration of admissionsin days for each individual were recorded as crude measures of illness severityto allow investigation of this phenotype with IQ score.

Logistic regression was used to calculate odds ratios (ORs) and 95%CIs for the different diagnoses given IQ score both before and after adjustmentfor potential confounders. Although a small number of subjects died duringfollow-up, analysis using Cox regression, which takes into account such losses,made no differences in the results. Logistic regression was therefore retainedas the method of choice. The relationship between IQ score and age at onsetwas examined using a linear regression model. The relationship between IQscore and number and severity of admissions was examined using Spearman rankcorrelation coefficients.

Variables previously shown to be associated with risk of schizophreniaand examined as potential confounders in the relationship between IQ and riskof disease were psychiatric diagnosis at conscription,²³ druguse,²⁴ place of upbringing,²⁵ andpaternal age.²⁶ Disturbed behavior in childhood,family history of psychiatric illness, and father's occupation were also examined.Variables that were considered potential mediators of the relationship betweenIQ and risk of disease (ie, lying on the causal pathway) included drug use²⁴ and personality variables concerned with interpersonalrelationships.²⁷ Only 3% of the sample hadmissing data for any of the questions.

Of the 50 053 male subjects, 362 (0.72%; 95% CI, 0.65%-0.80%) haddeveloped schizophrenia by 1996, 108 (0.22%; 95% CI, 0.15%-0.25%) had developedbipolar disorder, 113 (0.23%; 95% CI, 0.19%-0.28%) had developed severe depression,and 223 (0.45%; 95% CI, 0.39%-0.51%) had developed nonaffective psychoses.IQ test results were available for all cases and were missing for 86 (0.2%)of noncases (P = .43). Of the 7 variables initiallyinvestigated as potential confounders or mediating variables, only 4 had anyeffect on the adjusted results and were used in the final analyses. A summaryof these in relation to IQ is presented in Table 1. For the purposes of Table 1only, IQ was split into 3 categories,and disturbed behavior was treated as a dichotomous variable, using the 95thpercentile as a cutoff point for coding. The 4 intelligence subtests wereall correlated with each other. The strongest correlation was between verbalIQ and general knowledge (r = 0.75), and the weakestwas between visuospatial ability and mechanical knowledge (r = 0.50).

The 9-point aggregate IQ test score (Table 2) was associated with risk of developing schizophrenia (adjustedOR, 1.26; 95% CI, 1.19-1.33; P<.001). Of the 4intelligence subtests, reduced performance for verbal IQ, visuospatial ability,and mechanical ability were associated with increased risk of schizophreniathat persisted after adjusting for the other subtest measures (Table 3). The largest increase in risk of developing schizophreniawas for subjects with the lowest IQ scores (Figure 1).

Omitting subjects with an IQ score below the mean, subjects with anaverage IQ had a significantly greater risk of developing schizophrenia thanthose with the highest IQ (OR, 1.3; 95% CI, 1.04-1.54; P = .02). Examination of whether a nonlinear relationship (within thelogistic regression model) between IQ score and schizophrenia provided a betterfit for the data was made by inclusion of a quadratic term (likelihood ratiotest [LRT], χ²₁ = 2.88, P =.09).

Although schizoaffective disorders are included in the schizophreniagroup, we also looked at the specific relationship between IQ and this disorder.There were 35 subjects who developed schizoaffective disorder (0.07%; 95%CI, 0.05%-0.07%). Lower IQ scores were associated with an increased risk ofdeveloping schizoaffective disorder (adjusted OR, 1.40; 95% CI, 1.16-1.67; P<.001). In the subtest analyses, reduced performancefor verbal IQ, visuospatial ability, and mechanical ability were associatedwith increased risk of schizoaffective disorder, similar to schizophrenia.These became nonsignificant after adjusting for the other subtest measures.

There was no evidence that aggregate IQ score during adolescence wasassociated with subsequent risk of developing bipolar disorder (adjusted OR,0.99; 95% CI, 0.89-1.09; P = .78). However, lowerIQ score was associated with risk of developing severe depression (adjustedOR, 1.19; 95% CI, 1.09-1.31; P<.001) and riskof other nonaffective psychoses (adjusted OR, 1.24; 95% CI, 1.15-1.33; P<.001).

Examination of the 4 intelligence subtests showed that there were nosignificant associations for any of the subtests and risk of developing bipolardisorder in either the crude or adjusted analyses. Performances for all 4tests, however, were associated with risk of developing severe depressionor other nonaffective psychoses. After adjusting for the other subtest scores,the associations between severe depression and mechanical ability and betweenother nonaffective psychoses and both mechanical ability and verbal IQ persisted.

Of subjects with bipolar disorder, 25 had an additional diagnosis ofa psychotic illness (schizophrenia or other nonaffective psychoses) at somepoint during the follow-up period. The relationship with aggregate IQ scorefor these individuals was also investigated (OR, 0.97; 95% CI, 0.80-1.18; P = .76).

Comparison of mean ± SD aggregate IQ score for subjects developingbipolar disorder (5.1 ± 2.1) compared with the 3 other diagnostic categoriesgrouped together (4.3 ± 2.1) was performed using analysis of variance(F_1,791 = 12.1, P<.001). There wereno significant differences in IQ score among any of the 3 other diagnosticcategories (schizophrenia, 4.3 ± 2.1; nonaffective psychoses, 4.3 ±2.1; severe depression, 4.5 ± 2.1).

Results for the relationship between IQ score and age at onset, numberof admissions, and duration of hospitalization for each of the diagnosticcategories are presented in Table 4.There were no significant associations between these clinical markers andthe disorders investigated. A nonlinear relationship between IQ score andage at onset was also investigated to try to further distinguish between premorbideffects of IQ and possible prodromal effects on score results by inclusionof a quadratic term into the model (schizophrenia LRT, χ² =0.15, P = .70; bipolar disorder LRT, χ² = 2.8, P = .10; severe depression LRT, χ² = 1.5, P = .22; nonaffective psychoses LRT, χ² = 0.80, P = .37).

Premorbid IQ score was associated with risk of developing schizophrenia,schizoaffective disorder, other nonaffective psychoses, and severe depressionin men. However, risk of developing bipolar disorder was not associated withpremorbid IQ levels. The effect of IQ score on risk of these disorders wasnot mediated through or confounded by any of the other variables investigatedin this study.

These findings add to the results of Reichenberg et al,⁸ whofound no association between adolescent IQ score and subsequent developmentof nonpsychotic bipolar disorder. In our study, the bipolar group is likelyto have included subjects both with and without psychosis, because formalseparation of these was not possible due to the classification system usedduring the study period. However, in the subset of bipolar subjects who hadalso been given a psychotic diagnosis at some point, there was no evidencefor an association with premorbid IQ score, although the number of subjectsin this group meant that statistical power was rather limited.

The evidence that reduced intellectual performance is associated withbipolar disorder after onset of the illness suggeststhat this may be due to either damage caused by disease onset or the effectsof psychotropic medication on test performance. It is also possible that thepresence of affective or psychotic symptoms may affect cognitive performance,although impaired performance has been observed even after adjustment forresidual symptoms.^10- 11

The relationship between premorbid intellectual function and risk ofschizophrenia is already well established. The lack of evidence in our studyfor nonlinearity within the logistic regression model (which models the logarithmof odds) indicates that risk of schizophrenia increases exponentially withchange in IQ score. However, our results also indicate that risk of schizophreniawas significantly increased in subjects with average IQ compared with thosewith high IQ scores. Risk is spread over the whole of the IQ spectrum, andit is not just low IQ score that is a marker of increased risk of developingschizophrenia. Put another way, higher IQ score could be said to protect againstpsychotic illness.

This study also confirms an association between low IQ score and increasedrisk of schizoaffective disorder. The effect size seems comparable to thatof schizophrenia, suggesting that etiologically, schizoaffective disordermay be more akin to schizophrenia than bipolar disorder. A similar associationwas observed in one recent study⁸ but not inanother less well-controlled one that used the NART to estimate premorbidIQ.¹³ Although all the subtest estimates inour study were nonsignificant, the trends were similar to those seen for schizophrenia,with the greatest effect being for the test of mechanical ability. The smallnumber of subjects with schizoaffective disorder means that statistical powerwas limited for these analyses.

A strong association between IQ score and other nonaffective psychoseswas also present. Although a similar finding was reported previously for thiscohort,⁴ in this secondary analysis subjectswith affective psychoses were analyzed separately under bipolar disorder orsevere depression, and new cases were also identified for analysis as a resultof the longer follow-up period. Our results support the view that study ofpsychoses in general rather than schizophrenia in particular may be one wayof increasing power to identify such shared risk factors among psychotic disorders.

This is the first population-based, longitudinal study to our knowledgethat examines the relationship between premorbid IQ score and risk of developingdepression severe enough to warrant hospital admission. Our results indicatethat premorbid IQ score has a similar effect on risk for severe depressionas for schizophrenia, schizoaffective disorder, and other nonaffective psychoses.Also, as for these other disorders, the strongest impairments seem to be inperformance-based rather than verbal-based tests. Depression is a markedlyheterogeneous disorder, and depression severe enough to warrant hospital admissionis likely to represent an extreme of the disorder in which neurodevelopmentalabnormalities, including intellectual development, may be more prone to occur.

One of the strengths of this study is the ability it affords us to assessconfounding in the relationship between IQ and any of the outcomes investigated.There was no evidence of any strong confounding in this relationship. Furthermore,individual adjustments for drug use, disturbed childhood behavior, and socialpersonality traits had no effect on the associations observed, indicatingthat they do not lie on the causal pathway between IQ score and risk of psychoticdisorders. It is possible that IQ may either have a more direct effect ondevelopment of psychosis, perhaps by influencing cognitive interpretationof stimuli and events, or that IQ score may be a measurable marker of subtlecerebral disease that itself influences development of psychotic symptoms.There is substantial evidence that neuroanatomical abnormalities occur inschizophrenia, depression, and bipolar disorder.^10,28- 30 Theresults from the numerous neuroimaging studies are somewhat difficult to interpretgiven the often small sample sizes and marked heterogeneity in sample selectionand methods, as well as the difficulties in separating premorbid and postmorbideffects. Nevertheless, there seems to be a substantial overlap in the patternof abnormalities observed in the different disorders. However, there is alsoevidence to suggest important differences between bipolar disorder and bothschizophrenia and depression, for example, in the hippocampus and amygdala,that might partly explain the differences in premorbid IQ scores observedin this study.^10,28- 29

The possibility of bias as an explanation for these results needs tobe considered. Outcome misclassification is likely to be low for schizophreniagiven that more than 90% of people with schizophrenia are admitted to a hospitalat some point during their illness.³¹ Subjectsin the cohort with severe depression or bipolar disorder, however, are likelyto be somewhat less representative of all such cases in the population, becauseadmission to a hospital is less invariable. The results of such misclassification,if nondifferential with regard to IQ score, would be to underestimate anytrue associations between IQ and risk of developing these disorders. However,if subjects with low IQ are more likely than subjects with high IQ to be admittedto a psychiatric hospital, an overestimation of association between low IQscore and disease could occur. If this were the case, then we might expecta similar effect for subjects with bipolar disorder as for those with severedepression (ie, that an association with low IQ score would be present forboth disorders). Although it is possible that subjects with depression aremore likely to be admitted if they have low rather than high IQ, with an oppositerelationship for bipolar subjects whereby admission is more likely for highrather than low IQ cases, this seems an unlikely scenario. The marked differencein the association with IQ score suggests that selection bias is not likelyto be the main explanation for the disparity in results between these 2 disorders.

The use of clinical rather than operationally defined diagnoses wouldalso tend to reduce differences between bipolar disorder and other psychoticdisorders if such misclassification were random with respect to IQ score.However, information bias cannot be ruled out as a potential explanation ifdifferential misclassification is present, for example, if subjects with bipolardisorder and low IQ scores were more likely than those with higher IQ scoresto mistakenly receive a diagnosis of schizophrenia.

Structural abnormalities, such as ventricular enlargement and reducedbrain volume, detected by neuroimaging techniques are more marked in subjectsin whom schizophrenia commences in early childhood than in adolescence oradult life.³² If such structural abnormalitiesare a marker of a more global cerebral insult that occurs from early life,it is reasonable to expect that IQ score may be more likely to be reducedin early onset or more severe cases. However, no association was observedbetween IQ score and age at onset or severity of illness as measured by numberor duration of admissions. No adjustment for measures of social disadvantageat admission were made, although one would expect this confounder to resultin, if anything, an increased association. Similarly, no association was observedbetween NART scores and age at onset of schizophrenia in 2 case-control studies,^13,33 although the latter reported an associationbetween academic record and duration of hospitalization. It is possible thatIQ score does not correlate strongly enough with anatomical abnormalitiesto make such an association detectable. Furthermore, other factors likelyto contribute to IQ score may act independently of gross cerebral diseaseand reduce the power to observe these associations.

Examining markers of clinical severity also allows us to investigatethe possible presence of selection bias. If low IQ were actually a risk factorfor severity of illness rather than presence of illness per se, an apparentassociation between IQ and risk of psychotic disorders would arise if severecases are more likely to result in hospital admission. The lack of associationwith illness severity suggests that such a bias is not an adequate explanationfor these results.

We could not exclude the distinct possibility that IQ score is a riskfactor for very early onset, that is, those with onset before conscription.The 34 cases with a psychotic disorder at conscription were, of course, excludedgiven the study design. Whereas for schizophrenia there seems to be no relationshipbetween IQ score and age at onset, for severe depression, there is a suggestionthat subjects with a low IQ may have a younger age of disease onset. One explanationmay be prodromal effects of depression that result in reduced IQ scores atconscription.

Finally, one limitation to this study is the use of a nonpublished batteryof tests, that used by the Swedish Army. Our descriptive labeling of the testsis to some extent arbitrary, and caution should be observed equating thesewith performance and verbal subtests of, for example, the Wechsler batteries.This limits comparability of these results with other studies. However, thefact that the entire population of men was administered the same battery allowedus to standardize the scores, which should compensate for this and aids interpretation.

In conclusion, this study indicates that premorbid IQ is likely to bea risk factor for psychotic illnesses in general rather than schizophreniain particular. However, premorbid IQ score does not seem to have an effecton risk of developing bipolar disorder. Although extrapolating observationalfindings to underlying molecular mechanisms can only be theoretical in nature,these results nevertheless suggest that some aspects of the neurodevelopmentaletiology of bipolar disorder are different from those for schizophrenia, nonaffectivepsychoses, and severe depression.

Corresponding author: Stanley Zammit, MRCPsych, Department of PsychologicalMedicine, University Hospital of Wales, Cardiff CF14 4XN, Wales (e-mail: zammits@cardiff.ac.uk).

Submitted for publication August 25, 2003; final revision received October30, 2003; accepted November 18, 2003.

This research is funded by a Clinical Training Fellowship awarded toDr Zammit by the Medical Research Council UK.