Monthly birth distribution in subjectswith deficit and nondeficit schizophrenia in 3 northern hemisphere samplesof convenience.
Monthly birth distribution in subjectswith deficit and nondeficit schizophrenia in 2 northern hemisphere prevalencesamples.
Monthly birth distribution in subjectswith deficit and nondeficit schizophrenia in 5 northern hemisphere incidencesamples.
Odds of June/July birth. In order,the studies are based on 3 samples of convenience (DSM-IV Field Trial, Maryland Psychiatric Research Center outpatients, andFrench multicenter study), 2 population-based prevalence samples (EpidemiologicalCatchment Area and Nithsdale), and 5 population-based studies of incidencecases or cases that approximated incidence samples (Camberwell, Roscommon,Dumfries and Galloway, Cantabria, and Suffolk County). Size of the box isproportional to the number of subjects in the study.
Messias E, Kirkpatrick B, Bromet E, Ross D, Buchanan RW, Carpenter WT, Tek C, Kendler KS, Walsh D, Dollfus S. Summer Birth and Deficit SchizophreniaA Pooled Analysis From 6 Countries. Arch Gen Psychiatry. 2004;61(10):985-989. doi:10.1001/archpsyc.61.10.985
In some reports, summer birth has been associated with deficit schizophrenia.
Deficit schizophrenia and nondeficit schizophrenia also differ in several
To conduct a combined analysis of the published and unpublished data
sets from the northern hemisphere that relate deficit and nondeficit schizophrenia
to month of birth.
Studies of season of birth in which it was possible to make a deficit/nondeficit
Published studies with samples of convenience and all known population-based
studies with the deficit/nondeficit categorization were included. The studies
came from 6 countries.
Three published studies of samples of convenience, 2 population-based
prevalence studies, and 5 population-based studies that approximated incident
samples were included. Month of birth was compared for deficit and nondeficit
schizophrenia, using meta-analytic fixed-effects models.
A group x month goodness-of-fit χ2 showed a significant
difference between deficit and nondeficit subjects in season of birth (P < .001) in the studies that approximated
incidence. This difference was largely due to an increase in deficit schizophrenia
births in June and July (odds ratio, 1.9; 95% confidence interval, 1.3-2.9).
Similar results were found in the prevalence studies. A similar pattern was
found in 2 of the 3 samples of convenience, but when combined, these 3 samples
did not show a significant deficit/nondeficit difference.
Deficit schizophrenia has a season of birth pattern that differs from
that of nondeficit schizophrenia. This analysis supports the notion of a separate
disease within schizophrenia.
Schizophrenia is a clinically heterogeneous disorder and may consistof a group of diseases with overlapping clinical features but different etiologiesand, to some extent, different pathophysiologies. The association of differentrisk factors with different clinical features would be an important step indefining subgroups that are more homogeneous relative to their pathophysiology.
Winter birth was first reported to be a risk factor for schizophreniain 1929, and there have been many replications of this association in differentsites.1 Clinical characteristics that havebeen associated with winter birth include paranoid subtype and a more benigncourse of illness, although these have not always been found.2,3
The clinical features associated with winter birth differ from thoseof patients with deficit schizophrenia, a group defined by primary, enduringnegative symptoms and a relatively severe form of the illness.4 Consistentwith this suggestion of a deficit/nondeficit difference in season of birth,a summer birth excess in patients with deficit schizophrenia has also beenfound in several studies.5- 8 Confirmationof a different risk factor profile for deficit and nondeficit groups wouldsupport the hypothesis that the deficit group represents an etiologicallydistinct form of schizophrenia and would be consistent with other evidencethat deficit and nondeficit patients with schizophrenia also differ relativeto course of illness, biological correlates, and treatment response.9
An important limitation of the existing studies of the summer birtheffect is sample size. The deficit schizophrenia group comprises 15% to 20%of schizophrenia cases in epidemiological samples. If, for instance, 30% ofnondeficit patients are born in the summer, one would need 86 deficit and344 nondeficit cases to have 80% power to detect an odds ratio of 2.10 However, previous studies of summer birth and deficitschizophrenia have often used smaller samples.
In the present study, we pooled data from the previously published studiesas well as unpublished data. We performed a separate analysis for samplesof convenience, prevalence studies, and studies approximating incidence, aswell as a pooled analysis to produce a total estimate of the season effectin deficit schizophrenia.
Data came from studies that made the deficit/nondeficit categorizationwithin schizophrenia on the basis of the Schedule for the Deficit Syndrome(SDS), the Proxy for the Deficit Syndrome (PDS),6- 9,11 ora consensus medical record review diagnosis. The SDS12 isa semistructured interview for the identification of primary and enduringnegative symptoms and is the standard method for identifying patients withdeficit schizophrenia. Administration of the SDS is usually not feasible inlarge epidemiological samples, so medical record review and the PDS11 have also been used. The present analysis includesdata from 6 studies in which the PDS was used.8,13,14 ThePDS quantifies the combination of typical negative symptoms and diminishedemotionality, and it generates a score for each subject. On the basis of prevalenceassumptions, cutoff points for deficit and nondeficit groups are defined.The validity of the categorizations is then tested by comparing the 2 diagnosticgroups’ clinical features. This approach has also been validated bycomparison with SDS diagnoses.11 Details onthe validation procedures are included in the original publications.8,11,13,14 Methodsfor the medical record reviews are discussed below. Where possible, the presentanalyses were conducted with samples meeting DSM criteriafor schizophrenia; however, for some of these studies, DSM criteria were not used.
Studies were categorized based on the sampling method: convenience samples,population-based prevalence samples, and population-based samples with incidentcases or samples that approximated incidence. Data came from 2 prevalenceepidemiological samples in the northern hemisphere6,7 andfrom 2 articles with results from 3 samples of convenience.5,15 Datafrom the studies approximating incidence came from previously published studies(Camberwell,8 Dumfries and Galloway,14 and Cantabria13),the Suffolk County Mental Health Project (a previous publication was basedon a partial Suffolk County Mental Health Project sample5),and unpublished data from the Roscommon Family Study.16 Studieswere conducted with institutional review board approval for each site.
Samples of convenience came from the Maryland Psychiatric Research Center,the DSM-IV17 FieldTrial,5 and a French multicenter study.15 All 3 of these studies used the SDS in patients with DSM-III18 schizophrenia.
The first population-based study of prevalent cases came from Nithsdale,6,7 a town in the region of Dumfries andGalloway in Scotland. There was overlap between this sample and the Dumfriesand Galloway incidence study described below. However, this was probably smallbecause the study in Dumfries and Galloway used a registry to identify patients,whereas in the Nithsdale study, on May 1, 1996, all patients with an International Classification of Diseases, Ninth Revision (ICD-9)19 diagnosis of schizophrenia in Nithsdalewere identified by the key informant method.
The other prevalence data7 came fromthe 5-site Epidemiological Catchment Area study. In that study, psychiatricsymptoms were assessed with the National Institute of Mental Health DiagnosticInterview Schedule (DIS), administered by lay interviewers.20 Becausethere was little agreement between the Diagnostic Interview Schedule diagnosisand the psychiatrist’s ascertainment of schizophrenia, this analysisused an alternative definition based on the number of psychotic symptoms.There is a suggestion that this strategy yielded a categorization closer tothe clinician’s assessment.21 For details,see the original publication.7
In both prevalence studies, the deficit/nondeficit categorization wasmade using the PDS.
The Roscommon Family Study was a population registry–based familystudy of affective and psychotic disorders in Roscommon County, western Ireland,22 and therefore it approximated an incidence sample.Schizophrenia diagnoses used a best-estimate procedure and DSM-III-R23 criteria. The deficit /nondeficitcategorization was based on review of clinical information by 2 independentraters16; interrater reliability for this categorizationwas calculated in a subset of 12 subjects (92% agreement, κ = 0.82).Raters were blind as to the month of birth. A total of 137 subjects with schizophreniaand simple schizophrenia were included; in this sample, simple schizophreniaappeared to be genetically related to schizophrenia.24
Cases from the Camberwell Study were based on the Camberwell CumulativePsychiatry Case Register, which recorded contacts with psychiatric servicesin an area of southern London, England.25 Recordswere reviewed using the Operational Criteria Checklist for Psychotic Illness,and diagnoses were generated using a computerized algorithm (OPCRIT26). A previous publication on season of birth in thissample used Research Diagnostic Criteria8;data presented here used DSM-III diagnoses for schizophrenia.The PDS was used to distinguish deficit and nondeficit subjects.
This sample included all patients coming in contact with psychiatricservices in Dumfries and Galloway from 1979 to 1998 who had a diagnosis ofschizophrenia or schizoaffective disorder. The Operational Criteria Checklistfor Psychotic Illness was completed based on review of medical records foreach subject26; subjects had an OPCRIT DSM-IV diagnosis of schizophrenia. The PDS was used tocategorize subjects as having deficit or nondeficit schizophrenia.14
The Cantabria First Episode Schizophrenia Study27 includedall patients diagnosed as having a psychotic disorder in treatment facilitiesin the Autonomous Community of Cantabria in Spain. A Spanish translation ofthe Present State Examination28 was used togenerate a CATEGO diagnosis. Subjects with a diagnosis of schizophrenia orparanoid psychosis were included; DSM diagnoses werenot available. The PDS was used to distinguish deficit and nondeficit subjects.13
The Suffolk County Mental Health Project assessed patients with a firstepisode of psychotic symptoms in Suffolk County, New York, at the time offirst treatment and at 6- and 24-month follow-ups.29 Structuredrating scales and notes from interviews were reviewed. At least 2 of 3 raters(R.W.B., W.T.C., and B.K.) reviewed each subject’s information; if those2 raters did not agree on the categorization, the third rater reviewed theinformation and a consensus was reached. Raters were blind to month of birth.Of 182 subjects with a DSM diagnosis of schizophrenia,151 were categorized as nondeficit, 27 as deficit, and 4 were not categorizedowing to insufficient information. A previous publication of Suffolk Countybirth data was based on a partial sample.11
We used a goodness-of-fit χ2 test for the 2 × 12to compare the distribution of birth between deficit and nondeficit subjects.As previous studies showed a summer increase, the data suggested the effectwas restricted to a difference in June and July; the odds of June/July birthwere calculated for each sample, and a pooled estimate was produced in 2 stages:first by study design (samples of convenience, prevalence samples, and studysamples approximating incidence) and second by pooling all available estimates.Because of a lack of data, a time-series analysis was not possible.
Along with the odds ratio for summer birth (June/ July birth) and itsstandard error, a heterogeneity test was performed to assess differences amongthe studies. Meta-analytic estimates were calculated using a fixed-effectsmodel, and each estimate was entered in the calculation. A sensitivity analysiswas also performed in which we omitted each study in turn. We attempted tominimize publication bias by including all available studies, including unpublisheddata; however, it is still possible that studies reporting an associationhave been more likely to be published, leading to publication bias. Analyseswere performed using Stata version 7.0.30
A total of 1594 subjects were included in the 9 studies (407 in thesamples of convenience, 331 in the prevalence samples, and 856 in the incidencesamples).
The birth distribution of the samples of convenience studies is presentedin Figure 1. For these studies, thegoodness-of-fit test was not statistically significant (χ211 = 18.05, P = .08). Thetest for heterogeneity in these studies approached significance (χ22 = 5.53, P < .06).For June/July birth, the pooled estimate was not statistically significant(odds ratio, 1.59; 95% confidence interval, 0.93-2.74).
The birth distribution for the population-based prevalence studies ispresented in Figure 2. For these studies,the goodness-of-fit test was statistically significant (χ211 = 60.9, P < .001).The test for heterogeneity was not significant (χ21 = 0.23, P < .63), allowing the pooling of the samples.The pooled odds ratio for June/July birth from these 2 studies was 1.64 (95%confidence interval, 1.04-2.59).
The birth distribution for the population-based samples approximatingincidence is presented in Figure 3.For these studies, the goodness-of-fit test was statistically significant(χ211 = 42.95, P < .001).The test for heterogeneity was not significant (χ24 = 1.43, P < .84), allowing for the calculation ofa pooled estimate; for June/July birth, this estimate was statistically significant(odds ratio, 1.95; 95% confidence interval, 1.31-2.91).
The sensitivity analysis for all the studies combined, with the pooledestimate after omission of each study, showed a consistent pattern of association;that is, no single study, when omitted, rendered the association nonsignificant(data not shown). The pooled odds ratio estimate for June/July birth, includingall studies independent of study design, was statistically significant (oddsratio, 1.93; 95% confidence interval, 1.46-2.55) (Figure 4).
This pooled analysis of data from 6 countries in the northern hemisphereshowed a significant association between deficit schizophrenia and summerbirth. A similar pattern was found in studies of samples of convenience, population-basedprevalence samples, and population-based samples that approximated incidence.Information on month of birth only, as opposed to day of birth, was availableacross studies, and our analyses found an increase in June/July. However,it is likely that a more seasonal pattern would have been apparent with moredetailed information. In future studies of season of birth, it would be usefulto define the period of risk as June and July, if information on month ofbirth only is available.
Three different methods for categorizing patients into deficit and nondeficitgroups were used: administration of the SDS, medical record review, and thePDS. There was also variation in the diagnostic criteria for schizophrenia.However, any misclassifications or misdiagnoses due to this variance in methodsshould bias these studies toward a failure to find an association. The consistencyof the pattern across all of these data sets supports the existence of theassociation that we found.
All of the studies included in our analyses were conducted in the northernhemisphere. An unpublished multicenter prevalence study by McGrath and colleaguesfrom Australia failed to show an association between deficit schizophreniaand summer birth. These data are difficult to interpret because the winterbirth effect size is larger in sites further away from the equator31 and the latitude of these Australian sites is approximately28° to 32° south. In contrast, the approximate latitudes for the northernhemisphere sites we included in the present analysis range from 41° to55°. A study of deficit/nondeficit births from higher latitudes in thesouthern hemisphere would be of considerable interest.
Although survival bias (survival in the usual sense as well as survivalas a case) can lead to different results in incidence vs prevalence studies,the similarity in results between the prevalence studies and the studies approximatingincidence suggests survival bias did not have a substantial impact on theassociation between summer birth and deficit schizophrenia. In future studiesof the summer birth effect, population-based prevalence studies appear likelyto yield results that are not distorted by survival bias. The overlap betweenthe Nithsdale study and the Dumfries and Galloway study was a limitation ofour analysis, but this overlap was small, and examination of both samplesgave us an opportunity to examine the season of birth effect size in overlappingpopulations with prevalence vs incidence sampling.
From the present data, it is not possible to conclude whether the associationbetween deficit schizophrenia and summer birth reflects an increase in summerbirth in the deficit group compared with the general population or a decreasein summer birth in the nondeficit group. However, publications on these prevalencestudies have provided data that suggest the deficit group has an increasein summer births compared with the general population.7,14 Becauseschizophrenia as a whole is associated with winter birth and the smaller deficitgroup is robustly associated with summer birth, the nondeficit group appearsto be abnormal as well. The finding of a summer birth excess in deficit schizophreniadoes not contradict the winter birth excess1 becausethe deficit group represents a relatively small percentage (15%-20% of firstepisode patients) of schizophrenia. The winter birth effect is probably strongerthan previously thought because both deficit and nondeficit cases have beenincluded in previous studies of the winter birth effect, but the winter birtheffect appears to apply exclusively to nondeficit patients. It is importantto note that even with an association with summer birth, only a minority ofpatients with deficit schizophrenia is born in June/July. An age-incidencebias, which was proposed as an explanation for the winter birth effect butwas subsequently refuted,32 would not accountfor an association with birth in the middle of the year.
Our results support the concept of a double dissociation in deficitvs nondeficit schizophrenia and the risk factor of season of birth, with thedeficit group associated with summer birth and the nondeficit group with winterbirth. This difference strongly suggests differences in etiology between the2 groups. Data on family history and the prevalence of Borna disease virusseropositivity provide other evidence for etiological differences.8,16,33 Seasonal variationsin infectious agents,1 sunlight exposure andvitamin D,34 and the availability of nutrients3 have been proposed as possible explanations for theseasonality of births in schizophrenia. However, to date, no specific agenthas been identified. Nonetheless, summer birth may be a useful variable instudies of gene-environment interactions. More generally, etiological studiesof schizophrenia would benefit from making the deficit/nondeficit categorizationwhenever possible.
This epidemiological analysis adds to the evidence of clinical and neurobiologicaldifferences between deficit and nondeficit patients.4,5,11,29,35- 44 Aswe have argued elsewhere, this evidence, taken together, is consistent withthe proposition that deficit schizophrenia is a separate disease within thesyndrome of schizophrenia.9
Correspondence: Brian Kirkpatrick, MD, MarylandPsychiatric Research Center, PO Box 21247, Baltimore, MD 21228 (email@example.com).
Submitted for Publication: October 25, 2002;final revision received April 21, 2004; accepted April 26, 2004.
Funding/Support: This study was supported inpart by grants MH44801 (Dr Bromet), MH41953 (Dr Kendler), MH40279 (Dr Carpenter),and MH60487 (Dr Messias) from the National Institutes of Health, Rockville,Md.
Acknowledgments: James Tonascia, PhD, Departmentof Biostatistics, Bloomberg School of Public Health, Johns Hopkins University,Baltimore, Md, and Robert McMahon, PhD, Maryland Psychiatric Research Center,Baltimore, both provided statistical consultation.