Previous literature

There were already grounds for suspecting this conclusion. In their meta-analysis Geddes & Lawrie (Reference Geddes and Lawrie1995) noted significant heterogeneity of findings with respect to study design. The literature as a whole suggested an odds ratio of two for exposure to ‘obstetric complications’ on subsequent development of schizophrenia, but this depended upon the preponderance of studies that had used a retrospective case-control design. In the words of an early proponent (Reference LewisLewis, 1989) of the birth complications hypothesis:

The possibility that the mother of a child who has developed psychosis will be biased with respect to the events of the pregnancy when asked to recall them, or even before this, is a real one. In addition, Geddes & Lawrie (Reference Geddes and Lawrie1995) noted evidence (in a funnel plot) of a publication bias in favour of positive findings. The two cohort studies (Reference Done, Johnstone and FrithDone et al, 1991; Reference Buka, Tsuang and LipsittBuka et al, 1993) that avoided the recall bias gave an odds ratio of one, with the implication that obstetric complications and the development of schizophrenia are simply unrelated.

Meta-analysis is no defence against such biases. Geddes et al (Reference Geddes, Verdoux and Takei1999) re-analysed the data from 12 studies, seven of which depended upon maternal recall, and five of which, including the original study of Kendell et al (Reference Kendell, Juszczak and Cole1996), relied upon retrieval of obstetric records, but omitted the cohort studies. This time the pooled odds ratio was 1.4. The estimate reflects not only an aggregated bias (the ‘Lewis effect’) across maternal recall studies but, as Kendell et al (Reference Kendell, McInneny and Juszczak2000, this issue) now make clear, problems in the selection of the controls in retrospective comparisons of obstetric records.

The British Perinatal Mortality Survey recorded systematic data on the antenatal courses and births of 16980 individuals born in the week 3-9 March 1958.

Those who had developed schizophrenia by either broad or narrow criteria by the age of 28 years were not more likely than the cohort as a whole to have experienced events predicting mortality in the perinatal period (Reference Done, Johnstone and FrithDone et al, 1991). Nor were those 945 mothers who were recorded as suffering from influenza in the second trimester in the autumn of 1957 more likely to have children who later developed schizophrenia (Reference Crow and DoneCrow & Done, 1992; Crow, Reference Crow1997b ,Reference Crow c ). A subsequent analysis of the many variables recorded in this cohort (Reference Sacker, Done and CrowSacker et al, 1995) suggested that those indices that distinguished patients from controls related more to the characteristics of the mother (for example she more often had psychosocial problems, more often weighed less than 51 kg, and had fewer antenatal attendances) than to complications of the birth itself. Low birth weight (<2500 g) was more frequent in the patient group, as it was also in the recent cohort study of Jones et al (Reference Jones, Rantakallio and Hartikainen1998) who, as noted by Kendell et al (Reference Kendell, McInneny and Juszczak2000, this issue), concluded that the differences they found “appeared to be due largely to the characteristics of the child, not the delivery”.

Core problem of aetiology

The cohort studies, consistent with the conclusions of Kendell et al (Reference Kendell, Juszczak and Cole1996), thus give no support to the hypothesis that obstetric complications in general or any complication in particular is a cause of later schizophrenic illness. The importance of this conclusion is that it raises the question of whether there are any environmental contributions to the aetiology of schizophrenia. Only two — birth complications and prenatal exposure to influenza — have been seriously considered in the recent literature and both are now cast in considerable doubt.

This clears the air and brings the essential peculiarity of psychotic illness into focus — we are confronted with a phenomenon that is intrinsic, and therefore one might suppose ‘genetic’ in origin, and apparently universal in human populations (Reference Jablensky, Sartorius and ErnbergJablensky et al, 1992) but that yet is associated with a substantial biological (fecundity) disadvantage. In what sense is this a disease? What sort of genes are these? Why are they not rapidly selected out of the population?

A clue to the answer lies in the nature of the brain changes. The three best established — an increase in ventricular size, a modest reduction in cortical mass and diminution or reversal of the asymmetrical torque across the antero-posterior axis (Reference Bilder, Wu and DegreefBilder et al, 1994; Reference DeLisi, Sakuma and KushnerDeLisi et al, 1997) — do not identify sub-populations, and must clearly be related. They cannot be attributed to birth injury or other environmental insult. They are present early in illness course, are ‘developmental’ in origin and reflect the nature of the genetic variation. Whereas it is often argued, particularly in the light of the apparent failure of the genetic linkage approach (Reference DeLisi and CrowDeLisi & Crow, 1999), that psychosis is the outcome of many genes of small effect, the morphology of the brain, as yet ill understood, suggests otherwise.

The brain changes and their independence of environmental influence, along with universal incidence and persistence against a biological disadvantage (the ‘central paradox’), indicate a homogeneity to the phenomena of psychosis that is close to the core characteristics of the species. The genetic variation therefore may differ from that associated with other ‘disease-related’ genes, for example, cystic fibrosis and Huntington's disease. One can ask is variation (apparently relevant to survival) that persists across populations independent of the environment the same as that which results from random mutation and inheritance within families according to Mendelian rules? This provokes the parallel question of whether variation within a species is of the same nature as that which separates species, a question that relates to the nature of species transitions.

Perhaps only when we are free from the concept of ‘schizophrenia’ as a categorical disease entity with multiple environmental causes yet to be discovered can we appreciate the true significance of psychosis as a component of the variation that identifies the species (Reference CrowCrow, 1998a ). Such an insight helps us to understand the nature of the symptoms. The answer that I have given to the question first raised by Huxley et al (Reference Huxley, Mayr and Osmond1964) of what the genetic advantage that balances the disadvantage associated with schizophrenia is is language. The capacity for language defines the species (Reference BickertonBickerton, 1995), it arose some time between 100 000 and 150 000 years ago, and was the result of a relatively discrete genetic change that occurred in a population in east Africa (Reference Stringer and McKieStringer & McKie, 1996). This event, perhaps because it introduced a differentiation of function of the hemispheres, allowed language to evolve.

Schizophrenia, according to this view, is the price that Homo sapiens pays for language (Reference CrowCrow, 1997a ). It follows that schizophrenic symptoms can be understood as disorders of language, and more fundamentally as the key to the cerebral organisation of language. Nuclear symptoms can be understood as anomalies of the transition from thought to speech, and as an indication that a mechanism of ‘indexicality’ that distinguishes between what is self- and other-generated is a necessary component of the bihemispheric organisation of language (Reference CrowCrow, 1998b ). Delusions can be understood as disorders of the transition from speech to meaning, that is as semantic deviations. The fact that nuclear symptoms and other types of delusion are so frequently associated, and that both these classes of psychotic phenomena are sometimes associated with ‘thought disorder’, indicates that the neural bases of syntax and semantics are not independent.

The reason why the disadvantageous ‘genotypes’ associated with psychotic symptoms are not selected out must be that the variation is inherent in the genetic mechanism, the characteristics of which reflect an origin in the transition (through a ‘speciation event’) from a precursor hominid to modern Homo sapiens (Crow, Reference Crow1998c ,Reference Crow d ). It has been suggested that epigenetic modification (e.g. methylation) is particularly associated with species transitions (Reference Vrana, Guan and TilghmanVrana et al, 1998). Therefore, an alternative to the view that the inherited variation is polygenic and Mendelian is that it is speciation-related and ‘epigenetic’. A recent twin study can be interpreted as consistent with this concept (Reference CrowCrow, 1999).

Why it matters

Why I think a resolution of the question of obstetric complications is important is that so long as it is believed that there are significant environmental causes of schizophrenic illness the central problem posed by the phenomena of psychosis is obscured. The singularity of the intrinsic variation and its wider significance for an understanding of language and the nature of speciation events is unappreciated. If as Kendell et al conclude “ the evidence that schizophrenia is associated with a raised incidence of obstetric complications is weaker than has been assumed”, some undergrowth that stands in the way of an understanding of the nature of the problem has been cleared away. As Darwin remarked: “False facts are highly injurious to the progress of science for they often endure long; but false views… do little harm, as everyone takes a salutary pleasure in proving their falseness, and when this is done, one path toward error is closed and the road to truth is often at the same time opened” (Reference DarwinDarwin, 1871).