Schizophrenia, catatonic schizophrenia and poliovirus – a role for natural experiments?

This is the authors' abstract. We don't add key points for this paper.

Precisely characterising the distribution of schizophrenia incidence across temporal, spatial and cultural contexts is now a key aim of psychiatric epidemiology, and the dogma of uniform incidence and prevalence has been successfully countered.1 Yet the exact nature and meaning of temporal patterns in schizophrenia incidence has remained elusive. In principle, secular trends in disease incidence are relevant for understanding aetiology and for the planning of services. Research estimates on the extent and direction of temporal change in schizophrenia incidence may be inconsistent, and the existing evidence on trends in schizophrenia incidence over time is mixed, with a recent review indicating insufficient evidence for statistically significant temporal change.2 Tanskanen and colleagues’ paper in this Journal3 reflects the intersection of a focus on the possibly changing incidence of schizophrenia and its catatonic subtype, and the aetiological contribution of prenatal and life course infectious disease to cases of schizophrenia occurrence. This is an increasingly pertinent topic in psychiatry, and a matter of current research interest.4-6 Finnish national register data are used to analyse associations of both poliovirus and measles incidence with the incidence rate of schizophrenia, and of catatonic subtype of schizophrenia specifically, between 1972 and 2014. The authors report a reduction of around 20% in the age-specific incidence of all schizophrenia over this period of follow-up, with a more marked decline in the incidence of catatonic schizophrenia subtype specifically. Statistically significant associations of catatonic schizophrenia with polio were evident in all age groups, but not with measles. This study has important strengths. The sample is sufficiently large, and likely to be representative of the entire population of Finns diagnosed with schizophrenia. The availability of linked schizophrenia register data to non-psychiatric hospital data allows measurement of polio and measles alongside that of schizophrenia. Given consensus that some degree of heterogeneity characterises most schizophrenia samples, examining trends in subtypes (whether defined diagnostically as is done by Tanskanen et al., or based on symptom domains) can plausibly generate explanatory hypotheses for changes in incidence. The findings of this paper are relevant for developing research in psychiatry. Firstly, the finding of reduced overall incidence of schizophrenia is consistent with some previous findings.7 Given that reduction in incidence of catatonia subtype of schizophrenia is more marked, it is plausible that reduction in incidence of this schizophrenia subtype accounts for the reduction in schizophrenia incidence overall observed in previous research. Tanskanen et al.’s paper also contributes further evidence consistent with some causal impact of polio on the occurrence of catatonic schizophrenia. Widespread polio vaccination in European countries was accompanied by a decline incidence of schizophrenia. Polio incidence peaks in late summer/early autumn, consistent with the season of birth association seen in schizophrenia, assuming a second trimester infection.8 Using a somewhat different design, the same team found that second trimester exposure to polio may increase risk for schizophrenia, in models which adjusted for place of birth as well as sex. Suvisaari and colleagues found association between polio incidence and birth of individuals later diagnosed with schizophrenia 5 months later (consistent with a second trimester effect9). An infectious aetiology for individual cases of schizophrenia is consistent with neurodevelopmental models of schizophrenia causation, and supported by evidence on the impact of obstetric complications and prenatal physiological insult/stress on psychosis risk.10 An implication of this link is that policies which impact these factors should be expected to impact incidence (in those age groups born during the period of changing prenatal risk factors). Tanskanen and colleagues cautiously state that the relationship which they observe between the stark decline in the catatonic subtype of schizophrenia after polio eradication implies a causal relationship. Such a conclusion rests, to some extent, on knowledge of the policy context of the time, including in the prevalence of relevant risk factors. In a socio-cultural context of increasing economic development, and possibly reducing disadvantage, it may be difficult to firmly conclude this. Catatonia is also an outcome of some clinical heterogeneity.11 Catatonia as a syndrome presents in the context of a wide array of diagnoses. Across these presentations, symptom measurement studies indicate considerable clinical differences among different catatonic presentations. There is a significant change in catatonic symptoms over time, which is often disregarded by researchers. In the context of Tanskanen et al.’s work, it is possible that assessing the impact of polio on catatonia should be treated with caution, based on the conditioning on the diagnosis of schizophrenia in the design. For example, it is possible that examining association of polio incidence with catatonia occurring in all diagnostic contexts (not just in schizophrenia) would have delivered different results. Allied to this, investigators should consider measurement accuracy a key component of study validity. Assessing trends in incidence in relation to risk factors implies invariance of ascertainment (measurement accuracy) over time and between regions. This assumption is rarely researched directly and tends to be mentioned as a study limitation. Projects to validate diagnostic categories testing for spatial and temporal differences in performance could generate greater understanding by contextualising results of register-based epidemiological studies. This is particularly relevant as diagnostic practices around psychosis and service models change over time.12 Correlations between trends (as presented in Tanskanen et al.’s paper) can present challenges for interpretation, but also opportunities. For example, it is unclear whether estimating correlations in this way can take adequate account of other time-changing factors, including changes in policy, distribution of risk factors such as adversity, socioeconomic disadvantage, and cannabis use. Natural experimental methods may offer one alternative design to assess the causal impact of polio on schizophrenia, alongside other methods. Finding causes using observational data is hard, including in circumstances where exposures may be straightforward to conceptualise and measure. This is predominantly because of the counterfactual paradox—that, in order to attribute causality to an exposure, one must be able to observe the case in the absence of that exposure. Randomised controlled trials make use of probability theory to generate results which, over many repetitions of the design, ensure that all common causes of both exposure and outcome are equally distributed between exposed and unexposed groups. Causal inference methods in epidemiology, and also some methods which do not take an explicitly causal inference approach, can be considered to aim at recovering estimates generated by a randomised study (the so-called ‘target trial’ concept13). Many observational studies in epidemiology are strongly affected by residual confounding, owing to incomplete/inaccurate measurement of the confounders we know about, and absence of information on confounders we don't know about. Natural experiments refer to instances where policy changes, such as vaccinations, changes in alcohol or drug policy, or health promotion programmes, are presumed to change exposure status within a population of interest, allow examination of the impact of this exposure on an outcome. Contrasting outcome rates in exposed and unexposed groups before, and then after, some specified time point, can account for factors which are invariant over time, delivering a so-called ‘difference in differences’ design.14 Causal inference in this framework must assume that groups are comparable be prima facie comparable, and that outcome trends in unexposed group(s) in the ‘post’ period is a good counterfactual for the exposed group had they not been exposed, termed the parallel trend assumption. In essence, then, difference in differences designs test for interaction between the exposure status and pre-post condition, and under assumptions described above, providing intention to treat estimates at a policy level (in this case, measures to eradicate polio). So, this method should not be used to attribute polio infection in an individual to the onset of schizophrenia, but may clarify the impact of policy to influence polio incidence, on schizophrenia incidence at an aggregated level. What are the implications of Tanskanen's results? Polio is unlikely to return, but other enteroviruses may. Policies to control the distribution of infections can impact the subsequent occurrence of multi-system disorders. Knowledge of the historical context for policy changes (eg other changes alongside eradication of polio) is necessary for robust interpretation of these results. Theory-guided examination of schizophrenia rates, including of subtype rates, in relation to policy change is possible and probably under-utilised method in understanding the incidence of schizophrenia. For example, reduction in the occurrence of infectious disease could underlie reduction in schizophrenia incidence over time. Improvement in obstetric care over time could also underlie reduction in incidence. Researchers should remain attentive to secular changes in policies and risk factors which can impact incidence. National register studies are relevant to aetiological research in psychiatry, but should be understood in the context of much richer information on changing diagnostic practices over time, and changes in the development and delivery of mental health services. Assessment of risk factors, such as migration, cannabis use, abuse, and obstetric complications and schizophrenia, should be interpreted in the context of policy impact on these risk factors and could diversify recommendations made on the basis of research. Greater understanding of distinct patterns of incidence of schizophrenia, and its subtypes, has possibly struggled to deliver on improved specificity of services and policy responses. Greater focus on explicit causal assessment of policy interventions could provide stronger evidence on how public health responses, at different levels, may influence schizophrenia risk, in line with Rose's distinction between causes of incidence and causes of cases.15 While real change over time in the incidence of schizophrenia could strongly impact psychiatry, schizophrenia is not disappearing and this calls for renewed attention to causes at different levels. The authors declare no conflicts of interest in relation to this article. The peer review history for this article is available at https://publons.com/publon/10.1111/acps.13311. The peer review history for this article is available at https://publons.com/publon/10.1111/acps.13311.