Risk factors and peripheral biomarkers for schizophrenia

This 'umbrella review' aimed to systematically appraise the meta-analyses of observational studies on risk factors and peripheral biomarkers for schizophrenia spectrum disorders.

In recent years, genetic contributions to schizophrenia have been identified, as part of a wider effort to understand the genetic contribution to a range of psychiatric disorders. However, genetic factors are only part of the story; environmental risk factors are also important.

There have been many observational studies exploring potential risk factors for schizophrenia, but previous reviews of these have not evaluated the epidemiological credibility of the evidence, or examined potential biases. A new article attempts to address this limitation.

Belbasis and colleagues conducted a systematic review of meta-analyses (what they call an umbrella review) on risk factors and peripheral biomarkers for schizophrenia and other psychotic disorders, assessing the potential for bias and identifying associations supported by the most robust epidemiological evidence.

The authors also systematically searched for Mendelian randomisation (MR) studies, which provide a stronger basis for causal inference as they are not subject to reverse causality and potentially offer protection against confounding.

This ‘umbrella review’ aimed to systematically appraise the meta-analyses of observational studies on risk factors and peripheral biomarkers for schizophrenia spectrum disorders.

Methods

PubMed was systematically searched from inception to 5 January 2017 to identify meta-analyses of observational studies of schizophrenia spectrum disorders in adults and either environmental (i.e., non-genetic) risk factors or peripheral biomarkers, or MR studies of schizophrenia spectrum disorders. For each meta-analysis the authors extracted relevant information (e.g., examined risk factors, number and study design of component studies, study-specific risk estimates). They also recoded whether the meta-analyses performed a quality assessment of component studies. For each MR study they also extracted relevant information (e.g., sample size, effect size, 95% confidence interval, P-value, genetic instrument).

Effect sizes were converted to a common metric (odds ratio) and the summary effect size for each meta-analysis estimated under both a fixed- and random-effects model. The 95% prediction interval was also calculated, which accounts for between-study heterogeneity and evaluates the uncertainty for the effect that would be expected in a new study exploring the same association. Between-study heterogeneity was quantified using the I² metric, and small study effects examined using Egger’s test. The excess significance test was also applied to determine whether there were more statistically significant studies than would be expected given the average power of studies in the meta-analysis.

To identify associations with robust evidence the authors applied the following criteria:

For evidence to be convincing >1,000 cases were required as well as strong statistical evidence (P-value < 10^-6 in a random-effects model), no evidence of small-study effects or excess significance bias, a 95% prediction interval excluding the null, and no substantial between-study heterogeneity (<50%)
Highly suggestive evidence required >1,000 cases, a P-value < 10^-6 in a random-effects model, and a statistically significant effect in the largest study
Suggestive evidence required >1,000 cases and a P-value <0.001 in a random-effects model
For associations with convincing or highly suggestive evidence, a sensitivity analysis including only evidence from prospective cohort studies was used to evaluate temporal relationship.

Results

Overall, the authors searched 3,499 articles and 41 met the eligibility criteria. These examined a total of 98 associations. These included 41 environmental risk factors, of which 11 were studied in at least 1,000 cases (stressful events during adulthood, Borna disease virus infection, general academic achievement, handedness, cannabis use, tobacco smoking, traumatic brain injury, obstetric complications, advanced paternal age, childhood adversity, and urbanicity). There were 57 studies of peripheral biomarkers, of which 12 were studied in at least 1,000 cases (serum BDNF, serum vitamin B₁₂, serum CRP, serum interleukin-6, serum antigliandin IgA and IgG, serum anti-TTG2 IgA, serum leptin, serum folate, serum TNF-alpha, serum morning cortisol, and plasma adiponectin).

Only 1 association (obstetric complication) met the criteria for convincing evidence
4 associations (stressful events during adulthood, cannabis use, childhood adversity, and serum folate) met the criteria for highly suggestive evidence
7 associations (Borna disease virus infection, tobacco smoking, advanced paternal age, urbanicity, serum BDNF, serum CRP, and serum interleukin-6) met the criteria for suggestive evidence
In the sensitivity analysis, the evidence for childhood adversity and cannabis use remained highly suggestive
The authors also identified five MR studies comprising six different analyses, which examined cannabis use, serum CRP and serum vitamin D. A protective effect was found for higher serum CRP, while the studies of cannabis use showed conflicting results.

Only 5 of the 98 factors identified (history of obstetric complications, childhood adversities, cannabis use, stressful events during adulthood, and serum folate level) showed robust evidence that they confer a higher risk for developing schizophrenia spectrum disorders.

Conclusions

In total, Belbasis and colleagues evaluated almost 100 associations between risk factors and peripheral biomarkers and psychotic disorders on the schizophrenia spectrum. Over two-thirds were supported by nominally statistically significant evidence, but most were supported by only weak evidence, either due to a small number of cases or a relatively large P-value (or both). Only a small number of associations were supported by convincing or highly suggestive evidence.

A potential limitation is that the authors only searched PubMed for relevant meta-analyses, and therefore may have missed studies in journals not indexed in PubMed. However, given the medical nature of the topic it’s likely that most or all relevant studies will have been captured. There also did not seem to be a concerted attempt to identify relevant unpublished studies, so the reviewed meta-analyses may not be all of those that exist. These limitations are worth bearing in mind.

One of the interesting aspects of this study is the inclusion of evidence from Mendelian randomisation studies; a method that is rapidly growing in popularity. Given the fast-moving nature of the field, the review inevitably missed some more recent studies, but it nevertheless highlighted the likely importance of this method in future attempts to identify modifiable risk factors for schizophrenia and other psychiatric disorders.

It is particularly noteworthy that for one risk factor (serum CRP) the evidence from observational studies (which was only suggestive) was in the opposite direction to that indicated by Mendelian randomisation. The question then is which approach to trust, when they provide opposite findings. Certainly observational studies are notoriously unreliable, and have produced many spurious associations in the past that do not reflect the genuine operation of a causal risk factor.

Time will tell, but if this review highlights one thing it is the need for alternative approaches to identifying causal risk factors for schizophrenia that are potentially modifiable. When tested against stringent criteria for robust evidence, only one met the criteria for convincing evidence. The list of potential risk factors (i.e., those for which there was highly suggestive or suggestive evidence) is considerably longer, but observational studies will ultimately only take us so far.

When it comes to identifying causal risk factors for any health condition, observational studies will ultimately only take us so far.

The Mental Elf