Customize your JAMA Network experience by selecting one or more topics from the list below.
The association between residence in an urbanized area and risk for schizophrenia is among the oldest findings in psychiatric epidemiology,1 has been replicated many times, and is likely to be at least in part causal.2 However, little is known about its mediators, that is, through what mechanism(s) does living in cities predispose to psychotic illness?
The study of mediation for epidemiologic risk factors is important for 2 reasons. First, it can provide deeper scientific understanding of the causes of the association. Second, it can identify points of potential intervention in the causal chain.
A large literature supports the association between low cognitive ability and risk for schizophrenia and suggests that much of that association is causal.3 The question asked by Lewis et al,4 the degree to which the association of urban living with risk for schizophrenia might be mediated through lower cognitive ability, is an important and plausible one that has not, as they note, to their knowledge been previously explored. Furthermore, their work provides a good example of how, to elucidate epidemiological findings, psychiatric research often needs to move to cross-level analyses, in this case from psychosocial risk factors to cognition.
Working in a large, representative male Swedish population cohort, Lewis et al4 studied an expanded version of schizophrenia: nonaffective psychosis (NAP). Cognitive performance was assessed using high-quality measures of IQ from Swedish conscription examinations. They found, as expected, a robust association between IQ and subsequent risk for NAP. They explored, in a prospective cohort design, 2 potential epidemiologic associations of urbanization assessed at birth: population density and neighborhood deprivation (ND). No appreciable association was found between population density and IQ; therefore, they focused the remainder of their mediational analyses on ND.
In their main analyses, they calculate that the odds ratio (OR) between ND and NAP was 1.08, which they decomposed into a direct effect (OR = 1.06) and an indirect effect mediated through IQ: 1.02. From this, they estimated that 23% of the association of ND with NAP occurs through the association of ND with IQ. However, the authors, to their credit, note that this calculation assumes that the effect of ND on NAP is entirely causal. The plausibility of this assumption is questionable. Families are not randomly selected into neighborhoods of low or high deprivation,5 and parental cognitive potential is likely one of the factors influencing such selection. This is suggested by the strong negative association between ND and parental education in their sample. Molecular epidemiologic findings support this supposition by showing risk variants for low educational attainment clustering in historically deprived regions.6
The authors then propose a strong test to assess the effect of such confounding: controlling for paternal IQ (maternal IQ was not available). As seen in the Figure, it is quite likely that paternal IQ will be associated with ND, offspring IQ, and offspring risk for NAP, given the familial nature of IQ7 and the strong inverse association between IQ and schizophrenia risk.3 Controlling for paternal IQ does not change the total association between ND and NAP, but the indirect and direct effects are now 1.01 and 1.07, respectively. However, this small change in ORs produces a large effect on the estimated proportion of the ND mediated through IQ, which now falls to 9%. This lower estimate of the mediational effect is, in our judgment, likely to be more plausible than that obtained in their primary analysis because it is based on a more realistic model of the ND-NAP association that includes a key likely confounder.
A conceptual path model illustrating the analytic approach used by Lewis et al4 to test the mediation by IQ of the effect of neighborhood deprivation on risk for nonaffective psychosis in the presence of potential confounding by paternal IQ. This model depicts a direct path from neighborhood deprivation to nonaffective psychosis and a pathway mediated by IQ as well as potential paths from paternal IQ to neighborhood deprivation, proband IQ, and nonaffective psychosis. Of note, in their multilevel model, deprivation was assessed at the level of the community, while IQ and nonaffective psychosis were assessed at the level of the individual. This model resembles supplemental eFigure 3 presented by Lewis et al.4
Examining these ORs points out another possible methodologic concern. Decomposing weak associations (here into mediated and unmediated components) is challenging, especially for outcomes as rare as NAP and for associations as modest as those between ND and NAP. Their large sample certainly helps but does not entirely alleviate this problem. We see this most clearly in the ORs where controlling for familial confounding shrinks the indirect effect of ND mediated through IQ 50% (from 1.02 to 1.01) and the percentage of the effect that is mediated by IQ by two-thirds (from 27% to 9%). From a methodologic perspective, the statistical stability of such results is of some concern as suggested by the broad confidence intervals.
As noted by the authors, controlling for confounding in multilevel models is a complex task8 because it differs from single-level model mediation.9 That is, the authors are looking at mediation of a between neighborhood effect (ND) by an individual level variable (IQ) on an individual level outcome (NAP). Because ND varies only between neighborhoods, all individuals within the same neighborhood are exposed to the same ND. Therefore, any effect of ND on NAP, mediated or not, can exist only between neighborhoods. This analytic approach is more likely to overestimate than underestimate the mediational effect of IQ.
The authors were cautious in their interpretations, noting the problems of causal inference in their mediational model. To control for possible reverse causality (early psychosis affecting IQ), they excluded cases with psychosis onset prior to the IQ tests and also controlled for a possible prodrome effect by excluding cases diagnosed within 2 years of testing.
In summary, this study4 has a number of methodologic and conceptual strengths. The sample is large and representative. The measurement of the 3 key constructs, ND, IQ, and NAP, are strong. The goal of clarifying causal pathways for classic epidemiologic risk factors is a laudable one. Some caution is needed in interpreting their model, which moves from a community-level risk to an individual-level moderator and outcome. While likely present, the magnitude of the mediational effects of IQ on the ND-NAP association is probably quite modest, and even in this large sample, imprecisely known.
Corresponding Author: Kenneth S. Kendler, MD, Virginia Institute for Psychiatric and Behavioral Genetics of VCU, PO Box 980126, Richmond, VA 23298-0126 (firstname.lastname@example.org).
Published Online: March 11, 2020. doi:10.1001/jamapsychiatry.2020.0010
Conflict of Interest Disclosures: None reported.
Identify all potential conflicts of interest that might be relevant to your comment.
Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.
Err on the side of full disclosure.
If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.
Not all submitted comments are published. Please see our commenting policy for details.
Kendler KS, Ohlsson H. A Multilevel Mediational Model of the Association of Neighborhood Deprivation With Risk for Nonaffective Psychosis. JAMA Psychiatry. Published online March 11, 2020. doi:10.1001/jamapsychiatry.2020.0010
Coronavirus Resource Center
Create a personal account or sign in to: