Association of Residential Mobility Over the Life Course With Nonaffective Psychosis in 1.4 Million Young People in Sweden

Key Points Question Is residential mobility over the life course associated with the subsequent risk of developing nonaffective psychotic disorders? Findings In this prospective cohort study of 1.4 million young people living in Sweden, followed up each year from birth to age 29 years, more frequent moves during childhood and adolescence were associated with increased risk for nonaffective psychosis, peaking between age 16 to 19 years, with moves longer than 50 km independently associated with greater risk. There was less evidence that moving in adulthood increased risk, except among those who moved 4 or more times. Meaning The risk of nonaffective psychosis in young people was associated with greater residential mobility during formative periods of childhood and adolescence, which is consistent with the possibility that the disruption of social networks, peer support, and identity formation are relevant to the etiology of psychosis.


eMethods.
eResults. eFigure 1. Positively skewed distribution of population density at birth in cohort eFigure 2. Predicted probability of non-affective psychosis by number of moves aged 20 years and older, by university attandence eTable 1. Cohort characteristics by missing data type eTable 2. Correlation between number and cumulative distance of moves at different age periods eTable 3. Effect modification of number of moves after 20 years old on non-affective psychosis risk, by university attendance eTable 4. Model fit comparisons for linear versus non-linear distance functions eTable 5. Risk of non-affective psychoses after mutual adjustment for moves in all periods This supplementary material has been provided by the authors to give readers additional information about their work.

Sample
From our initial cohort (N=1,472,446) we excluded 30,368 (2.1%) participants partially or wholly missing geographical data on residential location, and 1,695 (0.1%) participants missing covariate information, leaving a final sample of 1,440,383 participants with complete residential mobility data.

Further exposure variable information
All people are required to inform Statistics Sweden of any residential move by law. Residential moves can be detected in the Swedish Register of the Total Population (RTP) by comparing a participant's SAMS area from one year to the next. By law, people must notify Statistics Sweden when they move residences, and changes in SAMS area are recorded annually in the Register of the Total Population. This methodology makes it possible to detect moves between small areas (SAMS) across the life course, but does not capture multiple moves within a given year, or moves within the same SAMS area. For each participant in each age period (0-6, 7-15, 16-19, 20+ years), we also estimated cumulative distance moved by each participant (in kilometers [km). To do so, we constructed a matrix of the Euclidian distance between any two SAMS areas in ArcGis (v10.4), based on their centroid XY coordinates, giving rise to a look-up table of 84.6m possible inter-SAMS moves.

Covariate information
Parental information was obtained from linkage between the Multigenerational register and other registers of interest (i.e. RTP, National Patient Register). Maternal age was strongly correlated with paternal age (corr=0.69), but had less missing data (0.09% vs. 0.56%), and so was preferred. Parental history of severe mental illness [SMI] was coded as a binary variable for either biological parent ever having been diagnosed with non-affective psychosis, bipolar disorder with psychosis or bipolar disorder without psychosis since 1969, when the National Patient Register began (ICD-9 codes: 295.x, 296.x; ICD-10 codes: F20-31). Parental death before the participant's 16 th birthday was also defined as a binary variable, and included either the biological or adoptive parents. Population density (in people per square kilometer [ppkm2]) was estimated for all SAMS between 1982-1995 based on total population size divided by area. All participants were assigned their SAMS population density at birth. Due to substantial positive skew (eFigure 1), a continuous log transformed variable was used in analyses to control for this confounder. Participant educational attainment may have been associated with propensity to move during the follow-up period 31 and is known to be related to psychosis risk. 35 To control for this confounder, we obtained educational attainment upon leaving compulsory school at age 15/16 years from the National School Register (see MacCabe et al 35 for full details). In the year students turn 16 years old, they are typically examined on 16 compulsory subjects. Those examined on 7 or more subject areas are given an overall grade point average (GPA). GPAs greater than or equal to 10 are converted to an A to E classification, with GPAs less than 10 recorded as failing compulsory school. Here, we grouped participants into the following categories of compulsory school educational attainment: fail (which included those with a GPA<10, and those examined on less than 7 subjects), grades D & E, grade C (the modal grade),

Statistical analyses
We fitted discrete time proportional hazards models to estimate the effect of residential moves during childhood adolescence and early adulthood on risk of non-affective psychotic disorders Discrete time survival analysis was necessary because our exposure (number of times moved) was only made available by Statistics Sweden for discrete intervals (years; see above). To do so, we constructed our cohort as a person-period dataset on the attained-age scale, such that each year of attained age was modelled as a separate observation within participants, with the outcome of interest (diagnosis of non-affective psychosis) coded as 1 in the year of exit, or 0 otherwise. Given strong evidence of non-linear associations between age and psychosis risk, were modelled as time-fixed covariates (invariant across levels of attained age); number of moves during the follow-up period (16-19, 20-29 years) were modelled as time-varying covariates.

Missing data
Participants excluded due to missing data did not differ from included participants in terms of the proportion diagnosed with non-affective psychosis (see eTable 1). They were, however, more likely to be younger and have a foreign background than included participants. Participants excluded due to missing geographical data (30,368 (2.1%)) were also more likely to have a parental history of non-affective psychosis, be women, belong to the highest or lowest income quintiles and have an older mother at birth, but were less likely to have experienced death of a parent during childhood. Those missing covariate data (N=1,695; 0.1%) were more likely to have moved at least once between 0-6 and 7-15 years old and come from higher income quintiles, but less likely to have moved after 16 years than those included in the study; these differences were also reflected in cumulative distances moved.

Correlations between number of and distances moved within and between age periods
Correlations ranged from 0.02-0.32 for number of moves in one age period to another, to 0.04-0.20 for cumulative distance moved in one age period to another, and 0.39-0.54 for within-period correlations between number of moves and distance moved (see eTable 2).  -0.003* -0.010* 0.054* 0.537* 0.054* 0.041* 0.139* 1 1 y is years. Distance moved is cumulative distance moved in kilometers *p<0.01; due to large sample all correlations met this statistical significance threshold. Shaded correlations denote most pertinent comparisons to the results of this paper a Models Adjusted for age, quadratic age, sex, foreign background, family history of severe mental illness, parental death before 16 years, disposable income quintile, mother's age at participant birth, population density at birth, distance and number of moves (categorical) in previous age periods (except 0-6 years old) and number of moves in current period b Restricted to those moving once or never in each period