Association of Air Pollution Exposure With Psychotic Experiences During Adolescence

Key Points Question Is exposure to air pollution associated with adolescent psychotic experiences? Findings In this nationally representative cohort study of 2232 UK-born children, significant associations were found between outdoor exposure to nitrogen dioxide, nitrogen oxides, and particulate matter and reports of psychotic experiences during adolescence. Moreover, nitrogen dioxide and nitrogen oxides together explained 60% of the association between urban residency and adolescent psychotic experiences. Meaning The association between urban residency and adolescent psychotic experiences is partly explained by the higher levels of outdoor air pollution in urban settings.

believed that you were being sent special messages through the television or radio, or that a programme has been arranged just for you alone? Have you ever thought you were being followed or spied on? Have you ever heard voices that other people cannot hear? Have you ever felt like you were under the control of some special power? Have you ever known what another person was thinking, like you could read their mind? Have you ever seen something or someone that other people could not see? The item choice was guided by the Dunedin Study's age-11 interview protocol 6 and an instrument prepared for the Avon Longitudinal Study of Parents and Children. 7 Interviewers coded each experience 0, 1, 2 indicating respectively "not present", "probably present", and "definitely present". A conservative approach was taken in designating a child's report as a symptom. First, the interviewer probed using standard prompts designed to discriminate between experiences that were plausible (e.g., "I was followed by a man after school") and potential symptoms (e.g., "I was followed by an angel who guards my spirit"), and wrote down the child's narrative description of the experience. Second, items and interviewer notes were assessed by a psychiatrist expert in schizophrenia, a psychologist expert in interviewing children, and a child and adolescent psychiatrist to verify the validity of the symptoms (but without consulting other data sources about the child or family). Third, because children were twins, experiences limited to the twin relationship (e.g., "My twin and I often know what each other are thinking") were coded as "not a symptom". Children were only designated as experiencing psychotic symptoms if they reported at least one definite, clinically-verified symptom. At age 12, 5.9% (N=125) of children reported experiencing psychotic symptoms (referred to as childhood psychotic symptoms).
The same items and clinical verification procedure was used when participants were interviewed at age 18, this time enquiring about psychotic symptoms they may have experienced since age 12. At age 18, 2.9% (N=59) of participants reported experiencing psychotic symptoms since age 12 that were clinically verified (referred to as adolescent psychotic symptoms). These rates are similar to those reported for community samples of children and adolescents in other studies using clinical verification procedures. 8,9 The comparatively low prevalence of psychotic symptoms at age 18 versus age 12 is also consistent with findings from other studies showing an attenuating rate of psychotic symptoms from childhood to adulthood. 10,11 Furthermore, psychotic symptoms in this cohort have previously been shown to have good construct validity, sharing many of the same genetic, social, neurodevelopmental, and behavioural risk factors and correlates as adult psychotic disorders. 12 To obtain a broader measure of adolescent psychotic experiences during the age 18 interviews, participants were asked six items about unusual feelings and thoughts in addition to the seven hallucination/delusion items. These items drew on item pools since formalised in prodromal psychosis screening instruments including the Prevention through Risk Identification, Management and Education (PRIME)-screen 13 and the Structured Interview for Psychosis-Risk Syndromes (SIPS). 14 These additional items included: I have become more sensitive to lights or sounds; I feel as though I can't trust anyone; I worry that my food may be poisoned; People or places I know seem different; I believe I have special abilities or powers beyond my natural talents; My thinking is unusual or frightening. Interviewers coded each of the 13 items (7 original hallucination/delusion items plus 6 additional unusual experiences items) 0, 1, 2, indicating respectively "not present", "probably present" and "definitely present". Responses to each of the 13 items (none, probable, definite) were summed to create a psychotic experiences scale (potential range=0-26, actual range=0-18, M=1.19, SD=2.58). The psychotic experiences measure did not involve clinical verification, meaning that this is a self-report measure capturing a broader range of mild, moderate and potentially clinically pertinent hallucinations, delusions, and other unusual feelings and thoughts. Since there were low numbers of adolescents with high psychotic experiences scores (e.g., only 1.0% [N=21] of participants had a psychotic experiences score of 13 or more), scores were placed into an ordinal scale to tackle the skewed distribution while retaining more information than a binary score. Just over 30% of participants had at least one psychotic experience between ages 12 and 18: 69.8% reported no psychotic experiences (coded 0; N=1,440), 15.5% reported 1 or 2 psychotic experiences (coded 1; N=319), 8.1% reported 3-5 psychotic experiences (coded 2: N=166), and 6.7% reported 6 or more psychotic experiences (coded 3: N=138). This 30.2% prevalence is similar to the prevalence of self-reported psychotic experiences in other community samples of teenagers and young adults. [15][16][17] Ambient air pollution. Pollution exposure estimates were linked to the latitude-longitude coordinates of participants' residential addresses (or where the participant spent most of their time) plus two additional addresses that the participants' reported spending their time in 2012, when the twins were aged 17 years. The most common locations were home, school, work, and shops, as described in the eTable 1. Pollution data for the primary addresses were available for 97.5% (N=2014) of the age-18 cohort (see eTable 1). Pollution estimates were modelled using CMAQ-urban, which is a coupled regional Chemical Transport model and street-scale dispersion model. CMAQ-urban uses a new generation of road traffic emissions inventory in the UK to model air quality down to individual streets, providing hourly estimates of pollutants at 20x20 metre grid points throughout the UK (i.e., address-level). Full details on the creation and validation of this model have been described previously. 18,19 Participants' exposure to several pollutants has been estimated by averaging the levels of the pollutant across the year at up to three locations that participants reported spending most of their time in, and then averaging this across the locations (i.e., [annual pollution exposure in Location 1 + Location 2 + Location 3] ÷ 3). Data for the proportion of time spent at each location were not available for participants, therefore the pollution measures are simply averaged rather than weighted. Pollutants include NO2 (nitrogen dioxide: regulated gaseous pollutant linked to road traffic and industrial activity), NOx (nitrogen oxides: measure of road traffic and industrial activity composed of NO2 and nitric oxide [NO]), and PM2.5 and PM10 (particulate matter with an aerodynamic diameter <2.5 µm and <10µm, respectively: regulated pollutants composed of inorganic aerosols, carbonaceous aerosols and dusts). Annualized average pollution levels in the E-Risk cohort (in micrograms per cubic metre [µg/m 3 ]) were, for NO2: M=20.18µg/m 3 , SD=9.68, range=2.31-67.89µg/m 3 ; for NOx: M=25.79µg/m 3 , SD=16.28, range=2.48-151.08µg/m 3 ; for PM2.5: M=11.24µg/m 3 , SD=2.18, range=4.05-20.56µg/m 3 ; and for PM10: M=16.40µg/m 3 , SD=2.71, range=8.42-33.27µg/m 3 . These levels are representative of the UK. 20 (See further statistical analysis of this dataset in the next section). In the E-Risk cohort, 4.5% (N=91) of participants were exposed to levels of NO2 that exceeded WHO guidelines (40µg/m 3 ); 29.7% (N=598) of participants were exposed to NOx levels that exceeded WHO guidelines (30µg/m 3 ); 80.8% (N=1,627) of participants were exposed to PM2.5 levels that exceeded WHO guidelines (10µg/m 3 ); and 9.3% (N=187) of participants were exposed to PM10 levels that exceeded WHO guidelines (20µg/m 3 ). Since there were substantial differences between air pollutants in terms of the numbers of participants who exceeded the WHO thresholds, WHO cut-offs were not used in the main analyses. Instead, to capture the worst levels of air pollution and create parity between the measures, air pollutants were dichotomized at the top quartile of exposure for this sample (these quartile cut-offs were: NO2=26.0 µg/m 3 ; NOx=33.0µg/m 3 ; PM2.5=12.4 µg/m 3 ; PM10=17.6 µg/m 3 ). Though all air pollutants were highly correlated (all r's=0.56-0.97, p's<0.001), we examined the associations of each pollutant with adolescent psychotic experiences in case of differential effects.
Urbanicity. Urbanicity was derived from the ONS's Rural-Urban Definition for Small Area Geographies (RUC2011) classifications. 21 The ONS classifications utilised 2011 census data, and were designed for application to small geostatistical units (e.g. Output Areas). Detailed information on how the ONS created the RUC2011 classifications of urbanicity is available on the ONS webpages (https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/239477/RUC11methodologypap eraug_28_Aug.pdf). Briefly, RUC2011 was created by laying a grid of hectare cells (100m 2 ) over England and Wales. Postcode addresses were assigned to cells, and residential densities were then calculated for increasing radii around each cell, providing each residential property with a density profile. This was combined with Output Area and contextual data, allowing each settlement to be assigned to one of ten urbanicity categories (Rural categories: sparse/non-sparse hamlets and isolated dwellings, sparse/non-sparse villages, sparse/nonsparse rural town and fringe; Urban categories: sparse/non-sparse city and town, and minor/major conurbations [conurbations are densely populated, large urban regions resulting from the expansion and coalescence of adjacent cities and towns]). ONS urbanicity scores were then assigned to every E-Risk family via the family's postcode when children were aged 5, 7, 10, 12 and 18. Given the low numbers within some rural categories, urbanicity was collapsed into three levels (1: "rural" = all rural categories [ Family-level factors. Family socioeconomic status (SES) was measured via a composite of parental income, education, and occupation when participants were aged 5. The latent variable was categorized into tertiles (i.e., low-, medium-, and high-SES). 22 Family psychiatric history and maternal psychotic symptoms were both assessed when participants were aged 12. In private interviews, the mother reported on her own mental health history and the mental health history of her biological mother, father, sisters, brothers, as well as the twins' biological father. 23,24 This was converted to the proportion of family members with a history of any psychiatric disorder (coded 0-1.0; M=0.37, SD=0.27). For maternal psychotic symptoms, mothers were interviewed using the Diagnostic Interview Schedule (DIS) 25 for DSM-IV 26 which provides a symptom count for characteristic symptoms of schizophrenia (e.g. hallucinations, delusions, anhedonia): 16.6% of mothers had at least one symptom of schizophrenia.
Adolescent substance use. Adolescent tobacco smoking, cannabis dependence and alcohol dependence were assessed during face-to-face interviews at age 18 using the DIS. 25 Smoking status was determined based on whether the participant reported ever having been a daily smoker (yes/no); 26.2% (N=541) of participants met this criterion. Cannabis and alcohol dependence was determined based on DSM-IV criteria. At age 18, 4.3% (N=89) of participants met criteria for cannabis dependence and 12.8% (N=263) met criteria for alcohol dependence.
Neighborhood deprivation. Neighborhood-level deprivation was constructed using A Classification of Residential Neighborhoods (ACORN), a geodemographic discriminator developed by CACI Information Services (http://www.caci.co.uk/). Detailed information about ACORN's classification of neighborhood-level deprivation has been provided previously. [28][29][30] Briefly, CACI utilized over 400 variables from 2001 census data for Great Britain (e.g., educational qualifications, unemployment, housing tenure) and CACI's consumer lifestyle database. Following hierarchical-cluster-analysis, CACI created five distinct and homogeneous ordinal groups ranging from "Wealthy Achiever" (coded 1) to "Hard Pressed" (coded 5) neighborhoods. Neighborhoodlevel deprivation scores for the E-Risk families were then created by identifying the ACORN classification for that family's postcode when children were aged 18. E-Risk families are representative of UK households across the spectrum of neighborhood-level deprivation: 27.0% of E-Risk families live in "wealthy achiever" neighborhoods compared to 25.3% of households nation-wide; 7.2% vs 11.6% live in "urban prosperity" neighborhoods; 26.8% vs 26.9% live in "comfortably off" neighborhoods; 13.2% vs 13.9% live in "moderate means" neighborhoods; and 25.8% vs 20.7% live in "hard-pressed" neighborhoods. 45,46 Neighborhood crime rates. Crime data in 2011 (the first year for which full street-level data was available), including information on the type of crime, date of occurrence, and approximate location, were accessed online as part of an open data sharing effort about crime and policing in England and Wales. Street-level crime data was extracted for each of the geospatial coordinates marking the family's home (for a full description see: https://data.police.uk/about/#location-anonymisation). Neighborhood crime rates were calculated by mapping a one-mile radius around each E-Risk Study participant's home and tallying the total number of crimes that occurred in the area each month (M=247, SD=274, range=1-1868). These monthly crime rates were calculated for 2011, and then collapsed into quartiles. This measure covers various forms of crime, including violent offenses (e.g., assaults), sexual offenses (e.g., rape), robberies, burglaries, theft, arson, and vandalism.
Neighborhood social conditions. Social conditions (i.e., social processes) were estimated via a postal survey sent in 2008 to residents living alongside E-Risk families when children were aged 12. 28,31 In Britain, a postcode area typically contains 15 households, with at most 100 households (e.g., large apartment block). This type of postcode-level resolution represents a marked advantage over many existing neighborhood studies in which much larger census tract or census block units of analysis are used. Our objective was to obtain multiple reporters (e.g., 2 or more) for each family's neighborhood (here defined to the street or apartment block level). Considering that the typical response rate for neighborhood surveys is approximately 30%, 32 questionnaires were sent to every household in the same postcode as the E-Risk families, excluding the E-Risk families themselves (addresses were identified from electoral roll records). The number of surveys sent per postcode ranged from 15 to 50 residences per neighborhood (M=18.96, SE=0.21). Excluding undelivered surveys (N=600), the overall response rate was 28.1% (5601/19926), similar to that previously found. 32 Survey respondents typically lived on the same street or within the same apartment block as the children in our study. Surveys were returned by an average of 5.18 (SD=2.73) respondents per neighborhood (range=0-18 respondents). There were at least three responses for 80% of neighborhoods and at least two responses from 95% of the neighborhoods (N=5,601 respondents). 31 Most respondents had lived in the neighborhood for more than 5 years (83%), and only 1% of respondents had lived in the neighborhood for less than 1 year. In the present study, analyses control for social cohesion and neighborhood disorder. Social cohesion 33 (5 items, each coded 0-4) was assessed by asking residents whether their neighbors shared values and trusted and got along with each other, etc. Neighborhood disorder 34 (14 items, each coded 0-2) was assessed by asking residents whether certain problems affected their neighborhood, including muggings, assaults, vandalism, graffiti and deliberate damage to property, etc. Items within each neighborhood characteristic scale were averaged to create summary scores from each of the 5601 resident respondents. Neighborhood characteristic scores for each E-Risk family were then created by averaging the summary scores of respondents within that family's neighborhood.

eTable 1. Types of Locations That Participants Reported Spending Most of Their Time at Age 18
Note: Cumulative pollution exposure estimates were derived by averaging the pollution estimates at the three locations that participants reported spending most of their time. Not all participants provided a second (N=1,899) and third (N=1,297) address, therefore cumulative pollution exposure estimates incorporated one or two addresses for these participants.

Location type
Top three locations Indicates association between annualized average levels of air pollutants and adolescent psychotic experiences among participants who did not move house between ages 12 and 18 Note: CI, confidence interval; OR, odds ratio. *p<0.05 **p<0.01 ***p<0.001. Model 1 -unadjusted association between air pollutants (annualized average of ambient air pollutants across top three locations that participants spent their time) and adolescent psychotic experiences, for the 71.4% of participants who did not move house between ages 12 and 18. Note: CI, confidence interval; OR, odds ratio. **p<0.01 Model 1 -association between air pollutants (annualized average of ambient air pollutants across top three locations that participants spent their time) and adolescent psychotic experiences, unadjusted for confounders but mutually adjusted for co-pollutants (that is, NOx and PM2.5 are simultaneously entered as covariates). Model 2 -adjusted for all individual-, family-, and neighborhood-level confounders simultaneously. Analyses conducted on participants with full data in Model 2: N=1,705. Analyses account for the non-independence of twin observations.