P27 Can we better capture longitudinal exposure to the neighbourhood environment? a latent class growth analysis of the obesogenic environment in new york city, 1990–2010

• Published in: Journal of epidemiology and community health (1979) 2017-09, Vol.71 (Suppl 1), p.A63
• Main Authors: Berger, N, Kaufman, TK, Bader, MDM, Sheehan, DM, Mooney, SJ, Neckerman, KM, Rundle, AG, Lovasi, GS
• Format: Article
• Language: English
• Subjects: Census; Censuses; Classification; Exposure; Food; Income; Latent class analysis; Nets; Obesity; Population; Population number; Sensitivity analysis; Statistical analysis; Studies
• ISSN: 0143-005X; 1470-2738
• DOI: 10.1136/jech-2017-SSMAbstracts.129

BackgroundThe growing availability of (non-)commercial historical datasets opens a new avenue of research on how long-term exposure to the neighbourhood environment affects health. However, traditional tools for longitudinal analysis (e.g. mixed models) are limited in their ability to operationalise long-term exposure. This study aims to summarise longitudinal exposure to the neighbourhood using latent class growth analysis (LCGA). Using the National Establishment Time-Series (NETS) 1990–2010, we analysed the trajectory of change in New York City (NYC) in the number of unhealthy food businesses – a potential indicator of an obesogenic environment.MethodsThe NETS is a commercial dataset providing retail business information in the United States. NYC data were acquired for the period 1990–2010. Businesses were grouped into researcher-defined categories based on Standard Industrial Classification codes and other fields such as business name. All businesses were re-geocoded to ensure accurate localisation. We defined access to BMI-unhealthy businesses (characterised as selling calorie-dense foods such as pizza and pastries) as the total number of BMI-unhealthy businesses present in each NYC census tract (n=2,167) in January of each year. We conducted LCGA in Mplus to identify census tracts with similar trajectories of BMI-unhealthy businesses. We used model fit statistics and interpretability to determine the number of classes. Using the final models, we assigned census tracts to latent classes. We predicted class membership with socio-demographic variables from the Census (population size, income, and ethnic composition) using multinomial logistic regressions and reported predicted probabilities with 95% CI. Sensitivity analyses were undertaken.ResultsThe final models include 5 and 10 latent classes, respectively. The 5-class solution indicates an overall increase in the number of BMI-unhealthy businesses over time and shows a pattern of fanning out: the higher the value in 1990, the greater the increase over time. Classes are associated with 1990 population size, income, proportion of Black residents (all p