(434b) Sensor Data Analysis for Environmental Exposure Assessment

Authors: 
Chapizanis, D., Aristotle University of Thessaloniki
Sarigiannis, D., Aristotle University of Thessaloniki
Handakas, E., Aristotle University of Thessaloniki
Kontoroupis, P., Aristotle University of Thessaloniki
Karakitsios, S., Aristotle University of Thessaloniki

The exposome represents the totality of exposures from conception onwards, simultaneously identifying, characterizing and quantifying the exogenous and endogenous exposures and modifiable risk factors that predispose to and predict diseases throughout a person’s life span. Unraveling the exposome implies that both environmental exposures and genetic variation are reliably measured simultaneously. The spread of smartphone applications and fitness monitors provides new and less expensive methods for tracking time-location-activity patterns in exposure studies. Small sensor devices and smartphone apps can help exposure assessors to determine both location and activity of individuals with greater ease. As part of the Health and Environment-wide Associations based on Large population Surveys (HEALS) research project, this study examines the feasibility of using a series of sensors for tracking personal location and activities. This is the first step of the development of a methodological approach to estimate the exposome, which encompasses the totality of human environmental exposures at an individual level.

25 participants in the city of Thessaloniki, Greece during a summer week, wore a series of devices such as

(a) a temperature logger to detect changes between indoor and outdoor conditions, (b) a commercially available fitness monitor to capture motion and intense of activity, (c) a GPS device to track location and speed along with (d) Moves, a smartphone application that enables tracking of location and activity. Additionally, a time activity diary was filled out on paper by participants each day. Location, motion and intensity of activity data were used as input to an Artificial Neural Network (ANN) model, aiming at deriving a time-activity model based solely on sensor data. The independent variables that fed the ANN input layer were consisted of a) personal temperature, Temp, derived from the wearable temperature sensor, b) the change of temperature with time, dTemp/dt, c) personal speed, derived from the GPS devices wore by the participants, Speed, e) the observed temperature, derived from a meteorological station located in the historical centre of Thessaloniki, Tempout, d) and the ratio of the personal temperature to the observed one, Temp/Tempout. Moreover, information on day light - whether it is day or night based on time - was transformed into a categorical element (day or night) which was also included as an input variable. The initial database was divided into training and validation set (85% and 15% of the total record entries, respectively) and the models developed from the training set were tested using the validation set.

Using a Monte Carlo analysis, distributions of participant’s movement and activities - derived from the sensors experiments - were extrapolated to a larger population. The final distribution of a representative sample helped us to define the way with which people are moving in time and space (what time they start/finish work/school, their speed) as well as their different types of activity (sleeping, working, resting etc.) within the boundaries of a city. This was valuable information that was then translated into moving agents inside the GAMA platform.

By importing GIS files on the GAMA platform, we projected a simplified neighbourhood of Thessaloniki with its road (shapefile consisted of lines) and buildings (shapefile consisted of polygons) network as well as information on land use (nature of buildings: residential, office, park etc.). In this example land use information is arbitrarily defined by the modeller. The main advantage of geographical data agentification is the possibility to manage geographical objects, such as buildings, roads and people, inside the simulation by giving them a dynamic: an internal state and a behaviour. For the purposes of this paper and from now on, the term “agents” is only referring to people agents.

Agents are clustered in three age groups (children (n = 60), adults (n = 100) and elderly (n = 60)). When the model is initialized, agents are randomly distributed/allocated to a residential place which will serve as their house for the whole simulation. Following the same approach, children and adults are arbitrarily assigned to a school or an office respectively. Depending on their age, agents are programmed to move (walk on the road) from a household to either their office or their school whereas agents that belong to the elderly group stay at home. The model captures a representative day with a step of 10 minutes (one day is composed of 144 simulation cycles). Useful information on agent’s routine was derived by the sensors experiments. The agents’ trajectories, derived by the coded routine, are captured as points (1 point captured in every cycle) and are exported as a GIS shapefile together with a database that contains the agents’ coordinates as they move in time through different locations/microenvironments. The main assumptions taken into account during the development of the ABM model are highlighted below:

  • The population of the model is arbitrarily defined by the modeller. In the near future such kind of information will be based on official censuses data.
  • Simplified neighborhoud of Thessaloniki: land use information is randomly defined by the modeller. In the near future, land use information will be exported by official sources so that there is an accurate representation of the urban landscape.
  • Random distribution of agents – random allocation to households and random assignement to an office/school. In the near future such kind of information will be based on official census data.
  • Agents choose the shortest path to reach their target (office, school, their home). In the future, the model will include spatially explicit information on habitat, roads, topography, local resources, and the effects these have on individual behavior.

The activity of an individual participant, “agent”, was projected on a single layer, superposed onto urban air quality maps. These maps are based on daily average outdoor concentration data of particulate matter at various aerodynamic diameters (PM10, PM2.5 and PM1), measured and modelled for the city ofThessaloniki, taking into account traffic and non-traffic sources for the same period of time. In addition, indoor, outdoor and in-transit concentration data of particulate matter at various aerodynamic diameters were also measured using an optical counter (Grimm 11-R). This combination of modelling and measuring techinques lead us to a pool of representative values for outdoor and indoor exposure. People’s exposure to air pollutants was then evaluated by assigning pollutant concentrations to a person depending on his/her coordinates as they move in time through different locations/microenvironments. The population uptake of different PM size fractions was further investigated using a human respiratory tract (HRT) deposition model.

PM levels and size distribution varied among the different parts of the urban agglomeration, as well as among different micro-environments and transportation means. Based on movement and activity, changes in exposure levels were calculated not only for the overall agent population but for individual agents as well. By estimating the daily time-activity patterns (as predicted by the coupled sensors-ANN-ABM platform) of vulnerable subgroups such as children or the elderly, we were able to estimate their personal exposure and uptake profile, taking into account a) PM size distribution across the different locations encountered and b) differences in intensity of activity.

Overall, exposure to PM and especially deposition of PM of lower fractions across the HRT are higher for children, as a result of the higher fraction time spent outdoors and in transport, the higher bodyweight-normalised inhalation rate, the more intensive activity profile and age-dependent differences in respiratory physiology.

This study represents the first step towards improving the calculation process of population exposure to environmental substances so that we would be able to draw better conclusions on the association between environment and health. Data collected by “smart” devices can help provide more accurate exposure assessment for exposure simulation modelling and epidemiology studies. Opportunities will come closer for further wide spread use to assess, at least part of the exposome. Such kind of investigations offer valuable information on the utility of several commercial devices as modular add-ons to exposure studies.