event_models_ita: Default italian-trained models.
In UBESP-DCTV/imthcm: Integrated Mdelling Tool - Helath and Cost Modules

Description Usage Format Details

Defaul models are GAM models trained on the italian data used for EpiAir2 study (@source http://www.epiair.it/). In particular, GAM models were trained on data belonging to Venice city. This dataset contains information on health events and climate of the city of Venice in the period from 2006-01-01 to 2009-12-31.

1	event_models_ita

A list of 3 list each of lenght 7:

summer

Model for the summer period. It takes into account both the lagged concentration of PM2.5 and O3. It predictes the daily mean number of health events with relative 95

mort_all: Predicted number of death for all causes.
mort_cardiac: Predicted number of death for cariac diseases.
mort_resp: Predicted number of death for respiratory diseases.
mort_cer: Predicted number of death for cerebrovascular diseases.
hosp_cardiac: Predicted number of hospitalizations for cardiac diseases.
hosp_resp: Predicted number of hospitalizations for respiratory diseases.
hosp_cer: Predicted number of hospitalizations for cerebrovascular diseases.

non_summer

Model for the non-summer period. It takes into account only the lagged PM2.5 concentration. It predictes the daily mean number of health events with relative 95

mort_all: Predicted number of death for all causes.
mort_cardiac: Predicted number of death for cariac diseases.
mort_resp: Predicted number of death for respiratory diseases.
mort_cer: Predicted number of death for cerebrovascular diseases.
hosp_cardiac: Predicted number of hospitalizations for cardiac diseases.
hosp_resp: Predicted number of hospitalizations for respiratory diseases.
hosp_cer: Predicted number of hospitalizations for cerebrovascular diseases.

full_year

Model for the full year period. It takes into account only the lagged PM2.5 concentration. It predictes the daily mean number of health events with relative 95

mort_all: Predicted number of death for all causes.
mort_cardiac: Predicted number of death for cariac diseases.
mort_resp: Predicted number of death for respiratory diseases.
mort_cer: Predicted number of death for cerebrovascular diseases.
hosp_cardiac: Predicted number of hospitalizations for cardiac diseases.
hosp_resp: Predicted number of hospitalizations for respiratory diseases.
hosp_cer: Predicted number of hospitalizations for cerebrovascular diseases.

Three different type of GAM models were trained:

One model for the summer period, which was defined as the period that goes from 04-01 to 09-30.
One model for the non-summer period, which was defined as the period that goes from 01-01 to 03-31 and to 10-01 to 12-31.
One model for the all year.

All the three models were implemented assuming a Poisson distribution for health outcome, since the goal was to model health outcomes as count events. A correction factor was applied to each model to take into account overdispersion, i.e. the possibility that each outcome could occur an high number of times for some days. For all the three models the following variables were considered as covariates: the mean of daily PM2.5 concentration of the simulated day and of the three previous days (lag 0-3), the year, the month, the day and all the possible combinations of them, the daily average temperature (Celsius), modeled with a penalized spline, and the daily average barometric pressure (hPa), modeled with a penalized spline. Among all the pollutants, only one was included in the model since the correlation between air pollutants is known to be relevant. In this case, one pollutant acts as proxy for the others by including the effect on health outcomes caused by the variation of other pollutants. The reason behind the choice of an additional model for the summer period relies in the different effect that pollutants can have in different period of the year. Indeed, O3 has a significant effect on health outcomes only in the summer period. Thus, the mean of daily O3 concentration of the simulated day and of the three previous days (lag 0-3) was included only in the summer model, while the non-summer model takes into account only PM2. Finally, a third model that predict the number of health outcomes for the all year was created as a replacement for the summer model if O3 are not provided as inputs.

UBESP-DCTV/imthcm documentation built on Dec. 2, 2019, 9:26 a.m.