We chose to fit autoregressive over-dispersed Poisson regression
models due to the nature of the neonatal outcomes variables,
daily counts of total births and daily counts of preterm births.
Trends and seasonality were controlled using functions -sine and
cosine- with annual and six-monthly periodicities, days of the
week were also introduced in the models. The chemical air pollutants
(PM2.5 and PM10, O3 and NO2) and the noise levels (Leqd
and Leqn) were introduced into the model as linear components.
Traditionally, the relation between ambient temperature and
morbi-mortality variables shows a V-shaped pattern (Basu and
Samet, 2002; Ye et al., 2012; Gasparrini et al., 2015). However, in
adverse birth outcomes there is a lack. There are studies that
consider the effect of heat and cold jointly on births outcomes
(Kloog et al., 2015). However, others studies (Schifano et al., 2013)
showed that heat, as measured by the maximum temperature
during warmer periods, was associated in the short term with
preterm birth. This association with temperature was not observed
in the cold periods.
For this reason, the effect of temperature has been divided into
two branches in our analysis, warmer periods and cold periods.
The definition of these periods is established according to the
thresholds for cold and heat calculated for Madrid City (Linares
et al., 2014, 2015). These values are:
Daily Maximum Temperature Threshold434 °C for the heat
wave definition.
Daily Minimum Temperature Thresholdo 2 °C for the cold
wave definition.
With the aim of parametrizing this fact, two new variables
were created as follows:
Thot¼Tmax– 34 °C if TmaxZ34 °C
Thot¼0if Tmaxo34 °C
Tcold¼ 2 °C-Tmin if Tminr 2 °C
Tcold¼0if Tmin4 2 °C
Only short-term lags, from 1 to 7 days of the environmental
variables were introduced in the models performed. First, single
models for the total number of births and for total number of
preterm births with all the environmental and control variables
were performed separately. In a second stage, a model for total
births adjusted for preterm births was performed in order to explore
whether the same environmental variables were influenced
the results through the total number of births. It means, if the
short-term effect of the environmental variables on preterm births
is influencing the global effect on the total number of births.
The results of the final models were expressed in the form of
relative risks (RRs) and their 95% confidence intervals only for
statistically significant variables (po0.05). These RRs are calculated
for an interquartile increase. Based on the RR, we then calculated
the attributable risk (AR) associated with this increase by
the following equation:
AR¼[(RR—1)/RR]x100 (Coste and Spira, 1991).
All analyses were performed using the SPSS v22 (SPSS: an IBM
company) and Stata/SE 11.2 (StataCorp LP) software programmes.
6.