Unit:
Τομέας Κοινωνικής Ιατρικής - Ψυχιατρικής και ΝευρολογίαςLibrary of the School of Health Sciences
Dissertation committee:
Ευαγγελία Σαμόλη, Επίκουρη Καθηγήτρια, Ιατρική Σχολή, ΕΚΠΑ
Ελένη-Κλεάνθη Κατσουγιάννη, Καθηγήτρια, Ιατρική Σχολή, ΕΚΠΑ
Παγώνα Λάγιου, Καθηγήτρια, Ιατρική Σχολή, ΕΚΠΑ
Παναγιώτα Τουλούμη, Καθηγήτρια, Ιατρική Σχολή, ΕΚΠΑ
Βασιλική-Αναστασία Σύψα, Αναπληρώτρια Καθηγήτρια, Ιατρική Σχολή, ΕΚΠΑ
Βασιλική Μπενέτου, Επίκουρη Καθηγήτρια, Ιατρική Σχολή, ΕΚΠΑ
Φώτιος Σιάννης, Επίκουρος Καθηγήτής, Τμήμα Μαθηματικών, ΕΚΠΑ
Original Title:
Μεθοδολογικές προσεγγίσεις για τη διερεύνηση των επιπτώσεων της ταυτόχρονης έκθεσης σε διάφορους ατμοσφαιρικούς ρύπους στην ανθρώπινη υγεία. Η προσέγγιση της ενιαίας ατμόσφαιρας (One atmosphere approach)
Translated title:
Methodological approaches for the assessment of cumulative health effects of exposure to multiple pollutants. One atmosphere approach
Summary:
Assessment of the cumulative effect of correlated exposures is an open methodological issue in environmental epidemiology. Most previous studies have applied regression models with interaction terms or dimension reduction methods. The combined effect of pollutants has been also evaluated through the use of exposure scores that incorporate weights based on the strength of the component-specific associations with health outcomes or their contribution in air quality. Nevertheless, there is lack of studies that compare the different statistical methodologies under varying conditions.
We compared three methodological approaches addressing multi-pollutant short term exposures in time series Poisson regression models with overdispersion aiming to contribute on the ongoing methodological discussion. The approaches compared were: an additive (in terms of the model’s linear predictor) main effects model, a refinement of least absolute shrinkage and selection operator (LASSO) named adaptive LASSO as a dimension reduction method that uses penalized variable selection before the application of a main effects model and a weighted exposure score taking into account all the pollutants under study. We used two values for pollutants weights: 1) the pollutant-specific concentration-response functions derived from published reviews and 2) a scaled version of them that incorporated the effect estimates’ uncertainty.
We assessed the performance of the methods in terms of their ability to estimate the “true” cumulative effect of short-term exposure to six regulated pollutants on all-cause and respiratory mortality, i.e. the daily number of deaths from all, excluding external causes and from respiratory, non-malignant causes, by simulations under various scenarios for the pollutants’ correlations (low, moderate, high). The concentration-response function for the “true” cumulative effect was assumed equal to 0.01 and 0.02 for all-cause and respiratory mortality, respectively. Simulations were based on time series data from Athens, Greece in 2007-2012 using a Multivariate Normal distribution to simulate daily pollutants’ concentrations and a Negative Binomial distribution to generate daily number of deaths under a quassi-Poisson distribution. The variance-covariance matrix of the Multivariate Normal distribution was estimated under the three different scenarios for the correlation among pollutants: 1) observed correlations using the real data (moderate correlation); 2) the half of the observed correlations (low correlation); and 3) twice the observed correlations (high correlation). For each outcome and pollutants’ correlation scenario we performed 1000 repetitions and we estimated the bias, the coverage probability and the mean square error. Finally, we applied the three multi-pollutant approaches to the true time series data from Athens and we compared the results among them.
The exposure score provided the least biased estimate under all correlation scenarios for both mortality outcomes. The maximum value of bias was 0.020 with √(MSE )= 0.020 under high correlation and all-cause mortality scenario for the scaled score, while the respective figures for respiratory mortality were bias = 0.014 and √(MSE ) = 0.014. The adaptive LASSO performed well in the case of low and medium correlation between exposures. The maximum value of bias was 0.027 with √(MSE ) = 0.027 under moderate correlation and respiratory mortality scenario. Finally, the main effect model resulted in severe bias with the maximum value being equal to 9.937 in absolute value and the respective √(MSE ) = 11.748 under high correlation and respiratory mortality scenario.
In the real data application, the cumulative effect estimate was similar between approaches for all-cause mortality ranging from 0.7% increase per interquartile range (IQR) for the score to 1.1% for main effects model, while for respiratory mortality conclusions were contradictive and ranged from 0.6% decrease for adaptive LASSO to 2.8% increase for scaled score. The cumulative effect’s estimates for respiratory mortality presented large uncertainty independently of the approach used possibly due to the scarcity of the outcome.
In conclusion, the use of a weighted exposure score to address cumulative effects of correlated metrics may perform well in terms of bias under different exposure correlation and variability in the health outcomes. However, the performance of methods under variable lag structures per pollutant or non- linear associations between pollutants and outcomes should be also assessed and evaluated.
Main subject category:
Health Sciences
Keywords:
Air pollution, Cumulative effect, Multi-pollutant models, LASSO, Exposure score
Number of references:
118
File:
File access is restricted only to the intranet of UoA.
Rodopoulou Sofia PhD.pdf
1 MB
File access is restricted only to the intranet of UoA.
