Multiscale measurement error models for aggregated small area health data

Modeling Provincial Covid-19 Epidemic Data Using an Adjusted Time-Dependent SIRD Model

In this paper, we develop a forecasting model for the spread of COVID-19 infection at a provincial (i.e., EU NUTS-3) level in Italy by using official data from the Italian Ministry of Health integrated with data extracted from daily official press conferences of regional authorities and local newspaper websites. This data integration is needed as COVID-19 death data are not available at the NUTS-3 level from official open data channels. An adjusted time-dependent SIRD model is used to predict the behavior of the epidemic; specifically, the number of susceptible, infected, deceased, recovered people and epidemiological parameters. Predictive model performance is evaluated using comparison with real data.

Multivariate Air Pollution Prediction Modeling with partial Missingness

Missing observations from air pollution monitoring networks have posed a longstanding problem for health investigators of air pollution. Growing interest in mixtures of air pollutants has further complicated this problem, as many new challenges have arisen that require development of novel methods. The objective of this study is to develop a methodology for multivariate prediction of air pollution. We focus specifically on tackling different forms of missing data, such as: spatial (sparse sites), outcome (pollutants not measured at some sites), and temporal (varieties of interrupted time series). To address these challenges, we develop a novel multivariate fusion framework, which leverages the observed inter-pollutant correlation structure to reduce error in the simultaneous prediction of multiple air pollutants. Our joint fusion model employs predictions from the Environmental Protection Agency's Community Multiscale Air Quality (CMAQ) model along with spatio-temporal error terms. We have implemented our models on both simulated data and a case study in South Carolina for 8 pollutants over a 28-day period in June 2006. We found that our model, which uses a multivariate correlated error in a Bayesian framework, showed promising predictive accuracy particularly for gaseous pollutants.

Zero-inflated multiscale models for aggregated small area health data

It is our primary focus to study the spatial distribution of disease incidence at different geographical levels. Often, spatial data are available in the form of aggregation at multiple scale levels such as census tract, county, state, and so on. When data are aggregated from a fine (e.g. county) to a coarse (e.g. state) geographical level, there will be loss of information. The problem is more challenging when excessive zeros are available at the fine level. After data aggregation, the excessive zeros at the fine level will be reduced at the coarse level. If we ignore the zero inflation and the aggregation effect, we could get inconsistent risk estimates at the fine and coarse levels. Hence, in this paper, we address those problems using zero inflated multiscale models that jointly describe the risk variations at different geographical levels. For the excessive zeros at the fine level, we use a zero inflated convolution model, whereas we consider a regular convolution model for the smoothed data at the coarse level. These methods provide a consistent risk estimate at the fine and coarse levels when high percentages of structural zeros are present in the data.

Comparing multilevel and multiscale convolution models for small area aggregated health data

In spatial epidemiology, data are often arrayed hierarchically. The classification of individuals into smaller units, which in turn are grouped into larger units, can induce contextual effects. On the other hand, a scaling effect can occur due to the aggregation of data from smaller units into larger units. In this paper, we propose a shared multilevel model to address the contextual effects. In addition, we consider a shared multiscale model to adjust for both scale and contextual effects simultaneously. We also study convolution and independent multiscale models, which are special cases of shared multilevel and shared multiscale models, respectively. We compare the performance of the models by applying them to real and simulated data sets. We found that the shared multiscale model was the best model across a range of simulated and real scenarios as measured by the deviance information criterion (DIC) and the Watanabe Akaike information criterion (WAIC).