Modeling and Prediction of Oyster Norovirus Outbreaks along Gulf of Mexico Coast

Abstract

Background: Oyster norovirus outbreaks often pose high risks to human health. However, little is known about environmental factors controlling the outbreaks, and little can be done to prevent the outbreaks because they are generally considered to be unpredictable.

Objective: We sought to develop a mathematical model for predicting risks of oyster norovirus outbreaks using environmental predictors.

Methods: We developed a novel probability-based Artificial Neural Network model, called NORF model, using 21 years of environmental and norovirus outbreak data collected from Louisiana oyster harvesting areas along the Gulf of Mexico coast, USA. The NORF model involves six input variables that were selected through stepwise regression analysis and sensitivity analysis.

Results: We found that the model-based probability of norovirus outbreaks was most sensitive to gage height (the depth of water in an oyster bed) and water temperature, followed by wind, rainfall, and salinity, respectively. The NORF model predicted all historical oyster norovirus outbreaks from 1994 through 2014. Specifically, norovirus outbreaks occurred when the NORF model probability estimate was > 0.6, whereas no outbreaks occurred when the estimated probability was < 0.5. Outbreaks may also occur when the estimated probability is 0.5-0.6.

Conclusions: Our findings require further confirmation, but they suggest that oyster norovirus outbreaks may be predictable using the NORF model. The ability to predict oyster norovirus outbreaks at their onset may make it possible to prevent or at least reduce the risk of norovirus outbreaks by closing potentially affected oyster beds.

Citation: Wang J, Deng Z. 2016. Modeling and prediction of oyster norovirus outbreaks along Gulf of Mexico coast. Environ Health Perspect 124:627-633; http://dx.doi.org/10.1289/ehp.1509764.

Figure 1

Oyster harvesting areas along Louisiana coast, USA. [MODIS Surface Reflectance image from 2001 was retrieved from https://ladsweb.nascom.nasa.gov/ maintained by the NASA EOSDIS Level 1 and Atmosphere Archive and Distribution System Distributed Active Archive Center (LAADS DAAC), NASA’s Goddard Space Flight Center, Greenbelt, Maryland. The data product for the image was provided by NASA.]

Figure 2

Relationships between the frequency distribution of norovirus outbreaks (y-axis) and normalized environmental predictors (x-axis): (A) gage height (GH) and daily change in gage height (DCGH), (B) temperature (T), (C) salinity (S), and (D) wind (W).

Figure 3

Comparison between the NORF model–predicted probabilities of norovirus outbreak and the observed norovirus outbreak probabilities (0 or 1) in oyster-harvesting areas along Louisiana Gulf Coast: (A) Areas 6 and 7 with outbreaks in February and December 1996, (B) Areas 1, 6, and 7 with outbreaks in March 2002, and (C) Area 3 with an outbreak in December 2007 The red horizontal line denotes the threshold probability of 0.6 for norovirus outbreaks, implying that a norovirus outbreak would occur if the model-predicted probability is > 0.6. Likewise, the yellow horizontal line indicates the threshold probability of 0.5 for non-outbreak, meaning that there would be no norovirus outbreaks if the model predicted probability is < 0.5.

Figure 4

Sensitivity of the NORF model output to environmental predictors. Solid bars indicate percent changes in the probability (model output) predicted by the NORF model due to positive changes to model input variables; the hollow bars indicate percent changes in the model output due to negative changes to model input variables.

Figure 5

Comparison between the NORF model–predicted probabilities and the observed probabilities (0 or 1) of the norovirus outbreaks in oyster-harvesting areas along the Louisiana Gulf Coast: (A) Areas 2 and 3 with outbreaks in March 2010, (B) Area 7 with an outbreak in March 2010, (C) Area 13 with outbreaks in March 2010, and (D) Area 26 with an outbreak in April 2012 and Area 30 with an outbreak in December 2012.