Climate variability, weather and enteric disease incidence in New Zealand: time series analysis

Abstract

Background: Evaluating the influence of climate variability on enteric disease incidence may improve our ability to predict how climate change may affect these diseases.

Objectives: To examine the associations between regional climate variability and enteric disease incidence in New Zealand.

Methods: Associations between monthly climate and enteric diseases (campylobacteriosis, salmonellosis, cryptosporidiosis, giardiasis) were investigated using Seasonal Auto Regressive Integrated Moving Average (SARIMA) models.

Results: No climatic factors were significantly associated with campylobacteriosis and giardiasis, with similar predictive power for univariate and multivariate models. Cryptosporidiosis was positively associated with average temperature of the previous month (β = 0.130, SE = 0.060, p <0.01) and inversely related to the Southern Oscillation Index (SOI) two months previously (β = -0.008, SE = 0.004, p <0.05). By contrast, salmonellosis was positively associated with temperature (β = 0.110, SE = 0.020, p<0.001) of the current month and SOI of the current (β = 0.005, SE = 0.002, p<0.050) and previous month (β = 0.005, SE = 0.002, p<0.05). Forecasting accuracy of the multivariate models for cryptosporidiosis and salmonellosis were significantly higher.

Conclusions: Although spatial heterogeneity in the observed patterns could not be assessed, these results suggest that temporally lagged relationships between climate variables and national communicable disease incidence data can contribute to disease prediction models and early warning systems.

Figure 1. Time series of raw and log transformed monthly incidence (after differencing) of campylobacteriosis (A-B), salmonellosis (C-D), cryptosporidiosis (E-F), and giardiasis (G-H) in New Zealand, 1997-2008.

Figure 2. Autocorrelation plots, partial autocorrelation plots of the residuals and scatter plot of residuals against the predicted values of the seasonal autoregressive moving average SARIMA model fitted to the natural logarithm differenced disease incidence.

Campylobacteriosis SARIMA (1, 0, 0) (2, 0, 0)₁₂ (A-C), salmonellosis SARIMA (1, 0, 0) (1, 0, 0)₁₂ (D-F), cryptosporidiosis SARIMA (1, 0, 0) (1, 0, 0)₁₂ (G-I), giardiasis SARIMA (1, 0, 0) (1, 0, 1)₁₂ (J-L). The x-axis gives the number of lags in months and the grey shaded areas represent the 95% confidence interval.

Figure 3. SARIMA model of forecasting weather variation in New Zealand (A-C-E-G).

Actual monthly incidence /100000 population (black line), rates predicted by the chosen SARIMA models for each disease (grey dashed line) and rates predicted for the validation period ( January to December 2008) (red dashed line). (B-D-F-H) Cumulative monthly incidence /100000 population of the actual rates (black line) and rates predicted by the chosen SARIMA models for each disease (red dashed line) from January to December 2008 (validation period). Campylobacteriosis (A-B), salmonellosis (C-D), cryptosporidiosis (E-F), giardiasis (G-H). The y axis gives the monthly incidence and the x axis represents time in months.