Limits...
Temporal cross ‐ correlation between influenza ‐ like illnesses and invasive pneumococcal disease in The Netherlands

View Article: PubMed Central - PubMed

ABSTRACT

Background: While the burden of community‐acquired pneumonia and invasive pneumococcal disease (IPD) is still considerable, there is little insight in the factors contributing to disease. Previous research on the lagged relationship between respiratory viruses and pneumococcal disease incidence is inconclusive, and studies correcting for temporal autocorrelation are lacking.

Objectives: To investigate the temporal relation between influenza‐like illnesses (ILI) and IPD, correcting for temporal autocorrelation.

Methods: Weekly counts of ILI were obtained from the Sentinel Practices of NIVEL Primary Care Database. IPD data were collected from the Dutch laboratory‐based surveillance system for bacterial meningitis from 2004 to 2014. We analysed the correlation between time series, pre‐whitening the dependent time series with the best‐fit seasonal autoregressive integrated moving average (SARIMA) model to the independent time series. We performed cross‐correlations between ILI and IPD incidences, and the (pre‐whitened) residuals, in the overall population and in the elderly.

Results: We found significant cross‐correlations between ILI and IPD incidences peaking at lags ‐3 overall and at 1 week in the 65+ population. However, after pre‐whitening, no cross‐correlations were apparent in either population group.

Conclusion: Our study suggests that ILI occurrence does not seem to be the major driver of IPD incidence in The Netherlands.

No MeSH data available.


Related in: MedlinePlus

Schematic of pre‐whitening approach for cross‐correlation analysis. Both time series are transformed to stabilize the variance; the best‐fit SARIMA model to the independent variable is then used to filter the dependent variable. The white noise of the best‐fit SARIMA and the residuals of the filtered dependent time series are cross‐correlated. Lags with a significant correlation will indicate the time lag between the independent and dependent time series
© Copyright Policy - creativeCommonsBy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5304567&req=5

irv12442-fig-0002: Schematic of pre‐whitening approach for cross‐correlation analysis. Both time series are transformed to stabilize the variance; the best‐fit SARIMA model to the independent variable is then used to filter the dependent variable. The white noise of the best‐fit SARIMA and the residuals of the filtered dependent time series are cross‐correlated. Lags with a significant correlation will indicate the time lag between the independent and dependent time series

Mentions: Weekly ILI and IPD incidences were cross‐correlated using the cross‐correlation function (CCF). Then, the CCF was used after applying the pre‐whitening method,42 which corrects for autocorrelation within time series: this is achieved by fitting a seasonal autoregressive integrative moving average (SARIMA) model43 to the independent time series—ILI—and filtering the dependent time series—IPD—with the best‐fit model of the independent time series (Figure 2). This removes the autocorrelation present in the independent series from the dependent series. Any remaining correlation between the series then can no longer be due to a common autocorrelation structure, including the seasonality, as the seasonal pattern in SARIMA is modelled as an autocorrelation between measurements exactly 1 year apart. The method for SARIMA fitting and for pre‐whitening is explained in more detail in Supplement, section B and C, respectively. Finally, a CCF was plotted between the residuals of the best‐fit SARIMA model to the ILI data, and the filtered IPD data (Figure 2). For the analysis in the elderly, the same procedures were used over the period of week 38 in 2004 to week 25 in 2014: the first weeks in 2004 were excluded because in many weeks, no ILI cases were found.


Temporal cross ‐ correlation between influenza ‐ like illnesses and invasive pneumococcal disease in The Netherlands
Schematic of pre‐whitening approach for cross‐correlation analysis. Both time series are transformed to stabilize the variance; the best‐fit SARIMA model to the independent variable is then used to filter the dependent variable. The white noise of the best‐fit SARIMA and the residuals of the filtered dependent time series are cross‐correlated. Lags with a significant correlation will indicate the time lag between the independent and dependent time series
© Copyright Policy - creativeCommonsBy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5304567&req=5

irv12442-fig-0002: Schematic of pre‐whitening approach for cross‐correlation analysis. Both time series are transformed to stabilize the variance; the best‐fit SARIMA model to the independent variable is then used to filter the dependent variable. The white noise of the best‐fit SARIMA and the residuals of the filtered dependent time series are cross‐correlated. Lags with a significant correlation will indicate the time lag between the independent and dependent time series
Mentions: Weekly ILI and IPD incidences were cross‐correlated using the cross‐correlation function (CCF). Then, the CCF was used after applying the pre‐whitening method,42 which corrects for autocorrelation within time series: this is achieved by fitting a seasonal autoregressive integrative moving average (SARIMA) model43 to the independent time series—ILI—and filtering the dependent time series—IPD—with the best‐fit model of the independent time series (Figure 2). This removes the autocorrelation present in the independent series from the dependent series. Any remaining correlation between the series then can no longer be due to a common autocorrelation structure, including the seasonality, as the seasonal pattern in SARIMA is modelled as an autocorrelation between measurements exactly 1 year apart. The method for SARIMA fitting and for pre‐whitening is explained in more detail in Supplement, section B and C, respectively. Finally, a CCF was plotted between the residuals of the best‐fit SARIMA model to the ILI data, and the filtered IPD data (Figure 2). For the analysis in the elderly, the same procedures were used over the period of week 38 in 2004 to week 25 in 2014: the first weeks in 2004 were excluded because in many weeks, no ILI cases were found.

View Article: PubMed Central - PubMed

ABSTRACT

Background: While the burden of community‐acquired pneumonia and invasive pneumococcal disease (IPD) is still considerable, there is little insight in the factors contributing to disease. Previous research on the lagged relationship between respiratory viruses and pneumococcal disease incidence is inconclusive, and studies correcting for temporal autocorrelation are lacking.

Objectives: To investigate the temporal relation between influenza‐like illnesses (ILI) and IPD, correcting for temporal autocorrelation.

Methods: Weekly counts of ILI were obtained from the Sentinel Practices of NIVEL Primary Care Database. IPD data were collected from the Dutch laboratory‐based surveillance system for bacterial meningitis from 2004 to 2014. We analysed the correlation between time series, pre‐whitening the dependent time series with the best‐fit seasonal autoregressive integrated moving average (SARIMA) model to the independent time series. We performed cross‐correlations between ILI and IPD incidences, and the (pre‐whitened) residuals, in the overall population and in the elderly.

Results: We found significant cross‐correlations between ILI and IPD incidences peaking at lags ‐3 overall and at 1 week in the 65+ population. However, after pre‐whitening, no cross‐correlations were apparent in either population group.

Conclusion: Our study suggests that ILI occurrence does not seem to be the major driver of IPD incidence in The Netherlands.

No MeSH data available.


Related in: MedlinePlus