2014.ISDS.Abstracts.Final.pdf


ISDS Annual Conference Proceedings 2014. This is an Open Access article distributed under the terms of the Creative Commons Attribution-
Noncommercial 3.0 Unported License (http://creativecommons.org/licenses/by-nc/3.0/), permitting all non-commercial use, distribution, 
and reproduction in any medium, provided the original work is properly cited.

66

ISDS 2014 Conference Abstracts

Vetsyn: Veterinary Syndromic Surveillance Streamlined 
into one R Package
Fernanda C. Dórea*, Stefan Widgrén and Ann Lindberg

Department of Epidemiology and Disease Control, National Veterinary Institute, Uppsala, Sweden

Objective
To describe an R package that was designed to provide ready 

implementation of veterinary syndromic surveillance systems, from 
classified data to the generation of alerts and an html interface.

Introduction
The field of veterinary syndromic surveillance (VSS) is developing 

fast, with countries exploring a great variety of data sources [1]. 
After implementing two VSS systems we have demonstrated that 
the steps from classified data to full system implementation can be 
streamlined, and published a guideline for implementation [2]. All 
the steps described have been made available in an R package (https://
github.com/nandadorea/vetsyn). We aim to demonstrate the utility 
and potential of this streamlined approach.

Methods
The vetsyn package takes advantage of object-oriented 

programming in R. Using S4 classes, we have defined a syndromic 
class, which stores in a single object all the data needed and produced 
during syndromic monitoring. The object contains the following slots:

• @observed: a matrix storing the observed event counts of one
syndrome per column, and one time point (day or week) per row (t x 
s). This is constructed with a single function starting from a dataset of 
health events, where the events have been classified into syndromes.

• @dates: a data frame storing date related information for each
time point in the dataset (rows in all other slots).

• @baseline: a t x s matrix which serves as a training baseline for
outbreak-signal detection algorithms. During the implementation 
of the system a function can be used to construct this baseline from 
historical data, removing outliers based on parametric (generalized 
linear models) or non-parametric methods (moving percentiles). 
After the system is implemented the baseline continues to grow by 
automatically removing detected outbreak-signals from observed data 
[3].

• @ alarms: a t x s x a array (t x s x algorithms for detection), in
which a third dimension is added to accommodate for the parallel 
use of multiple aberration detection algorithms. Users can choose 
to use control charts (EWMA, CUSUM and Shewhart) or the Holt-
Winters exponential smoothing to detect outbreak-signals, and those 
can be coupled with parametric (generalized linear models) or non-
parametric (differencing) pre-processing to remove temporal effects. 
The user can also set up multiple detection limits, in which case for 
each syndrome, time-point, and algorithm used, the stored result is a 
detection score, rather than a binomial indicator of whether an alarm 
happened.

• @UCL: a t x s x a array, which stores, for each syndrome, time-
point and algorithm used, the upper confidence limit of the detection 
method chosen, that is, the minimum value that would have generated 
an alarm.

One function helps the user set up email alerts to be generated in 
case of an alarm. Another function allows the user to generate an html 
interface where all syndromes monitored are tabulated, an alarm score 
is plotted for each syndrome (Figure), and time series graphs show 
any alarms detected.

Conclusions
We have combined several statistical analyses methods already 

tested for use in animal data streams, and R functions to support data 
management and output visualization. A full tutorial is available, 
which further helps streamlining the process, facilitating the 
implementation of syndromic surveillance systems by veterinary 
epidemiologists.

Figure: Example of the html interface generated to display the information 
stored in a syndromic object created with the vetsyn package.

Keywords
Animal health; syndromic surveillance; R programming

References

[1] Dupuy C, Bronner A, Watson E,et al. Inventory of veterinary syndromic 
surveillance initiatives in Europe (Triple-S project): Current situation 
and perspectives. Prev Vet Med 2013 (111) 3-4 220-9

[2] Dórea FC, Lindberg A, McEwen BJ, et al. Syndromic surveillance 
using veterinary laboratory requests: a practical guide informed by 
experience with two systems. Prev Vet Med 2014 .DOI: 10.1016/j.
prevetmed.2014.04.001

[3] Dórea FC, McEwen BJ, McNab BW, et al. Syndromic surveillance 
using veterinary laboratory data: data pre-processing and algorithm 
performance evaluation. J Royal Soc Interface. 2013 DOI:10.1098/
rsif.2013.0114

*Fernanda C. Dórea
E-mail: fernanda.dorea@sva.se    

Online Journal of Public Health Informatics * ISSN 1947-2579 * http://ojphi.org * 7(1):e19, 2015