Welcome to surveillance project!

The R-package 'surveillance' is a framework for the development and the evaluation of outbreak detection algorithms in univariate and multivariate routine collected public health surveillance data. It is hosted on CRAN.

• The intention of the R-package surveillance is to provide open source software for the temporal and spatio-temporal visualization, modelling and monitoring of epidemic phenomena. This includes count, binary and categorical data time series as well as continuous-time processes having discrete or continuous spatial resolution.
  
• Potential users of the package are biostatisticians, epidemiologists and others working in, e.g., applied infectious disease epidemiology. However, applications could just as well originate from environmetrics, reliability engineering, econometrics or social sciences.
• The main application is in the detection of aberrations in routine collected public health data seen as univariate and multivariate time series of counts.
• surveillance provides an S4 class data structure and framework for methodological developments of change-point algorithms for time series of counts.
• Prospective outbreak detection procedures for count data time series:
• cdc - Stroup et al. (1989)
• farrington - Farrington et al. (1996)
• rki - The system used at the Robert Koch Institute, Germany
• bayes - A Bayesian predictive posterior approach, see Höhle (2007)
  
• hmm - An online version of the Hidden Markov Model approach by Le Strat and Carrat (1999)
• rogerson - Surveillance for time varying Poisson means as documented in Rogerson and Yamada (2004).
  This approach has been extended to cover time varying proportions in a binomial setting.
• cusum - An approximate CUSUM method for time varying Poisson means as documented in Rossi et al (1999)
• glrnb - Likelihood and generalized likelihood ratio detectors for time varying Poisson and negative binomial distributed series documented in Höhle and Paul (2008).
  
• outbreakP - Semiparametric surveillance of outbreaks by Frisén and Andersson (2009)
• Online change-point detection in categorical time series:
• categoricalCUSUM - includes change-point detection based on regression models for binomial and beta-binomial distributed response. Furthermore, multi-categorical models includes the multinomial logistic model, proportional odds model and the Bradley-Terry models, see Höhle (2010).
• pairedbinCUSUM - paired-binary approach taken in Steiner et al. (1999)
  
• Retrospective modelling of univariate and multivariate count data time series is also available as estimation and prediction routines for the models described in
  • algo.hhh - Held et al. (2005) and Paul et al. (2008)
  • algo.twins - Held et al. (2006)
• For evaluation purposes, the package contains example datasets drawn from the SurvStat@RKI Database maintained the RKI, Germany. More comprehensive comparisons using simulation studies are possible by methods for simulating point source outbreak data using a hidden Markov model. To compare the algorithms, benchmark numbers like sensitivity, specificity and detection delay can be computed for entire sets of surveillance time series. Furthermore, a Markov Chain approximation for computing the run-length distribution of the proposed likelihood ratio CUSUMs is available as function LRCUSUM.runlength.
• Retrospective continuous time modelling and simulation of epidemic phenomena (experimental from version 1.3-0 on):
  • twinSIR - continuous-time and discrete-space modelling as described in Höhle (2009). The epidata class provides the appropriate data structure for such data.
  • twinstim - continuous-time and continuous-space modelling as desribed in Meyer et al. (2010) and as caputred by the epidataCS data class.
• Prospective space-time cluster detection:
• stcd - (experimental) Point process based approach by Assuncao & Correa (2009)
  
• The package comes with ABSOLUTELY NO WARRANTY; for details see http://www.gnu.org/copyleft/gpl.html (GPL). This is free software, and and you are welcome to redistribute it under the GPL conditions.

The surveillance package (version 1.2-1) is available for download from CRAN.
Current package development, help-forum and bugtracking is hosted through R-Forge:

https://r-forge.r-project.org/projects/surveillance/

From this page snapshots of the current development version are available. For example -- if you are running a recent R version -- you can obtain a binary snapshot of the development version as

install.packages("surveillance",repos="http://r-forge.r-project.org").

You can also manually download the binary snapshot or the source tarball.

New features:

• See NEWS file in the current distribution

• A good (but slightly old) introduction to the package is provided in the paper surveillance: An R package for the surveillance of infectious diseases, Computational Statistics (2007), 22(4), pp. 571-582. [preprint]
• A more recent description can be found in the book chapter Aberration detection in R illustrated by Danish mortality monitoring (2010), M. Höhle and A. Mazick, To appear in T. Kass-Hout and X. Zhang (Eds.) Biosurveillance: A Health Protection Priority, CRC Press. [preprint]. Note: As ISO 8601 handling is not fully implemented in R on Windows the demo("biosurvbook") will only run with package version >= 1.2, where a workaround was implemented.
• An overview of statistical methods and implementational usage is given the course notes of the short course on Statistical surveillance of infectious diseases held at the Department of Statistics, Universidade Federal de Minas Gerais (UFMG), Belo Horizonte, Brazil, Nov 27-28, 2008.
• Invited talk held at the ESCAIDE satellite workshop on Computer supported outbreak detection and signal management (R-File, Data from SurvStat@RKI)
• Application of the package in veterinary public health surveillance is described in Statistical approaches to the surveillance of infectious diseases for veterinary public health [preprint].
• Read the package vignette or look here for further preprints.
  
• Sometimes one picture says more than 1000 words:

• Michael Höhle, Department of Statistics, University of Munich, Germany (Project Admin)
• Michaela Paul, Institute of Social and Preventive Medicine, University of Zurich, Switzerland
• Sebastian Meyer, Institute of Social and Preventive Medicine, University of Zurich, Switzerland
• Former student programmers: C. Lang, Andrea Riebler, Valentin Wimmer
• Contributions by: T. Correa, M. Hofmann and S. Steiner

The development of models and algorithms implemented in surveillance was financially supported by :

• Munich Center of Health Sciences (MC-Health, 2007-2010)
• Swiss National Science Foundation (SNF, since 2007)
• German Science Foundation (DFG, 2003-2006)