[Proposal for next NOSOVE course in June 2014, please vote in the poll]
Teacher: Dorea Fernanda
Syndromic
surveillance can be characterised as a process involving the continuous
analysis of health data to provide immediate feedback.
The increasing
amount of health data recording in electronic format presents extra challenges
for data analysis, but also new opportunities to extract information from data
in real and near-real time. In this course participants will have a chance to
learn, through several hands-on exercises, how to use freely available software
to set up automated, autonomous routines of data analysis.
The theory and exercises will cover all the basic steps
to successfully develop, evaluate and implement a syndromic surveillance system
capable of detecting temporal aberrations (which may indicate the occurrence of
outbreaks) when monitoring cases load from a given animal health data source,
such as laboratory submissions, clinical cases, etc. These steps can be summarized
as:
- basic text mining
methods for automated classification of records into syndromes;
- retrospective evaluation
of data to create baseline profiles following the removal of excessive
noise and aberrations, and the identification of temporal effects;
- prospective evaluation
of detection algorithms; and finally
- real-time monitoring and implementation.
As all software
used are freely available, participants will be able to readily apply the
skills learned into their work or research.
Course specifications
Participants are expected to have a basic
knowledge of biostatistics. No previous knowledge of the software to be used is
expected.
The course exercises will use the statistical programming environment R,
and RapidMiner,
the world-leading open-source system for data mining available freely from
Rapid-I (http://rapid-i.com/content/view/26/84/).
Participants will receive download and installation instructions upon
registration, and they should bring their own laptop computer. Datasets will be
provided as part of the course, but participants are welcome to bring a dataset
of their own to explore some of the techniques learned.
Workshop Content
Day 1 – Introduction
to the tools: Basics of using R and
Rapid Miner.
Day2 – Syndromic
surveillance, Step 1: Automated classification of records into syndromes. Participants
will learn the basics of text-mining and will practice implementing supervised (rule-based)
and unsupervised (naïve Bayes, Decision Trees) machine learning methods in
order to create automated routines to classify records into syndromes.
Day3 – Syndromic
surveillance, Step 2: Retrospective evaluation of data available. Basic concepts of time
series analysis will be covered. Participants will practice on a dataset in
order to learn how to identify statistical characteristics of the time series
which can impact aberration detection, in special temporal effects such as day
of week and seasonal patterns. Step 3: Prospective evaluation of data. Participants will practice the use
of basic temporal aberration detection algorithms, such as control charts
(cumulative sums, Shewhart type control charts and Exponentially Weighted
Moving Average). More advanced methods (such as Holt-Winters exponential
smoothing and removal of temporal effects using regression models) will then be
employed to deal with specific characteristics that can be found when
monitoring different data sources.
Day4 – Syndromic
surveillance, Step 4: Implementation.
Participants will learn how to combine tools (Rapid Miner and R) in order to
implement a syndromic surveillance system, from data acquisition to final
reports.
