Sentinel

A toolkit for visualization, early warning detection and prediction of disease outbreaks (Malaria, Diarrhea, etc.)

28-02-2017:

Real values were replaced by fake random data

Getting started:

• Open sentinel.Rproj
• Run app.R

Goals:

• Build a package that could receive any kind of algorithms for epidemiologists to play with.
• Automate data processing and conversion for any kind of format and backend.

Sentinel as a reporting tool (in report folder):

• Produce pdf LateX documents to summarize situation of sentinel sites (plots, trends and tables).
• Produce word docx documents to display historical alerts and summary report on SMS sent to our postgres database for each sentinel sites.
• Produce an interactive version of each report in pdf:

Main report summary:

• Produce a two weeks outline of alerts taken from the main report:

• HTC report summary (user can select an HTC site an chart will be displayed)

• Malaria/Fever report summary (user can select a site an chart will be displayed)

• Diarrhea report summary (user can select a site an chart will be displayed)

• ILI report summary (user can select a site an chart will be displayed)

• AFP report summary (user can select a site an chart will be displayed)

• Missing sent report summary (with cell highlighting when SMS sent <=3)

System and Packages requirements:

• R >= 3.2.3 and RStudio >= 0.99.896
• Have a look at libraries.R

Algorithms used to trigger alert:

Percentile algorithm:

Percentile algorithm is used to trigger alert in sentinel network. An alert is triggered when during n (consecutive or not) week(s) , diseases occurrence exceeds 90th percentile calculated using the entire historical time series. This calculation of 90th percentile excludes the current week.

Default values for percentile algorithm are:

• n number of consecutive weeks we observe = 3
• value of the percentile = 90th

MinSan algorithm:

MinSan: The Ministry of Health defined a simple rule that tells that if occurrences of a given disease exceed a certain proportion (slope parameter (value)) during n ( 3 or 4 ) consecutive weeks or not then an alert is triggered.

CSum algorithm:

CSum algorithm consists of comparing 52 latest weeks with smoothed mean of past years. Comparison is made week by week. For example,week 02 of this month is compared with the smoothed mean of week 02 for past years excluding the 52 latest. Parameters can be tuned such as number of past years from which mean will be calculated, sd, degree of smoothing (number of weeks for mean calculation) .It is only efficient when disease presents weekly strong seasonal pattern.