Analysis of Surveillance Data Philippe Dubois From Denis Coulombier, Julia Fitzner, Augusto Pinto & Marta Valenciano, WHO-HQ/LYON April 8 – 12, 2013 Phom Penh, Cambodia
Data characteristics Data validation Descriptive analysis Hypothesis testing Analysis of Surveillance Data
Data Characteristics Various sources of notification Various levels of qualification Continuous data collection subject to change
Surveillance Data Validation Frequency distributions missing values expected distribution digit attraction Cross-Tabulations age, sex, logical errors by source: collect bias ?
Notifications of All Notifiable Diseases by Date of Onset, USA, 1989
Birth weight Distribution, in Pounds Fermattes Hospital, Haiti, February 1994
Descriptive Approach Time Place Persons Generating hypotheses
Analyzing Time Characteristics Graphical analysis The 3 data components secular trend seasonal variations accidental variations
Notifications of Foodborne Outbreaks in France,
Components of Surveillance Data Signal secular trend seasonal variations accidental variations
Decomposition of Surveillance Data Signal
Descriptive analysis of Components Moving averages empirical method for reducing variability same area under the curve Logarithmic scale dynamic analysis of changes difficult to interpret
Calculation of moving averages 1993 Jan Feb Apr Jun Aug Jul Sep Oct Dec Mar May Nov /5=724,4 3690/5= /5= month window 890
Notification of giardiasis in Delaware, 03/ /1995 Crude Weekly Data
12 Week Moving Average Notification of giardiasis in Delaware, 03/ /1995
52 Week Moving Average
Notification of giardiasis in Delaware, 03/ /1995 Aggregated data
Size of the Moving Average Window empirical approach: the visual impression inversely proportional to the number of cases increases as the variance increases 1. Showing cyclical variations by removing accidental variations
Effect of the Moving Average Window Size Weekly Notifications of Salmonellosis, Georgia, weeks 7 weeks 5 weeks 10 weeks
Size of the Moving Average Window Cycle span 52 for weekly data 12 for monthly data 4 for quarterly data 2. Showing secular trend by removing cyclical variations
Cases of Gonorrhea in Michigan Week 10 of 1994 and 208 Previous Weeks
Cases of Gonorrhea in Michigan Week 10 of 1994 and 208 Previous Weeks
Descriptive Analysis of the 3 Components Moving average empirical method variability reduction same area under the curve Logarithmic scale dynamic analysis of changes difficult to interpret
Annual Rate of Tuberculosis, United States,
MALARIA- By year, United States,
MALARIA- By Year, United States, MALARIA- By year, United States,
GONORRHEA - By race and ethnicity, United States, Arithmetic scale
GONORRHEA - By race and ethnicity, United States, Logarithmic scale Source: Summary of Notifiable Diseases, United States 1993 GONORRHEA - By race and ethnicity, United States,
Typhoid Notifications in France
Interpreting the results Role of chance Role of bias True disease pattern
Conclusions Analysis to draw attention Validation by investigation
Questions? Comments? Discussions?