Intro to Biosurveillance Flashcards
What are the goals of dx surveillance?
- Describe current burden of dx
- Monitor trends -> impact of interventions, cost-effectiveness, control & eradication
- Identify outbreaks & new pathogens -> emerging/re-emergeing dx
What are some relevant metrics for morbidity?
What relevant metrics of Mortality?
How do we identify cases?
- Cases can be found in communities & hospitals -> enrollment in surveillance may vary
- Mild cases harder to see -> detection based on community-based surviellance
What are different approaches to surveillance?
Active or Passive ->
Reporting by medical professionals vs engaging in data collection
Notifiable diseases have passive surveillance but PH importance.
Sentinel sites vs population based ->
Few locations, good quality data
More costly, but more generalizable to populations
Community- vs Clinical- based
Communities report, but may need incentives
Reporters are physicians, key for rare diseases
Zero reporting can be ?
States NOT reporting vs reporting they have 0 cases
temporal clustering?
Seeing if there is higher levels than expected?
- Comparing 2 or more disease patterns on time-series plots can be hard
Time series analysis is …
- Used to describe or predict temporal distribution of dx
- Rq long series of observations
- Inappropriate for new health surveillance situations
What are soem general aims for analysis of temporal distributions of health events?
- Rapid identification of a cluster of events
- Identification of risk factors
- Generation of dx hypotheses
What is used to look at temporal clusters ?
SCAN test
Describe SCAN test?
- Particularly useful in rare dx
- Assumption of constant size of population at risk
- Estimate the number of events in a given time- window
If we calculate the probability of detecting 5 cases (the max observed) using binomial distribution and work out that it isn’t significant what does this mean?
We are NOT seeing a cluster if p value is not significant
What do control charts do?
- Set an upper and lower limit -> if value within the limit it’s “under contorl” -> if values go above upper threshold -> worrying
(Not good for controlling seasonal effects)
What problems changing from yesterday?
- Baseline changes due to random fluctuations might trigger alarms
- Day to week effects can be huge
What is a moving average? (CHANGING THE RED LINE)
- a bit of both Worlds
- Take a “window size” and predict the value based on the window
What is disadvantage of moving average?
gets ‘used ‘ to outbreaks as mean evens it out and makes them less obvious
- Upper limit and alarm become less substantial
=> motivation for CUSUM
What is CUSUM?
Detects the shifts from the mean more quickly than a control chart
Cumulative sum of deviations from a reference value (generally the mean)
General aims for cluter analyis ?
– Rapid identification of a cluster of events
– Identification of risk factors
– Generation of disease hypotheses
(same as for temporal clustering)
Vet specific differences?
– Limited life-span
– Limited movement
– Herd characteristics
Spatial biosurveillance - Population approach?
– Baselines represent populations (i.e. from census data)
– Expect counts to be proportional to baselines
– Compare disease rate inside and outside region
Spatial biosurveillance - expectation approach?
– Baselines represent expected counts (i.e. from models)
– Expect counts to be equal to baselines
– Compare regions count to its expected count
A general inc will be missed in which approach?
Population approach -> attention to spatial scales
Example of expectation approach?
– Assume you have number of cases in each zip code
– You also have an expected mean and sd for each zip code
– Is any zip code higher than expected?
Two main problems of our zip code scenario?
– We are assuming each zip code is independent
* can’t detect cluster of adjacent zip codes → scale!!
– Multiple hypothesis testing
* Independently testing many zip codes, some will come up as alarms
just due to chance!!
* Bonferroni correction too conservative
Z score: ?
= (cases-mean)/sd
is linked to normal distribution
if we have null hypothesis?
underlying dx rate is spatially uniform
alternative?
underlying dx rate is higher inside region S than outside
