Unformatted text preview: Environmental Health
Lecture 2 Cluster Investigation
Dr. Bartlett and Dr. Geary Olsen
1 Elizabeth Lyons MSU Graduate
2 3 4 Environmental Health: Cluster Investigation A lot of what health departments do is respond to citizen complaints regarding clusters. Usually cancer clusters. They also participate in longterm studies.
5 Environmental Health: Cluster Investigation Somebody calls the health department: Three kids at our school have childhood leukemia Your vet clinic is right next to the playground, and we can smell pesticide coming from your clinic. We think the pesticide fumes are causing our kids to get leukemia.
6 Cluster Investigation Epidemiology Cluster - A number of persons, animals, or things gathered or situated close together. A cluster is a closely grouped series of events or cases of a disease, with well-defined distribution patterns, in relation to time or place (or both). Time cluster, space cluster, time-space cluster Random = happening by chance. P= probability. A P-value of .01 would mean that the probability of a event occurring by chance would be 1 in 100. (NO!)
7 Cluster Investigation Epidemiology If the null hypothesis is true, then P = .01 represents the probability that a difference as extreme as that observed (or more extreme) would occur (just by chance). With regards to clusters of disease, a P-value or Significance Level is meaningful and interpretable as a probability statement only if the observations were drawn at random from a defined population. 8 Time-Space Cluster investigation Why it is hard to use P values to determine if you have a cluster. The problem is that sometimes clusters happen "naturally" just by chance. Is your cluster due to something causing an increase rate of disease? Or is it one of these "chance" clusters?
9 10 11 What can cause a cluster? "Statistical" Clusters: - Given enough time and enough potential groupings, eventually there will be subsets of the data (a particular town, month, farm, sector, county, etc.) that may (by chance alone) have a higher rate of a disease than the entire population. Illinois Subdivision
12 A posteriori vs. A priori "After the fact" vs. "before the fact" What is the probability that, through the eons of time, I would be standing here before you with two arms, two eyes, 10 fingers and only one nose? The probability is 1.000.... (It has already happened!) But that's not the point. The question was posed after observing the data. If I had 8 fingers, I would have asked a different question. When the question is posed based on what you observe in the data, then p values from statistical 13 tests of association are no longer valid. Random is a process, not a result. Consider the tables of random numbers. Look at about 100 of them until you see one that somehow doesn't look random. 14 15 16 17 18 Statistical test of association Chi-square goodness of fit test Expect 10% of the numbers to be "7" 110 number in the cluster Would expect 11 to be "7" Observed 25 to be "7" P = .01 but is it valid? 19 Texas sharpshooter
Texas sharpshooter who shoots at the side of a barn and then draws a bull's eye around the bullet hole. If you define a cluster (draw the bull's eye) based on what you observe in the data (the bullet hole), then statistical tests can not be used to confirm the existence of the cluster (time-space association with a particular risk factor). Why are you studying "7" disease? Why are you studying it here?
20 "Statistical" Clusters:
Given the large number of diseases and risk factors (employment, organizations, housing location, etc), some will appear to be associated with disease just by chance alone. "Given enough time, it is probable for the improbable to happen." Albert Einstein 21 Other causes for clusters Biological clusters - are clusters of disease which have a biological basis. This is what we are looking for! Confounding (and time-space clusters) Legionnaire's Disease in Michigan Reporting Bias (and time-space clusters) Rabies hysteria or apathy
22 Example: www.RUSick2.msu.edu For foodborne outbreaks: Statistical Clusters: Given enough food items and enough time, clusters of foodborne disease will occur which do not have a biological basis and do not represent common-source foodborne outbreaks. Pranksters or malicious intent Confounding Strawberries every spring Restaurant next to "an event" 23 Cluster Investigation Cancer clusters- these cannot be investigated like acute infectious disease clusters. 1. Long and indefinite incubation/induction period for the disease. 2. Routes of the cancer causing agents are usually through the environment not through personal contact, or consumption of food or beverage. "Hot pursuit" case control studies usually done in acute infectious disease outbreaks are not useful in cancer cluster analysis. 24 Cluster Investigation The following list of characteristics can help to identify a situation where a case-control study or a multicommunity investigation might be useful. 1. There must be at least five cases to a cluster and they must have a high relative risk (RR). 1. What is high? 2.0? 10.0? 2. A unique and well known etiological agent is known to be the cause and the pathophysiologic mechanism for that agent is known. 3. The agent is in the environment and can be measured there. 25 4. The agent is persistent in the infected/exposed people but rare in normal populations and its unique physiologic response in the exposed can be measured. 5. There is a heterogeneity of exposure (range from high to low) within the neighborhood so effects can be easily measured. 6. The route of exposure can be easily recalled. 7. Multi-community studies can be done by looking at (otherwise) similar exposed and unexposed communities. 8. Endemic space cluster, not a space-time cluster that exists for a while then vanishes.
26 Cluster Investigation
Minnesota Cluster Analysis Track Record 1. Information and education- 95% 2. Public initiated surveys- 4% 3. Validation, evaluation, feasibility and education- 1% 4. In-depth study- <1% 27 28 Cluster Investigation
Evaluation of false positive reports is the price we must pay in order to identify the true biological cluster. Examples: The first few AIDS cases Foodborne outbreaks of E.coli 0157:H7 Most foodborne outbreaks (Note: These are all infectious agents!) Bitter Harvest (PBB) 29 ...
View Full Document
- Fall '11
- Epidemiology, Cluster Investigation Epidemiology, cluster investigation