Case-Control Studies for Outbreak Investigations

Goals Describe the basic steps of conducting a

case-control study Discuss how to select cases and

controls Discuss how to conduct basic data

analysis (odds, odds ratios, and matched analysis)

Provide examples of recent outbreak investigations that have used the case-control study design

Quick Review of Case-Control Studies Analytic studies answer “what is the

relationship between exposure and disease?”

Case-control design often conducted with relatively few diseased individuals (so is efficient)

Case-control design useful when studying a rare disease or investigating an outbreak 

Case Selection Depends on how the study investigator

defines a case Case definition: “a set of standard criteria

for deciding whether an individual should be classified as having the health condition of interest” (1)

Clinical criteria Restricted to time, place, person

characteristics Simple, objective, and consistently applied

Case Selection

Sources for identifying case-patients: Medical records Laboratory results Surveillance systems Registries Mass screening programs Case-patients identify other persons

who have similar illness

Case Selection Example August 2001: Illinois Department of Health

notified of a cluster of cases of diarrheal illness associated with exposure to a recreational water park in central Illinois (2)

Local media and community networks used to encourage ill persons to contact the local health department

Case-patients asked if there were any other ill persons in their household or if anyone attending the water park with them was ill

Control Selection Most difficult part of a case-control

study! We would like to be able to

conclude that there is an association between exposure and disease in question

Way the controls are selected is major determinant of whether this conclusion is valid (3)

Control Selection (1)

Controls are persons who do not have the disease in question

Should be representative of population from which cases arose (source population)

If a control had developed the disease, would have been included as a case in the study

Should provide good estimate of the level of exposure one would expect in that population

Control Selection Sources for controls:

Same health-care institutions or providers as cases Same institution or organization as cases (e.g., schools,

workplaces) Relatives, friends, or neighbors of cases Randomly from the source population (1)

May choose multiple methods of control selection Source will depend on the scope of the outbreak May choose multiple controls per case to increase

likelihood of identifying significant associations (usually no more than 3 controls per case)

Control Selection Example Persons served by the same health-care

institution or providers as the cases August 2001: cluster of Ralstonia pickettii

bacteremia among neonatal intensive care unit (NICU) infants at a California hospital (4)

Controls were NICU infants who: 1. Had blood cultures taken during either cluster

period (July 30-August 3 and August 19-30);2. Had blood cultures that did not yield R. pickettii; and 3. Had been in the hospital for at least 72 hours.

Attempted to recruit 2 controls per case-patient

Control Selection Example Members of the same institution or

organization 2004: outbreak of varicella in a primary

school in a suburb of Beijing, China (5)

Case-control study to identify factors contributing to high rate of transmission and assess effectiveness of control measures

Controls included randomly-selected students in grades K-2 of the primary school with no history of current or previous varicella

One control recruited for each case-patient

Control Selection Example Relatives, friends, or neighbors

August 2000: increase noted in Salmonella serotype Thompson isolates from Southern California patients with onset of illness in July (6)

Preliminary interviews found many case-patients had eaten at Chain A restaurant in 5 days before illness onset

Case-control study conducted to evaluate specific food and drink exposures at Chain A restaurants

Controls were well friends or family members who shared meals with cases at Chain A during exposure period

Control Selection Example Random sample of the source population

January-June 2004: aflatoxicosis outbreak in eastern Kenya resulted in 317 cases and 125 deaths (7)

Case-control study conducted to identify risk factors for contamination of implicated maize

Randomly selected 2 controls from each case patient’s village

Spun a bottle in front of village elder’s home and walked to fifth house in direction indicated by the bottle (or third house in sparsely populated areas)

Random number list was used to select one household member

Control Selection Example

Multiple methods of control selection In waterpark outbreak in Illinois

previously mentioned, recruited 1 control per case using 3 methods (2)

Case-patients asked to identify another healthy person

Used local reverse-telephone directory based on residential address of case-patients

Canvassed local schools and community groups

Selection Bias Bias: distortion of relationship between

exposure and disease Systematic difference in way you select

your controls compared to way you select your cases that could be related to the exposure could introduce bias

Bias related to the way cases or controls are chosen for a study is ‘selection bias’

Selection Bias Example Case-patients more likely to work on lower

floors of an office building and employees on the lower floors are more likely to leave the building to go out for lunch

If control population is mostly employees from upper floors, conclude there is a real difference between cases and controls associated with eating at a local deli

But the difference is due to where they worked in the building, which resulted in how often they ate out

Selection Bias Example Outbreak at a gym and a majority of the

case-patients are females Majority of the controls are male Found an association between illness

and an aerobics class Outbreak was caused by the steam in

the sauna in the women’s locker room Relationship between illness and the

aerobics class due to the fact that women are more likely to take an aerobics class than men

Matching Validity is dependent on the similarity of

cases and controls in all respects except for exposure

“Match” cases and controls on characteristics like age and gender Matching factors should be important in

disease development, but not the exposure under investigation

Since matching variable will not be associated with either case or control status, it cannot confound, or distort, the exposure-disease association.

Analysis of data must take matching into account

Matching Individual matching (aka matched pairs)

Matches each case with a control that has specific characteristics in common with the case

Used when each case has unique and important characteristics

Group matching (aka frequency matching, category matching)

Proportion of controls with certain characteristics to be identical to the proportion of cases with these same characteristics

Requires that all cases be selected first so investigator knows the proportions to which the controls should be matched

If 30% of cases were male, would select so that 30% of controls were male

Matching Can be time efficient, cost effective, and

improve statistical power The more variables that are chosen as

matching characteristics, the more difficult it is to find a suitable control to match to the case Once a variable is used for matching, no

relationship can be discerned between this variable and the disease

Don’t match on anything you think might be a risk factor!

Individual Matching Example Outbreak of tularemia in Sweden in

2000 (8)

Selected two controls for each case Matched for age, sex, and place of

residence Identified through computerized

Swedish National Population Register (stores name, date of birth, personal identifying number, address of all citizens and residents)

Group Matching Example Outbreak of Escherichia coli

associated with petting zoo at 2004 North Carolina State Fair (9)

Recruited 3 controls for each case Group-matched by age groups (1-5 years,

6-17 years, and 18 years and older) Identified from list provided by fair

officials of 23,972 persons who purchased tickets to the fair online, at kiosks, or in malls

Conducting the Investigation

Gather demographic information and exposure histories from cases and controls

After you have collected the data you need, you can begin the analysis and calculate measures of association

Analyzing the Data

Odds ratio is calculated to measure the association between an exposure and a disease outcome

Calculating Odds Odds measure occurrence of an

event compared to non-occurrence of same event

Variables with two levels (binary variables) used to calculate an odds ratio Examples of binary variables: yes/no

responses (disease/no disease, exposed/not exposed)

Calculating Odds

Odds of exposure among cases calculated by dividing number of exposed cases by number of unexposed cases

Odds of exposure among controls calculated by dividing number of exposed controls by number of unexposed controls

An Odd Measure – How are odds different from probability or risk? In a bag containing 20 poker chips: 4 red and 16 blue… Probability is the number of times something occurs

divided by the total number of occurrences Probability of getting red is 4/20 (or 1/5 or 20%) Probability of getting blue is 16/20 (or 4/5 or 80%).

Odds are the number of times something occurs divided by the number of times something does not occur

Odds of getting red are 4/16 (or 1/4) Odds of picking blue are 16/4 (or 4/1) May refer to the odds of getting blue as 4 to 1 against

getting red Odds = probability/(1-probability)

If probability for picking red is 20%, odds are 0.20/(1-0.20) or 1/4

Probability = odds/(1+odds) If odds of picking red is 1/4, probability is

0.25/(1+0.25)=0.20

Calculating Odds

Case Control

Exposed a b

Not Exposed c d

Odds of Exposure

a/b c/d

A 2x2 table shows distribution of cases and controls:

Calculating Odds Ratios

Odds ratio is odds of exposure among cases divided by odds of exposure among controls

Exposure among cases is compared to exposure among controls to assess if and how exposure levels differ between cases and controls

Calculating Odds Ratios

Odds ratio calculated by dividing odds of exposure among cases (a/c) by odds of exposure among controls (b/d)

Numerically the same as dividing the products obtained when multiplying diagonally across the 2x2 table (ad/bc)

Also known as “cross-products ratio”

Calculating Odds Ratios To interpret odds ratio, compare value to 1:

If odds ratio = 1: odds of exposure is the same for cases and controls (no association between disease and exposure)

If odds ratio > 1: odds of exposure among cases is greater than among controls (a positive association between disease and exposure)

If odds ratio < 1: odds of exposure among cases is less than among controls (a negative, or protective, association between disease and exposure)

Calculating Odds Example Outbreak of Hepatitis A among patrons of

a single Pennsylvania restaurant (10)

240 case-patients and 134 controls identified

OR = (218/22) = (218x89) = 19.6 (45/89) (45x22)

Case Control

Exposed 218 45

Not Exposed 22 89

Matched Analysis If individual matching, 2x2 table set up differently Examine pairs in table, so have cases along one

side and controls along the other, and each cell in the table contains pairs

CONTROLS

CAS ES

ExposedNot

ExposedTotal

Exposed e f e + f

Not Exposed g h g + h

Total e + g f + h

Matched Analysis Cell e contains number of matched case-control

pairs where both case and control were exposed Concordant cell (and cell h) because case and control

have same exposure status Cell f contains number of matched case-control

pairs where cases were exposed but controls were not exposed

Discordant cell (as cell g) because case and control have different exposure status

Only discordant cells give useful data: the matched odds ratio calculated as cell f divided by cell g 

Matched Odds Ratio = f/g

Odds vs. Risk Odds are qualitatively different from risk

(calculated in a cohort study) Case-control studies select participants based on

disease status and then measure exposure among the participants

Can only approximate risk of disease given exposure Values needed to calculate risk are not available

because entire population at risk is not included in the study

Finding and accessing all who did not get sick would be difficult or impossible

Case-control study allows us to use only a subset of controls and calculate the odds ratio as an estimate of the risk

Example Case-Control Study: E. coli at fast-food restaurant November 1999: children’s hospital notified Fresno

County Health Department (California) of 5 cases of E. coli O157 infections during a 2-week period (11)

All case patients had eaten at popular fast-food restaurant chain A in 7-day period before onset of illness

Local health officials and clinicians throughout California asked to enhance surveillance for E. coli O157 infections

States bordering California asked to review medical histories of persons with recent E. coli O157 infections and arrange for subtyping of isolates

2 sequential case-control studies conducted in early December 1999

Example Case-Control Study: E. coli at fast-food restaurant First study conducted to determine the restaurant

associated with the outbreak Case defined as patient with:

An infection with the PFGE-defined outbreak strain of E. coli O157:H7, diarrheal illness with more than 3 loose stools during a 24-hour period, and/or hemolytic uremic syndrome (HUS) during the first 2 weeks of November 1999; or

Illness clinically compatible with E. coli O157:H7 infection, without laboratory confirmation but with epidemiologic connection to the outbreak

Control defined as person without a diarrheal illness or HUS during the first 2 weeks of November 1999

Example Case-Control Study: E. coli at fast-food restaurant Controls age-matched and systematically

identified using computer-assisted telephone interviewing or residents in the same telephone exchange area as case patients.

Attempted 2 controls per case Enrolled 10 cases and 19 matched controls Only chain A showed statistically significant

association with illness among cases and controls

Example Case-Control Study: E. coli at fast-food restaurant Second case-control study involving patrons of

chain A restaurants conducted to determine specific menu item or ingredient associated with illness (11)

Case defined as above but restricted to those who had eaten at chain A and who could be matched with “meal companion-controls”

8 cases and 16 meal companion-controls enrolled

Consumption of a beef taco was found to be statistically associated with illness

Traceback investigation implicated an upstream supplier of beef, but farm investigation was not possible

Example Case-Control Study: Listeriosis with deli meat July and August 2002: 22 cases of listeriosis

were reported in Pennsylvania, a nearly 3-fold increase over baseline (12)

Subtyping identified cluster of cases caused by single Liseteria monocytogenes strain

CDC asked health departments in northeast United States to conduct active case finding, prompt reporting of listeriosis cases and retrieval of clinical isolates for rapid PFGE testing

Conducted case-control study to identify cause of increase in cases

Example Case-Control Study: Listeriosis with deli meat Case-patient defined as person with culture-

confirmed listeriosis between July 1 and November 30, 2002, whose infection was caused by the outbreak strain

Control defined as person with culture-confirmed listeriosis between July 1 and November 30, 2002, whose infection was caused by any other non-outbreak strain of L. monocytogenes, and who lived in a state with at least 1 case patient

Interviewed with standard questionnaire including more than 70 specific food items to gather medical and food histories during the 4 weeks preceding culture for L. monocytogenes.

Example Case-Control Study: Listeriosis with deli meat Study obtained data from 38 case-patients and

53 controls Infection strongly associated with consumption

of precooked turkey breast products sliced at the deli counter of groceries and restaurants

Based on traceback investigation, 4 turkey processing plants investigated: outbreak strain of L. monocytogenes found in plant A and in turkey breast products from plant B

Both plants suspended production and recalled more than 30 million pounds of products, resulting in one of the largest meat recalls in US history 

Conclusion Important to keep in mind the hypothesis you are

testing Consideration of underlying population that gave

rise to cases will help select appropriate controls Improper selection of controls can introduce bias

and result in a spurious association between exposure and illness

If controls are representative of the source population, case-control studies are an efficient way to conduct an analytic study to determine the relationship between exposures and a disease

References1. Gregg MB. Field Epidemiology. 2nd ed. New York, NY:

Oxford University Press; 2002.2. Causer LM, Handzel T, Welch P, et al. An outbreak of

Cryptosporidium hominis infection at an Illinois recreational waterpark. Epidemiol Infect. 2006;134(1):147-156.

3. Gordis L. Epidemiology. 2nd ed. Philadelphia, PA: WB Saunders Company; 2000.

4. Kimura AC, Calvet H, Higa JI, et al. Outbreak of Ralstonia pickettii bacteremia in a neonatal intensive care unit. Pediatr Infect Dis J. 2005;24:1099-1103.

5. Ma H, Fontaine R. Varicella outbreak among primary school students--Beijing, China, 2004. MMWR Morb Mortal Wkly Rep. 2006;55(suppl):39-43.

References6. Kimura AC, Palumbo MS, Meyers H, Abbott S, Rodriguez R, Werner SB.

A multi-state outbreak of Salmonella serotype Thompson infection from commercially distributed bread contaminated by an ill food handler. Epidemiol Infect. 2005;133:823-828.

7. Azziz-Baumgartner E, Lindblade K, Gieseker K, et al and the Aflatoxin Investigative Group. Case-control study of an acute aflatoxicosis outbreak, Kenya, 2004. Environ Health Perspect. 2005;113:1779-1783.

8. Eliasson H, Lindbäck J, Nuorti JP, et al. The 2000 tularemia outbreak: a case-control study of risk factors in disease-endemic and emergent areas, Sweden. Emerg Infect Dis 2002;8:956-960.

9. Goode B, O’Reilly C. Outbreak of Shiga toxin producing E. coli (STEC) infections associated with a petting zoo at the North Carolina State Fair – Raleigh, North Carolina, November 2004. NC Dept of Health and Human Services: June 29, 2005. Available at: www.epi.state.nc.us/epi/gcdc/ecoli/EColiReportFinal062905.pdf.

References10.Wheeler C, Vogt TM, Armstrong GL, et al. An outbreak

of hepatitis A associated with green onions. N Engl J Med. 2005; 353:890-897.

11.Jay M, Garrett V, Mohle-Boetani JC, et al. A multistate outbreak of Escherichia coli O157:H7 infection linked to consumption of beef tacos at a fast-food restaurant chain. Clin Infect Dis. 2004;39:1-7.

12.Gottlieb SL, Newbern EC, Griffin PM, et al and the Listeriosis Working Group. Multistate outbreak of listeriosis linked to turkey deli meat and subsequent changes in US regulatory policy. Clin Infect Dis. 2006;42:29-36.