Descriptive and analytic study types Epidemiology Fall semester 2007
Cross sectional surveys
Descriptive & analytic epidemiology II Case-control studies
A patient series Carcinoma of the penis and cervix “… Case 3. – Presented with 5-year history in November, 1969, aged 47. He had massive penile condylomata with squamous carcinomatous change and invaded ingual nodes. Died in 1977. His wife presented with carcinoma of the cervix in 1971 at the age of 43. She had a squamous cell carcinoma and stage III disease. Died 27 months later.”
Cartwright and Sinson, 1980; Lancet: 1: 97
Occurrence of gonorrhoea England and Wales
In correlational studies, measures that represent characteristics of entire populations are used to describe disease in relation to some factors of interest such as age, calendar time, utilization of health services or consumption of a food, medication or other product. (H&B p 102) Occurrence Correlation (0 < r < 1) No correlation (r = 0) Correlation (-1 < r < 0)
Gonorrhoea incidence per 100,000 women
Calendar year Beral, Lancet, 25 May 1974
Exercise Birth cohort England Scotland
Measure of mortality from cervical cancer in England/Wales & Scotland per birth cohort Adapted from Beral, Lancet, 1974
1902-6 1907-11 1912-16 1917-21 1922-26 1927-31 1932-36 1937-42 1943-47 1948-52
• Insert data for cervical cancer mortality in hand-outs • Discuss what is shown in the figure • Can this method be used to test hypotheses?
Kaposi’s sarcoma Fig. 1. Kaposi’s sarcoma and nonHodgkin’s lymphoma incidence among men, per 100 000 people per year, agestandardized to the 1970 U.S. population, shown on a linear and log scale to illustrate both the absolute and relative changes in nine Surveillance, Epidemiology and End Results (SEER) registries and in the San Francisco area registry only, from 1973 through 1998. Years with no cases were set arbitrarily at 0.12 cases in the log scale.
“The presence of a correlation does not necessarily imply the presence of a valid statistical association. Conversely, lack of a correlation in such studies does not necessarily imply the absence of a valid statistical association.”
(H&B p. 104)
Helicobacter pylori antibody prevalence Greenland 1998
Cross sectional surveys
”A third type of study is the cross-sectional or prevalence survey, in which exposure and disease status are assessed simultaneously among individuals in a well-defined population.”
(H&B p. 108)
Koch et al. 2004
HP seroprevalence pattern childhood 100 90 80 70 60 50 40 30 20 10 0
Risk factors for HP seropositivity (adjusted), Greenland N (% seropositive)
Ethiopia* Greenland Sweden*
No. of children in household
* Lindkvist et al. 1996
(0.75 – 11.2)
(0.38 – 6.07)
(1.52 – 26.4)
(0.37 – 5.58)
(1.45 – 50.7)
(0.1 – 3.37)
(0.3 – 5.68)
(0.03 – 0.9)
No. of older siblings 0
9 10 11 12 13 14 15
• • •
At what age are Greenlanders infected? At what age are children in other countries infected? What are possible causes of the differences?
Differences in infection patterns Cohort effect
Distance to nearest older sibling
Risk factor assessment in correlational studies
Cross sectional surveys
• Measure association at time of study • Information on prior exposures may be biased • Time of outcome (seropositivity) unknown
Healthy Diseases Prevalence
80 20 20%
95 5 5% (H&B p. 110)
Cross sectional surveys Job
Cross sectional studies
• Studies, in which outcome and exposure are determined simultaneously • The observed outcomes are prevalent
A Healthy Diseased Prevalence
80 10 11%
• Data on risk factor associations will accordingly represent both survival and etiology
• It cannot be ruled out that the exposure under observation has changed after and maybe because of the outcome • It cannot always be determined which came first, the exposure or the outcome
95 15 14%
• Cannot test causation, but suggest associations to be tested in analytical studies (H&B p. 110)
Descriptive & analytic epidemiology
Descriptive epidemiology Advantages • Cheap & quick • May provide important overview Disadvantages • No information on the individual • No control for confounding • May involve bias • Results may be ambiguous • Can not test (causal) hypotheses
Analytic epidemiology Disadvantages • Expensive • Laboreous • May involve bias Advantages • Information on the individual • Control for confounding • Results less ambiguous • Can test (causal) hypotheses
Analytical studies Case-control studies
How are you going to test, if
• Definition of case-control studies
• Amyl nitrite (’Poppers’) is the cause of Kaposi’s sarcoma?
• Know the difference between case-control and cohort studies
• Hospitalisation with hip fracture is the cause of lung embolus?
• Describe odds ratio
• Smoking is the cause of lung cancer?
• Know the principles for selection of cases and controls
• Use of childcare centres is a cause of respiratory tract infections?
– Case-control – Case-control
– Cohort study
• Describe advantages and disadvantages of CCstudies
• Viagra acts against erectile dysfunction? – Randomised controlled study
The question (hypothesis) determines the method
Analytical study types
• The method dependes (among others) of
• Determination of causes and effects
– – – –
Type of disease Frequency of disease Characteristics of affected persons Diagnostic methods
• Observational – Cohort studies – Case-control studies
• Interventional studies – Randomised, controlled studies
Outcome in cohort study: Relative risk
Cohort studies • Cohort: Cohors (latin): 1/10 of a legion • Prospective (!) • Starting point a population of healthy
TIME Sick Population
Persons without disease
A A Relative risk = + B C C+D
Healthy Sick Healthy
With other words: Cohort studies measure
Case control studies
• Risk of disease among the exposed compared with the risk of disease among the non-exposed TIME
• The absolute risk may be calculated for both groups!
Exposed Non-exposed Exposed Non-exposed
Cases (with disease) Population Controls (without disease)
Why case control studies?
• If exact information of exposure and cases in a complete population is not available
• Became popular with the change from infectious disease epidemiology to chronic diseases. Why?
• Rare disease • Long disease latency • Fewer persons necessary to show effect
– – – –
Western life-style diseases (cancer, heart diseases) Diseases with long latent period Most applicable when disease is rare Study many possible risk factors / causes
• Today the most used analytical study type in epidemiology
Retrospective – prospective?
• CC-studies are often referred to as synonymous with retrospective studies, and cohort studies as prospective. Correct?
All incident cases (why incident?)
• No, retrospective refers to if all cases are identified at time of study start, which is the most usual in CCstudies
End of study
• However, do prospective CC-studies exist?
Prospective/retrospective refers to time of registration of cases
Outcome in case-control study: Odds ratio
Calculation of odds ratio
• Odds: measure of frequency of exposure in group • Measure of association: if the exposure is a cause of disease, then sick persons (cases) should be exposed more often than controls!
• 2x2 table • OR = a/c / b/d = ad/bc • P = Chi-square test/Fisher’s exact test
• Odds have no unit Disease • Odds among cases = Number of cases exposed to risk faktor Number of cases not exposed • Odds among controls = Number of controls exposed to risk faktor Number of controls not eksposed
• Odds ratio: Odds for cases/odds for controls
Calculate odds ratio:
Observation by chance?
• OR • • • •
= a/c / b/d
Odds for cases = 10/40 Odds for controls = 80/20 OR = 0.25/4 Interpretation in words?
• Could an OR of 0.0625 occur by chance (’is it significant’)? • A X2 test tests if the observed numbers differ (significantly) from the expected numbers • What are the expected numbers in each cell? Disease
= ad/bc = ??? = 0.25 =4 = 0.0625
Observed/expected numbers Disease
Χ2 test a
If Χ2>3.84, then p<0.05!
• Does the observed deviate from the expected (’is this significant’)? What is Χ2?
Odds ratio in practice. Salmonella in Wales 1989
Salmonella outbreak in Wales 1989
31 cases office workers 6 cases canteen staff In total 37 cases Hereof 3 attended doctor, other identified through interviews or faecal tests
Odds ratio • Odds =
50 ~ p < 0.0001
All risk factors in Wales outbreak
Number of persons exposed Number of persons not exposed OR
• Odds for having eaten in the canteen January 22 for cases = 6/31 = 0.193
• Odds for having eaten in the canteen January 22 for controls = 9/48 = 0.188
• Odds ratio = Odds for cases/odds for controls
• Odds ratio for having eaten in the canteen at January 22 = 0.193/0.188 = 1.03
Odds ratio vs. relative risk
OR <> RR
• Why is relative risk not used in case-control studies?
• 1.400 in building, 37 cases • How many got sick out of those having eaten in the canteen?
Lunch Jan. 22
6/15 / 31/79
= 1.02 ???
• Number of eating and number of sick unknown • Sample of a population • Therefore the rate (=risk) cannot be calculated in casecontrol study But
• Because the calculation is nonsense! • If disease is rare, then OR ~ RR
If disease is rare (a and c small), then OR ~ RR Disease
a (a + b) c (c + d)
6 (6 + 19,781) 19 (19 + 35,313)
a b c d
• As we have only got a random sample and not a whole population, we must calculate a measure of how certain our estimate (OR) is: the confidence interval • The values between which the ’true’ population estimate with 95% confidence is found
Generalisability – Confidence intervals
6 x 35,313 19,781 x 19
= = 0.56
Lunch Jan. 22
0.33 – 3.18
Lunch Jan. 23
1.21 – 7.09 1.65 – 16.4
0.99 – 5.8
0.38 – 7.01
With other words: Case-control studies measure
• The extent of exposure among the sick compared with the extent of exposure among the healthy
• Temporal associations between exposure and outcome difficult to evaluate INFORMATION
• Odds ratio expresses this
• Odds ratio is not the same as risk, as the risk in the population is unknown (a sample of the population is drawn),
Cases (with disease) Population Controls (without disease)
• Measurement methods back in time maybe not relevant today (HPV)
But • If the disease is rare (relatively), then OR ~ RR!
Finding cases, examples
• • • • •
• Hospital source
Demands precise definition Time, place, and person Colon cancer in DK 1973-88 Myocardial infarction among 60-70 old males in DK 1973-88 Salmonella in Wales – >3 loose stools/day between January 22 and 26. – Stayed in office building in Wales between January 22 and 26.
– Easy, but bias possible
• Certain localisation (restaurant outbreak) – Evident in actual case
• Toxic shock syndrome – Fever, rash, scalded skin, hypotension, involvement of 3 or more organ systems (GI, muscles, mucous membranes, urinary tract, liver, blood, CNS), negative tests for various other mikrobes than staphylococci
• Population source (lung cancer, register) – Often costly, but used in DK because of good registers
• Working definition, may be refined during study work up (ex. SARS) • Likely, possible
Prevalent or incident cases?
• Must cases reflect all persons with the disease?
Number of AIDS-cases in DK 1989 – Prevalence, measure of disease burden
Number of newly diagnosed AIDS-cases in DK 1989 – Incidence, measure of risk
In a CC study of risk of AIDS, what measure to use?
• Myocardial infarction – All cases in Copenhagen County 1989, or – Males 45-74 år hospitalised 1989 on Herlev Amtssygehus?
By including both incident and prevalent cases risk factors and factors determining disease course (cause/prognosis), and interpretation is difficult
• Big difference in biology (familiar hypercholesterolemia / calcification of blood vessels)
The hen and the egg: coffee may be a risk factor for gastric ulcer, but if you have a gastric ulcer, you drink less coffee because of stomach pains
• Validity most important, not generalisability!
Important that exposure precedes outcome, therefore use incident cases
Choice of controls
Example of control selection
• Crucial point - problematic!
• Question: do certain genetic polymophisms result in an increased risk of serious bacterial infections in childhood?
• Must reflect the question whether the frequency of an exposure observed among cases is different than that among comparable individuals without the disease • A representative sample of the population that the sick persons come from, must have the same risk of exposure as cases • Must be selcted at random from the population (randomised) • A control is a case without the disease
• Cases: children hospitalised at least once with a serious bacterial infection (meningitis, septicaemia, etc.) before age 2 years, born in DK by Danish parents and of normal birth weight and birth length, without concurrent diseases and having lived constantly in DK before age 2 years • Controls? – The same, just without any hospitalisation for the same diagnoses
Types of controls
• Hospital controls • Population controls • Special groups
– Friends, family members, spouses, neighbours
• Advantages and disadvantages by all types
– Easy to find – Minimize recall bias (what is that and why??) – Subjected to the same specific and unspecific factors that made the cases attend this hospital
• Con’s – Sick by definition, do not represent the distribution of exposures in the background population (e.g. they smoke and drink more) – Yields a biased estimate (in what direction?) – In a study of smoking and bladder cancer many smokers among controls resulting in ’dilution’ of estimate (weaker or negative effect) – In a study of coffee and bladder cancer less coffee drinkers among controls, resulting in enhancement of estimate (stronger or positive effect) – Controls from Frederikssund County Hospital vs. Steno Diabetes Center?
What categories of patients may be used as controls?
• Controls to lung cancer patients, patients with
• Typically, when cases come from a particular population
– – – – –
Bronchitis? Heart diseases? Hip fractures? Stomach ulcers? Asthma?
• Examples – Households – Random digit dialing – Registers/voters lists – CPR
• The diseases of the controls may be associated with the risk factors under study (positively/negatively), which is not desirable
– Larger expenses – Hard to get hold of people (working, not at home – selection bias) – Recall bias – Less motivation – Problems with random digit dialing in the USA?
More control groups?
• Neighbours, friends, family
• Ideally one per case group, but sometimes desirable with more groups. When?
• Advantages – Cooperative – Confounder control (how?)
• Disadvantages – More alike cases (result?) – Dilution of estimate
• When no ideal control group can be selected (e.g. patient groups) • Breast cancer patients: Gynaecological cancers, noncancer gynaecological patients, emergency operations
Number of controls per case?
Finding cases and controls
• 1:1 best • What is gained by more controls per case? • If cases are hard to find, increased statistical strength
• Avoid bias
But • 1:3 (=1 case + 3 controls) less statistical strength than 2 + 2 (2:2) • Max. 1:4 (can be shown statistically), more controls waste of time and money
• Example: Study of PKU-cards – Controls card before and after case card in the box in he freezer – Advantages? – Can all cards be used as controls?
Information on exposure
PARTICULAR STUDY TYPES: NESTED CASE-CONTROL
• Numerous possibilities
• A case-control study ’nested’ into a cohort
– – – –
Registers Hospital files Telephone interviews Etc.
• CC-study of Hodgkin’s lympoma and birth weight – Cases interviewed about birth weight in hospital – Controls information on birth weight from the Central Birth Register – OK?
• No, information must be obtained in the same way and from the same place/source from cases as well as controls, otherwise risk of bias
• Eg. A study of genetic factors and myocardial infarction: • Cohort: Østerbro-undersøgelsen (thousands of persons) • Case-control: 200 cases and 200 controls from the cohort having genetic analyses done • Why? Expensive or difficult to carry out genetic analyses on all subjects in the cohort or difficult to obtain detailed information for all
DENSITY CASE-CONTROL STUDY
• Controls selected to represent proportion of person-time for exposed an non-exposed controls
• Determines effect of two or more interventions
• Knowledge of person-time of cases necessary
• Each case receives all interventions and acts as his/hers own control
• The chance of being seleted as control is proportional to person-time experience in the source population
• Only relevant when exposure varies from time to time within a person
• If sampled properly, the Odds ratio from a density case control study design gives an estimate of incidence rate in study population
• Evaluates exposures that trigger short-term effects (mobile phones and traffic accidents, sexual intercourse and myocardial infarction)
ALTERNATIVE: CASE-COHORT STUDY • Controls have the same chance of being selected irrespectively of person-time spent • Fraction of total number of people in study population rather than person-time • Gives an estimate of risk ratio instead of incidence rate as in density CC study • Control may also be a case!
Take home messages • If exact information of exposure and cases in a complete population is not available, a case-control study may be carried out • Well suited in case of rare diseases, long latency time and multiple risk factors • Cheap and effective – the most frequent analytical study design today • Association measured in Odds ratio (odds among cases divided by odds among controls) with confidence intervals to express the statistical uncertainty • Selection of control group difficult – controls should ideally be cases who just haven’t developed the disease • Causes hard to determine, only associations
Next lesson: pitfalls and challenges in case-control studies…