Cross-sectional studies

Christina-Evmorfia Kampitsi
Postdoc, PhD in Epidemiology, MMedSc (Public Health), Registered Dietitian
Institute of Environmental Medicine

Acknowledgements: Giorgio Tettamanti, Hanna Mogensen

Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 2

Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 3

Evidence hierarchy

Systematic Reviews
& meta-analyses

RCTs

Cohort studies

Case-control studies

Cross-sectional studies

Case series, Case reports

Ideas, opinions, editorials, anecdotal

Karolinska Institutet - a medical university 4

Cross-sectional study

• A study that observes the association between exposure and outcome

at a certain point in time

• Exposed and unexposed are identified and outcome is measured in the

sample at the same time

A snapshot in time

Karolinska Institutet - a medical university 5

A snapshot in time

Unexposed

Exposed

Sick

Karolinska Institutet - a medical university 6

Cohort study (open cohort)
= Outcome+

No. of persons

Time
Karolinska Institutet - a medical university 7
Case-control study
= Case

No. of persons
= Control

(incidence density sampling) Time

Karolinska Institutet - a medical university 8
Cross-sectional study
= Outcome+

No. of persons

Time
Karolinska Institutet - a medical university 9
Cross-sectional study

• Also called prevalence study

→ The disease prevalence is compared between groups of different exposures

• Often used to investigate chronic diseases

• Exposure is assessed simultaneously with the disease, therefore we cannot

make assumptions about causation

Karolinska Institutet - a medical university 10

Examples of “standard” cross-sectional studies

• Identify all men and women aged 18-45 living in the Stockholm county in
May 2023
• Send them a questionnaire about:
→ Body size
→ Diabetes
→ Depression
→ Diet
→ Smoking
→ Alcohol consumption
• RQ: Is smoking associated with depression? Are those who smoke more
likely to have depression compared to those who do not?
Karolinska Institutet - a medical university 11
Examples of “standard” cross-sectional studies

• Identify all men and women aged 18-45 living in the Stockholm county in
May 2023
• Invite them to a clinic to measure their blood pressure
• Send them a questionnaire about:
→ Body size
→ Diabetes
→ Diet
→ Smoking
• RQ: Is diabetes associated with blood pressure? Are those who have
diabetes more likely to have high blood pressure compared to those who do
not?
Karolinska Institutet - a medical university 12
Examples of “standard” cross-sectional studies

• Standard questions in a “standard” cross-sectional study

→ Do you smoke at least 1 cigarette daily?
→ Do you drink alcohol on a weekly basis?
→ Do you eat vegetables daily?
→ Do you have diabetes?
→ Do you have depression?
→ Do you…

Karolinska Institutet - a medical university 13

Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 14

Measures of disease occurrence
Healthy Population Total Population
Cases

Incidence Cases

Prevalence
Example
How common is diabetes in Sweden?
- The number of Swedes who got diabetes in 2023 Incidence
- The proportion of the Swedish population with diabetes at the end of
2023 Prevalence
Karolinska Institutet - a medical university 15
Measures of disease occurrence

Recovery
Incidence

Prevalence

Death
(Mortality)
Karolinska Institutet - a medical university 16
Measures of disease occurrence

• Prevalence
• Cumulative incidence The only measure of
disease occurrence in a
• Incidence rate cross-sectional study

Karolinska Institutet - a medical university 17

Exercise: Suitable health outcomes for cross-
sectional studies
• Which health outcomes might be suitable for investigating with a cross-
sectional study and why?
→ Depression?
→ Sleeping problems?
→ Childhood cancer?
→ Mortality?
• Which measure of disease occurrence would be used?
→ Incidence?
→ Prevalence?

Karolinska Institutet - a medical university 18

Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 19

Biases in epidemiological studies

• Cohort studies
→ Confounding
→ Exposure and outcome misclassification: exposure misclassification likely non-
differential
→ Selection bias: predominantly loss to follow-up
• Case-control studies
→ Confounding
→ Exposure and outcome misclassification: differential exposure misclassification
can occur (recall bias)
→ Selection bias: often different participation rates among cases and controls (also
related to exposure!)
• Cross sectional studies?
Karolinska Institutet - a medical university 20
Design advantages

• Cost-efficient

• Fast data collection

• Can measure under-diagnosed cases not captured in registers

→ High blood pressure
→ Diabetes
→ Psychiatric disorders (anxiety, depression)

Karolinska Institutet - a medical university 21

Design disadvantages

• Vulnerable to selection bias

• Disease-related modification of exposure (reverse causation)

• Length bias sampling

Karolinska Institutet - a medical university 22

Selection bias in etiological studies: which
statement is correct?
• Selection bias occurs if there is a difference in response rate between
exposed and unexposed

• Selection bias occurs if there is a difference in response rate in relation

to BOTH exposure and outcome status

Karolinska Institutet - a medical university 23

Selection bias affecting the OR?

Exposure: late evening screen time | Outcome: sleeping problems

D= 0 D=1
E=0 100 100 True OR = 3.0
E=1 50 150

Participation rate: 50% among unexposed; 80% among exposed

D= 0 D=1
E=0 50 50 Observed OR = 3.0
E=1 40 120
Karolinska Institutet - a medical university 24
Selection bias affecting the OR?

Exposure: late evening screen time | Outcome: sleeping problems

D= 0 D=1
E=0 100 100 True OR = 3.0
E=1 50 150
Participation rate: 90% among exposed or people with the disease; 50%
among healthy unexposed
D= 0 D=1
E=0 50 90 Observed OR = 1.7
E=1 45 135
Karolinska Institutet - a medical university 25
Selection bias affecting the OR?

Exposure: late evening screen time | Outcome: sleeping problems

D= 0 D=1
E=0 100 100 True OR = 1.5
E=1 80 120
Participation rate: 90% among exposed or people with the disease; 50%
among healthy unexposed
D= 0 D=1
E=0 50 90 Observed OR = 0.8
E=1 72 108
Karolinska Institutet - a medical university 26
Likely reasons leading to selection bias

• Exposed individuals:
→ Want to participate (irrespective of outcome status)
→ Accurate description of outcome frequency among exposed

• Unexposed individuals:
→ Less likely to participate
→ Those who participate more likely to have the outcome, since they are more
motivated = overestimation of outcome frequency among unexposed

• This can lead to a (false) negative association between exposure and

outcome, but different scenarios are possible!
Karolinska Institutet - a medical university 27
Selection bias in cross-sectional studies

• In case-control studies, the disease of interest is clear

→ Case-control study on breast cancer

• In cross-sectional studies, it is often not

→ Cross-sectional study on common diseases and common exposures

Karolinska Institutet - a medical university 28

Reverse causation

• Also called disease-related modification of exposure

• Time order of events: what came first?

• Exposure Disease OR Exposure Disease

• The disease itself might make the individuals modify their exposure!

Karolinska Institutet - a medical university 29

Reverse causation
Cross-sectional study on the association between late evening screen-time
and sleeping problems

Results: higher proportion of cases among individuals who do not use screens
during late evenings
But: Those with sleeping problems may avoid screens in late evenings!

Results: higher proportion of cases among individuals who do use screens

during late evenings
But: Those with sleeping problems may use screens in late evenings because
they cannot fall asleep!

Karolinska Institutet - a medical university 30

Reverse causation
Cross-sectional study on the association between air pollution and asthma in
children

Results: higher proportion of asthmatic children in areas with low residential

pollution
But: Families with an asthmatic child may choose to move to less polluted
areas!

People might change their exposures/behaviors due to the disease!

Karolinska Institutet - a medical university 31

Length bias sampling

• Since cross-sectional studies are a snapshot in time

→ More likely to capture cases with long duration of the disease
→ Less likely to capture cases with short duration of the disease

• Length bias sampling: exposure related to the duration of the disease and
not the cause of the disease

• Occurs in studies of prevalent cases

• Biased estimates of the association between exposure and outcome

Karolinska Institutet - a medical university 32

Length bias sampling
No. of persons Cases
= Long duration
4 cases = Medium duration
= Short duration

2 cases The same incidence but different

prevalence!

What if exposure is related to the

duration of the disease but
1 case does not cause the disease?

Karolinska Institutet - a medical university

Time 33
Length bias sampling

• If the exposure does not alter the disease risk but causes the disease to be
prolonged when contracted, the prevalence of the exposure will be elevated
among cases
→ A positive exposure-disease association will be observed even though exposure
has no true effect on disease risk
• If the exposure does not alter the disease risk but causes the disease to be
rapidly fatal when contracted, then the prevalence of the exposure will be
very low among cases
→ A negative exposure-disease association will be observed even though exposure
has no true effect on disease risk

Karolinska Institutet - a medical university 34

Exercise: what sort of bias?

• Researchers wanted to examine whether hairdressers were at increased risk

of allergy. They conducted a cross-sectional study by sending out a
questionnaire about allergic symptoms to all individuals in their study
population, which included hairdressers, teachers, and bank personnel in a
certain geographical region. The results showed that prevalence of allergy
was lowest among hairdressers.
• Which bias could have played a role here?

Karolinska Institutet - a medical university 35

Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 36

Cross-sectional study analysis

• Binary outcome
→ Logistic regression: Odds ratios

• Continuous outcome
→ Linear regression: Beta coefficients

• Categorical outcome
→ Multinomial/ordered logistic regression
• Rarely used
• More difficult to interpret

Karolinska Institutet - a medical university 37

Logistic regression

• Calculating (prevalence) Odds Ratios

• A very common way to analyze cross-sectional data
• Requires a binary outcome variable
• Example:
→ Exposure: consuming at least one soda can/day, yes/no
→ Outcome: Diagnosis of diabetes, yes/no

Odds of disease among exposed

Odds ratio =
Odds of disease among unexposed

Karolinska Institutet - a medical university 38

Linear regression

• Used when outcome is continuous

• Examples:
→ Blood pressure
→ Cholesterol levels
• Calculating beta coefficients
• Compare differences of means among different exposure groups

Karolinska Institutet - a medical university 39

Exercise: calculate a prevalence OR

• A study assessed the association between smoking and high blood pressure
• Calculate the prevalence OR and interpret the results

High BP Normal or low BP

Smokers 200 800
Non-smokers 100 900

Karolinska Institutet - a medical university 40

Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 41

Is a cross-sectional study the same as an ecological
study?
• NO!

• Cross sectional study: individual data on exposure and outcome

• Ecological study: only group level data on exposure and outcome

Karolinska Institutet - a medical university 42

Example of an ecological study

Chocolate
Consumption,
Cognitive Function,
and Nobel Laureates
NEJM, 2012
Karolinska Institutet - a medical university 43
Is chocolate consumption associated with winning
the Nobel prize? A cross-sectional study
• Questions
→ How many grams of chocolate do you eat on average every week?
→ Have you ever won the Nobel prize? (yes/no)
• How will we analyze it?
→ Logistic regression
• Association (on the individual level)?
→ Likely not…
• Causation?
→ Nope!
• Other problems?
→ Rare outcome!
Karolinska Institutet - a medical university 44
Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 45

When is a cross-sectional design appropriate?

• Descriptive studies
→ Prevalence of diabetes
• Relatively common chronic diseases
→ High blood pressure, diabetes
→ Not appropriate for brain tumors or pancreatic cancer
• Relatively common exposures
• Stable exposures (exposures and outcomes with built-in temporality)
• Heritability studies (twin studies)

Karolinska Institutet - a medical university 46

Example: The Stockholm Public Health Cohort

• Public health surveys of 2002, 2006, 2010 and 2014

• Study population: Population of Stockholm county >=18 years

Karolinska Institutet - a medical university 47

Standard cross-sectional study

• Aim: study the effect of smoking on sleeping problems

• Smoking: current smoking

• Sleeping problems: current sleeping problems

• Not possible to evaluate whether smoking is a risk factor for sleeping

problems

Karolinska Institutet - a medical university 48

Non-standard cross-sectional study

• Aim: study the effect of smoking on sleeping problems

• Smoking
→ Have you ever smoked?
→ Are you currently smoking?
→ When did you start smoking?
• Sleeping problems
→ Have you ever had sleeping problems?
→ Are you currently having sleeping problems?
→ When did your sleeping problems start?
• Now we can determine temporality and it is possible to evaluate whether
smoking is a risk factor for sleeping problems
Karolinska Institutet - a medical university 49
Cross-sectional studies to create a cohort study

• The Stockholm Public Health Cohort Study

→ Participants were followed up with questionnaires in 2007 and 2010
→ With these different surveys a cohort study is conducted

Karolinska Institutet - a medical university 50

Cross-sectional studies to create a cohort study

• Cross-sectional study Stockholm Public Health Survey of 2002

→ Do you smoke?
→ Have you ever had sleeping problems?
• Remove those with sleeping problems at baseline (2002)
• Follow-up questionnaire in 2007
→ Do you currently have sleeping problems?
• RQ: Is smoking in 2002 associated with sleeping problems in 2007?
• Is this a cohort study?
→ NO! We cannot identify those who got the disease between 2003-2006 (who
don’t still have it at 2007)
→ It is a cross-sectional study that is not affected by reverse causation
Karolinska Institutet - a medical university 51
Cross-sectional studies to create a cohort study

• Cross-sectional study Stockholm Public Health Survey of 2002

→ Do you smoke?
→ Have you ever had sleeping problems?
• Remove those with sleeping problems at baseline (2002)
• Follow-up questionnaire in 2007
→ Do you currently have sleeping problems?
→ Have you had sleeping problems in the last 5 years?
• RQ: Is smoking in 2002 associated with sleeping problems in 2007?
• Is this a cohort study?
→ Yes! We can identify all cases of sleeping problems between 2002 and 2007 (if
not lost to follow-up)
Karolinska Institutet - a medical university 52
Cross-sectional studies to create a cohort study

• Cross-sectional study Stockholm Public Health Survey of 2002

→ Do you smoke?
→ Do you have diabetes?
• Remove those with diabetes at baseline (2002)
• Follow-up questionnaire in 2007
→ Do you have diabetes?
• Is this a cohort study?
→ Yes! Diabetes is a chronic disease (we identify all cases of diabetes between
2002 and 2007 if not lost to follow-up)

Karolinska Institutet - a medical university 53

Example: STAGE (Study of Twin Adults: Genes and
Environment)
• Study created in 2005: Questionnaire

• Study population: All twins in the Swedish Twin Registry, born between 1959
and 1985

• 1300 questions regarding common complex diseases and common

exposures

Karolinska Institutet - a medical university 54

Sections of the STAGE questionnaire

Karolinska Institutet - a medical university 55

Stage: Prevalence studies

Karolinska Institutet - a medical university 56

Stage: Genetic studies

• Twin concordance
• Heritability studies
→ Genetic and environmental effects on same-sex sexual behavior: a population
study of twins in Sweden
→ Genetic susceptibility to burnout in a Swedish twin cohort
→ Genetic and environmental influences on adult attention deficit hyperactivity
disorder symptoms: a large Swedish population-based study of twin

Karolinska Institutet - a medical university 57

Exercise: Which study design would you use?

• Do wealthy countries have a higher average life expectancy?

• How many heroin users have hepatitis B in New York?
• How many heroin users developed hepatitis B in New York in 2023?
• Is there an association between having a public health law on wearing
seatbelt and death rates in traffic accidents?
• Which risk factors are associated with lung cancer?

Karolinska Institutet - a medical university 58

Outline

• What is a cross-sectional study?

• Measures of disease occurrence
• Biases in cross-sectional studies
• Analyze a cross-sectional study
• Cross-sectional studies vs ecological studies
• When are cross-sectional studies useful?
• Summary

Karolinska Institutet - a medical university 59

Summary: cross-sectional studies

• Snapshot in time, exposure and outcome measured simultaneously

• Cost-efficient, fast data collection, can measure under-diagnosed diseases
not captured in registers
• No temporality, can only calculate prevalence, vulnerable to biases
• Selection bias, reverse causation, length bias sampling
• Often analyzed with logistic regression calculating prevalence ORs
• Appropriate for descriptive studies, relative common exposures, stable
exposures, relative common chronic diseases

Karolinska Institutet - a medical university 60