Workshop 3: Epidemiological study designs
Overview
Teaching: 10 min
Exercises: 20 minQuestions
What are the different epidemiological study designs
How do we use appropriate study design for our research
Objectives
Learn to spot different types of study design
Quick summary of epidemiological study designs
- Case series
- Case control studies
- Cross sectional surveys
- Cohort studies
- Randomised Controlled Trials (Intervention studies)
Case series
- Description or summary of cases
- Provides description of exposure OR outcomes
- Does not have any comparison group to base analysis
Case control study design
- Start with identifying people with and without disease
- cases
- controls
- Likelihood or odds of exposure for cases
- Likelihood or odds of exposure for controls
- The comparison of odds is Odds Ratio
- Open to recall bias
- Open to selection bias
- Can study multiple exposures for rare diseases
- rare disease: prevalence 1 per 10,000 or less
- Use logistic regression to assess
Cross sectional Surveys
- Over a fixed period of time or a fixed territory
- Individuals surveyed or measured as a cross-section
- Used for finding prevalence of a condition or exposure
- Exposure and outcomes are assessed at the same time
- Measures of association using cross-sectional surveys are reported as prevalence odds ratio
- Open to recall bias
- Healthy worker effect
Cohort studies
- Start with identifying individuals on the basis of their exposure
- Exposed and non-exposed cohorts
- Follow up over time
- Study the emergence of outcomes for both cohorts
- Use survival analysis and proportional hazards model to study
- Effect measure is given by Hazard Ratio or Risk Ratio
- Can study multiple outcomes for a fixed set of exposures
- Time consuming and expensive
- Can establish that causes would have preceded the emergence of outcomes
Randomised Controlled trials
- Experimental study design within health
- Randomisation = use random numbers table to assign
- Random numbers to assign people with and without intervention
- Follow up individuals with and without intervention
- Study the emergence of outcomes for both groups
- Analysis of variance is used to analyse RCTs
- Time consuming and expensive
- Limited generalisability or external validity
Challenge
In this challenge, read the following abstract(s) and identify what study design did the authors use to conduct their study. Write the study aim. Can you think of another way to study the same phenomenon? The article is from: Jasmine A. McDonald, Parisa Tehranifar, Julie D. Flom et.al. Hair product use, age at menarche and mammographic breast density in multiethnic urban women. Environ Health. 2018; 17: 1.
Background
Select hair products contain endocrine disrupting chemicals (EDCs) that may affect breast cancer risk. We hypothesize that, if EDCs are related to breast cancer risk, then they may also affect two important breast cancer risk factors: age at menarche and mammographic breast density.
In two urban female cohorts (N = 248): 1) the New York site of the National Collaborative Perinatal Project and 2) the New York City Multiethnic Breast Cancer Project, we measured childhood and adult use of hair oils, lotions, leave-in conditioners, root stimulators, perms/relaxers, and hair dyes using the same validated questionnaire. We used multivariable relative risk regression models to examine the association between childhood hair product use and early age at menarche (defined as <11 years of age) and multivariable linear regression models to examine the association between childhood and adult hair product use and adult mammographic breast density.
Early menarche was associated with ever use of childhood hair products (RR 2.3, 95% CI 1.1, 4.8) and hair oil use (RR 2.5, 95% CI 1.2, 5.2); however, additional adjustment for race/ethnicity, attenuated associations (hair products RR 1.8, 95% CI 0.8, 4.1; hair oil use RR 2.3, 95% CI 1.0, 5.5). Breast density was not associated with adult or childhood hair product or hair oil use.
If confirmed in larger prospective studies, these data suggest that exposure to EDCs through hair products in early life may affect breast cancer risk by altering timing of menarche, and may operate through a mechanism distinct from breast density.
Challenge 2
Aim: The average mercury load in children under 7-years old was determined in a populated but not overly industrial coastal area in China. Methods: 395 blood samples, 1072 urine samples, and 581 hair samples were collected from 1076 children, aged 0 to 6 years, from eight representative communities of Xiamen, China. Mercury levels in the samples were surveyed. Results: The 95% upper limits of mercury in blood, urine, and hair for the children were 2.30, 1.50 and 2100.00 μg/kg, respectively. Levels tended to increase with age. Correlation analyses showed that mercury levels in blood and urine correlated with those in hair (n = 132), r = 0.49, p < 0.0001 and r = 0.20, p = 0.0008; however, blood mercury levels did not correlate with urine levels (n = 284), r = 0.07, p = 0.35. Conclusions: Surveying the average mercury load in children 0 to 6 years, and the 95% upper limit value of mercury in their blood, urine, and hair should help guide risk assessment and health management for children.
What kind of a study is this? What are its limitations?
Challenge 3
Risk Factors for Deep Vein Thrombosis and Pulmonary Embolism John A. Heit, MD; Marc D. Silverstein, MD; David N. Mohr, MD; et al
Background Reported risk factors for venous thromboembolism (VTE) vary widely, and the magnitude and independence of each are uncertain.
Objectives To identify independent risk factors for deep vein thrombosis and pulmonary embolism and to estimate the magnitude of risk for each.
Patients and Methods We performed a population-based, nested, case-control study of 625 Olmsted County, Minnesota, patients with a first lifetime VTE diagnosed during the 15-year period from January 1, 1976, through December 31, 1990, and 625 Olmsted County patients without VTE. The 2 groups were matched on age, sex, calendar year, and medical record number.
Results Independent risk factors for VTE included surgery (odds ratio [OR], 21.7; 95% confidence interval [CI], 9.4-49.9), trauma (OR, 12.7; 95% CI, 4.1-39.7), hospital or nursing home confinement (OR, 8.0; 95% CI, 4.5-14.2), malignant neoplasm with (OR, 6.5; 95% CI, 2.1-20.2) or without (OR, 4.1; 95% CI, 1.9-8.5) chemotherapy, central venous catheter or pacemaker (OR, 5.6; 95% CI, 1.6-19.6), superficial vein thrombosis (OR, 4.3; 95% CI, 1.8-10.6), and neurological disease with extremity paresis (OR, 3.0; 95% CI, 1.3-7.4). The risk associated with varicose veins diminished with age (for age 45 years: OR, 4.2; 95% CI, 1.6-11.3; for age 60 years: OR, 1.9; 95% CI, 1.0-3.6; for age 75 years: OR, 0.9; 95% CI, 0.6-1.4), while patients with liver disease had a reduced risk (OR, 0.1; 95% CI, 0.0-0.7).
Conclusion Hospital or nursing home confinement, surgery, trauma, malignant neoplasm, chemotherapy, neurologic disease with paresis, central venous catheter or pacemaker, varicose veins, and superficial vein thrombosis are independent and important risk factors for VTE.
Decide:
- What was the study design?
- Is the conclusion justified? What do you think?
Key Points
Case series, case control studies, cohort studies are commonly used epidemiological study designs
Case control studies are great for cancer & other rare diseases
Cohort studies are great for rare exposure but common diseases