Lecture 1 & 2: Introduction & Grand writing and brief refresher
Cross-sectional study:
- Estimate exposure and outcome at one time point
- Association measure: Prevalence Proportion Ratio (PPR)
- Disadvantage: reverse causality
Prospective study / cohort study:
- Exposure assessed before outcome, exposure is by choice or circumstance
- Measure of association: incidence proportion ratio (IPR) or incidence rate ratio (IRR)
o Incidence proportion (IP)/cumulative incidence (CI) =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑜𝑝𝑙𝑒 𝑤ℎ𝑜 𝑑𝑒𝑣𝑒𝑙𝑜𝑝 𝑑𝑖𝑠𝑒𝑎𝑠𝑒 𝑖𝑛 𝑎 𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑒𝑑 𝑝𝑒𝑟𝑖𝑜𝑑
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑜𝑝𝑙𝑒 𝑎𝑡 𝑟𝑖𝑠𝑘 𝑜𝑓 𝑔𝑒𝑡𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 𝑑𝑖𝑠𝑒𝑎𝑠𝑒 𝑎𝑡 𝑡ℎ𝑒 𝑠𝑡𝑎𝑟𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑒𝑟𝑖𝑜𝑑
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑜𝑝𝑙𝑒 𝑤ℎ𝑜 𝑑𝑒𝑣𝑒𝑙𝑜𝑝 𝑑𝑖𝑠𝑒𝑎𝑠𝑒
o Incidence rate (IR)=𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑟𝑠𝑜𝑛−𝑦𝑒𝑎𝑟𝑠 𝑤ℎ𝑒𝑛 𝑝𝑒𝑜𝑝𝑙𝑒 𝑎𝑟𝑒 𝑎𝑡 𝑟𝑖𝑠𝑘 𝑜𝑓 𝑔𝑒𝑡𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 𝑑𝑖𝑠𝑒𝑎𝑠𝑒
- Weaknesses:
o Expensive to set up and follow-up time
o Exposure assessment in large population
o Confounding
o Those collaborating and lost to follow-up might not be representative
1
,(Retrospective) case-control study:
- Outcome assessed and looking back at the exposure
- Measure of association: odds ratio = ad/bc
- Weaknesses
o People have to remember what they did before the outcome occurred (recall
bias=information bias)
o Cases may not be representative for all people (selection bias)
o Controls may not be representative for the population from which the cases come
(selection bias)
o Confounding
o Reverse causation
o Not able to calculate relative risk
Randomized controlled trial
- Measure of association: IPR or IRR
- Weaknesses:
o Ethical issues
o Often short-term
o Fixed exposure (treatment)
o Often ‘intermediary’ endpoints (e.g. BP, body weight)
o Selected (high-risk) populations
2
,Lecture 3: Observational study designs and variants
Cohort characteristics:
- Compare the occurrence/incidence of disease/outcome in groups of people with and without
a particular exposure
- When studying etiological factors: participants free of the disease/outcome of interest at the
start of the follow up (to be sure: exclude the first year of follow-up)
- Measurement of exposure not biased by disease/outcome status
- When studying prognostic factors: cohorts are not disease free
(prognostic/survival cohorts)
- A prospective cohort study uses baseline data at start of the study
- A retrospective cohort study uses records of past exposure instead of
baseline data (better name might be historical cohort study)
With cohorts, the researchers begin with a large cross sectional study (e.g.
survey) or baseline data of an intervention. They then re-survey the same population after a time
period. This converts a cross-sectional or intervention study into a cohort design.
Case-control study: a study where a group of people with disease (cases) are compared to a group
without the disease (controls), selected to represent the population from which the cases came.
Controls are representative with regard to exposure, thus sampled irrespective of exposure.
Case-control studies are needed:
- For rare diseases
- When latency time is long
- When your budget is limited (but case-control can also be expensive)
- For more detailed measurements (e.g. interviews, biomarkers (urine or blood))
Example: effect of folate intake on colorectal adenomas (pre-cursor of colorectal cancer). For this
research endoscopy is needed, as this is not ethical for a big group, a case-control design is used.
People are invited that had intestinal complaints who did undergo an endoscopy.
High dietary folate intake seemed to increase OR. This outcome could be due to bias:
- Selection bias
o Recruitment by nurse or physician not ‘at random’, for example only the healthiest
people and people who visit the clinic vary often are included (health conscious
people and people with many complaints)
o Response rates for cases and controls: the exposure distribution in both groups
should represent the exposure distribution in their source population
- Information bias
o Recall bias: retrospective data collection
o Most participants are aware of their case-control status which causes possibly
systematic difference in filling out the food-frequency questionnaire
o Misclassification cases/controls
▪ Not very likely because most participants had a total colonoscopy
- Confounding
o A lot of factors can be of influence such as age, gender, smoking, alcohol, BMI,
education, use of NSAIDs, contraceptives, HRT, outpatient clinic, intake of fat, etc.
However, none of these factors changed the crude estimates by more than 10% (so
no confounding).
o It was found that dietary vitamin B2 was an effect modifier: folate is only protective
for those with high vitamin B2 intake. With a lower B2 intake, a higher dietary folate
intake increases risk of carcinomas.
3
, Case-cohort study and nested case-control study are ‘hybrid’ designs embedded within existing
cohort study. They combine aspects of a case-control and cohort study which causes them to be
efficient (use part of a larger cohort) and there is no recall bias due to prospective data collection.
A case-cohort study is conducted in a cohort study. The cases are all those diagnosed with a
particular disease and the controls are a random sample (sub-cohort) of the whole population
selected before the cases develop. Detailed information collected for this sub-cohort and all of the
people who develop the disease of interest. So, data of cases is compared with that of a random
sample of controls drawn from the start of the study. The sub-cohort can be used for comparison
with multiple different case group.
Source population = cohort, study population = sub-cohort
Advantages case-cohort design
- Efficient: data from only part of the parent cohort is included (versus cohort study)
- Reduces selection bias: cases and non-cases are sampled from same the population (versus
case-control study)
- Reduced information bias: risk factor exposure can be assessed before the disease occurs
(versus case-control study)
- The sub-cohort can be used to study multiple outcomes (versus (nested) case-control study)
- Sub-cohort can be used to calculate person-time risk (versus (nested) case-control study)
Disadvantages case-cohort design
- Reduces power because of reduced sample size (versus full cohort)
- Sampling should be done completely unbiased (not related to exposure status)
- Exposure should not change over time
- More challenging statistical analyses (versus full cohort)
- When not established at the start of the full cohort (e.g. later blood sampling): potential
information/selection bias (versus full cohort)
A nested case-control study is conducted in a cohort study. The cases are all those diagnosed with a
particular disease and the controls are selected from those without disease at the diagnosis time of
the cases or disease free at the end of follow-up (FU). So, a nested case-control study compares the
exposure of the cases within a cohort to controls sampled from the cohort.
Selection of controls can be done in two ways:
- ‘Cumulative sampling’
o Controls are sampled at the end of FU out of those who did not get the disease (~
classical case-control)
o When FU is short (no loss of FU, no change in exposure)
o OR instead of RR
4
Cross-sectional study:
- Estimate exposure and outcome at one time point
- Association measure: Prevalence Proportion Ratio (PPR)
- Disadvantage: reverse causality
Prospective study / cohort study:
- Exposure assessed before outcome, exposure is by choice or circumstance
- Measure of association: incidence proportion ratio (IPR) or incidence rate ratio (IRR)
o Incidence proportion (IP)/cumulative incidence (CI) =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑜𝑝𝑙𝑒 𝑤ℎ𝑜 𝑑𝑒𝑣𝑒𝑙𝑜𝑝 𝑑𝑖𝑠𝑒𝑎𝑠𝑒 𝑖𝑛 𝑎 𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑒𝑑 𝑝𝑒𝑟𝑖𝑜𝑑
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑜𝑝𝑙𝑒 𝑎𝑡 𝑟𝑖𝑠𝑘 𝑜𝑓 𝑔𝑒𝑡𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 𝑑𝑖𝑠𝑒𝑎𝑠𝑒 𝑎𝑡 𝑡ℎ𝑒 𝑠𝑡𝑎𝑟𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑒𝑟𝑖𝑜𝑑
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑜𝑝𝑙𝑒 𝑤ℎ𝑜 𝑑𝑒𝑣𝑒𝑙𝑜𝑝 𝑑𝑖𝑠𝑒𝑎𝑠𝑒
o Incidence rate (IR)=𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑟𝑠𝑜𝑛−𝑦𝑒𝑎𝑟𝑠 𝑤ℎ𝑒𝑛 𝑝𝑒𝑜𝑝𝑙𝑒 𝑎𝑟𝑒 𝑎𝑡 𝑟𝑖𝑠𝑘 𝑜𝑓 𝑔𝑒𝑡𝑡𝑖𝑛𝑔 𝑡ℎ𝑒 𝑑𝑖𝑠𝑒𝑎𝑠𝑒
- Weaknesses:
o Expensive to set up and follow-up time
o Exposure assessment in large population
o Confounding
o Those collaborating and lost to follow-up might not be representative
1
,(Retrospective) case-control study:
- Outcome assessed and looking back at the exposure
- Measure of association: odds ratio = ad/bc
- Weaknesses
o People have to remember what they did before the outcome occurred (recall
bias=information bias)
o Cases may not be representative for all people (selection bias)
o Controls may not be representative for the population from which the cases come
(selection bias)
o Confounding
o Reverse causation
o Not able to calculate relative risk
Randomized controlled trial
- Measure of association: IPR or IRR
- Weaknesses:
o Ethical issues
o Often short-term
o Fixed exposure (treatment)
o Often ‘intermediary’ endpoints (e.g. BP, body weight)
o Selected (high-risk) populations
2
,Lecture 3: Observational study designs and variants
Cohort characteristics:
- Compare the occurrence/incidence of disease/outcome in groups of people with and without
a particular exposure
- When studying etiological factors: participants free of the disease/outcome of interest at the
start of the follow up (to be sure: exclude the first year of follow-up)
- Measurement of exposure not biased by disease/outcome status
- When studying prognostic factors: cohorts are not disease free
(prognostic/survival cohorts)
- A prospective cohort study uses baseline data at start of the study
- A retrospective cohort study uses records of past exposure instead of
baseline data (better name might be historical cohort study)
With cohorts, the researchers begin with a large cross sectional study (e.g.
survey) or baseline data of an intervention. They then re-survey the same population after a time
period. This converts a cross-sectional or intervention study into a cohort design.
Case-control study: a study where a group of people with disease (cases) are compared to a group
without the disease (controls), selected to represent the population from which the cases came.
Controls are representative with regard to exposure, thus sampled irrespective of exposure.
Case-control studies are needed:
- For rare diseases
- When latency time is long
- When your budget is limited (but case-control can also be expensive)
- For more detailed measurements (e.g. interviews, biomarkers (urine or blood))
Example: effect of folate intake on colorectal adenomas (pre-cursor of colorectal cancer). For this
research endoscopy is needed, as this is not ethical for a big group, a case-control design is used.
People are invited that had intestinal complaints who did undergo an endoscopy.
High dietary folate intake seemed to increase OR. This outcome could be due to bias:
- Selection bias
o Recruitment by nurse or physician not ‘at random’, for example only the healthiest
people and people who visit the clinic vary often are included (health conscious
people and people with many complaints)
o Response rates for cases and controls: the exposure distribution in both groups
should represent the exposure distribution in their source population
- Information bias
o Recall bias: retrospective data collection
o Most participants are aware of their case-control status which causes possibly
systematic difference in filling out the food-frequency questionnaire
o Misclassification cases/controls
▪ Not very likely because most participants had a total colonoscopy
- Confounding
o A lot of factors can be of influence such as age, gender, smoking, alcohol, BMI,
education, use of NSAIDs, contraceptives, HRT, outpatient clinic, intake of fat, etc.
However, none of these factors changed the crude estimates by more than 10% (so
no confounding).
o It was found that dietary vitamin B2 was an effect modifier: folate is only protective
for those with high vitamin B2 intake. With a lower B2 intake, a higher dietary folate
intake increases risk of carcinomas.
3
, Case-cohort study and nested case-control study are ‘hybrid’ designs embedded within existing
cohort study. They combine aspects of a case-control and cohort study which causes them to be
efficient (use part of a larger cohort) and there is no recall bias due to prospective data collection.
A case-cohort study is conducted in a cohort study. The cases are all those diagnosed with a
particular disease and the controls are a random sample (sub-cohort) of the whole population
selected before the cases develop. Detailed information collected for this sub-cohort and all of the
people who develop the disease of interest. So, data of cases is compared with that of a random
sample of controls drawn from the start of the study. The sub-cohort can be used for comparison
with multiple different case group.
Source population = cohort, study population = sub-cohort
Advantages case-cohort design
- Efficient: data from only part of the parent cohort is included (versus cohort study)
- Reduces selection bias: cases and non-cases are sampled from same the population (versus
case-control study)
- Reduced information bias: risk factor exposure can be assessed before the disease occurs
(versus case-control study)
- The sub-cohort can be used to study multiple outcomes (versus (nested) case-control study)
- Sub-cohort can be used to calculate person-time risk (versus (nested) case-control study)
Disadvantages case-cohort design
- Reduces power because of reduced sample size (versus full cohort)
- Sampling should be done completely unbiased (not related to exposure status)
- Exposure should not change over time
- More challenging statistical analyses (versus full cohort)
- When not established at the start of the full cohort (e.g. later blood sampling): potential
information/selection bias (versus full cohort)
A nested case-control study is conducted in a cohort study. The cases are all those diagnosed with a
particular disease and the controls are selected from those without disease at the diagnosis time of
the cases or disease free at the end of follow-up (FU). So, a nested case-control study compares the
exposure of the cases within a cohort to controls sampled from the cohort.
Selection of controls can be done in two ways:
- ‘Cumulative sampling’
o Controls are sampled at the end of FU out of those who did not get the disease (~
classical case-control)
o When FU is short (no loss of FU, no change in exposure)
o OR instead of RR
4
