Case Control Studies PDF

CASE CONTROL STUDIES Lecturer: Prof Ghufran Ahmed Jassim  Learning Outcomes By the end of the lecture, students will be able to: Describe a case control study and explain how it is conducted Evaluate the advantages and limitations of case control studies...

CASE CONTROL STUDIES Lecturer: Prof Ghufran Ahmed Jassim  Learning Outcomes By the end of the lecture, students will be able to: Describe a case control study and explain how it is conducted Evaluate the advantages and limitations of case control studies Identify and interpret the measure of association that can be calculated from a case control study Identify three ways of dealing with confounding  Did the observer assign the exposure? yes no Intervention Observational studies studies Random Comparis allocatio on group? n? yes no yes no Non- Randomise Analytic Descriptive randomised d trial study study trial Ecologi Case- Cohor Cross- cal control t sectional study study study study  Did the observer assign the exposure? yes no Intervention Observational studies studies Random Comparis allocatio on group? n? yes no yes no Non- Randomise Analytic Descriptive randomised d trial study study trial Ecologi Case- Cohor Cross- cal control t sectional study study study study  There are two possible ways of measuring association between exposure and outcome Describe a case control study and explain how it is conducted  The case-control study approach A study that compares two groups of people: those with the disease or condition under study (cases) and a very similar group of people who do not have the disease or condition (controls). Purpose is to examine association between an exposure(s) and an outcome Four key steps 1. Identify cases – the people with the disease or outcome 2. Identify the controls – the people who do not have the disease or outcome 3. Measure exposures (e.g. potential risk factors for the outcome) among the cases and controls 4. Analyse whether or not the cases are more likely to have been exposed to a risk factor than the controls DESIGN OF A CASE-CONTROL STUDY Not Not Exposed Exposed Exposed Exposed Disease No Disease “CASES” “CONTROLS”  The case-control approach Individuals are selected from a defined population on the basis of their disease/condition status Start here Smokers Cases: Lung cancer Non smokers Smokers Controls: no lung cancer Non-smokers Study  Case-control study design Population Identify cases (i.e. all people in the population who have the condition) Identify controls (a representative sample of the study population without the condition) Smokers Non-smokers  Where a cohort study is difficult… If the outcome is rare And/or if the outcome has long latency period A cohort study will need to have a large sample size, have a long follow-up…….. And therefore be very costly and difficult to conduct  Conducting a case-control study Must clearly state the research question in order to develop clear definitions of cases and controls Case Definition Definition should be replicable and applied to all cases Can be based on clinical or laboratory definition Must state inclusion and exclusion criteria  Example: case definition Quantification of risk factors for herpes zoster Data source: General practice database Case definition: Individuals with a code for herpes zoster in general practice records or Hospital Episode Statistics Inclusion criteria: – Patients aged 18 years and over – Under follow-up between 1 January 2000 and 31 December 2011 – No evidence of previous zoster – At least 12 months of follow-up prior to first diagnosis of zoster to exclude past cases of zoster recorded retrospectively after registration at a GP Taken from Forbes HJ, Thomas SL, Clayton T. Quantification of risk factors for herpes zoster: population based case-control study. BMJ 2014;348:g2911.  Identifying cases Must have a clear case definition – Definition should be replicable and applied to all cases – Can be based on clinical or laboratory definition Must describe carefully how cases are selected  Selecting controls Selection of appropriate controls is often the most demanding and difficult part of a case-control study Controls should be representative of the population from which the cases have arisen. But they should be without the disease or outcome.  Selecting controls Need to understand from what population the cases arose from Sources of “population controls” or “population-based controls” include Population registers Electoral rolls General practice databases  Measuring exposures Data on exposures can be measured in a variety of ways e.g., by interview, reviewing medical records, using biological samples. As with all studies you need to use a method that is valid and reliable to measure the exposure (and indeed the outcome). Case-controls studies are often not suitable to use when the exposure is rare.  Advantages Useful for rare diseases (i.e. outcomes) Useful for diseases with long latency Often cheaper and quicker than cohort studies Can study association between multiple exposures and an outcome Can conduct expensive or time-consuming tests, which may not be possible with a cohort study  Summary of sources of error All apply to case-control studies Source Type of error Selection of participants Sampling error (i.e., sampling) Selection bias Measurement – instrument Inaccuracy (poor validity) (e.g. a self-administered questionnaire, monitor, interview) Poor reliability Measurement – observer Between observers Within observers  Selection bias Controls are not representative of the population that cases come from Particularly arises if using hospital controls Hospital controls are usually people who are patients at the same hospital(s) as the cases who do not have the disease Have to ensure that – There are no health-care access issues that prevent hospital controls being representative of the population – The disease for which they were admitted is not related to risk factors for the outcome of interest – The distribution of exposures in the hospital controls may differ to the distribution of exposures in the population that cases came from  Observer bias (a.k.a. interviewer bias) Often information on exposures is collected by interview Interviewers knowing whether they are talking to a case or a control may change how they collect data on the exposure To minimise observer bias: Train interviewers and use standardised questioning Blind interviewers to whether a person is a case or control Limit knowledge among interviewers about the hypothesis being tested (e.g., don’t tell them which exposure is of most interest)  Recall bias Cases may describe their level of exposure differently than controls, even if there is no difference Having a disease may make people more aware of an exposure or the importance they attach to it Minimise recall bias by blinding cases and controls to the research question  Confounding As with all observational studies, an apparent association between an exposure and outcome may be due in part or whole to a third factor Identify and interpret the measure of association that can be calculated from a case control study  Analysis of case-control studies Cases Controls Total Exposed Unexposed  Analysis of case-control studies Cases of Controls Total ovarian cancer First degree Yes 129 191 320 relative with breast/ovarian No 495 1296 1791 cancer Total 627 1508 2135 Taken from Jordan SJ, Green AC, Whiteman DC, Moore SP, Bain CJ, Gertig DM, et al. Serous ovarian, fallopian tube and primary peritoneal cancers: a comparative epidemiological analysis. 2007;122:1598- 1603.  Analysis of case-control studies Cases of Controls Total ovarian cancer First degree Yes 129 191 320 relative with breast/ovarian No 495 1296 1791 cancer Total 627 1508 2135 Odds of exposure = number of exposed people/ number of unexposed people  Analysis of case-control studies Cases of Controls Total ovarian cancer First degree Yes 129 191 320 relative with breast/ovarian No 495 1296 1791 cancer Total 627 1508 2135 Odds of exposure = number of exposed people/ number of unexposed people ? ? 2 min Think/pair/share  Analysis of case-control studies Cases of Controls Total ovarian cancer First degree Yes 129 191 320 relative with breast/ovarian No 495 1296 1791 cancer Total 627 1508 2135 Odds of exposure = number of exposed people/ number of unexposed people Odds of having a first degree relative with cancer among women with ovarian cancer = 129/495 = 0.26 Odds of having a first degree relative with cancer among women without ovarian cancer = 191/1296 = 0.15 Odds ratio = 0.26/0.15 = 1.73  Analysis of case-control studies Cases of Controls Total ovarian cancer First degree Yes 129 191 320 relative with breast/ovarian No 495 1296 1791 cancer Total 627 1508 2135 Suggests the odds of having a first degree relative with breast/ovarian cancer is higher among women with ovarian cancer relative to women without cancer suggests association between having a relative with cancer and getting cancer Confounding  Effects of confounding 1. Create an apparent association when one does not exist E.g., odds ratio for association between asthma and Covid-19 ICU admission is 2.84. If we controlled for educational level somehow, the odds ratio might be 1 2. Over- or under-estimation of the size of the true association E.g. odds ratio for association between asthma and Covid-19 ICU admission is 2.84. If we controlled for educational level, the odds ratio might be 2.00 3. Hide a true association if it exists E.g., if we found the odds ratio for association between asthma and Covid-19 ICU admission is 1 when we don’t control for educational level, and it becomes 2.84 when we do control for educational level. Asthma and Covid-19 ICU admission are associated, but the association is hidden unless we control for educational level 4. Reverse the direction of the association (Simpson’s paradox) E.g., odds ratio for association between asthma and Covid-19 ICU admission is 2.84. If we controlled for educational level, the odds ratio might become 0.9 (i.e., the association goes from a positive association to a negative association)  Dealing with confounding 1. Design a study in a way that minimises the effect of confounding factors Restriction: restricting the sample to people with or without the confounding variable, Matching: matching cases and controls for potential confounding factors, makes them more similar with respect to potential confounding factors. Randomisation (covered in Randomised Controlled Trials lecture) Stratified analysis: splitting the sample into strata according to their level of the confounding variable and estimating the association between the exposure and outcome for each strata Multivariable modelling: use a statistical modelling technique to estimate the association between the exposure and outcome while controlling for confounding variables. Provides an “adjusted” estimate of the effect, e.g. adjusted odds ratio  Dealing with confounding 2. Use statistical methods for adjusting the effects of confounding Multivariable analysis using regression techniques Stratification (or post-stratification): splitting the sample into strata according to their level of the confounding variable (e.g. smokers and non-smokers) and estimating the association between the exposure and outcome for each strata  Multivariable regression Adjusts the estimated effect of an exposure on an outcome for the effect of other potentially confounding factors. Hence, derive an estimate of the independent effect of the exposure of interest Provides “adjusted” effect (e.g., adjusted odds ratio)  Poll After adjusting for age, sex, educational level, marital status and region of birth, are people with type 1 diabetes more likely to get Covid-19 requiring mechanical ventilation than those without type 1 diabetes? 1. Yes 2. No When you adjust for age and sex the odds ratio is 2.32. When you adjust for age, sex, educational level, marital status and region of birth the odds ratio is 3.13. Because the odds ratio changed when additionally adjusted for educational level, marital status and region of birth, indicates that these factors do confound the association between the exposure (type 1 diabetes) and outcome (Covid-19 requiring mechanical ventilation)  Summary Advantages of case-control studies Useful for rare diseases (i.e. outcomes) Useful for diseases with long latency Often cheaper and quicker than cohort studies Can study association between multiple exposures and an outcome Can conduct expensive or time-consuming tests, which may not be possible with a cohort study Limitations of case-control studies Selection bias, particularly when sampling controls Recall bias Observer bias Confounding Reverse causality, particularly if using prevalent cases Not suitable for measuring rare exposures  IN SUMMARY HOW ARE THEY DONE? Select a group of cases of the disease Assess their exposure to the risk factor(s) of interest Select a control group of healthy persons who are comparable otherwise with the cases Assess the exposure of the controls Calculate an odds ratio – Measure of the effect of a factor on risk of disease COHORT VS. CASE CONTROL STUDY Disease + Exposed Disease - Study population Disease + Non-exposed Disease - 1 2 Summary Cohort Case-control Start with disease-free Recruit people with people disease Record risk factor exposure Find comparable control Follow-up to record group incidenc Record risk factor Use this to calculate exposure retrospectively relative risk Calculate odds ratio  Sample MCQ Walsh and colleagues conducted a case-control study to identify risk factors for myocardial infarction (MI) among young adults in Ireland. They identified a sample of cases and controls and recorded information about their exposure to lifestyle risk factors including smoking, alcohol intake and physical activity using a questionnaire. Based on the information provided, who are the cases in this study? A. Young adults with MI who are exposed to lifestyle risk factors B. Young adults with MI who are not exposed to lifestyle risk factors C. Young adults without MI who are not exposed to lifestyle risk factors D. Young adults with MI who may or may not be exposed to lifestyle risk factors E. Young adults without MI who may or may not be exposed to lifestyle risk factors  Sample MCQ Walsh and colleagues conducted a case-control study to identify risk factors for myocardial infarction (MI) among young adults in Ireland. They identified a sample of cases and controls and recorded information about their exposure to lifestyle risk factors including smoking, alcohol intake and physical activity using a questionnaire. Based on the information provided, who are the cases in this study? A. Young adults with MI who are exposed to lifestyle risk factors B. Young adults with MI who are not exposed to lifestyle risk factors C. Young adults without MI who are not exposed to lifestyle risk factors D. Young adults with MI who may or may not be exposed to lifestyle risk factors E. Young adults without MI who may or may not be exposed to lifestyle risk factors Thank you for listening and… Please add any questions to the forum on Moodle Recommended Reading Sections: 6.1 Why do a case-control study? 6.2 Key elements of study design 6.3.1 The odds ratio 5.7.5 Methods for dealing with confounding