A short introduction to epidemiology Chapter 2b: Conducting a case-control study

A short introduction to epidemiology

Chapter 2b: Conducting a case-control study

Neil PearceCentre for Public Health

ResearchMassey University

Wellington, New Zealand

Chapter 2 (additional material)

Case-control studies

• This presentation includes additional material on conducting a case-control study

• More information on data analysis is given in chapter 9



• Reasons for doing a case-control study

• Basic study design

• Selection of cases

• Selection of controls– control sampling strategies– sources of controls– issues in control selection

Birth End of Follow up

Death

other death

lost to follow up

“non-diseased”

symptoms

severe disease

A Hypothetical Incidence Study

Exposed Non-exposed Ratio

Cases 1,813 952

Non-cases 8,187 9,048

Total 10,000 10,000

Person-years 90,635 95,163

Incidence rate 0.0200 0.0100 2.00

Incidence proportion

0.1813 0.0952 1.90

Incidence odds 0.2214 0.1052 2.11

A Hypothetical Case-Control Study

1813/8187 a/c adOdds ratio = ---------------- = ----- = ----

952/9048 b/d bc

1813/952 a/b ad= ---------------- = ----- = ----

8187/9048 c/d bc

Reasons for Doing a Case-Control Study• It may be inefficient to have to obtain exposure

information on all people in the source population• It is sufficient to obtain information on all of the

2,765 deaths and a control sample (e.g. 2,765 controls) of the 17,235 survivors

• We therefore only need to get exposure information on 5,530 people instead of 20,000

• This gain in efficiency is much greater when the disease is “rare” (e.g. if it were 1/10th as “common” then we would have 277 cases and 277 controls)

Reasons for Doing a Case-Control Study

• Rare disease

• long induction time

• smaller study size permits collection and analysis of more detailed exposure information

• cohort difficult to enumerate (registry-based studies)







Basic Case-Control Study Design• Every study is based on a particular source

population followed over a particular period of time (the risk period)

• Ideally the study base should be made explicit• We study all cases of the outcome and a sample

of controls drawn from the source population• The case-control design thus involves all of the

potential biases involved in a full cohort study, as well as additional biases involved in sampling controls

• Information bias is not an inherent feature of such studies

Cohort-Based (Nested) Case-Control Studies

• Enumerate the cohort (source population) and its experience over time (the risk period)

• Ascertain all cases generated by this study base

• Sample controls from the person-time (or persons) that generated the cases

Registry-Based Case-Control Studies

• Ascertain all cases appearing in the registry during a specified period of time

• Sample controls from the source population for the registry







Selection of Cases

Cohort-based• All cases (or deceased cases) generated by the

cohort study

• Living cases may be added from other sources (e.g. hospital records, cancer registrations)

Registry-based• All eligible cases appearing in the “registry”

during a specified period of time







Control Sampling Strategies

• Cumulative incidence sampling

• Case-base sampling

• Density sampling


Death

other death

lost to follow up

“non-diseased”

symptoms

severe disease


Exposed Non-exposed Ratio

Cases 1,813 952

Non-cases 8,187 9,048

Total 10,000 10,000

Person-years 90,635 95,163

Incidence rate 0.0200 0.0100 2.00

Incidence proportion

0.1813 0.0952 1.90

Incidence odds 0.2214 0.1052 2.11

Cumulative Incidence Sampling

• “Traditional” method of control selection in nested case-control studies

• Controls are sampled from the “non-cases”, those free of disease at the end of the follow-up period, i.e. the survivors

• I.e. controls are sampled from the denominators for (cohort) odds ratio analyses


Exposed Non-exposed Total

Cases 1,813 962 2,765

Non-cases 8,187 9,048 17,235 Total 10,000 10,000 20,000

Odds ratio

Incidence odds

1813/8187 952/9048 2.11

A Hypothetical Case-control Study


Cases 1,813 952 2,765

Controls 1,313 1,452 2,765

Odds ratio

Odds 1813/1313 952/1452 2.11

Cumulative Incidence Sampling

• Estimates the (cohort) odds ratio (without any rare disease assumption)

• Estimates the risk ratio and rate ratio approximately (with a rare disease assumption)

• May involve matching on age, etc• Exposure is usually only considered up to the

“time” (year or age) that the case occurred

Case-cohort Sampling

• Controls can be selected from those at risk at the beginning of the follow-up period, I.e. from the entire source population

• I.e. controls are selected from the denominators for (cohort) risk ratio analyses



Cases 1,813 952 2,765

Total 10,000 10,000 20,000

Odds ratio

Odds 1813/10000 952/10000 1.90



Cases 1,813 952 2,765

Controls 1,383 1,383 2,766

Odds ratio

Odds 1813/1383 952/1383 1.90

Case-cohort Sampling

• Estimates the risk ratio (without any rare disease assumption)

• Requires minor modifications to the standard formulas for confidence intervals and p-values

• May involve matching on age, etc

• Once again, exposure is usually only considered up until the “time” that the case occurred


Death

other death

lost to follow up

“non-diseased”

symptoms

severe disease

Density Sampling

• Controls are selected longitudinally throughout the course of the study, i.e. from the person-time of the study base

• I.e. controls are sampled from the denominators for the rate ratio analyses

• In general, controls are selected from the “risk set” of persons at risk at the “time” that each case occurred



Cases 1,813 952 2,765

Person-years 90,635 95,163 185,798

Rate ratio

Incidence rate

1813/90635 952/95163 2.00



Cases 1,813 952 2,765

Controls 1,349 1,416 2,765

Odds ratio

Odds 1813/1349 952/1416 2.00

Density Sampling

• The “time” variable is usually taken to be age rather than calendar time (year)

• Estimates the rate ratio (without any rare disease assumption)

• Matching may also be done on other time-related factors, although this is usually not necessary

• Usual method of sampling in registry-based studies

Selecting Controls

Cohort-based studies

• Sample of the cohort (preferably by density sampling on age)

Registry-based studies

• Sample of the source population for the Registry (usually by density sampling on year, perhaps with matching on age)

Selecting Controls in Registry-Based Studies

• Cases chosen from all lung cancer cases at hospitals in the City

• Controls chosen from general population of the City?

Selecting Controls in Registry-Based Studies• All lung cancer cases at all hospitals in the City

• Controls chosen from general population of the City?

• Restrict cases to those living in the City (exclude those who have come to the City for treatment)

• Restrictions that apply to one group (e.g. having a telephone, being on Electoral Roll, having health insurance) should also be applied to the other


• Cases chosen from all lung cancer cases at the main hospital in the City

• What is the source population for these cases?


• Cases chosen from all lung cancer cases at the main hospital in the City

• What is the source population for these cases?

• “All those who would have come to the main hospital in the City for treatment if they had developed lung cancer”

Issues in Control Selection

• Controls are usually sampled at random from the entire study base

• However, it is sometimes desirable to restrict the controls to a sample of a subset of the study base

• In particular, we may select controls from persons with other diseases generated by the same study base (e.g. other deaths, other cancers, other hospital admissions)

“Other Disease” Controls

• All other diseases

• All other diseases except those known to be related to exposure

• A disease “known to be unrelated to exposure”

Reasons for Using “Other Disease” ControlsThe cohort (source population) is not enumerated• E.g. if the cases are identified from hospital

admissions (e.g. for lung cancer) then the study base is “all persons who would have been admitted to this hospital if they had developed lung cancer”

• Controls might be selected from other admissions to the same hospital

Reasons for Using “Other Disease” ControlsComparability of information• E.g. in a case-control study of non-Hodgkin’s

lymphoma and pesticide exposure, cases might be more likely to recall brief exposures

• We might therefore select controls from “other cancer” registrations rather than from the entire source population for the Cancer Registry

Selection Bias in Case-Control Studies

• In a case-control study, the controls are a sample of the source population

• Selection bias can occur if the sample is non-random, and the selection of controls is related to exposure status

• In other words, selection bias can occur if the controls are not representative of the exposure in the source population

Selection Bias in Case-Control Studies: Solutions• Selection bias can occur if the selection of

controls is related to exposure status• In the analysis, we can control for the

determinants of control selection (e.g. social class)

• An exception is when we have chosen “other disease” controls and the other diseases are directly caused by the main exposure of interest: this selection bias cannot be removed

General Population and “Other Cancer” ControlsGeneral population

• Represents study base

• May be more prone to recall bias if cases are more likely to recall exposures

• Difficult to keep interviewer blind, and may get interviewer bias

“Other cancers”

• Other diseases may be caused by exposure (selection bias)

• Equal motivation and recall in cases and controls

• Easier to keep interviewer blind

Reasons for Matching

Practical efficiency

• e.g. if we are using hospital controls then it is usually more efficient to select a control admitted on the same day as the case, rather than sampling at random from all admissions for the year

Reasons for Matching

Statistical efficiency

• e.g. if we select general population controls at random in a lung cancer case-control study then the cases will be mostly “old” and the controls will be mostly “young”. It will therefore be difficult to stratify on, and control for, age

Reasons for Not Matching

Practical efficiency

• matching can be costly and time-consuming and is usually not necessary since we can adjust for the major matching factors (e.g. age, gender, smoking status) in the analysis

Reasons for Not Matching

Statistical efficiency

• Matching on a weak risk factor (or a non-risk factor) that is strongly correlated with the main exposure can dramatically reduce efficiency

Matching

Only match on risk factors that are:

• Not of intrinsic interest in themselves (e.g. age)

• Strong risk factors for disease

• Not too difficult to match on

Common misconceptions about case-control studies

• Fundamentally different type of study design that proceeds from disease to exposure (I.e. reverse causality)

• Inherently less valid (more biased) than cohort studies

• Require a rare-disease assumption• Odds ratio only approximates the rate ratio or

risk ratio (under the rare disease assumption)

A short introduction to epidemiology

Chapter 2b: Conducting a case-control study

Neil PearceCentre for Public Health

ResearchMassey University

Wellington, New Zealand

Documents

A short introduction to epidemiology Chapter 2b: Conducting a case-control study