Case-Control Study (Design Conduct and Analysis)
-
Upload
rajeshwari -
Category
Documents
-
view
199 -
download
3
Transcript of Case-Control Study (Design Conduct and Analysis)
CASE CONTROL STUDY – DESIGN, CONDUCT AND ANALYSIS
Dr. C . RAMESH
ASSOCIATE PROFESSORDEPARTEMENT OF EPIDEMIOLOGY AND BIOSTATISTICS
KIDWAI MEMORIAL INSTITUTE OF ONCOLOGYHOSUR ROAD, BANGALORE
Mobile: 98454 62496
“ Epidemiology is the simplest and most direct method of
studying the causes of disease in humans and many major
contributions have been made by studies that have
demanded nothing more than ability to count, to think
logically & have an imaginative idea ”
-- SIR RICHARD DOLL
EPIDEMIOLOGY
Study of disease & Health in Human Population Pathological Process State of well being - Study of distribution of disease & its Determinants
Descriptive Analytical
Aims of Epidemiologic Research
Describe Explain Predict Control
Health Etiology Disease DistributionStatus occurrence of Disease
Utility of cancer registry data
A well organized and continuing Cancer Registry can serve as an efficient tool for
1. Medical Audit – Hospital Performance
2. Planning Hospital facilities
3. Evaluation of Patient Care
4. Research – Epidemiology, Clinical trials, Prognostic factors, etiology, cancer control
5. Education – Public and Professional
Epidemiological Study Design
Main types of epidemiological study design Intervention (Experimental studies)
– Clinical Trials– Field Trials
• Individual level• Aggregated level (community trials)
Observational (Non-experimental studies)– Cohort studies– Case-control studies– Cross sectional surveys– Routine data-based studies
• Individual level• Aggregated level (ecological studied)
CASE – CONTROL STUDY
Exposed Cases
Unexposed
Exposed
Unexposed
Direction of inquiry
Study Populn
Controls
Case – Control study
1. Directionality Outcome to exposure
2. Timing Retrospective for exposure
Case ascertainment either retrospective or
concurrent
3. Sampling Almost always on outcome with matching controls to
cases
HISTORICAL NOTE
Very few Studies prior to 1920
1. LANE-CLAYPTON (1926) : Reproductive factors in Breast Cancer (Method for selecting matched controls)
2. Cornfield (1951) : Showed that RR can be estimated from either a Case control or cohort
3. Cornfield (1954) : Direct Standardization to control extraneous factors in the Analysis
4. Mantel – Haenzel (1959) : Estimation of RR from Stratified data and gave Chi Square for association
5. Cornfield (1962) : Introduced the multiple logistic function (for analysis of factors related to CHD in Framingham Heart Study)
Case Control Study Important Discoveries
– 1950’s : Cigaratte smoking and lung cancers
– 1970’s : DES and Vaginal adeno Ca.
– 1980’s : Asprin and Rayes Sydrome
: Tampon use and Toxic shock syndrome
: L.tryptophan and eosinophilia – myalgia Syndrome
: AIDs and Sexual Practices
– 1990’s : Vaccine effectiveness
: Diet and Cancers
Case Control Study
ADVANTAGES
Well suited for Rare disease / Long Latency Quick to mount and conduct, inexpensive Requires comparatively few subjects Existing records occasionally be used
– no risk to subjects Allows study of multiple factors
Case Control StudyDISADVANTAGES Relies on Recall records for past exposures Valid information difficult/impossible Control of extraneous factors – incomplete Selection of appropriate comparison group difficult Rates of disease in E & E individuals cannot be Determined. Detailed study of mechanism rarely possible.
Useful in the study of exposures that cannot be randomized for logistic or ethical reasons (ex. Water hardness, alcohol consumption during pregnancy)
Case Control Study
Characteristics of Cases
1. Representativeness
Ideally, cases are random sample of all cases of interest in the source population (eg. Registry data)
Commonly, they are selection of available cases from Hospital
CASE SELECTION
Group of individuals who have disease which is in as far as possible a homogenous etiologic entity
Incident Prevalent Decedant
Usually used Only advantage Occasionally Readily Used in Prelim. Available Study based on Medi. Rec.
Diseases manifested by sudden death
Case Control Study
Selection of Controls
Most important and most difficult task
• No single type suitable for all studies• No firm criteria for an acceptable group
Case Control Study
Characteristics of Control Who is the best control ?
What universe should controls come from ?
If cases are random sample of cases in the popln.,
controls should be a random sample of all non cases
in the popln. Sampled at the same time
Case Control Study
Qualities needed in controls1. Comparability More important than
representative ness
2. Should be at Risk of the disease
3. Resemble case in all respects except for presence of disease (and any as yet undiscovered risk factor for disease)
Case Control Study
No. of Control Group:
1. One control group best suited to needs of particular study
2. II group if I group has known or suspected deficiency which can be offset by II Group
3. Stay within the bounds of 4:1 Small in statistical power as ratio greater than 4
Case Control Study
SOURCE OF CONTROLS
Hospital Patients General Population Restricted Popn. Group (Neighbourhood, associate or Relative of cases)
HOSPITAL CONTROLS
* Readily available may be in hosp. for* Have time to spare & co-operative condition which has* Have ‘mental set’ similar to casesetiological features* Similar to cases w.r.t determinants in common (include
of Hospitalization diff Dignostic Category)
Case Control Study
General Population Control
Comparability when 1. Extremely expensive,Popn. Based cases time consumingselected
2. Often not co-operativeEven with Hosp. cases - Response is poorCausative factors are not 3. Factors may be present Inordinately prevalent leading to seeking Med.Care
Matching
Refers to selection of Comparison series (controls)
Effects on study efficiency Not on validity
Individual Matching Principles - Identical
Frequency Matching Purpose
Control confounding Increase information per obsern.in the post stratification Analysis
Stratification Matching
Excess results in Over matching
Data Analysis
Purpose: 1. Assess random variation 2. Control confounding
3. Evaluation of interaction
Data Editing Data Reduction Effect estimation
- Accuracy - Distn of obsern Testing of Stat. Hyp.- Consistency - Contingency Tables - point estimate- Completeness for key factors - Interval Estimates
Regression Method For M.V.Analysis
Case Control Study – Data Analysis
Depends on the design of the study
Unmatched or frequency matched studies
Individual matched studies
VALIDITY
Lack of Systematic Error
Inference on actual Inference on subjectssubjects outside study Population(Internal Validity) (External Validity)
Many biases * Distinction often difficult
Unmatched (and frequency matched studies)
Exposed Unexposed Total
Cases a b n1
Controls c d no
Total m1 m0
Odds of exposure in the cases = a/b
Odds of exposure in the controls= c/d
Odds (of exposure) ratio = (a/b) / (c/d) = ad / bc
Example for Unmatched Analysis SCHOOLING
Never (E+) Ever (E-) Total
Cx. Cancers 119 317 436Controls 68 319 387Total 187 636 823
O.R = (119/317)/(68/319)= 1.76
Chi.Square = 11.04, p = 0.0009
95 % CI can be estimated using S.E of the logaritham of an odds ratio (OR).
S.E ( in OR) = Sqrt [1 + 1 + 1 + 1 ] a b c d
95 % C.I = (1.24,2.46)
Individual Matched Studies (Paired – Analysis)
Controls
Cases Exposed Unexposed total
Exposed r s aUnexposed t u b
Total c d N/2
N = Total number of Paired individualMatched O.R = s / t (Provided t <> 0)
Example of Matched Study
Controls
Cases Exp. Unexp. Total
Exp. 468 87 555 Unexp. 73 4 77
Total 541 91 632 (N/2)
Matched O.R = 87/73=1.19McNeman’s Chi-Square =1.23, p=0.2795 % , CI=(0.86, 1.65)
Case Control Study
The OR is good estimate for the RR when disease is rare ( prevalence < 20 %)
Can be extended to N > 1 Controls Statistical testing is by simple Chi-square (unmatched
analysis) or by McNemar’s Chi-square (matched pair analysis)
Can be extended to multiple strata using M-H Chi-square.(M-H Chi square gives a weighted average of the OR’s in different strata, where those from larger strata are given more weight)
Regression Modelling
Can be used to adjust for the effect of confounders Dependent (out come variable) =
Function [Independent (explanatory variable)] Main advantage – It does not require us to define,
which independent variable is the exposure and which ones are the potential confounders, since all independent variables are treated in the same way
Case Control Study
Logistic Regression: Suitable for unmatched (frequency matched) case control studies.
Conditional L.R : suitable for individually matched Case Control Studies
Adjusted Vs. Crude RR
Stratum 1 Stratum 2 Crude
C & No Intn. 1.02 3.00 4.00
No C & No intn. 1.83 1.83 1.83
No C & Intn. 0.82 0.85 4.00
Strong Intn. But 1.10 9.00 4.00
C.irrelevant
COST OF CASE CONTROL STUDIES
Data Collection Phase - 75 % of the total cost of study
Planning - 10 %Analysis - 15 %
Evaluation of an optimal allocation procedure based on the relative cost of Cases and Controls reduced the total
study cost by at most 2%
(ERICA BRITTAIN ET.AL, AJE)
Case Control StudyCONCLUSION
RETROSPECTIVE CASE CONTROL STUDY
Important Research Strategy commonly encountered in Medical literature
When thoughtfully designed, carefully executed - Provides important clinical information
“Backward Logic” accompanied by several methodological hazards.
Conflicting or incorrect conclusions - Directly attributed to methodological deficiencies.
Case Control Study
Difference between Bias and ConfoundingBias creates an association that is not true,
confounding describes an association that is true but potentially misleading
GOOD STUDY DESIGN PROTECTS AGAINST ALL FORMS OF ERROR
Thank you