7/26/10

CLINICAL EPIDEMIOLOGY 1
Ida Marie Tabangay Lim, MD

VARIABLES BIAS RANDOMIZATION BLINDING

VARIABLES

The funcEon of epidemiology research is in part, to examine the relaEonship between exposure and outcome. To this eect various studies are conducted gathering data, analysing the data and interpreEng the data.

Variable
DATA- comprise observaEons on one or more variables VARIABLE -Any enEty which can vary or take on dierent values Two important traits:
ExhausEve
Should include all possible answerable responses

Mutually exclusive
No respondent should be able to have two aSributes simultaneously

7/26/10

Examples
No of lymph nodes:
1 2 3

Types of Variables
Quan0ta0ve/ Numerical: Discrete-a variable can only take certain whole numerical values
Ex number of students in a classroom Number of asthma aSacks per year

Variable - Sex:
Male Female

Qualita0ve/Categorical: Nominal -categories are simply named but not ordered

Ex Blood type A, B,AB,O

Blood sugar level:

< 100 100-200 >200

Variable Degree of malnutri2on

Mild Moderate Severe

Ordinal ordered
Ex degree of pain-none, mild moderate severe

Con0nuous -there is no limitaEon on the values that the variable can take
Ex. Height , weight

Independent vs dependent variable

Independent
Treatment or intervenEon exposure

Dependent
Outcome eect

Dependent variable The variable that is used to describe or measure the problem under study Independent variables The variables that are used to describe or measure the factors that are assumed to cause or at least inuence the problem

Example:
Study of the rela2onship between human papilloma virus infec2on and head and neck cancer Dependent variable- Independent variable

BIAS

7/26/10

Bias
Any systemaEc error in an epidemiological study which results in an incorrect esEmate of the associaEon between exposure and disease SystemaEc variaEon of measurements from the true value

Bias
trend in the collecEon, analysis, interpretaEon, publicaEon or review of data that can lead to conclusions that are systemaEcally dierent from the truth deviaEon of results or interferences from the truth, or processes leading to such deviaEon

Quality of an estimate
Precision & validity

No precision

Precision but no validity

Random error

(low precision) ( low validity)

Systematic error

Random error !

Systematic error (Bias) !

In epidemiological studies, an invesEgator would like to minimise both systemaEc error (bias) and random error (chance). Reducing systemaEc errors lead to an increase in the validity of the study, while reducing random errors increase the power of the study. Knowledge of systemaEc error (bias) therefore become an important issue in epidemiological studies.

Example
Measuring height: Measuring tape held dierently by dierent invesEgators 180 179 loss of precision
error
178 177 176 175 174 173

Tape shrunk/wrong
systemaEc error bias (cannot be corrected aderwards!)

7/26/10

Bias can occur during any stage of a study:

literature review of the study quesEon selecEon of the study sample measurement of exposure and outcome analysis of data interpretaEon of the analysis publicaEon of the results By careful use of proper technique in the design, data collecEon, and analysis stages bias can be prevented or minimised

Three general types of bias:

SelecEon bias InformaEon bias Confounding bias

SelecEon bias
Systematic errors in the process of identifying the study population Preferential selection of subjects related to their
case/control status exposure status

SelecEon bias
can occur in the design phase of studies. It may also occur during the execuEon of study when some subjects are included and not others, based on the procedures used to select subjects Errors in the esEmaEon of eect happens when characterisEcs of the subjects selected for the study are systemaEcally dierent from those in the target populaEon, a distorEon of the measured eect will then result.

The common element of such biases is that the relaEon between exposure and disease is dierent for those who parEcipate in the study and those who would be theoreEcally eligible for the study but do not parEcipate. SelecEon bias is a theoreEcal possibility whenever correlates of the outcome capable of inuencing study parEcipaEon are existent in some individuals at the beginning of the study. These correlates may be unmeasured or even unrecognised by the invesEgator.

7/26/10

Types of SelecEon Bias

Berkson bias (Admission rate) Neyman bias (prevalence/incidence) Popularity bias Referral bias Volunteer bias DiagnosEc access bias Procedure selecEon bias Missing clinical data bias StarEng Eme bias Migratory bias Membership bias Non-respondent bias Centripetal bias Filter bias

SelecEon bias

Admission bias, prevalence/incidence bias detecEon bias, volunteer bias and loss to follow-up bias are common forms of this type of bias.

Admission bias (Berksons bias )occurs when case control and cross secEonal studies are done exclusively in hospital sehngs where the populaEon studied not accurately reects the target populaEon.

Prevalence/incidence(Neyman) bias happens when mild or asymptomaEc cases as well as fatal short disease episodes are missed when studies are performed late in disease process.

Volunteer bias occurs when those who volunteer to parEcipate in a study dier systemaEcally with regard to either exposure or disease status from those who did not volunteer

7/26/10

Minimising selection bias

Clear definition of study population Explicit case and control definitions Cases and controls from same population

InformaEon (observaEon) bias

InformaEon bias occurs in the data collecEon stage of studies. It happens when esEmated eect is distorted either by an error in measurement or by misclassifying the subject for exposure and/or outcome variables.

InformaEon (observaEon) bias

Systematic error in the measurement of information on exposure or outcome Differences in accuracy
of exposure data between cases and controls of outcome data between different exposure groups

Study subjects are classified in the wrong category

Two main types

ReporEng bias
Recall

Observer bias
Interviewer bias Biased follow-up

Recall bias happens e.g. when people, having had adverse health outcomes, remember and report past exposure dierently from those who did not experience any adverse health outcome.

7/26/10

Interviewer bias results when systemaEc dierences occur in the soliciEng, recording, or interpreEng of informaEon from study subjects.

Biased follow-up Unexposed are less likely diagnosed for disease than exposed Example
Cohort study to inves0gate risk factors for mesothelioma Dicult histological diagnosis Histologist more likely to diagnose specimen as mesothelioma if asbestos exposure known

Other Types of InformaEon Bias

Ques2onnaire bias results when leading quesEons or other aws in quesEonnaire result in a dierence in accuracy between compared groups.
interviewer bias quesEonnaire bias recall bias diagnosEc suspicion bias exposure suspicion bias ASenEon bias(Hawthorne eect) Instrument bias InsensiEve measure bias (Type II errors) RuminaEon bias(underlying cause bias) End digit preference bias Family informaEon bias

Response Bias occurs when subjects give inaccurate responses. Measurement Bias occurs when instruments are faulty Hawthorne Eect - a process tends to show improvement when being observed Observer error

7/26/10

Minimizing information bias

Standardise measurement instruments Administer instruments equally to cases and controls (exposed/unexposed) Use multiple sources of information Questionnaires Direct measurements Registries Case records Use multiple controls

Confounding bias (confounding)

essenEally a mixing of eects that occurs when a factor (confounder) associated with the exposure of interest is also associated with development of the disease or outcome of interest independently of exposure Therefore, a distorted esEmate of the exposure eect results because the exposure eect is mixed with the eect of extraneous variables.

Confounder
It must be predicEve of disease occurrence independent of its associaEon with the exposure of interest, but cannot be an intermediate in the casual chain of associaEon between exposure and disease development. It can eect the associaEon between exposure and disease posiEvely or negaEvely; the distorted esEmate resulEng from confounding can overesEmate or underesEmate the true eect or even change the apparent direcEon of eect.

To be confounding, an extraneous variable must have the following characterisEcs:

It must be a risk factor for disease It must be associated with the exposure under study in the populaEon studied It must not be an intermediate step in the casual path between the exposure and the disease

Confounders
Confounders act by being associated with both a risk factor and outcome in a way that makes the two seem related.
Poor Maternal Nutrition Low Socioeconomic Class Low Birth Weight

7/26/10

Example of Confounder - Sex

How to Check for Confounders

First calculate Odds RaEo for the exposure variable.

Males Females

Next calculate odds raEo for dierent strata of the confounding variable If the odds raEos are not materially dierent then there is no confounding.

How to Deal with Confounders

Design stage RestricEon (Strict inclusion criteria) Matching RandomizaEon Analysis stage Do analysis by adjus2ng for several strata of the confounding variable MulEple regression analysis

How to Deal with Confounders

Think about possible confounders at the design stage, and gather data on all possible confounders. A quick test about a possible confounder is to check whether it is unevenly distributed between study and comparison groups. Suspect confounding if the odds raEo gets altered ader adjusEng for another factor.

RandomizaEon RandomizaEon and Blinding

AllocaEon procedure that assigns subjects into exposure groups being compared so that each subject has the same probability of being in one group as in any other Strategy which ensures that experimental and control groups are similar except for the intervenEon being studied However, it does not guarantee that unmeasured variables are evenly distributed among the dierent exposure groups

7/26/10

RandomizaEon
The determinaEon of assignment to treatment group is based on probability alone and is not inuenced by the preference of the paEent or physician

Purpose of randomizaEon

To achieve equality of baseline characterisEcs of treatment groups so that comparison of the treatments is considered fair

Types of randomizaEon

1. Simple Each paEents treatment is determined at random independently without constraints Ex. toss of coin Odd or even numbers Table of random numbers Advantage: simple and opEmal in terms of its robustness against selecEon and accidental bias Drawback, possibility of imbalance bet groups which may be a problem for small sample size (< 200)

2. Block randomizaEon or restricted randomizaEon

Used when the study requires the numbers in each group to be very close at all Emes(ex researcher wants equal numbers of males and females) Purpose is to protect against imbalanced treatment assignment with respect to prognosEcally important paEent subgroups

Advantage of permuted block randomizaEon: Promotes group balance at the end of the trial Promotes periodic balance Disadvantage:
SuscepEble to selecEon bias

10

7/26/10

3. StraEcaEon SeparaEon of trial parEcipants into groups RandomizaEon is done within each of the important but dierent groups One can achieve approximate balance of important characterisEcs without sacricing the advantages of randomizaEon

Example
StraEed according to year level in Medicine

MinimizaEon
Group allocaEon does not rely solely on chance but is designed to reduce any dierence in distribuEon of known or suspected determinants of outcome

MinimizaEon
Used when it is crucial to achieve close similarity between treatment groups for several variables

Covariate-adapEve randomizaEon
This is used when one needs to have a balance across each of the variables when there are a number of variables that may inuence the outcome.

Outcome-adapEve randomizaEon
VariaEon of tradiEonal randomizaEon designed to address ethical issues in RCTs involving human subjects Advantage & Disadvantage: staEsEcal advantages of randomizaEon are retained while more paEents are assigned to superior treatments

11

7/26/10

AllocaEon concealment
SequenEally numbered opaque sealed envelopes(SNOSE) SequenEally numbered containers Pharmacy controlled Central randomizaEon

Blinding

Blinding
Lack of knowledge of the idenEty of the exposure and to where the study parEcipants are allocated to

Blinding
Used to prevent biases that arise from trial subjects, invesEgators and other health care givers, outcome evaluators and data analysis

Types
Open label- no blinding Single blind- study subjects are not aware of the interventon they will receive Double blind- studysubjects and invesEgators are unaware Triple blind- study subjects, invesEgators and outcome assesors Quadruple- also the data analysts are unaware Double dummy approach- strategy to make the intervenEons indisEnguishable from each other

12

7/26/10

Thank you

13