Measurement

1
Measurement

• Research Methods for Promotion of Lung Health
• Cairo, 13-22 October 2002

2
Measuring Disease OccurrenceGeneral Principles

• Methods of data collection must be defined
• Criteria for variables entered must be specified
• Methods and criteria must be rigorously applied

3
Types of Data Continuous
These are points located on a continuous scale of
values (for example, height or age and some
measures of function such as FEV1 ). The numbers
reflecting continuous data have no discrete
reality and flow into one another along a scale.
4
Types of Data Continuous
Example
Although we say someone is 52 years of age, this
is not absolutely accurate. It only
approximates the age, to the nearest year.
Adding decimal points (for example, 52.69 years)
can indicate the actual value more precisely.
Increasing the number of decimal points increases
the precision of the estimate. Because it is
not possible to be absolutely precise, such
measurements are usually recorded within
intervals (that is to say, someone is 52 years
of age, with an age lying between 52.00000.. and
52.999999..).
5
Types of Data Continuous
Outcomes that are continuous are often
functional measures and are reflections of the
state of health of the individuals studied.
Because they do not identify a state, they are
scales of function or health.
6
Types of Data Discrete (Categorical)
These represent groups of unique occurrences.
Information as to whether an individual or object
falls into one or other category is usually
indicated by an answer of yes or no to that
category. There may be more than two discrete but
independent possibilities, for example, number of
children in the family.
7
Types of Data Discrete (Categorical)
Example
For gender, one is either male or female there
is no other category. The categories are mutually
exclusive and do not overlap. If you are not
female, you are certain to be male.
8
Types of Data Discrete (Categorical)
Outcomes that are measured as discrete variables
are often states (such as death) or
diseases. The actual state being studied may
not always be defined in the same way by
different investigators.
9
Types of Data Discrete (Categorical)
Example
It is usually assumed that one either has or has
not got a disease such as asthma or tuberculosis
or pneumonia (i.e. they are discrete data).
However, the boundary between having and not
having any one of them is far from clear and what
one person might call pneumonia another might
not.
10
Types of Data Discrete (Categorical)
Categorisation is usually based on arbitrarily
assigned criteria in standard definitions. Strict
definition of these criteria and, when possible,
using standardized criteria is essential.
11
Sources of InformationMeasurement Instruments
Routinely-collected information
Clinical charts Standardized records Routine
reports
Specifically-collected information
Questionnaires Data collection forms Physical
tests
12
Problems With Measurement

• Ensuring comparability
• Ensuring precision
• Completeness and accuracy of recording

13
Problems With MeasurementEnsuring Comparability

• Standardization
• Definitions often from consensus
• Measurement techniques and instruments
• methods precisely defined
• carefully followed
• preferably those universally recommended

14
StandardizationTuberculosis examples

• Agreed definitions
• Recommended forms
• Quarterly reports
• Cohort evaluation

15
Standardized Definitions

• Tuberculosis case any patient treated for
  tuberculosis
• Smear positive case a patient with positive
  sputum smear, confirmed on a second
• Retreated case any patient previously treated
  for as much as one month
• Extrapulmonary disease outside the lung
  parenchyma

16
Recommended FormsPrinciples

• Useful for case management
• Clearly laid out
• Minimum necessary information
• Training in use / pilot introduction
• Regular evaluation of accuracy / completeness

17
Tuberculosis Treatment Card
RH Z E S
No tablets / dose
Day
1 2 3 4 5 6 7 8 9 10 11 1213 14
15 16 1718 19 2021 22 23 24 2526 27 28 29 30 31
Month
Enter x on day when medications swallowed under
direct observation
18
Evaulation of Measurements

• Precision
• Reliability
• Validity
• Completeness / Accuracy of Reporting

19
Sources of Error in Measurement

• Inappropriate tools for measurement
• Lack of standardization of tools
• Lack of training of personnel
• Lack of systematic evaluation
• Lack of monitoring of test results

20
Precision of MeasurementsTuberculin Testing,
Djibouti 1994
Per Cent
Excluding 0
Size of induration, mm
21
Precision of MeasurementPemba, Mozambique
Tuberculin 1990
Per cent
Induration, mm
22
Problems With MeasurementEnsuring Precision

• Causes of errors
• Carelessness of personnel
• Inadequacy of methods and tools
• Failure to follow standardized procedures

23
Measurement CharacteristicsReliability

• Ability of the test to give the same result when
  repeated on the same subject
• Random error reduces reliability and decreases
  probability of finding a real difference

24
Sputum Smear MicroscopyConcordance of Readings
David et al 1973
25
Test CharacteristicsValidity

• Ability of a test to truly identify a condition
  or disease
• Sensitivity
• Specificity
• Predictive Value

26
Test CharacteristicsValidity

• sensitivity, ability of a test to identify
  correctly those with a disease
• specificity the ability of a test to the ability
  identify those without disease
• Predictive value ability correctly to predict a
  positive or negative result

27
Tuberculin Reaction SizeXai Xai, Mozambique 1990
Per cent
Induration, mm
28
Results of Tuberculin TestingRoutine Testing,
Alberta Aboriginals
Per Cent
Excluding 0
Size of induration, mm
Int J Tuberc Lung Dis 19982S16
29
Validity of Tuberculin TestsDisease in
Aboriginals Following Test
Per cent
Years since positive
Int J Tuberc Lung Dis 19982S16
30
Validity of Tuberculin TestsLong-term Risk of
Disease
Annual Case rate
Tuberculin reaction, mm.
Int J Tuberc Lung Dis 19982S16
31
Accuracy of RecordingGrading Asthma Severity
By Doctor
By Guide
32
Problems With MeasurementCompleteness / Accuracy
of Recording

• Causes of errors
• Carelessness of personnel
• Lack of standardized forms
• Imprecision in recording and transcribing
  information

33
Error, Bias and Confounding

• Random error
• Systematic error
• Selection bias
• Information bias

34
Sources of Selection Bias

• Inappropriate population studied
• Inadequate participation
• Change of classification regarding exposure
• Selection of most accessible or of volunteers

35
Minimizing Selection BiasStudy Design

• Appropriate population selection
• High participation rate
• Demonstration of lack of difference between
  participants and non participants

36
Minimizing Selection BiasAnalysis

• Exclude from numerator and denominator
• Analyze by time at risk
• Worst and best case scenarios

37
Source of Information Bias

• Subject variation
• Observer variation
• Technical errors

38
Minimizing Information Bias

• Specify criteria in advance
• Analyze directly according to criteria
• Reduce numbers of observers
• Monitor performance of observers
• Use standardized tools for measurement

39
ConfoundingA Special Type of Bias

• A factor associated with both the outcome and an
  etiological factor
• Associated with outcome through its association
  with etiological factor

40
Tuberculosis and AgeNicaragua 1990
Per 100 000
Age group, years
41
Tuberculosis Notification RateNorway, by Age
Per 100 000
1927
1947
1980
Age, years
Nor Fore Lunge 198630
42
Impact of Error or Bias

• Random error will obscure a real difference
• Random error will require a larger sample size
• Bias will result in false difference
• It cannot be overcome by statistics if present