BIOSTATISTICS AND

EPIDEMIOLOGY
Saint Louis University
School of Natural Sciences
Department of Medical Laboratory Science
A.Y. 2019 – 2020

1
 Methods for
evidence-based
medicine
 Evaluate medical journals effectively
 Evaluate diagnostic procedures
 Assess agreement of various measures
 Evaluate information from survival analysis

2
 TOPIC  Evaluating diagnostic procedures
 Assessing agreement
OUTLINE  Survival Analysis
 Levels of evidence

3
SENSITIVITY AND
SPECIFICITY

4
 RECAP
 Epidemiology – studies the distribution of diseases and related
characteristics in a population

 Contingency tables – graphical presentation of characteristic

distribution

5
SENSITIVITY AND SPECIFICITY
• Objective: Correctly classifying individuals into by disease status;
“How well is a subject classified into disease or non-disease
groups?”

6
VARIATION IN BIOLOGIC VALUES
Without the With the
disease disease

7
VALIDITY
• The validity of a test is defined as its ability to distinguish
between who has a disease and who does not
• Sensitivity and Specificity

8
 SENSITIVITY SPECIFICITY
 Ability of the test to correctly  Ability of the test to correctly
identify those who have a identify those who don’t have a
disease disease

 Influenced by characteristics of
the affected population

Prevalence of the disease DOES NOT affect the sensitivity and

specificity of a test

9
SENSITIVITY AND SPECIFICITY
• Testing with dichotomous results
• Testing of continuous results
• Use of multiple tests
• Sequential (two-stage) testing
• Simultaneous testing
• Net sensitivity using two simultaneous tests
• Net specificity using two simultaneous tests

10
SENSITIVITY AND SPECIFICITY
• Knowledge of the disease status prior to calculation  index of
suspicion

• Screening and diagnostic test

• Screening test
• Diagnostic test
• Gold standard - test regarded as the most accurate method
available for classifying people as disease positive or negative

11
SCREENING TESTS

 Identifies asymptomatic individuals who may have a disease

 Detection of early disease or risk factors of disease

12
DIAGNOSTIC TESTS

 Establishing the presence or absence of disease when

subjects show signs or symptoms
 Basis for treatment decisions in symptomatic or screened
positive subjects
13
DICHOTOMOUS RESULTS

Sensitivity is the
proportion of
individuals with
disease which are
correctly identified as
positive

Gordis (2013) 14
DICHOTOMOUS RESULTS

Specificity is the
proportion of
individuals without
disease which are
correctly identified as
negative

Gordis (2013) 15
 Identification of true
positives and true
negatives

16
Gordis (2013)
CONTINUOUS RESULTS
Identification of a “cut-off point”  positive and negative

17
CONTINUOUS RESULTS

Results into
misdiagnosis 
increased false
positives

18
CONTINUOUS RESULTS

Results into
misdiagnosis 
increase of false
negatives

19
20
21
22
Increase sensitivity
Increasing the ability of the test to identify TP
Decreasing the ability of the test to identify TN

23
Increase specificity
Increasing the ability of the test to identify TN
Decreasing the ability of the test to identify TP

24
Gordis (2013)
25
100% Sensitivity 
higher chance of
having FP

Gordis (2013)
26
100% Specificity 
higher chance of
detecting TN

Gordis (2013) 27
The choice of a high or a low cut-off point would be
dependent on the objectives regarding identifying
true positives or true negatives

28
SENSITIVITY AND SPECIFICITY
USE OF MULTIPLE TESTS
 Sequential or two-stage testing
 Simultaneous testing

29
SEQUENTIAL OR TWO-STAGE TESTING
 Initial test – less expensive, less invasive or less uncomfortable
 Positive  recalled for further testing; more expensive, more
invasive or more uncomfortable procedure which may have a
greater sensitivity and specificity

30
SIMULTANEOUS TESTING
 Using two tests at the same time
 To determine net sensitivity  an individual must be identified as
positive by test A, test B and both tests
 To determine the net specificity  an individual must be identified
as negative by both tests

31
RECEIVER OPERATING
CHARACTERISTICS (ROC CURVE)
 A plot of the sensitivity or (true-positive rate) to the false-
positive rate

32
 Interpretation:
 The closer an ROC
curve to the upper left-
hand corner of the
graph – more accurate
it is
 True positive rate = 1;
False positive rate = 0

33
 Excellent: 0.90 – 1.00
 Good: 0.80 – 0.90
 Fair: 0.70 - 0.80
 Poor: 0.60 – 0.70
 Fail: 0.50 – 0.60

The accuracy of the test would depend on

how well the test separates the group
being tested into those with and without
the disease in question

34
The area under the curve
would indicate the test’s
overall performance

35
PREDICTIVE VALUE
The probability that a patient does or does not have a disease

Gordis (2013)
36
ASSESSING
AGREEMENT

37
RELIABILITY (REPEATABILITY) OF TESTS
• Intrasubject variation – variation within individual subjects

• Intraobserver and Interobserver variation – variation between

those reading the test result
• Highly subjective
• Two examiners often do not have the same result

38
RELIABILITY (REPEATABILITY) OF TESTS

Gordis (2013)
39
KAPPA STATISTIC
 To what extent does the agreement between the two observers
exceed the level of agreement that would result just from chance?

Gordis (2013)

40
KAPPA STATISTIC
 Quantifies the extent to which the observed agreement that the
observers achieved exceeds that which would be expected by
chance alone

 Proportion of the maximum improvement that could occur beyond

the agreement by chance alone

41
  >0.75 – excellent agreement
 0.40 – 0.75 – intermediate to
good agreement
 <0.40 – poor agreement

Gordis (2013)

42
SURVIVAL
ANALYSIS

43
PROGNOSIS AND NATURAL HISTORY OF
DISEASE

• Differences in time of
seeking medical care

• Differences in the natural

history of disease

• Patients who have died are

excluded from the study

Gordis (2013)
44
IMPORTANCE OF QUANTIFICATION OF THE
NATURAL HISTORY OF DISEASE
• Establishing the severity of the disease  priorities for clinical
services and public health programs
• Establishing prognosis
• Comparison of the effectiveness of new treatments/procedures
among individuals or subjects receiving it

45
SURVIVAL ANALYSIS
• Concerned with the time it takes an individual to reach an endpoint
of interest

• It is the length of time for the patients to reach the endpoint NOT
whether or not he or she reaches the endpoint

• Censored data

46
SURVIVAL ANALYSIS

47
DEFINITIONS
• Event – outcome of interest which is typically death but can be a
non-fatal or favorable outcome (i.e.: recovery, discharge from
hospital)

• Censored observation – used to describe participants who

withdraw from the study or who do not experience the outcome of
interest

48
METHODS OF DESCRIBING THE NATURAL
HISTORY OF A DISEASE
• Case-fatality
• Five-year survival
• Observed survival
• Median survival time
• Relative survival

49
OBSERVED SURVIVAL: LIFE TABLE
Analysis of the survival experience or data of patients in a study

Gordis (2013)

50
OBSERVED SURVIVAL: LIFE TABLE
Probability of surviving the first year

Gordis (2013) 51
OBSERVED SURVIVAL: LIFE TABLE
Probability of surviving for the consecutive years

Gordis (2013) 52
Gordis (2013)
53
Gordis (2013) 54
 CALCULATION OF A LIFE TABLE
Column 1 Interval since beginning of treatment

Column 2 The number of study subjects who were alive at the beginning of each
interval
Column 3 The number of study subjects who died during the interval

Column 4 Number of withdrawals or loss-to-follow up

Column 5 Number of people at risk of dying [Col 2 – 0.5(Col 4)]

Column 6 Proportion of people who died during the interval [Col 3/Col5]

Column 7 Proportion of people who did not die [1.0 – Col 6]

Column 8 Proportion of people who survived from enrollment to end

55
Gordis (2013) 56
 KAPLAN-MEIER METHOD
• Identification of the exact point in time when each death occurred
• Each death would terminate the previous interval and a new
interval (new row) is started

Gordis (2013)
57
KAPLAN-MEIER METHOD
• This would allow for some participants to be followed for longer
periods of time than others and for the event to have not yet
occurred in all participants

Gordis (2013)
58
KAPLAN-MEIER METHOD

59
Gordis (2013)

60
 CALCULATION OF A KAPLAN-MEIER TABLE
Column 1 Time of death from enrollment/treatment initiation

Column 2 Number of patients who were alive and followed up at the time of that
death, including those who died at that time
Column 3 Number who died at the time

Column 4 Proportion of those who were alive and followed [Col 3/Col 2]

Column 5 Proportion of those who were alive and survived [1.0 – Col 4]

Column 6 Cumulative survival / the proportion of those who were initially

enrolled and survived to that point

61
CENSORED DATA
Used to describe
participants who are lost to
follow-up

62
RIGHT-CENSORED LEFT-CENSORED
DATA DATA
Occurs when a subject Occurs when the event of
leaves the study before an interest has already
event occurs or the study happened before enrollment
ends before the event has
occurred

63
64
COMPARISON OF SURVIVAL ANALYSIS
• Objective: assessing the impact of a number of factors of interest on
survival (i.e. treatment, disease severity)

• Survival curves can be plotted separately for subgroups of patients

• Testing of significant differences in progression rates between

different groups

65
COMPARISON OF SURVIVAL ANALYSIS
• Log-rank test
• Non-parametric test that would address the null hypothesis that
there are no differences in survival times in the groups being
studies and compared events occurring at all time points on the
survival curve

• Regression model
• Quantifies relationships between one or more factors of interest
and survival

66
COX-PROPORTIONAL HAZARDS MODEL
• Used to test independent effects of a number of explanatory
variables on the hazard (endpoint or event)

67
HAZARDS RATIO
• An expression of the hazard or chance of events occurring in the
treatment arm as a ratio of the hazard of the events occurring in
the control arm

• Interpreted like the relative risk or odds ratio

• >1: the treatment group has a higher chance of experiencing an
event compared to the control group

68
SYSTEMATIC REVIEW
AND META-ANALYSIS

69
SYSTEMATIC REVIEW
• A formalized process of combining
information from all relevant studies
of the same health condition

• Clinical trials or observational

studies

70
OBJECTIVES OF A SYSTEMATIC REVIEW
• Refinement and reduction
• Efficiency
• Generalizability and consistency
• Reliability
• Power and precision

71
META-ANALYSIS
• Type of systematic review that focuses on numerical results which
aims to combine the results from several, relevant independent studies

• To provide an estimate of the overall or average effect of interest

• Direction of the magnitude of the overall effect, confidence result and

hypotheses test results

72
STATISTICAL APPROACH
1. Deciding on the effect of interest and evaluating it for each study
2. Check for statistical homogeneity and obtain an estimate of
statistical heterogeneity
3. Estimate the average effect of interest (with confidence intervals)
and perform the appropriate hypothesis test on the effect
4. Interpreting the results and presenting the findings

73
MEASURE OF EFFECT
 Numerical – difference in treatment means; 0 would mean that
there is no treatment effect

 Binary – risks of the outcome in the treatment group; the effect

may be the difference in risks or their ratio (Relative Risk) wherein
difference in risks (0) and RR (1) would mean no treatment effect

74
STATISTICAL HETEROGENEITY
 May be caused by:
 clinical differences between studies
 Methodological differences between studies
 Unknown study characteristics

75
STATISTICAL HETEROGENEITY
 Test of homogeneity – low power to detect statistical heterogeneity or
may give highly significant results when there are many studies involved

 Index or I2 - percentage of total variation across studies due to

heterogeneity; can be used to assess the impact of heterogeneity and
assess inconsistencies
 0% - no observed heterogeneity
 25% - low heterogeneity
 50% - moderate heterogeneity
 75% - high heterogeneity
76
ESTIMATING THE AVERAGE EFFECT OF
INTEREST
 The average estimate is usually the weighted mean of the estimated
effects from all the studies wherein the weight of each study is the
inverse of the variance of the estimate

 Fixed-effects model meta-analysis – no statistical heterogeneity

 Random effects model meta-analysis, stratifying subgroups and

performing individual fixed effects meta-analysis in each stratum and
meta-regression – statistical heterogeneity

77
 FIXED-EFFECT MODEL RANDOM EFFECTS MODEL
 True treatment effect is the same in  Separate studies represent a
every study random sample from a population
of studies
 Any observed variation in the
estimates from different studies is  Mean treatment effect of individual
only due to sampling error studies varies

 Within-study variability  Within-study and Between-study

variability

 Standard error of estimate is

greater, CI for the true average
effect is wider and P-value is larger

78
INTERPRETATION OF RESULTS
 Provide a summary of the following: sample size, baseline
characteristics, effect of interest (OR, RR etc.), related Cis

 Forest plot

79
80
81
82
IMPORTANT CONSIDERATIONS FOR META-
ANALYSIS
 Publication bias
 Clinical heterogeneity
 Quality differences
 Dependence

83
PUBLICATION BIAS
 The tendency to include in the
analysis only the results from
published papers which favor
statistically significant findings

 Detected through funnel plot

84
FUNNEL PLOT
A scatter diagram which usually has some measure of study size or
standard error on the vertical axis and the treatment effect (i.e. OR, RR)
on the horizontal axis

85
SENSITIVITY ANALYSIS
 Assesses the robustness of the common estimate; to determine
whether any particular study in a meta-analysis strongly influences the
average measure of effect

 Influence plot  any estimate which appears on visual inspection to

differ substantially from others may flagged as an influential study

 Cumulative meta-analysis  adding studies one by one in a specified

order and performing a separate meta-analysis on the accumulated
studies after each addition

86
LEVELS OF
EVIDENCE

87
88
89