Introduction to Clinical Epidemiology

1
Introduction to Clinical Epidemiology

• Dr Rodolphe Thiébaut,
• Institut de Santé Publique, dEpidémiologie et de
  Développement (ISPED), Bordeaux, France

2
Aknowledgements

• Pr Roger Salamon Pr Geneviève Chêne
• Staff from Unit for clinical and epidemiological
  research of Bordeaux University Hospital
• Staffs from INSERM Unit 593 (Epidemiology) and
  Unit 875 (Biostatistics)

3
Objectives and outline

• Objectives
• Methodology of clinical trials. What is
  essential?
• Epidemiological studies examples and theory
• Organisation of the talk
• Issues and solutions through examples

• Ask question whenever you want

4
Examples of clinical questions
5
(No Transcript)
6
Take home message

• The question should be clearly defined

7
How to answer ?

• Personal experience and biology of disease are
  valuable but not sufficient
• Mainly because of variability
• need of a systematic approach to answer clinical
  questions

8
Clinical epidemiology

• Science that develops and uses epidemiological
  methods to evaluate clinical innovations (A
  Feinstein)
• Aims
• provide clinicians with optimal strategies for
  diagnosis, treatment and prognosis, improvement
  of clinical decision making
• better clinical decision results from studies
  evidencing real facts, evidence-based medicine

9
Clinical research
Epidemiology
Clinical research
Epidemiology
patient-oriented research
population-oriented research

• patient-oriented research
• context

• population-oriented research
• methods

Clinical epidemiology
Clinical epidemiology
 A basic science for clinical medicine (Sackett
et al. 1991)
10
What we want?
11
Precision (lack of random error)

• Unprecise Precise
• Number of subjects necessary to include

12
Validity (lack of systematic error)

• Bias process at any stage tending to produce
  results that depart systematically from the true
  values Biased Unbiased

13
How?Issues and solutions
14
Example Congenital toxoplasmosis
15
Example Congenital toxoplasmosis

• France
• J.O n 40 du 16 février 1992
• Décret no 92-143 du 14 février 1992 relatif aux
  examens obligatoires prénuptial, pré et
  postnatal
• En outre, la sérologie toxoplasmique sera répétée
  chaque mois à partir du deuxième examen prénatal
  si l'immunité n'est pas acquise.

• UK

16
Example Congenital toxoplasmosis

• EFFECT ???
• Couvreur NEJM 1974
• Foulon AMJOG 1999

• EFFECT ???
• Gilbert IJE 2001
• EMSCOT BJOG 2002

17
Example Congenital toxoplasmosis
18
Example of confounding bias
Prenatal treatmente.g. spiramycin
Transmission of Toxoplasma gondii
Gestational age at seroconversion
19
(No Transcript)
20
Example Congenital toxoplasmosis
Prenatal treatmentSpiramycin vs. none
Transmission of Toxoplasma gondii
Unadjusted OR0.83 Adjusted OR1.47(source
SYROCOT technical report on transmission
analysis)
21
Key points

• Confounding an issue at the time of analysis
• Could be controlled by adjustment
• Other solutions
• Matching
• Randomisation (to be continued)

22
Example of indication (channelling) bias
23
Example of indication bias
HR non ajusté 4.3
Traitement IP
Progression clinique
HR non ajusté 7.8
Stade SIDA initial
JAIDS 200333380-6
24
Which treatment for these issues?
25
Comparaison (control)
???
???
26
(No Transcript)
27
Randomisation

• Goal
• To have comparable groups
• Avoid that differences between groups are
  something different than treatment
• Unbiased allocation of treatment
• !!! If strictly adhered to, alternation is
  unbiased
• Randomisation based on random numbers avoid the
  prediction of what treatment a patient will
  receive

28
Insu

• Objective to keep the comparability between the
  groups

29
BMJ 200132342-6
30
Types of epidemiologic studies

• Experimental studies
• Clinical trials
• Control
• Randomization
• Blinding (masking)
• Community intervention and cluster randomized
  trials

• Observational studies
• Cohort studies (exposed/non exposed)
• Case-control studies
• Cross-sectional studies
• Ecologic studies

31
What is the best design?
Case series lt case-control lt observ. cohort lt
randomized
32
What is the best design?
AIDS 1999132075-82
33
What is the best design?
AIDS 1999132075-82
34
Example of postmenopausal hormone use and CHD

• Lower rates of CHD (35 to 80) for women taking
  estrogen during postmenopausal period in
  observational studies

35
Example of postmenopausal hormone use and CHD
36

• But observational studies might help particularly
  when the exposure allocation is unrelated to the
  outcome

37
(No Transcript)
38
Example of attrition bias
39
Example of attrition bias
40
LOCF Last Observation Carried Forward
41
LOCF
Take values from  closest  patients
Take values from  closest  patients
AIDS 1998121155-61
42
LOCF

• Strong assumption the last value is
  representative of the future ones
• Most often wrong !!!

43
Even better maximum likelihood

• Statistical methods taking into account
  unbalanced data
• More precision and less biased estimates compare
  to  complete case analysis 

44
Take home message

• Sophisticated analyses do not replace good data
   garbage in, garbage out  (GIGO)

45
Intent-to-treat (ITT)
ANRS 005/ACTG 154 international RCT designed to
assess the effectiveness of pyrimethamine for the
primary prophylaxis of cerebral toxoplasmosis
Controlled Clinical Trials 199819233-48
46
Intent-to-treat (ITT)
ITT Takes into account all patients according to
their initial treatment group as designated by
the randomization procedure gt maintain
comparability
Controlled Clinical Trials 199819233-48
47
Intent-to-treat (ITT)
on-treatment analysis loses the benefits of
initial randomization because it accounts only
for the patients who continue the study medication
48
Intent-to-treat (ITT)
Controlled Clinical Trials 199819233-48
49
Statistical risk

• Alpha probability of concluding for A difference
  although THERE IS NOTgt p value a posteriori

• Beta probability of concluding for NO difference
  although THERE ISgt Power1-Beta

50
Lack of statistical power

• Why there is no significant association between
  prenatal treatment and clinical signs in CT?The
  3 hypotheses
• Reality no association
• Biased estimation
• Lack of statistical power (N Needed gt2300)

51
The multiple test issue
52
The multiple test issue
53
The multiple test issue
54
How to deal with this issue

• Define the question a priori
• Correct for multiple testing

55
P value
BMC Medical Research Methodology 2004413
56
P value
Highest decimal were more likely rounded
BMC Medical Research Methodology 2004413
57
Interpretation of results
58
(No Transcript)
59
(No Transcript)
60
Which is the worst risk factor?
Exposure to PI per 5 additional year
exp(5ln(1.17))2.19
61
Summary

• Define clearly the question
• Collect good data (GIGO) few missing data
• Best design?
• Controlled
• Randomized (initial comparability)
• Blinding ITT (subsequent comparability)
• Other important issues
• Number of included and followed subjects
• Interventions identical in all groups