Grant Writing: Study design in clinical research

1
Grant WritingStudy design in clinical research

• Huib Burger, MD, PhD
• Department of Epidemiology
• Department of Psychiatry
• University Medical Center Groningen
• June 14, 2007

2
Quiz

• Most common reason for rejection is
• A. uninteresting research question
• B. methodological flaws in the design
• C. unfeasible study
• D. excessive budget

3
Prevention of methodological flaws

• Carefully address your study design, i.e.
• Theoretical design research question the type
  of study
• Operational design

4
The study type

• Etiologic study (aim is causal explanation)
• Diagnostic or screening study (aim is prediction)
  
• Prognostic study (aim is prediction)
• Intervention study (aim is causal explanation)

5
Prognostic and diagnostic research is prediction
researchExamples

• Prognosis is it possible to predict early
  relapse in early-stage Hodgkin disease from age,
  gender, erythrocyte sedimentation rate
• Diagnosis can we accurately assess the
  probability of the presence of osteoarthritis
  from age, gender, joint stiffness and typical
  pain? 2. Is there added value of physical exam?

6
The research question

• Be sure that you exactly write down the question
  you want to answer
• Poor formulations ex
• the role of estrogens in breast cancer
• Better
• Is the risk of breast concer increased by higher
  levels of endogenous estrogens?

7
The research question

• Preferably define your research question in terms
  of
• Determinants
• Outcomes
• Confounders (causal studies)
• Effect modifiers
• Mediating variables

8
Operational design

• Observational / experimental
• Cohort / case-control
• Cross-sectional / longitudinal
• Prospective/retrospective
• Defend your choice!
• 5. Research plan (study population, measurements,
  etc.)

9
Study Population

• Be sure that you
• sample from the domain, i.e. the abstract
  formulation of the target population
• address representativeness of your sample for the
  target population, if relevant

10
Example

• Aim
• to predict mortality in preterm neonates admitted
  to neonatal intensive care units from birth
  weight, duration of gestation, congenital
  malformations, and several physiologic parameters
  
• Should one exclude newborns with inevitably
  lethal conditions?

The International Neonatal Network. The CRIB
(clinical risk index for babies) score a tool
for assessing initial neonatal risk and comparing
performance of neonatal intensive care units.
Lancet 1993 324 193-8
11
Study Population

• Describe clearly
• the sampling frame (general population, primary
  care)
• the methodology random, stratified, case-control
• The practice of sampling (e.g. approaching
  consecutive patients by a research nurse)

12
Study Population

• In- or exclusion criteria
• Strict (intervention studies) or loose
  (diagnostic and prognostic)
• Age or gender restrictions (males in CVD studies)
• (Co) morbidity
• Medication
• Language skills
• Provide motivation for your choices

13
Determinants

• Describe measurement (as in practice or not)
• Address precision (reproduceability) and validity
  of your instruments give key references
• Describe the source of the data (medical charts,
  registrations, lab files etc)
• Take care that among the determinants are the
  potential confounders (in causal research) and
  effect modifiers

14
Intervention

• Describe the exact nature of the intervention
• Single intervention e.g. medication, surgery,
  psychotherapy, placebo (explanatory RCTs)
• Strategy e.g. care as usual, stepped care,
  psychotherapy with medication on demand
  (pragmatic RCTs)

15
Intervention

• Motivate the choice and give argumentations why
  it would work
• Is the intervention feasible PILOT!
• Describe the process of randomisation (computer
  generated list, the use of internet, blinding,
  (stratified block randomisation, minimization)

16
Follow-up

• Motivate frequency and time period of follow-up
• What do you do to prevent lost to follow-up?
• How do you optimize compliance (intervention
  studies), e.g. placebo run-in.
• What is the procedure in case of unexpected
  adverse events, unexpected pathology, severe
  worsening of a condition, etc.

17
Outcomes

• Define unambiguously primary and secondary
  outcomes (relevance)
• Describe assessment carefully (clinical,
  registration data)
• Multiple outcomes
• Multiple testing problems
• May provide essential information

18
Example multiple endpoints
19
Sample size calculation

• Be realistic !
• Underpin your estimates consider pilot
• Ingredients alpha, beta, SD of outcome/base
  rate, clinically relevant effect size,
  non-response
• Many computer programs available e.g.
• PS from Vanderbilt University

http//biostat.mc.vanderbilt.edu/twiki/bin/view/Ma
in/PowerSampleSize
20
Data analysis

• Provide a strategy for your data-analysis
• Keep the study objectives in mind
  (explanation/prediction)
• Dont use unnecessary technical terms
• In principle all analyses must be specified
• Trend of less freedom (subgroup analysis)

21
Most importantly

• Make sure that you master
• The methodological area (e.g. clinical
  epidemiology, public health, medical technology
  assessment)
• The subject matter (e.g. oncology, cardiology)

22
Further reading on Clinical Epidemiology

• Burger H, Hofman A. Klinische Epidemiologie. In
  Van der Meer J, CDA Stehouwer (red). Interne
  Geneeskunde. Houten Bohn, Stafleu, Van Loghum,
  20051-15.
• Weiss NS. Clinical epidemiology. The study of the
  outcome of illness. New York Oxford University
  Press 1986.
• Feinstein, A.R. Clinical Epidemiology The
  Architecture of Clinical Research. Philadelphia
  W. B. Saunders Co, 1985.