Lecture 1 - Health Behaviour Change & RCTs

Question 1

Choosing a method

Answer 1

• consider:
  > nature of data (quant or qual)
  > design (experimental, or non-)
  > ensure validity & reliability

Question 1

quantitative methods

Answer 1

• numerical data to quantify variables and for statistical estimation/inference
• focus on what & how much
• test hypotheses/theories
• data analysed using statistical methods
• quantify behaviour
• establish general patterns
• test theories of hypotheses

Question 2

qualitative methods

Answer 2

• non-numerical
• focus on why & how
• generate theories based on data
• data are analysed via categorisation & interpretation
• explore new research areas/phenomena
• understand deep psychological processes
• study indiv experiences
• can be combined with quant = mixed methods design

Question 3

experiments

Answer 3

• manipulates one variable and measures DV
• extraneous variables = any variables other than the IV that may affect the DV
• EV become CV when their values change systematically with the levels of the IV
• effects minimised through true randomisation

Question 4

randomised controlled trials

Answer 4

• subset of possible experimental designs
• a planned experiment to assess intervention accuracy by comparing intervention to control groups
• allocation is random
• true randomisation:
  > eliminates bias in treatment
  > ensures differences can be attributed to the treatment
  > permits use of probability theory to express likelihood that any difference in outcome between treatment/intervention groups merely indicates chance

Question 5

why conduct RCTs

Answer 5

• no self selection, observer bias or influence by secular trends (before and after study)
• provides gold standard for proof of concept
  > eligibility criteria - very stringent
  > specified hypotheses
  > predefined intervention & control groups
  > primary and secondary outcomes/endpoints (e.g. behavioural change, HIV incidence) to address hypotheses
  > methods for enrolment and follow up
  > rigorous monitoring
  > analysis plans & stopping rules
  > comprehensive reporting of methods and data analysis

Question 6

History of RCTs

Answer 6

• first conducted in 1747 by James Lind examining impact of citrus in scurvy. Impacted people who travlled by boats & found it was lack of vit c (gave one group citrus & one nothing)
• first published RCT in medicine: streptomycin treatment of pumonary tuberculosis (1948)

Question 7

RCTs in non-medical behavioural research

Answer 7

• nudging (small nudge to lead to big change on behaviour) explored how different changes in how info is presented may change decision making
• e.g. if you are told you are using more energy than others you tend to reduce. If told you are using less, may increase.

Question 8

types of RCTs

Answer 8

• individually randomised trials
  > eligible indivs are randomised on an indiv basis. like many normal studies e.g. conventional medical RCTs and behavioural RCTs
  > self selection of persons volunteering for enrolment
• cluster randomised trials
  > clusters e.g. communities are randomised and all consenting people enrolled
  > used if not feasible to individually randomise people to interventions
  > less indiv level self-selection = inc generalisability
  > acceptability and reduced stigma (everyone in cluster has same treatment)
• BUT cluster randomisation more vulnerable to lack of comparability between study groups than individual randomisation
• cluster RCTs inc sample size requirements and are less efficient than indiv RCTs due to intra-cluster correlation

Question 9

steps in RCT

Answer 9

1. conduct power analysis - provides size of the sample that must be used to detect effects and ensure validity
2. register trial
3. enrollment - select and define study sample
4. baseline assessment - measure DV and mediating variables e.g. knowledge, norms, attitudes, blood levels etc
5. allocation - randomise
6. implement study conditions - implement HPP and control
7. follow up - follow pp’s and post HPP to monitor changes in IV and outcomes
8. analysis - compare changes in outcomes observed among pp’s in the HPP condition with those observed among participants in the comparison condition
9. prepare report describing process and trial findings
10. reporting trial findings

Question 10

sample size calculation

Answer 10

• a priori statistical analysis to determine no. pp’s needed in RCT to detect meaningful effect
• effect size = magnitude of dif between two groups in RCT
• sample size calc for individually randomised RCTs:
  > specify type I and II error (e.g. power >80% to detect signif at .05)
  > specify expected or meaningful difference in outcome rates between intervention and control groups
  > estimate losses to follow up on primary outcome
  > estimate required sample size at enrolment

Question 11

registering protocols for RCT

Answer 11

• ensures we provide info for pp’s, reduce publication bias, help editors and grant bodies to understand research
• registry purpose:
  > ethical obligation
  > provide info to potential pp’s and referring clinicians
  > reduce publication bias
  > help editors and others understand context of study results
  > promote more efficient allocation of research funds
  > help review boards determine appropriateness of study
• group that benefits:
  > patients, public, research community
  > patients/clinicians
  > users of medical literature
  > journal editors
  > granting agencies
  > ethicits, IRBs

Question 12

outcomes

Answer 12

• primary outcome - the outcome that an investigator considers most important among the many outcomes examined. need a primary interest and do sample size calc for this
  > must be defined
  > reduces risk of type I error from statistical testing of many outcomes
  > reduces risk of type II error by providing basis for estimation of sample size necessary for powered study
• secondary outcome - provide info on therapeutic effects of secondary importance, side effects or tolerability

Question 13

enrolment

Answer 13

• eligibility criteria establish parameters for determining inclusion and exclusion criteria
• eligibility is predefined to:
  > ensure pp’s meet criteria for intervention
  > also defined by age, gender, race, health etc
  > narrower the elligibility criteria the less generalisable
  > must consent to screening eligibility
• enrolment after eligibility established and provided consent

Question 14

baseline assessment

Answer 14

• assess all pp’s on measures for primary and secondary outcomes and mediating variables prior to treatment
• create a logic model depicting hypothesised path

Question 15

random allocation (randomisation)

Answer 15

• 2 features to consider:
  1. implementing valid randomisation procedure
  2. establishing procedures to safeguard the integrity of randomisation procedure so unintentional or intentional biases do not influence the pp allocation

Question 16

implementing a valid randomisation procedure

Answer 16

• must be by chance
  > e.g. random numbers, computerised, but not randomised based on characteristic
• to maintain balance between groups can:
  > indiv match
  > block randomisation in groups of 10 - 20
  > stratify randomisation
• simple randomisation = analogous to coin toss.
• restricted randomisation/blocking = done to ensure equal balance across groups e.g. block of 6 have 3 controls and 3 intervention pps
• stratified randomisation - indivs are identified based on categories and then randomised
• dynamic or adaptive methods - not pre-defined, only first pp truly random

Question 17

safeguarding the integrity of randomisation

Answer 17

• concealment of allocation strategies mask pp knowledge of their group
• priot to study main investigator can generate allocation sequence & not be shared with others

Question 18

implementing study group/design

Answer 18

• blinding minimises pp or researcher bias
  > single blinding - researcher but not pp knows randomisation group
  > double blinding - neither researcher nor pp know group of randomisation
  > triple blinding - researcher pp and statistician do not know group randomisation
  > unblinded - cannot conceal randomisation

Question 19

concealment of allocation vs blinding

Answer 19

• concealment of allocation - procedure to protect randomisation process before subject enters RCT. always feasible
• blinding - masking of treatments after randomisation once trial begins. not always feasible

Question 20

ensuring fidelity to protocol

Answer 20

• need to ensure fidelity to protocol - extent to which the outcome measures and intervention are administered in accord with the registered protocol
• any violations to protocol need to be recorded and reported in RCT publication (with sensitivity analyses conducted for deviations)

Question 21

follow up

Answer 21

• conducted at predetermined intervals to detect outcomes
• frequency/duration depend on:
  > type of outcome e.g. response to treatment, progression of disease
  > level of risk: higher = more frequent follow up
• losses to follow up must be minimised because:
  > losses are often selective e.g. high risk persons, low socio-economic status etc and this introduces bias
  > losses to follow up should be comparable in intervention and control groups to avoid biased comparisons if not this leads to attrition bias
  > losses to follow up lead to reducing power by reducing person-time observation

Question 22

analysis

Answer 22

• intention to treat
  > analyse all people randomised even if some did drop out before
  > analysis based on group pp was randomly allocated to
  > least biased & most conservativ
• as treated
  > analyse those who complete RCT
  > only inc those pps who completed treatment originally allocated to
  > potentially biased by selection of most compliant and lowest risk pop

Question 23

reporting RCT

Answer 23

• CONSORT flow diagram present progress of pp’s through phased of RCT

Question 24

are RCT results always valid?

Answer 24

- can provide conflicting results = important to carry out reviews/meta analyses - can be flawed in design & reporting etc - intervention vs control is internally valid but may not be externally valid - RCT could suffer conflict of interest