helen looker mph course october 21 st 2014. understanding clinical trials a b super drug wonder drug
TRANSCRIPT
Helen LookerMPH Course
October 21st 2014
Understanding Understanding Clinical TrialsClinical Trials
A B
SuperDrug
WonderDrug
Why clinical trials
Design features
Main problems
quart of cider
3 servings elixir of vitriol
6 spoonfuls of vinegar
3 portions of nutmeg
2 oranges, 1 lemon
Experimental treatments (per day)
Limitations of theory
Previous disasters
Spontaneous improvements
Importance of small effects
Intervention Impact of intervention
Sleep baby on its front SIDS frequency increased 200%
Anti-oxidants to reduce mortality
5% increase in mortality
Juvenile delinquents exposed to prison – scared straight
Increase in offending rates
School based driver education
Increase in accident rates
Service organisation
Quality of direct care
Ancillary care
Initial severity
Co-morbidity
Adherence
Social support
Lifestyle
Drugs
Surgery
Type of management
Public health interventions
Phase I clinical pharmacology
Phase II initial clinical assessment
Phase III rigorous testing
Phase IV post-marketing surveillance
Trials are experiments on people
Must be real doubt (clinical equipoise)
Obtain informed consent
Preserve clinical freedom
Historical controls◦health care has moved on◦same diagnostic criteria??
Concurrent control◦why were some not selected
Randomized controls◦ new treatment◦ placebo/ conventional treatment
Streptomycin and Pulmonary TB
Intervention Control
Recruited 55 52
Dead at 6 months
7%
27%
Published 1948
Like tossing a coin
Avoids choosing
Permits fair comparison◦ two groups the same at baseline
two groups the same at baseline◦ group A: 1x , 4 x , 2 x
◦ group B: 1x , 4 x , 2 x
A A A A A AB BB B B A B B
Factors which might influence outcome
◦ Illness severity at entry
◦Current treatment
◦Disease duration
◦Relevant previous medical history
New treatment
Control
% advanced disease
47
49
% married 65 70
mean age (years)
64.2
48.9
Severe
Moderate
Moderate
A
B
A
B
A
B
When powerful predictor of outcome
Construct groups low to high
Randomise within groups
Achieve balance on the predictor
Keep it simple
Inclusion criteria◦ likely to benefit from treatment
definitely has the disease patient is likely to respond
◦unlikely to be harmed no known adverse reactions/ contraindications
Aim◦well defined, homogeneous group◦ increased likelihood of detecting an effect ◦smaller, cheaper trials
Exclusion criteria◦clear preference for intervention or
control by patient or doctor
◦??? patient unlikely to adhere to treatment complete the follow-up
◦many other factors
Entry Criteria
◦Diastolic BP 90-109
◦Age 35-64
◦Men and women
Common exclusion criteria
◦Comorbidities
◦Recently cardiovascular disease
◦Age: 50+ yrs, some 85+ years
◦Severe hypertension
◦Dementia
◦Depression
◦Substance abuseVan Spall et al 2007Uijen et al 2007
Compared to primary care patients with CVD, trial participants were:
◦Younger
◦More men
◦Lower risk CVD
◦Fewer with history of CVD
Uijen et al 2007
Death rateTrial participant 3.6%
Eligible, not enrolled
7.1%
Not eligible 11.4%
Steg et al 2007
Global Registry of Acute Coronary Events
Systematic reviews
% of ineligible patients (median)
Asthma 94%
COPD 95%
Wound healing 85%
Well-defined
Easily delivered
Same for all patients
Prior evidence of effect
Measurable outcome
Drug
Surgery
Psychotherapy
Counselling
Complex intervention
Difficulty of standardising
Death
Symptoms Quality of life
Clinical measurement
Clinically relevant
Easily measured
Accurately measured◦ measured in the same way for intervention and control
groups
Specified in trial protocols◦ primary out identified◦ others called secondary outcomes
Single blind Vs.
Double blind
Increase in effect size
Poor randomisation 41%
Not double blind 17%
Schultz et al 1995
Healthy ones may emigrate
Sick ones may be admitted to hospital
Rule of thumb◦ less than 20% lost◦similar losses for intervention and control
Specify treatment
Define study group
Random allocation
Blinded outcome assessment
Fair interpretation
Like tossing a coin
Avoids choosing
Protects against unknown confounding
Permits fair comparison
Too few patients
Performance bias
Losing patients
Flawed analysis/interpretation
Expected effect size◦ The bigger the effect you are trying to measure the fewer
people needed
How certain do you need to be that a detected difference is true?◦ Typically aim for a significance criteria of 0.05 (ie if you
find a difference between groups you have a 95% confidence that the difference is true)
Power to detect a difference when there is a true difference ?◦ Typically select a range of 80-90%
Systematic differences in the care provided to the participants in the comparison groups other than the intervention under investigation.
Usually intervention get more attention than controls
Example in psychotherapy ◦ intervention group may get more attention or more
drugs than controls◦ Example in diabetes
◦ Increased clinical contact improves diabetes control
Why do patients drop out?
Treatment side-effects
Lack of desired improvement
Too much time/effort to continue
Maintained contact with study
Analyse by intention to treat◦ change in treatment may be related to
efficacy ◦ exclusion will lead to bias
Beware sub-group analysis◦ for every 20 groups explored one will be
spuriously significant
◦ 30% reported severity of RA
◦ Randomisation described in 11%
◦ 8% of reported double blind were not
◦ 6% nominated main outcome measure
◦ 1% report sample size calculation
◦ 42% made doubtful/invalid statements
Bero and Rennie 1996
857 significance tests
48 were significant
43 expected by chance
What You Should KnowWhat You Should Know New treatments need careful
assessment
The RCT is study design of choice randomisation enables fair comparison controls for confounding: known and unknown also need fair outcome assessment
Problems often occur unequal at baseline, lack of blinding , loss to follow-up, poor outcome measurement, flawed analysis
Papers may mislead
◦ Who is being studied
◦ What treatments are being used
◦ Are treatment groups comparable at baseline
◦ Are sensible outcome measures used
◦ Are outcomes assessed blind
◦ How many patients dropped out
◦ What do the results really mean