Randomized Trials:
Some Further Issues
Dr. Win Aye Hlaing
Lecturer
Department of Epidemiology
University of Public Health, Yangon
• Sample Size
• Type I Error
• Type II Error
• Power
• External Validity (Generalizability)
• Internal Validity
Sample Size
• two jars of beads, each containing 100
• some white and some blue
• jars are opaque
• cannot see the colors
• 10 from A – 9 B & 1 W
• 10 from B – 2 B & 8 W
• Can we conclude that 90% of the beads in jar
(A) are blue and 10% are white?
• Clearly, we cannot
• by chance
• 10 from A – 7 B & 3 W
• 10 from B – 7 B & 3 W
• It clearly could concluded color distribution be
the same
• We are looking at samples and trying to draw a
conclusion regarding a whole universe
• We want to draw a conclusion from the study
results that goes beyond the study population—is
• treatment (A) more effective than treatment (B)
• 2 × 2 table
• Two columns —therapy A differs from therapy
B, or
• therapy A does not differ from therapy B
• Two rows - our decision to conclude
• four possibilities – a,b,c,d
• Not differ (-) --- differ (+) ----Type I Error (α)(c)
• differ (+) --- not differ (-) ----Type II Error (β)(b)
• α = p value (probability)
• α ‹ 0.05
• Therapy differ (+) - (2) groups-----probability = 1
• b cell = β
• d cell = 1-β = Power
Estimate the Sample Size
1. must specify the expected difference in response rate
• existing therapy cures- 40%
• expect new therapy cure-50%, 60%
• expect 10% or 20% better than current therapy
• A new therapy
– no prior experience
– to search data in human populations
– To search data from animal studies
– No way of producing such an estimate
– Can make a guess [estimate] --- bracket the estimate
2. must have an estimate of the response rate
• E.g. current cure rate is 40%.
3. must specify the level of (α)
• P = 0.05 or 0.01
4. must specify the power (1-β)
• (TP)Power increase=FN decrease=FP decrease
• (1-β) increase = β decrease = α decrease
5. must specify the test - one-sided or two-
sided
• Current T/m — cure rate — 40%
• New therapy—cure rate—50% or 60%
• one-sided test
– Cure rate – difference between (2) T/m
– Cure rate in New? --- better
• two-sided test
– Cure rate – difference between (2) T/m
– Cure rate in New? --- better or worse
• not only for a difference that is better than the
current cure rate
• but also for one that is worse than the current
rate
• One-sided test
– Require smaller sample sizes
– Attractive
– E.g. Disease – 100% fatal
• Only be the direction of improvement
• One-sided test
Gehan Table (Sample Size Table)
• Lower of the two cure rates=higher and lower
Choose lower one
– E.g. Lower = 40%
– Higher = 60%
– Choose lower one --- 40% = 0.4
• Difference in cure rates = higher – lower
– E.g. Lower = 40%
– Higher = 60%
– Choose --- 60-40 = 20% = 0.2
• Choose the intesect row of 0.4 & column of
0.2
97
• We need 97 subjects in each of our study
groups (Sample Size)
Alternative
• If the sample size is too small
– Decide not to do the study (or)
– Decide to extend the study (or)
– Decide to collaborate with investigators at other
Institutions to increase the total no.of subjects
– Test -- single site (bias – difficult to detect)
-- multicentre (bias – readily to detect)
– Sample size – Use – in RCT, Cohort, Case-Control
Recruitment and Retension of Study
Participants
• A major challenge in RCT - to recruit the
sufficient no. of eligible and willing volunteers
• Failure to recruit – trial without sufficient no.
of volunteers to yield statistically valid results
• Trials may be significantly delayed – by limited
recruitment and costs of completing such
trials may be increased
• High level of vigilance is needed
– No coercion, either overt or convert
– By study investigators consciously or
subconsciously
– To enroll the subjects (Bias)
• Participants must be fully informed of the risks
& what arrange/m have been made
• If untoward effects occur (Consent)
• Appropriate arrangements - cover participants’
expenses such as transportation,
accommodations, etc.
payment of cash incentives to prospective
volunteers
will risk or overt coercion
biases and distortion may occur, if large
incentives (+)
• conducting RCT is that we do not know which therapy
is better
Losses to follow-up and other forms of noncompliance
major concern
Participants may lose interest in the study over time
drop out of studies
appropriate measures to prevent losses to follow-up
Ways of Expressing the Results of RCT
I. Reduction in Risk (Efficacy)
• can be expressed as rates of developing disease in the
vaccine and placebo groups
• Showed the extent of the reduction in disease by use
of the vaccine
• Risks are calculated per person-years of observation
• Effectiveness
– ideal conditions ≠ real-life situations
• evaluate efficacy of a treatment
– Efficacious
– Effective
II. Ratios of Risk (Relative Risk)
• A major objective of RCT - to have an impact
on the way clinical medicine and public health
III. NNT
• to estimate the number of patients who
would need to be treated (NNT)
• to prevent one adverse outcome such as one
death
– E.g. Mortality in Untreated = 17%
– Mortality in treated = 12%
– NNT = 1/17-12% = 1/5% = 1/0.05 = 20
• used in studies of various interventions
including both treatment and prevention
IV. NNH
• Number needed to harm
Intrepreting the Results of RCT
• Ultimate objective - to generalize the results
beyond the study population itself
• Evaluate the New Drug
• Defined pop is from Total pop
External Validity (Generalizability)
External Validity (Generalizability)
• The results in the study pop can be applied to
broader pop
• Study pop is representative of Defined pop
• Patients included in study – generalized the result
– applied the patients not included in study
• Generalize the results – not only just to all
patients with disease in community – but also
total pop of patients with disease
• Defined pop is representative of Total pop
Internal Validity
Internal Validity
• A RCT is internally valid:
– if the randomization has been done and
– the study is free of other biases
– and is without any of the major methodologic
problems
Limiting Factors
1. Most RCT do not provide the information
2. Droup outs from trials – should be kept to
minimum
• RCT are usually not representative of General
pop (external validity)
Comparative Effectiveness Research
(CER)
• Some randomized trials - to compare a new
therapy to a placebo
• Other randomized trials – to compare a new
treatment with an older accepted treatment
• two or more existing interventions are
compared - to determine which intervention
would work best in a given population or for a
given patient
• Another issue - costs of interventions
• E.g. many treatments of HIV infections are
very expensive - affordable in developed
countries - but not affordable in many
developing countries
• Newer and cheaper medications - to
determine the new, cheaper alternatives are
as effective as the more expensive
interventions --- “Equivalence Studies”
Four Phases in Testing New Drugs in
USA
• Phase I Trial
– Clinical studies
– Small – 20-80 subjects
– Effects are examined, including safety and side
effects
– Go --- Phase II
• Phase II Trial
– Clinical
– 100-300 subjects
– Evaluate the efficacy
– Further assess its relative safety
– Go --- Phase III
• Phase III Trial
– Large scale RCT
– For effectiveness and relative safety
– 1000-3000 or more subjects
– Very difficult
– More than one study center-Multicenter trial
– can be approved and licensed for marketing
• Phase IV Trial
– Certain adverse effects of drugs
– May not become manifest for many years
– May not be detectable
– Become evident only when use by large pop after
marketing
– Post-marketing Surveillance
– Are important for monitoring new agents-general
use by the public
• Phase IV are not Randomized studies
• Ascertain side effects of new t/m after the
drug is being market
• A very high quality system for reporting of
adverse effects is essential
• Very valuable in providing additional evidence
on benefits and help optimize the use of the
new agent
• The multicenter Hypertension Detection and
Follow-up Program (HDFP) study
• benefits of treating mild to moderate
hypertension
• 22,994 subjects who were eligible because
they had elevated diastolic blood pressure
• 10,940 were randomized to the stepped care
or to the referred care group
The Hypertension Detection and
Follow-up Program
The Hypertension Detection and
Follow-up Program
• Stepped Care - according to a precisely defined
protocol, under which treatment was changed
when a specified decrease in blood pressure
• Referred care - a group receiving no care for
hypertension
• known hypertensive subjects
• the investigators believed that ethically
unjustifiable
• referred back to their own physicians
• the patients in the stepped care group had
lower mortality than those in the referred care
group
• untreated group for comparison
• patients referred back to their own physicians
• but there was no monitoring of the care by
their physicians
REGISTRATION OF CLINICAL TRIALS
• beneficial results have been published
• showing negative results (for some reason)
have not been published
• Publication Bias (or) Non-Publication Bias
• more eager to publish results from studies
showing dramatic effects than
• results from studies showing no benefit from a
new drug
• Both researchers and journals - excited about
studies - a new treatment is inferior to current
treatment or that the findings are not clear
one way or the other
• adopt a policy
• all clinical trials of medical interventions must be
registered in a public trials registry before any
participants are enrolled in the study
• Medical include drugs, surgical procedures,
devices, behavioural treatments, and processes
of health care
• at no charge
• Before publication
ETHICAL CONSIDERATIONS
1. Randomization is ethical?
– Randomization is ethical only when we do not know
whether drug A is better than drug B
– it is ethical to use a placebo
2. Is it ethical not to randomize?
3. Truly informed consent can be obtained?
– Whether they are capable of giving truly informed
consent
4. Under what circumstances should a trial be
stopped earlier than originally planned?
– harmful effects or beneficial effects of the agent
become apparent early
CONCLUSION
• the gold standard
• for evaluating the efficacy of therapeutic,
preventive, and other measures
• in both clinical medicine and public health
6/26/2017 66
THANK YOU!!

Randomized trials ii dr.wah

  • 1.
    Randomized Trials: Some FurtherIssues Dr. Win Aye Hlaing Lecturer Department of Epidemiology University of Public Health, Yangon
  • 2.
    • Sample Size •Type I Error • Type II Error • Power • External Validity (Generalizability) • Internal Validity
  • 3.
    Sample Size • twojars of beads, each containing 100 • some white and some blue • jars are opaque • cannot see the colors
  • 5.
    • 10 fromA – 9 B & 1 W • 10 from B – 2 B & 8 W • Can we conclude that 90% of the beads in jar (A) are blue and 10% are white? • Clearly, we cannot • by chance
  • 7.
    • 10 fromA – 7 B & 3 W • 10 from B – 7 B & 3 W • It clearly could concluded color distribution be the same • We are looking at samples and trying to draw a conclusion regarding a whole universe • We want to draw a conclusion from the study results that goes beyond the study population—is • treatment (A) more effective than treatment (B)
  • 14.
    • 2 ×2 table • Two columns —therapy A differs from therapy B, or • therapy A does not differ from therapy B • Two rows - our decision to conclude • four possibilities – a,b,c,d
  • 15.
    • Not differ(-) --- differ (+) ----Type I Error (α)(c) • differ (+) --- not differ (-) ----Type II Error (β)(b) • α = p value (probability) • α ‹ 0.05 • Therapy differ (+) - (2) groups-----probability = 1 • b cell = β • d cell = 1-β = Power
  • 17.
    Estimate the SampleSize 1. must specify the expected difference in response rate • existing therapy cures- 40% • expect new therapy cure-50%, 60% • expect 10% or 20% better than current therapy • A new therapy – no prior experience – to search data in human populations – To search data from animal studies – No way of producing such an estimate – Can make a guess [estimate] --- bracket the estimate
  • 18.
    2. must havean estimate of the response rate • E.g. current cure rate is 40%.
  • 19.
    3. must specifythe level of (α) • P = 0.05 or 0.01
  • 20.
    4. must specifythe power (1-β) • (TP)Power increase=FN decrease=FP decrease • (1-β) increase = β decrease = α decrease
  • 21.
    5. must specifythe test - one-sided or two- sided • Current T/m — cure rate — 40% • New therapy—cure rate—50% or 60%
  • 22.
    • one-sided test –Cure rate – difference between (2) T/m – Cure rate in New? --- better • two-sided test – Cure rate – difference between (2) T/m – Cure rate in New? --- better or worse
  • 23.
    • not onlyfor a difference that is better than the current cure rate • but also for one that is worse than the current rate
  • 24.
    • One-sided test –Require smaller sample sizes – Attractive – E.g. Disease – 100% fatal • Only be the direction of improvement • One-sided test
  • 25.
    Gehan Table (SampleSize Table) • Lower of the two cure rates=higher and lower Choose lower one – E.g. Lower = 40% – Higher = 60% – Choose lower one --- 40% = 0.4
  • 26.
    • Difference incure rates = higher – lower – E.g. Lower = 40% – Higher = 60% – Choose --- 60-40 = 20% = 0.2
  • 27.
    • Choose theintesect row of 0.4 & column of 0.2 97 • We need 97 subjects in each of our study groups (Sample Size)
  • 29.
    Alternative • If thesample size is too small – Decide not to do the study (or) – Decide to extend the study (or) – Decide to collaborate with investigators at other Institutions to increase the total no.of subjects – Test -- single site (bias – difficult to detect) -- multicentre (bias – readily to detect) – Sample size – Use – in RCT, Cohort, Case-Control
  • 30.
    Recruitment and Retensionof Study Participants • A major challenge in RCT - to recruit the sufficient no. of eligible and willing volunteers • Failure to recruit – trial without sufficient no. of volunteers to yield statistically valid results
  • 31.
    • Trials maybe significantly delayed – by limited recruitment and costs of completing such trials may be increased • High level of vigilance is needed – No coercion, either overt or convert – By study investigators consciously or subconsciously – To enroll the subjects (Bias)
  • 32.
    • Participants mustbe fully informed of the risks & what arrange/m have been made • If untoward effects occur (Consent)
  • 33.
    • Appropriate arrangements- cover participants’ expenses such as transportation, accommodations, etc. payment of cash incentives to prospective volunteers will risk or overt coercion biases and distortion may occur, if large incentives (+)
  • 34.
    • conducting RCTis that we do not know which therapy is better Losses to follow-up and other forms of noncompliance major concern Participants may lose interest in the study over time drop out of studies appropriate measures to prevent losses to follow-up
  • 35.
    Ways of Expressingthe Results of RCT
  • 36.
    I. Reduction inRisk (Efficacy) • can be expressed as rates of developing disease in the vaccine and placebo groups • Showed the extent of the reduction in disease by use of the vaccine • Risks are calculated per person-years of observation • Effectiveness – ideal conditions ≠ real-life situations • evaluate efficacy of a treatment – Efficacious – Effective
  • 38.
    II. Ratios ofRisk (Relative Risk) • A major objective of RCT - to have an impact on the way clinical medicine and public health
  • 39.
    III. NNT • toestimate the number of patients who would need to be treated (NNT) • to prevent one adverse outcome such as one death – E.g. Mortality in Untreated = 17% – Mortality in treated = 12% – NNT = 1/17-12% = 1/5% = 1/0.05 = 20
  • 41.
    • used instudies of various interventions including both treatment and prevention
  • 42.
    IV. NNH • Numberneeded to harm
  • 43.
    Intrepreting the Resultsof RCT • Ultimate objective - to generalize the results beyond the study population itself • Evaluate the New Drug • Defined pop is from Total pop
  • 44.
  • 45.
    External Validity (Generalizability) •The results in the study pop can be applied to broader pop • Study pop is representative of Defined pop • Patients included in study – generalized the result – applied the patients not included in study • Generalize the results – not only just to all patients with disease in community – but also total pop of patients with disease • Defined pop is representative of Total pop
  • 46.
  • 47.
    Internal Validity • ARCT is internally valid: – if the randomization has been done and – the study is free of other biases – and is without any of the major methodologic problems
  • 48.
    Limiting Factors 1. MostRCT do not provide the information 2. Droup outs from trials – should be kept to minimum • RCT are usually not representative of General pop (external validity)
  • 49.
    Comparative Effectiveness Research (CER) •Some randomized trials - to compare a new therapy to a placebo • Other randomized trials – to compare a new treatment with an older accepted treatment • two or more existing interventions are compared - to determine which intervention would work best in a given population or for a given patient
  • 50.
    • Another issue- costs of interventions • E.g. many treatments of HIV infections are very expensive - affordable in developed countries - but not affordable in many developing countries • Newer and cheaper medications - to determine the new, cheaper alternatives are as effective as the more expensive interventions --- “Equivalence Studies”
  • 51.
    Four Phases inTesting New Drugs in USA • Phase I Trial – Clinical studies – Small – 20-80 subjects – Effects are examined, including safety and side effects – Go --- Phase II
  • 52.
    • Phase IITrial – Clinical – 100-300 subjects – Evaluate the efficacy – Further assess its relative safety – Go --- Phase III
  • 53.
    • Phase IIITrial – Large scale RCT – For effectiveness and relative safety – 1000-3000 or more subjects – Very difficult – More than one study center-Multicenter trial – can be approved and licensed for marketing
  • 54.
    • Phase IVTrial – Certain adverse effects of drugs – May not become manifest for many years – May not be detectable – Become evident only when use by large pop after marketing – Post-marketing Surveillance – Are important for monitoring new agents-general use by the public
  • 55.
    • Phase IVare not Randomized studies • Ascertain side effects of new t/m after the drug is being market • A very high quality system for reporting of adverse effects is essential • Very valuable in providing additional evidence on benefits and help optimize the use of the new agent
  • 56.
    • The multicenterHypertension Detection and Follow-up Program (HDFP) study • benefits of treating mild to moderate hypertension • 22,994 subjects who were eligible because they had elevated diastolic blood pressure • 10,940 were randomized to the stepped care or to the referred care group
  • 57.
    The Hypertension Detectionand Follow-up Program
  • 58.
    The Hypertension Detectionand Follow-up Program • Stepped Care - according to a precisely defined protocol, under which treatment was changed when a specified decrease in blood pressure • Referred care - a group receiving no care for hypertension • known hypertensive subjects • the investigators believed that ethically unjustifiable • referred back to their own physicians
  • 59.
    • the patientsin the stepped care group had lower mortality than those in the referred care group • untreated group for comparison • patients referred back to their own physicians • but there was no monitoring of the care by their physicians
  • 61.
    REGISTRATION OF CLINICALTRIALS • beneficial results have been published • showing negative results (for some reason) have not been published • Publication Bias (or) Non-Publication Bias
  • 62.
    • more eagerto publish results from studies showing dramatic effects than • results from studies showing no benefit from a new drug • Both researchers and journals - excited about studies - a new treatment is inferior to current treatment or that the findings are not clear one way or the other
  • 63.
    • adopt apolicy • all clinical trials of medical interventions must be registered in a public trials registry before any participants are enrolled in the study • Medical include drugs, surgical procedures, devices, behavioural treatments, and processes of health care • at no charge • Before publication
  • 64.
    ETHICAL CONSIDERATIONS 1. Randomizationis ethical? – Randomization is ethical only when we do not know whether drug A is better than drug B – it is ethical to use a placebo 2. Is it ethical not to randomize? 3. Truly informed consent can be obtained? – Whether they are capable of giving truly informed consent 4. Under what circumstances should a trial be stopped earlier than originally planned? – harmful effects or beneficial effects of the agent become apparent early
  • 65.
    CONCLUSION • the goldstandard • for evaluating the efficacy of therapeutic, preventive, and other measures • in both clinical medicine and public health
  • 66.