RANDOMIZED TRIALS:

ASSESSING EFICACY
OF
PREVENTIVE MEASURES
THERAPEUTIC MEASURES

umo 2012

1
 Objective of PH & Clinical Practice
• to modify natural H/O diseases
– to prevent /delay death /disability
– to improve health of individual / population

2
 history …
• 1537, Paré: unintentional / unplanned trial for treatment of
treatment of the wounded after the capture of the castle of
Villaine
• 1747 , James Lind: planned trial for scurvy
• 1883, Sir Francis Galton: Research question
– Are prayers answered or are they not?
• … Do sick persons, who pray or are prayed for, recover on the
average more rapidly than others?
• 1965, Joyce and Welldon: double-blind clinical trial of the
efficacy of prayer …. No indication of benefit
• 1988, Byrd: effectiveness of intercessory prayer in a coronary
care unit population… randomized double-blind protocol …
beneficial therapeutic effect 3
basic design of a randomized trial

Figure 7-1 Design of a randomized trial.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

4
© 2005 Elsevier
 SELECTION OF SUBJECTS
• inclusion and exclusion criteria
– precisely stated, and in writing
– no element of subjective decision-making
– must in principle be replicable by others

5
 Allocation of subjects to treatment groups:
alternatives for randomization
• Studies without Comparison
– case study or case series
– Earl Peacock …
– Edere …
– Hugo Muensch, Harvard University: Second Law: “Results
can always be improved by omitting controls”
• Studies with Comparison
– Historical Controls: problems?
– Simultaneous Nonrandomized Controls

6
 Simultaneous Nonrandomized Controls
• Results of a Trial of BCG Vaccination I
Selected from
TUBERCULOSIS DEATHS  
health conscious
families Number of children Number %
Vaccinated 445 3 0.67
Controls 545 18 3.30

• Results of a Trial of BCG Vaccination 2

alternated TUBERCULOSIS DEATHS  
children selected
Number of children Number %
Vaccinated 556 8 1.44
Controls 528 8 1.52

7
 Randomization

Figure 7-2 How to predict the next patient's treatment assignment in a randomized study.
(PEANUTS © UFS. Reprinted by permission.)c

Unpredictability of next assignment

Reproducible
documentation

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

8
© 2005 Elsevier
 •random starting point
A Table of Random
Numbers •write down direction
00–04 05–09 10–14 15–19 •assign treatment gp
00 56348 01458 36236 07253 •odd = A, even=B
01 09372 27651 30103 37004 •0-4 =A, 5-9 =B
02 44782 54023 61355 71692 •1-3 = A, 4-6=B, 7-9=C, 0 skip
03 04383 90952 57204 57810
•sealed envelope
04 98190 89997 98839 76129
05 16263 35632 88105 59090 •separate coordinating and
06 62032 90741 13468 02647 statistical center
07 48457 78538 22759 12188
08 36782 06157 73084 48094
09 63302 55103 19703 74741

… hope that randomization will increase the likelihood that the groups will be
comparable in regard to characteristics about which we may be concerned
… prognostic variables
…. randomization is not a guarantee of comparability, because chance may play a
role in the process, but over the long term, the groups will tend to be similar 9
Figure 7-3 Observational versus experimental studies. I, If the study is not randomized, the proportions of
patients with arrhythmia in the two groups may differ. II, If the study is randomized, the proportions of
patients with arrhythmia in the two groups are more likely to be similar.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

10
© 2005 Elsevier
Figure 7-4 Stratified randomization.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

11
© 2005 Elsevier
 DATA COLLECTION ON SUBJECTS

• Treatment (Assigned and Received)

• Outcome measurement
• Prognostic Profile at Entry
• Masking (Blinding)

12
crossover

Carryover effect
Enough washout period
Order of treatment
Psychological response
Surgical or cure

Figure 7-5 A-F, Design of a planned crossover trial. 13

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
© 2005 Elsevier
Figure 7-6 Unplanned crossover in a randomized study of cardiac bypass surgery: I.
Original study design.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

14
© 2005 Elsevier
Figure 7-7 A-D, Unplanned crossover in a randomized study of cardiac bypass surgery: II.
Reality: Unplanned crossovers.
•Current practice is to perform the primary analysis by intention to treat—
according to the original randomized assignment.
•Keep minimum number of unplanned crossovers
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
15
© 2005 Elsevier
 Factorial design

Two drugs are to be tested

Different anticipated
outcomes
Independent modes of action
Can economically use the
same study population

Figure 7-8 Factorial design for studying the effects of two treatments.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

16
© 2005 Elsevier
 Figure 7-9 Factorial design.
A, The effects of treatment A (blue cells) versus no treatment A.
B, The effects of treatment B (yellow cells) versus no treatment B.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

17
© 2005 Elsevier
Physicians' Health Study:

Figure 7-10 Factorial design used in a study of aspirin and beta-carotene.

Tested aspirin for primary prevention of cardiovascular disease and
beta-carotene for primary prevention of cancer 18
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
© 2005 Elsevier
Figure 7-11 Factorial design of the study of aspirin and beta-carotene in 2 × 2 table
format.
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
19
© 2005 Elsevier
Figure 7-12 Factorial design.
A, The effects of aspirin (blue cells) versus no aspirin.
B, The effects of beta-carotene (yellow cells) versus no beta-carotene.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

20
© 2005 Elsevier
 noncompliance
Coronary Drug Project: Five-Year Mortality in Patients Given Clofibrate or Placebo
  Number of Patients Mortality (%)
Clofibrate 1,065 18.2
Placebo 2,695 19.4

Coronary Drug Project: Five-Year Mortality in Patients Given Clofibrate or Placebo

According to Level of Compliance
  Number of Patients Mortality (%)
Clofibrate    
 Poor complier (<80%) 357 24.6
 Good complier (≥80%) 708 15.0
Placebo 2,695 19.4

21
 noncompliance
Coronary Drug Project: Five-Year Mortality in Patients Given Clofibrate or Placebo
According to Level of Compliance

CLOFIBRATE PLACEBO
Number of Number of
Compliance Patients Mortality (%) Patients Mortality (%)
Poor (<80%) 357 24.6 882 28.2
Good (≥80%) 708 15.0 1,813 15.1
Total Group 1,065 18.2 2,695 19.4

22
Sample size

23
Figure 8-1 Two opaque jars, each holding 100 beads, some blue and some white.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-2 Samples of 10 beads from jar A and 10 beads from jar B.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-3 Samples of 10 beads from jar A and 10 beads from jar B.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Four Possibilities in Testing Whether the Treatments Differ

1.  The treatments do not differ, and we correctly

   conclude that they do not differ.
2.  The treatments do not differ, but we conclude that
   they do differ.
3.  The treatments differ, but we conclude that they do
   not differ.
4.  The treatments do differ, and we correctly conclude
   that they do differ.

27
Figure 8-4 Possible outcomes of a randomized trial.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-5 Possible outcomes of a randomized trial: type I and type II errors.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-6 Possible outcomes of a randomized trial when the treatments differ.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-7 Possible outcomes of a randomized trial: summary.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
 TABLE 8-2   -- Summary of Terms
Term Definitions
α = Probability of making a type I error
  = Probability of concluding the treatments differ when in reality they do
not differ; rejecting the null hypothesis (H0) when it is true
β = Probability of making a type II error
  = Probability of concluding that the treatments do not differ when in
reality they do differ; fail to reject the null hypothesis (Ho) when it is false

Power = 1 - Probability of making a type II error

  =1-β
  = Probability of correctly concluding that the treatments differ

  = Probability of detecting a difference between the treatments if the

treatments do in fact differ

32
 TABLE 8-3   -- What Must Be Specified to Estimate the Sample Size
Needed in a Randomized Trial?

1.    The difference in response rates to be detected

2.    An estimate of the response rate in one of the groups
3.    Level of statistical significance (α)
4.    The value of the power desired (1 - β)
5.    Whether the test should be one-sided or two-sided

33
TABLE 8-4   -- Number of Patients Needed in Each Group to Detect Various
Differences in Cure Rates; α = .05; Power (1 - β) = .80 (Two-sided Test)

Differences in cure rates between the two treatment groups

Lower of the
Two Cure
Rates .05 .10 .15 .20 .25 .30 .35 .40 .45 .50 .55 .60 .65 .70
.05 420 130 69 44 36 31 23 20 17 14 13 11 10 8
.10 680 195 96 59 41 35 29 23 19 17 13 12 11 8
.15 910 250 120 71 48 39 31 25 20 17 15 12 11 9
.20 1,090 290 135 80 53 42 33 26 22 18 16 12 11 9
.25 1,250 330 150 88 57 44 35 28 22 18 16 12 11 —
.30 1,380 360 160 93 60 44 36 29 22 18 15 12 — —
.35 1,470 370 170 96 61 44 36 28 22 17 13 — — —
.40 1,530 390 175 97 61 44 35 26 20 17 — — — —
.45 1,560 390 175 96 60 42 33 25 19 — — — — —
.50 1,560 390 170 93 57 40 31 23 — — — — — —

34
TABLE 8-5   -- Number of Patients Needed in Each Group to Detect Various
Differences in Cure Rates; α = .05; Power (1 - β) = .80 (One-sided Test)

Differences in cure rates between the two treatment groups

Lower of
the Two
Cure Rates .05 .10 .15 .20 .25 .30 .35 .40 .45 .50 .55 .60 .65 .70
.05 330 105 55 40 33 24 20 17 13 12 10 9 9 8
.10 540 155 76 47 37 30 23 19 16 13 11 11 9 8
.15 710 200 94 56 43 32 26 22 17 15 11 10 9 8
.20 860 230 110 63 42 36 27 23 17 15 12 10 9 8
.25 980 260 120 69 45 37 31 23 17 15 12 10 9 —
.30 1,080 280 130 73 47 37 31 23 17 15 11 10 — —
.35 1,160 300 135 75 48 37 31 23 17 15 11 — — —
.40 1,210 310 135 76 48 37 30 23 17 13 — — — —
.45 1,230 310 135 75 47 36 26 22 16 — — — — —
.50 1,230 310 135 73 45 36 26 19 — — — — — —

35
Treatment effects
 Treatment effects
• New terms were used to describe both the
good and bad effects of therapy in the current
evidence based journals. Here are some terms
of which (Evidence Based Medicine and ACP
journals Club) have achieved consensus

Sackett.D.L et al, Evidence based medicine: How to practice & Teach EBM, 2nd edt. Chrrchill Living Stone; 2000, pp 250.
 Treatment effects
Event Rate. The proportion of patients in a
group in whom the event is observed.
Experimental Event Rate (EER). The
proportion of patients in experimental
group in whom the event is observed.
Control Event Rate (CER). The proportion of
patients in control group in whom the event
is observed.
Sackett.D.L et al, Evidence based medicine: How to practice & Teach EBM, 2nd edt. Chrrchill Living Stone; 2000, pp 250.
 Treatment effects (bad outcomes)

ARR (Absolute Risk Reduction). The absolute arithmetic

difference in rates of bad outcomes between experimental
and control participants in a trial, calculated as: |EER – CER|,
RRR (Relative Risk Reduction). The proportional reduction in
rates of bad outcomes between experimental and control
participants in a trial, calculated as: |EER – CER| / CER,
NNT (Number Needed to Treat). The number of patients
who need to be treated to avoid one additional bad
outcome, calculated as 1/ARR,
Sackett.D.L et al, Evidence based medicine: How to practice & Teach EBM, 2nd edt. Chrrchill Living Stone; 2000, pp 250.
 Calculation of occurrence of diabetic
retinopathy in IDDMs

Occurrence of diabetic retinopathy at 5 years among IDDMs in the

DCCT trial
Usual insulin regimen (CER) =38%
Intensive insulin regimen (EER) =13%
Worsening diabetic retinopathy
Absolute Risk
Absolute Risk Reduction(ARR)=
Reduction(ARR)=(EER
(EER––CER)=(13%-
CER)=(13%-38%)=25%
38%)=25%
Relative Risk Reduction (RRR)=(EER – CER)/CER=(13%- 38%)/38%=66%
Number Needed to Treat (NNT) = 1/ARR =1/25% = 4 patients with
intensive insulin
Sackett.D.L et al, Evidence based medicine: How to practice & Teach EBM, 2nd edt. Chrrchill Living Stone; 2000, pp 250.
 Treatment effects (good outcome)

RBI (Relative Benefit Increase). The proportional increase in

rates of good outcomes between experimental and control
participants in a trial, calculated as: |EER – CER| / CER
• ABI (Absolute Benefit Increase). The absolute arithmetic
difference in rates of good outcomes between experimental
and control participants in a trial, calculated as: |EER – CER|,
• NNT (Number Needed to Treat). The number of patients who
need to be treated to achieve one additional good outcome,
calculated as 1/ARR,
 Ways of expressing the results of randomized trials

• Efficacy = Reduction in Risk (RRR)

extent of the reduction in disease by use of the vaccine
Risks are often calculated per person-years of observation
• Other approaches
– ratio of the risks in the two treatment groups (the relative risk= RR)
– compare the survival curves for each of the groups and
determine whether they differ

42
 Ways of expressing the results of randomized trials

• NNT= 1/ARR= 1/(Rate in untreated gp - Rate in treated gp)

Number of patients who would need to be treated (NNT) to
prevent one adverse outcome such as one death
Thus, if, for example, the mortality rate in the untreated group is
17% and mortality in the treated group is 12%,
we would need to treat (1 /(17%-12%))= 1/5% = 1/0.05 = 20
• Risk of side effects by calculating the number needed to harm
(NNH) to cause one additional person to be harmed

43
GENERALIZABILITY OF RESULTS

Figure 8-8 Internal and external validity in a randomized trial.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
 The Hypertension Detection and Follow-up Program

Figure 8-9 Design of the Hypertension Detection and Follow-up Program (HDFP).

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-10 Cumulative all-cause mortality by blood pressure status and type of care received
in the HDFP. (Adapted from Hypertension Detection and Follow-up Program Cooperative
Group: Five-year findings of the Hypertension Detection and Follow-up Program: I. Reduction
in mortality of persons with high blood pressure, including mild hypertension. JAMA
242:2562-2571, 1979.)
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
© 2005 Elsevier
TABLE 8-6   -- Mortality from All Causes during the Hypertension Detection and
Follow-up Program

5-YR DEATH RATE

Mortality
Diastolic Blood Pressure Stepped Referred Reduction in
at Entry (mm Hg) Care (SC) Care (RC) SC RC SC Group (%)
90–104 3,903 3,922 5.9 7.4 20.3
105–114 1,048 1,004 6.7 7.7 13.0
≥115 534 529 9.0 9.7 7.2
Total 5,485 5,455 6.4 7.7 16.9

47
 MRFITS
The Multiple Risk Factor Intervention Trial

Figure8-11 Mean risk factor levels by year of follow-up for Multiple Risk Factor Intervention Trial Research Group
participants. BP, blood pressure; S1, first screening visit; SI, special intervention; UC, usual care. (From Multiple Risk Factor
Intervention Trial Research Group: Multiple Risk Factor Intervention Trial: Risk factor changes and mortality results. JAMA
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
248:1465-1477, 1982.) © 2005 Elsevier
 Figure 8-12 Cumulative coronary heart disease (CHD) and total mortality rates for Multiple Risk
Factor Intervention Trial Research Group participants. The heavy line indicates men receiving usual
care; the thin line indicates men receiving special intervention. (From Multiple Risk Factor
Intervention Trial Research Group: Multiple Risk Factor Intervention Trial: Risk factor changes and
mortality results. JAMA 248:1465-1477, 1982.)
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
© 2005 Elsevier
 Study of Breast Cancer Prevention Using Tamoxifen

Figure 8-13 Cumulative rates of invasive and non-invasive breast cancer occurring in participants receiving
placebo or tamoxifen. (From Fisher B, Constantino JP, Wickerham DL, et al: Tamoxifen for prevention of
breast cancer: Report of the National Surgical Adjuvant Breast and Bowel Project P-1 Study. J Natl Cancer
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
Inst 90:1371-1388, 1998.) © 2005 Elsevier
Figure 8-14 Cumulative rates of invasive endometrial cancer occurring in participants receiving placebo or
tamoxifen. (From Fisher B, Constantino JP, Wickerham DL, et al: Tamoxifen for prevention of breast cancer:
Report of the National Surgical Adjuvant Breast and Bowel Project P-1 Study. J Natl Cancer Inst 90:1371-
1388, 1998.)

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
 PHASES IN TESTING OF NEW DRUGS IN THE UNITED STATES

• Phase I studies: clinical pharmacologic studies

– small studies, 20 to 80 patients - toxic and pharmacologic effects
• Phase II studies: clinical investigations
– 100 to 200 patients for efficacy and relative safety
• Phase III studies: randomized controlled trials
– large-scale for effectiveness and relative safety
– often multicentered
– can be licensed for marketing
– certain adverse effects (carcinogenesis and teratogenesis) may not become manifest for
many years but may become evident only when the drug is in use by large populations
• Phase IV studies: postmarketing surveillance
– for monitoring new agents as generally used by the public

52
 ETHICAL CONSIDERATIONS
• Is randomization ethical?
– Withholding a drug from patients, particularly those with serious and life-
threatening diseases?
– ethical only when we do not know whether drug A is better than drug B
• may have some indication that one treatment is better than the other (and, often, this is the
rationale for conducting a trial in the first place), but we are not certain
– when we do not have adequate evidence to support the conclusion that drug A is
better…
– is it ethical to use a placebo?
• is it ethical to withhold a treatment that has been shown to be effective
• Is it ethical not to randomize?
• Is truly informed consent can be obtained?
• under what circumstances should a trial be stopped earlier than
originally planned?

53
 RANDOMIZED TRIALS FOR EVALUATING WIDELY ACCEPTED INTERVENTIONS

Figure 8-15 Design of a controlled trial of arthroscopic surgery for osteoarthritis of the knee .
(Based on Moseley JB, O'Malley K, Petersen NJ, Menke TJ, et al: A controlled trial of arthroscopic surgery for
osteoarthritis of the knee. N Engl J Med 347:81-88, 2002.)
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
© 2005 Elsevier
Figure 8-16 Mean values (and 95% confidence intervals) on the Knee-Specific Pain Scale. Assessments
were made before the procedure and 2 weeks, 6 weeks, 3 months, 6 months, 12 months, 18 months, and
24 months after the procedure. Higher scores indicate more severe pain.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-17 Mean values (and 95% confidence intervals) on the Walking-Bending Subscale of the Arthritis
Impact Measurement Scales (AIMS2). Assessments were made before the procedure and 2 weeks, 6
weeks, 3 months, 6 months, 12 months, 18 months, and 24 months after the procedure. Higher scores
indicate poorer functioning.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
Figure 8-18 Design of a randomized, controlled trial of group psychosocial support on
survival in patients with metastatic breast cancer.
(Based on Huston P, Peterson R: Withholding proven treatment in clinical research. N Engl J Med
345:912-914, 2001.)
Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)
© 2005 Elsevier
Figure 8-19 Kaplan-Meier survival curves for women assigned to the intervention group and
the control group. There was no significant difference in survival between the two groups.

Downloaded from: StudentConsult (on 7 February 2012 10:23 AM)

© 2005 Elsevier
 CONCLUSION
• The randomized trial is the gold standard for
– evaluating the efficacy of therapeutic, preventive,
and other measures
– in both clinical medicine and public health

59