1. Intention-to-Treat (ITT)
What is it and why should it be used?

2. Invoked Population Model – Randomization Model
Nonrandom Selection of Clinics in a Nonrandom Selection of Communities
Undefined Sampling Procedure for Patients (a variety of sources are used)
N = NA + NB patients
Randomization
NA patients
NB patients
Source: Lachin J. Cont Clin Trials, 1988.

3. Realities of Clinical Trials
Enrollment of ineligible participants
Incorrect diagnosis, e.g., TB
Errors in applying eligibility criteria
Knowing violation of eligibility criteria
Incorrect treatment assigned
Incorrect treatment given
Less than 100% adherence to study treatment
Refusal to take study treatment after randomization
Withdrawal from study treatment during the study
Failure to adhere to instructions for taking study medication
Use of prohibited concomitant treatments
Missing data (losses to follow-up)

4. What do you do with these deviations?
Many opinions on this.

5. ITT: Origin of Term Bradford Hill, Principles of Medical Statistics, 7th Edition, 1961
Hill raised the question “have any patients after admission to the treated or control group been excluded from further observation?”
Patients who cannot be retained on treatment.
Patients for whom a surgical procedure cannot be performed.
“Unless the losses are very few and therefore unimportant, we may inevitably have to keep such patients in the comparison and thus measure the intention to treat in a given way rather than actual treatment”.
Concerned about bias resulting from groups that were no longer comparable.

6. Strict Definition of ITT
“Includes all randomized patients in the groups to which they were randomly assigned, regardless of the treatment they actually received, and regardless of subsequent withdrawal from treatment or deviation from the protocol”. Working group of Biopharmaceutical Section of the American Statistical Association (Fisher LD et al, in Statistical Issues in Drug Development, Peace KE, editor, 1990.

7. Two General Approaches
“Intention to treat” (also referred to as “full analysis” set and “as randomized”)
Analyze all randomized subjects in their assigned groups utilizing as much information from each as possible
“Treatment received” (also referred to as “per protocol” or “as treated”)
Analyze only fully eligible and compliant subjects with no missing data, e.g., “valid” and “evaluable” subjects

8. There May Be a Middle Ground
Modified Intention to Treat (MITT) population
Participants included in assigned treatment group regardless of treatment actually received.
Ineligible participants based on measures made before randomization, but delayed, are excluded (e.g., patients with HIV who are in enrolled in an HIV prevention trial or patients without TB who are enrolled in a TB treatment trial).
Safety population
Participants who take the experimental treatment even if assigned control treatment.

9. Example: Randomized Trial of Prevention of HIV with Acyclovir in Couples Where One Partner is Co-infected with HIV and HSV-2
“The primary analysis was a modified intention-to-treat analysis of linked transmissions of HIV-1; unlinked transmissions, seroconversions that occurred among men when their female partners who were infected with HIV-1 were pregnant and not taking the study drug, and seroconversions that occurred after the death of the HIV-1 infected partner were excluded. The secondary analysis was intention-to-treat”
N Engl J Med 2010; 362:

10. Arguments for Intention to Treat
Consistent with randomization – get the right significance probability for hypothesis testing.
Addresses the question of practical interest – a comparison of treatment policies.
If the objective is to understand the implication of using a specific intervention in practice, this is the right analysis (e.g., non-adherence is a consequence of using a strategy in practice).

11. Arguments for Per Protocol (As Treated) Analysis
Better estimate of pure pharmaceutical effect of treatment (i.e., including non-compliers dilutes the treatment difference).
The relevant question is whether the treatment can work when used as intended, e.g., is it effective among patients who can tolerate it?
In a non-inferiority/equivalence study (as opposed to a superiority study) , this may be a more conservative analysis (less dilution toward no difference)
Key Point: Make sure you discuss the question
before you start the study.

12. Arguments for Both Trials can ask two questions:
“Can Drug A reduce tumor size”? (explanatory)
“Does prescribing Drug A to patients with tumors do more harm than good”? (management)
Can it work versus does it work study designs. Also referred to pragmatic and explanatory approaches by Shwartz and Lellouch (J Chronic Dis, 1967)
Sackett and Gent, N Engl J Med 1979.
In non-inferiority studies you should consider both
ITT and per protocol analyses.

13. Pre-Exposure Prophylaxis (PrEP) Trials to Prevent HIV Acquisition
Several trials, results not all consistent, does adherence explain all/part of the difference?
IprEx trial in men who have sex with men (36 vs 64 infections -- 44% reduction in incidence with FTC-TDF; 95% self-reported adherence but about 50% based on drug levels.)
FEM-PrEP trial in heterosexual women (33 vs 35 infections with FTC-TDF; 95% self-reported adherence but about 38% based on drug levels)
Partners trial in discordant couples (13 vs 52 infections – 75% reduction in incidence with FTC-TDF; 97% self-reported adherence and 82% adherence based on blood levels)

14. Obstacles to Intention to Treat (ITT)
Missing data
Due to the way the protocol was written
Losses to follow-up
Withdrawal of consent
Important to note that ITT not only requires all randomized participants be included in the analysis, but also requires that all randomized participants be followed and have the outcomes of interest measured no matter adherence to the protocol.

15. Did not start treatment Ineligible Unacceptable toxicity
Some Protocols Define Situations When Patients Should No Longer Be Followed: Off Study
Did not start treatment
Ineligible
Unacceptable toxicity
Disease progression
Incarceration
Lost to follow-up
Withdrawal of consent
Bad idea if ITT is goal – follow everyone until the end of the
study or until some defined follow-up period
has been achieved. “Off study” is a confusing and bad term.

16. Obstacles to Per Protocol Analysis
Defining adherence to treatment
What is an acceptable level of adherence and how do you measure it?
Do you count events that occurred within 2 days, 7 days, 30 days of treatment discontinuation?
Depends on the study and it may not always be clear where to draw the line.

17. Treatment Received Advantage: Undiluted treatment effect
Disadvantage: Comparison of groups may be biased and it is not predictable in which direction.

18. Intention-to-Treat
Advantage: Comparability of treatment groups; no bias resulting from exclusions.
Disadvantage: Possible dilution of treatment effect; loss of power unless sample size was increased to account for it.

19. Intent-to-Treat May be More Powerful
Not only larger sample size, but…
If the treatment under study has an effect even after discontinuation (e.g., disease progression slowed, lingering pharmacologic effect)

20. ICH Guidelines – Full Analysis Set
Exclusions may occur for failure to meet major entry criteria, failure to take at least one dose of medication, and for lack of any data after randomization
Exclusion of ineligibles may only occur if:
Criterion measured prior to randomization
Eligibility can be objectively assessed
There equal scrutiny for all patients
All violations of specific type are excluded

21. ICH Guidelines – Per Protocol Set Typical Criteria
Completion of pre-specified minimum exposure to treatment
Availability of measurements of primary outcomes
Absence of major protocol violations

22. Examples of Eligibility Errors in AIDS Trials
Liver enzyme tests are mixed up for Patient X and Patient Y; 2 weeks after randomization it is determined study drugs are contraindicated for Patient X
Patients have CD4+ cell counts mixed up and the wrong patient is randomized.
Patient X is randomized and is discovered 4 weeks later to be HIV negative
Qualifying lab measurements made 45 days before randomization instead of within 30 days

23. Policies for Handling Eligibility Errors
1st Priority is Prevention
Simple inclusion/exclusion criteria
Eligibility checks before randomization
Regular summary reports to monitor performance
Possible Policies
Don’t enroll until eligibility is verified
Enroll those possibly eligible and withdraw later if ineligible; decision to withdraw is blinded to treatment group and based on pre-randomization measurements
Enroll those possibly eligible and keep them
Peto R et al., Br J. Cancer 1976; 34:

24. What do you do about eligibility errors?
Document them.
Determine whether it is safe for patients to continue treatment.
If safe, assess whether patient should be allowed to continue treatment.
In most cases, follow the patients like other randomized patients so that an intent to treat analysis can be carried out.
Pre-specify a plan for handling them in the protocol.

25. Examples of “Adherence” Problems in AIDS Trials
Patient X reports taking a study medication which is not allowed by the protocol
Patient X dies after randomization but before study drug is picked up from pharmacy
Patient X quits taking study treatment 2 weeks after randomization because she decides he does not want to participate in a placebo controlled study
Patient X quits taking study drug 8 weeks after randomization due to side effects
Patient X stops taking study drug before outcome assessment because their condition is worsening.
Patient X is randomized twice because he did not like the first assignment

26. Anturane Trial N Engl J Med 1980; 302:250-256. Anturane Placebo No. of
Patients
No. of
Events
No. of
Patients
No. of
Events
P-value
Total
Ineligible patients
Eligible patients
Nonanalyzable deaths
Analyzable deaths
Analyzable cardiac deaths
Analyzable sudden cardiac
Analyzable sudden cardiac deaths in 1st 6 months
N Engl J Med 1980; 302:

27. Coronary Drug Project Mortality Results
Clofibrate
Placebo
No. patients
No. deaths in 5 years
Percent dead
p-value = 0.55
JAMA 231:360-81, 1975.

28. * (No. of capsules taken/No. that should have been taken) x 100
Coronary Drug Project –Adherence to Clofibrate (3 capsules, 3 times per day)
Percent
Adherence*
* (No. of capsules taken/No. that should have been taken) x 100
Averaged over all 4 month visits for 5 years for those alive after 5 years.
JAMA 231:360-81, 1975.

29. Coronary Drug Project Mortality According to Adherence to Clofibrate
Percent Dead
<80% 24.6
80%+ 15.0
Overall 20.0
p=0.0001
NEJM 303: , 1980.

30. The Obvious, But Naïve, Solution
Clofibrate Adherers
Placebo
Percent dead
p = 0.04

31. Coronary Drug Project Adherence to Clofibrate and Placebo (3 capsules, 3 times per day)
Percent
Adherence
Placebo
Percent
Adherence
JAMA 231:360-81, 1975.

32. Coronary Drug Project Mortality According to Adherence to Clofibrate and Placebo
<80%
80%
Overall
p=0.0001 p=
NEJM 303: , 1980.

33. Summary / Recommendations
Primary analysis should usually be ITT (need to continue collecting data to do this right) – it addresses a pragmatic policy/management question which is always relevant.
ITT analysis requires excellent trial conduct.
It is appropriate to carry out secondary “per protocol” or “as treated” analyses but these have to be interpreted with caution.
For analyses which are not intent-to-treat it is often difficult/impossible to quantify bias resulting from not comparing like with like
If exclusions after randomization are to be made as part of secondary “per protocol” analyses, they should be specified in the protocol
Think about what you want to estimate in advance.