Research method
Intention-to-Treat Analysis
Intention-to-treat analysis keeps participants in the arm to which they were randomised, including people who never started, dropped out, or deviated from protocol. The point is to preserve the baseline balance that randomisation created, so the contrast answers 'what happened after assignment?' rather than 'what happened in people who complied?' Per-protocol analysis of the same trial can look more significant, more null, or simply similar; it answers a different question and re-opens selection.
Trialists reach for ITT when they need an effect of being offered a programme, app, anaesthetic or digital medicine, not an effect among fans who finished it. It answers 'did random assignment change the primary outcome?' Its main limitation is missing data: ITT does not magically fix 73.5% versus 26.9% attrition, and a borderline ITT p-value next to a significant per-protocol result is a warning, not a licence to quote only the significant one.
Evidence
What the evidence shows
Drawn from 9 studies in this library. Each finding starts with a plain-language takeaway, then the denser detail. Supports means evidence for a finding; Challenges means evidence against a stated position; Qualifies marks scope with a short note on each study’s contribution. Challenged positions are labeled — they are not findings.
MOBITEL randomised 120 patients after acute heart-failure decompensation to phone telemonitoring versus control. ITT primary events were 17% versus 33% (relative risk reduction 50%, P=.06); per-protocol RRR was 54% (P=.04). NYHA class improved only in the tele group, and HF hospital stays were shorter (median 6.5 versus 10 days). Some patients were 'never beginners' unable to use the equipment — exactly the people ITT keeps in the denominator.
NHS Choices users randomised to MoodGYM versus wait-list (n=3,070) were analysed with an ITT mixed model: adjusted WEMWBS differences favoured the programme by about 2.5 points at 6 weeks and 2.9 at 12 weeks (≈Cohen d 0.34). Attrition was 73.5% in the intervention arm versus 26.9% in control. High differential dropout can bias even an ITT model if missingness is related to outcome.
Adjunctive smartphone CBT in antidepressant-refractory major depression used ITT on 164 randomised participants: at week 9 the Kokoro-app arm scored about 2.5 PHQ-9 points lower (SMD 0.40) and was more likely to meet response criteria, while remission and side-effect differences were not significant. Deprexis, in 396 adults recruited online, improved BDI by about 5.4 points (d=0.58) versus delayed treatment. Neither ITT result is a head-to-head against face-to-face CBT or a long-term relapse trial.
Study Role Design N Population Outcome Can a CBT phone app help stubborn depression? Supports RCTKokoro-app smartphone CBT plus antidepressant switch vs medication change alone N=164 · ITT; primary outcome at week 9 in 163/164 (99.4%) Adults in Japan with antidepressant-refractory major depression PHQ-9 at week 9 Deprexis online therapy reduced depression scores Supports RCTImmediate Deprexis vs delayed treatment; online recruitment N=396 · 81% assigned to immediate treatment Adults with depression symptoms recruited online for Deprexis BDI change over time (condition × time) A large surgical trial can be ITT-null. Among 1,670 patients having curative primary breast-cancer surgery (841 propofol, 829 sevoflurane), five-year overall survival was about 92% in both arms, with a hazard ratio near 1 and no statistically significant difference. Results were similar after per-protocol analysis and after adjustment for uneven triple-negative receptor status. Equivalence for rare deaths is still hard to prove; opioid co-analgesia also differed by arm.
ITT does not require individual randomisation. A 12-week open-label cluster-randomised pilot at 13 US primary-care sites found digital-medicine users had about 9 mm Hg greater SBP reduction by week 4, with more patients at BP goal by week 12. That is an effect of site-level assignment, unblinded, and not a hard cardiovascular-event result. A produce-app RCT that increased fruit intake by about three pieces per week also reported high attrition — another ITT-with-missing-data problem.
Study Role Design N Population Outcome Digital pills cut blood pressure faster Supports RCTOpen-label cluster-randomized three-arm pilot: 4-week DMO, 12-week DMO, usual care N=109 · Modified ITT (40 / 40 / 29) across 13 US primary-care sites Adults with uncontrolled hypertension and type 2 diabetes in US primary care SBP reduction, BP goal attainment, and LDL-C/HbA1c change to 12 weeks Audio vs text produce-app RCT Supports RCTThree-arm RCT: text-tailored vs audio-tailored vs questionnaire control N=146 · Final analytic sample Adults using a fruit-and-vegetable mobile app intervention Fruit and vegetable intake at 6 months
Open questions
Tensions and limits
Some items are genuine disagreements on the same question. Others mark different assays, populations, or outcomes — limits on how far one study travels — not a forced fight between papers.
ITT and per-protocol can disagree on whether a result 'counts.' MOBITEL's primary ITT contrast was P=.06 while per-protocol reached P=.04; the propofol–sevoflurane trial was null on both. Quoting only the significant per-protocol RRR in MOBITEL, or reading the anaesthetic HR near 1 as proven equivalence, treats two different estimands as if they were one.
Study Role Design N Population Outcome Phone telemonitoring after heart-failure flare Supports RCTMOBITEL multicentre RCT; home telemonitoring via mobile phones vs control N=120 · Stopped early at 120 of a planned 240 (eight centres; median age 66); many patients could not operate the phone Adults after acute heart-failure decompensation Cardiovascular death or HF re-hospitalisation over 6 months Propofol vs sevoflurane and breast cancer survival Supports RCT1:1 propofol vs sevoflurane for anaesthesia maintenance in curative primary breast cancer surgery N=1670 · ITT: 841 propofol, 829 sevoflurane Patients undergoing curative primary breast cancer surgery 5-year overall survival ITT mixed models do not travel with equal credibility across dropout patterns. MoodGYM's 2.5–2.9 point WEMWBS advantage sits on 73.5% versus 26.9% attrition and a wait-list control. Kokoro-app's 2.5-point PHQ-9 contrast had high engagement among those assigned CBT. Same analysis label, very different missing-data threats.
Study Role Design N Population Outcome MoodGYM improves population wellbeing Supports RCTFully automated MoodGYM vs waiting-list; ITT mixed models N=3070 · Randomized NHS Choices users; high differential attrition NHS Choices website users randomized to MoodGYM or wait-list WEMWBS well-being at 6 and 12 weeks Can a CBT phone app help stubborn depression? Supports RCTKokoro-app smartphone CBT plus antidepressant switch vs medication change alone N=164 · ITT; primary outcome at week 9 in 163/164 (99.4%) Adults in Japan with antidepressant-refractory major depression PHQ-9 at week 9
Common misconceptions
Intention-to-treat means dropouts are ignored, or that missing outcomes are automatically unbiased.
ITT keeps people in their assigned arm. MoodGYM still lost 73.5% of the intervention arm versus 26.9% of controls; an ITT mixed model can remain biased if those who leave differ in well-being. MOBITEL's 'never beginners' stay in the ITT denominator precisely because they did not use the kit.
If per-protocol is significant and ITT is not, the treatment works and ITT was just conservative.
Per-protocol in MOBITEL (RRR 54%, P=.04) analyses people who used monitoring as intended. That can overstate the effect of offering the programme, which is what ITT (RRR 50%, P=.06) targets. Significance switching at the 0.05 line is not a reason to discard the randomised population.
An ITT hazard ratio near 1 in 1,670 patients proves two anaesthetics are equivalent for survival.
Five-year survival was about 92% in both propofol and sevoflurane arms, and per-protocol agreed, but rare mortality is a hard equivalence claim. Opioid co-analgesia differed by arm, and the result applies to primary breast-cancer surgery, not all cancers.
Exam-style questions
Short-answer questions that ask you to explain or compare, not recall.
MOBITEL's ITT event rates were 17% versus 33% (P=.06) and per-protocol RRR was 54% (P=.04). Which number answers 'should we offer this monitoring after decompensation?' and why?
The ITT contrast, because offering the programme includes people who never start the equipment. Per-protocol asks what happened among users, which can look better (here, crossing P=.05) by dropping 'never beginners.' Median stays of 6.5 versus 10 days are supporting clinical detail, not a substitute for the ITT primary.
How should a student read MoodGYM's ≈2.5–2.9 point WEMWBS ITT advantage together with 73.5% versus 26.9% attrition?
The mixed-model ITT estimate favours MoodGYM (d≈0.34) among 3,070 randomised users, but differential dropout can bias that estimate if missingness relates to well-being. Wait-list controls may also inflate apparent benefit. ITT is the right analysis population, not a guarantee of unbiased completeness.
Kokoro-app moved PHQ-9 by about 2.5 points (SMD 0.40) in ITT without a significant remission difference. What did ITT protect, and what did it not prove?
ITT protected the comparison among all 164 randomised participants rather than only engaged app users. It showed a mean symptom-score benefit and more response, not remission, not fewer side effects, not long-term relapse prevention, and not superiority to face-to-face CBT.
Why is the propofol–sevoflurane ITT result a different lesson from the digital-medicine cluster pilot's 9 mm Hg SBP drop?
Both analyse as randomised, but one is a 1,670-person individual RCT with a rare 5-year death endpoint that did not move (HR near 1, ~92% survival both arms). The other is an open-label cluster assignment of sites, a short BP/LDL contrast, and not a hard-event ITT. Same label, different estimand, blinding and endpoint.
The studies
9 studies in this library bear on Intention-to-Treat Analysis, ordered by citations. The first 8 are shown.
- Deprexis online therapy reduced depression scores
An integrative online program (Deprexis) improved BDI depression symptoms versus delayed treatment in a large German internet RCT.
- Mediterranean diet and depression risk
Overall Mediterranean diet assignment did not significantly reduce depression; a nuts arm suggested benefit in diabetes subgroup analyses.
- Phone telemonitoring after heart-failure flare
Home mobile-phone telemonitoring after acute heart-failure decompensation cut primary events and hospital days versus usual care.
- Digital pills cut blood pressure faster
Patients with uncontrolled hypertension and type 2 diabetes using digital medicines had larger 4-week SBP drops than usual care.
- MoodGYM improves population wellbeing
In a large UK web RCT, automated MoodGYM raised mental well-being versus wait-list control despite high intervention attrition.
- Web program increased older adults’ measured activity in a trial
A Dutch RCT found a web-based intervention raised daily activity ~46% and improved weight and HbA1c markers in inactive 60–70-year-olds.
- Can a CBT phone app help stubborn depression?
Adding a smartphone CBT program to a medication switch improved depressive symptoms more than switching antidepressants alone.
- Audio vs text produce-app RCT
A tailored mobile app’s audio messages raised fruit intake more than text or control, while vegetable intake showed no overall main effect.
Show 1 more studiesShow fewer studies
- Propofol vs sevoflurane and breast cancer survival
In the CAN randomised trial, five-year overall survival after primary breast cancer surgery was essentially the same with propofol or sevoflurane maintenance anaesthesia.
Learn alongside