← Clinical Trials
Cardiovascular Diseases Phase 4 Non-Inferiority & Superiority NCT00153101

ONTARGET: Complete Statistical Analysis of Telmisartan and Ramipril in High-Risk Cardiovascular Disease

A statistical review of ONTARGET, which compared telmisartan alone and telmisartan combined with ramipril against ramipril alone in patients at high risk for cardiovascular events, together with its companion trial TRANSCEND, which compared telmisartan with placebo in patients intolerant to ramipril.

Start: November 2001  ·  Primary completion: June 2008  ·  Sponsor: Boehringer Ingelheim  ·  Results posted on ClinicalTrials.gov
About this page

This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

ONTARGET was a large randomized, parallel-group prevention trial in patients at high risk for cardiovascular events. It asked two different statistical questions at once: whether telmisartan alone was non-inferior to ramipril, and whether combining telmisartan with ramipril was superior to ramipril alone. Patients who could not tolerate ramipril were entered into TRANSCEND, a placebo-controlled comparison of telmisartan.

31,546
Enrollment
ONTARGET and TRANSCEND
1.01
Telmisartan vs ramipril HR
97.5% CI 0.93–1.10; margin 1.13
0.99
Combination vs ramipril HR
95% CI 0.92–1.07
0.92
TRANSCEND HR
95% CI 0.81–1.05
FeatureONTARGET (NCT00153101)
PhasePhase 4
ConditionCardiovascular diseases; patients at high risk for cardiovascular events
Primary purposePrevention
Design modelParallel-group, randomized
InterventionsTelmisartan; combination of telmisartan and ramipril; ramipril
Companion trialTRANSCEND: telmisartan vs placebo in patients intolerant to ramipril
Enrollment31,546
Primary endpointsCardiovascular composite (ONTARGET); renal composite in diabetic nephropathy (ONTARGET); cardiovascular composite (TRANSCEND)
Time frame56 months
DatesStart November 2001; primary completion June 2008; status completed
SponsorBoehringer Ingelheim (industry)

2. Clinical Question

ONTARGET was built around a comparison with an established active treatment. Ramipril served as the reference therapy, and the trial asked whether an alternative drug (telmisartan) could preserve its effect on major cardiovascular events, and whether adding telmisartan on top of ramipril could improve on it. TRANSCEND addressed a different, placebo-controlled question for patients who could not take ramipril at all.

Population

Patients at high risk for cardiovascular events. Those intolerant to ramipril were entered into TRANSCEND instead. The renal primary endpoint was assessed in the subset with diabetic nephropathy (diabetes with macro-albuminuria, UACR ≥300 mg/g creatinine at baseline).

Intervention

Telmisartan 80 mg daily alone, or telmisartan 80 mg daily combined with ramipril 10 mg daily.

Comparator

Ramipril 10 mg daily in ONTARGET; placebo in TRANSCEND.

Primary question

Is telmisartan non-inferior to ramipril, and is the combination superior to ramipril, for time to first cardiovascular death, non-fatal myocardial infarction, non-fatal stroke or hospitalization for congestive heart failure?

3. Trial Design

01
ScreenHigh cardiovascular risk
02
Ramipril toleranceTolerant → ONTARGET; intolerant → TRANSCEND
03
RandomizeParallel groups within each trial
04
Follow-upTime frame 56 months
05
AnalysisCox regression, time to first event

ONTARGET: three parallel arms

REFERENCE ARM · 8,576 at risk (safety)

Ramipril

  • Ramipril 10 mg daily
  • Telmisartan-matching placebo
  • Comparator for both active-arm contrasts
NON-INFERIORITY ARM · 8,542 at risk (safety)

Telmisartan

  • Telmisartan 80 mg daily
  • Ramipril-matching placebo
  • Tested against ramipril with a margin of 1.13
SUPERIORITY ARM · 8,502 at risk (safety)

Telmisartan + ramipril

  • Telmisartan 80 mg daily
  • Ramipril 10 mg daily
  • Tested against ramipril for superiority
TRANSCEND · 2,954 vs 2,972 at risk (safety)

Telmisartan vs placebo

  • Patients intolerant to ramipril
  • Telmisartan compared with placebo
  • Tested for superiority

The arm sizes shown above are the numbers of participants at risk in the serious-adverse-event tables of the ClinicalTrials.gov record. They describe the safety population rather than the randomized totals by arm, which the registry summary does not break down; the overall enrollment across both trials is 31,546.

Why a matching-placebo structure matters. The registry describes the treated groups as telmisartan 80 mg/ramipril 10 mg, telmisartan 80 mg/ramipril 10 mg placebo, and ramipril 10 mg/telmisartan 80 mg placebo. Each patient therefore took the same number of study medications, which is the usual way to keep an active-controlled comparison blinded when the drugs look different. Blinding is especially important in a non-inferiority trial, where any systematic bias that makes the arms look more alike pushes the result toward the conclusion being tested.

4. Analysis Populations

All posted efficacy analyses use a Full Analysis Set (FAS) or a defined subset of it. The choice of population is not a technicality here: in a non-inferiority comparison, the analysis population can change the direction of bias.

Analysis populationDefinition in the registryUsed for
FAS of ONTARGETFull Analysis Set of the ONTARGET trialCardiovascular primary endpoint; most secondary endpoints
FAS [DN]All randomised patients with diabetic nephropathy (UACR ≥300 mg/g creatinine) of the ONTARGET trialRenal primary endpoint; renal components in diabetic nephropathy
FAS of TRANSCENDFull Analysis Set of the TRANSCEND trialTRANSCEND primary endpoint
Patients without baseline diabetesONTARGET patients treated with the randomized regimens for 56 months without baseline diabetesNewly diagnosed diabetes
Patients with baseline MMSEONTARGET patients treated with the randomized regimens with a Mini Mental State Examination at baselineCognitive decline

A full analysis set is generally intended to be as close as possible to the intention-to-treat ideal of analysing every randomized patient in the group to which they were assigned. The FAS [DN] population is a subgroup defined by a baseline characteristic (albuminuria), so randomization is preserved within it; however, it is a much smaller population than the full trial, which is reflected in the wider confidence intervals for the renal endpoints.

5. Endpoints

Primary endpoints

EndpointRegistry definitionTime frame
ONTARGET cardiovascular compositeTime to first event analysis of cardiovascular death, non-fatal myocardial infarction, non-fatal stroke and hospitalization for congestive heart failure.56 months
ONTARGET 3-fold renal composite in diabetic nephropathyDoubling of serum creatinine, progression to end stage renal disease (ESRD) and all-cause mortality in diabetic nephropathy patients. ESRD is defined by initiation of dialysis, need for renal transplantation, or eGFR <15 mL/min/1.73 m². Diabetic nephropathy patients are diabetic patients with macro-albuminuria (UACR ≥300 mg/g creatinine) at baseline. These renal outcomes were not adjudicated (apart from death).56 months
TRANSCEND cardiovascular compositeTime to first event analysis of cardiovascular death, non-fatal myocardial infarction, non-fatal stroke and hospitalization for congestive heart failure, in ACE-inhibitor-intolerant patients with cardiovascular disease.56 months

Secondary endpoints with posted analyses

Reading a composite endpoint. A time-to-first-event composite counts only the first qualifying event for each patient. A patient who is hospitalized for heart failure and later dies contributes one event (the hospitalization) to the composite analysis. This makes composites more statistically efficient than any single component, but it also means that the composite hazard ratio can be driven by the most frequent component, which is not necessarily the most serious one.

6. Statistical Methodology

Cox proportional-hazards regression

Every posted analysis of a time-to-event endpoint in ONTARGET and TRANSCEND uses Cox regression, reporting a hazard ratio with a confidence interval and a p-value. The Cox model estimates the ratio of instantaneous event rates between two groups without requiring any assumption about the shape of the underlying hazard over time.

Conceptual form
h(t | group) = h0(t) × exp(β × treatment)   →   HR = exp(β)

Here h0(t) is an unspecified baseline hazard and β is the log hazard ratio for treatment. The model assumes the ratio of hazards is constant across follow-up (proportional hazards). The ClinicalTrials.gov record does not report covariates, stratification factors, or checks of the proportional-hazards assumption.

Each comparison is pairwise against ramipril

ONTARGET has three arms, but each posted analysis compares one arm with ramipril: telmisartan versus ramipril, and telmisartan plus ramipril versus ramipril. No direct telmisartan versus combination analysis is posted as a primary comparison. Ramipril is therefore the anchor for both questions, and each contrast has its own hypothesis type.

Non-inferiority framework

For the telmisartan versus ramipril comparison on the cardiovascular composite, the registry records the hypothesis as non-inferiority with a margin of 1.13 on the hazard-ratio scale, and reports a 97.5% confidence interval rather than a 95% interval.

Non-inferiority decision rule
H0: HRtelmisartan/ramipril ≥ 1.13   vs   H1: HRtelmisartan/ramipril < 1.13
Conclude non-inferiority if the upper confidence limit < 1.13

The margin expresses the largest relative increase in the hazard of the composite that would still be regarded as preserving an acceptable proportion of ramipril's effect. A margin of 1.13 tolerates, at most, a 13% higher estimated hazard with telmisartan.

Superiority framework

For the combination versus ramipril, for the renal primary endpoint, for TRANSCEND, and for nearly all secondary endpoints, the registry records a superiority hypothesis: the null hypothesis is HR = 1, and the question is whether the confidence interval excludes 1.

Chi-squared test and risk ratio for cognitive decline

Cognitive decline is the one posted endpoint analysed as a binary outcome rather than a time to event. The registry reports a risk ratio with a chi-squared test. A risk ratio compares the proportions of patients with the outcome at the end of the assessment period and does not use information about when the outcome occurred.

7. Primary Results: ONTARGET Cardiovascular Composite

The primary cardiovascular composite was analysed in the FAS of ONTARGET using Cox regression, with separate hypotheses for the two active comparisons.

7.1 Telmisartan vs ramipril (non-inferiority)

Hazard ratio, telmisartan vs ramipril

1.01

97.5% CI: 0.93–1.10   ·   P = 0.0019

Hypothesis: non-inferiority; margin 1.13

Clinical Biostats interpretation

What the estimate means. An HR of 1.01 indicates that the estimated hazard of the composite with telmisartan was essentially the same as with ramipril: a 1% higher estimated hazard, which is a negligible difference.

What the interval says. The 97.5% CI of 0.93–1.10 is compatible with anything from a 7% lower to a 10% higher hazard with telmisartan. Because its upper limit (1.10) lies below the prespecified margin of 1.13, the data exclude the degree of harm that was defined as unacceptable. That is the formal basis for non-inferiority.

What it does not mean. Non-inferiority is not equivalence and is not superiority. The result does not show that telmisartan and ramipril are identical, and it does not rule out a hazard up to 10% higher. The conclusion is only as meaningful as the margin: a reader who would regard a 10% increase as clinically important should interpret the result with that in mind.

Why the p-value is not the effect size. P = 0.0019 tests the non-inferiority null hypothesis (HR ≥ 1.13), not the usual null of no difference. It is small because the interval sits clearly below the margin, not because telmisartan had a large effect. The registry does not state whether this p-value is one-sided or two-sided; in non-inferiority testing the hypothesis is inherently one-directional.

Why a 97.5% interval. A 97.5% interval is wider than a 95% interval and corresponds to a stricter error rate. Trials with more than one primary comparison commonly use a narrower alpha per comparison; the registry records the interval level but not the rationale. Using a wider interval makes the non-inferiority test harder to pass, so the conclusion is conservative in that respect.

Cautions. In a non-inferiority trial, poor adherence, crossover between treatments or lack of assay sensitivity all tend to make arms look similar and so favour a non-inferiority conclusion. A full-analysis-set result is therefore usually read alongside a per-protocol analysis; the registry reports the FAS analysis. The single hazard ratio also assumes that the relative hazard was roughly constant over the 56-month time frame.

7.2 Telmisartan + ramipril vs ramipril (superiority)

Hazard ratio, combination vs ramipril

0.99

95% CI: 0.92–1.07   ·   P = 0.8462

Hypothesis: superiority

Clinical Biostats interpretation

What the estimate means. An HR of 0.99 corresponds to a 1% lower estimated hazard of the composite with the combination, which is effectively no difference from ramipril alone.

What the interval says. The 95% CI of 0.92–1.07 is narrow, reflecting the size of the trial. It is compatible with at most an 8% lower hazard and at most a 7% higher hazard. Because the interval includes 1, superiority was not demonstrated. Because it is also narrow, the data argue against any large benefit of adding telmisartan to ramipril for this composite.

What it does not mean. A non-significant superiority test does not by itself prove that the two regimens are equivalent. The posted hypothesis for this comparison was superiority, and no non-inferiority or equivalence test is posted for it. The narrow interval, however, does make a clinically important benefit on this endpoint implausible.

Why the p-value is not the effect size. P = 0.8462 means the observed HR is very close to what would be expected if the combination had no effect; it does not measure how large any effect is. The confidence interval carries that information.

Cautions. The comparison adds a second drug to an active background, so any added benefit had to be detected on top of ramipril's own effect. Interpretation of the combination should also consider its safety profile (Section 11) and its renal secondary results (Section 10), rather than the primary composite alone.

8. Primary Results: Renal Composite in Diabetic Nephropathy

The second ONTARGET primary endpoint was the 3-fold composite of doubling of serum creatinine, progression to ESRD and all-cause mortality, analysed in the FAS [DN] population of randomized patients with diabetic nephropathy. Both active arms were compared with ramipril under superiority hypotheses.

ComparisonHR95% CIP-valueHypothesis
Telmisartan + ramipril vs ramipril0.920.71–1.20 (two-sided)0.5461Superiority
Telmisartan vs ramipril0.880.68–1.140.3436Superiority
Clinical Biostats interpretation: combination vs ramipril

What the estimate means. An HR of 0.92 corresponds to an 8% lower estimated hazard of the renal composite with the combination.

What the interval says. The two-sided 95% CI of 0.71–1.20 is wide: it is compatible with a 29% lower hazard or a 20% higher hazard. Compared with the cardiovascular composite, the imprecision reflects a smaller population and fewer events. The data are therefore inconclusive about the direction of any effect on this endpoint.

What it does not mean. The point estimate below 1 is not evidence of renal benefit; a result this imprecise does not distinguish benefit from harm.

Why the p-value is not the effect size. P = 0.5461 reflects the combination of a modest point estimate and a wide interval. A larger study with the same HR could produce a much smaller p-value; the p-value summarises compatibility with no effect, not the magnitude of an effect.

Cautions. The registry states that the renal outcomes were not adjudicated, apart from death. Unadjudicated endpoints can carry more measurement noise, which usually biases estimates toward no difference. The composite also includes all-cause mortality, so it combines kidney-specific and non-specific events.

Clinical Biostats interpretation: telmisartan vs ramipril

What the estimate means. An HR of 0.88 corresponds to a 12% lower estimated hazard of the renal composite with telmisartan than with ramipril.

What the interval says. The 95% CI of 0.68–1.14 includes 1 and extends from a 32% lower to a 14% higher hazard. Superiority was not demonstrated, and the interval is too wide to support a conclusion of similarity either.

What it does not mean. Although the cardiovascular composite was tested for non-inferiority, this renal comparison was posted as a superiority test. Its non-significance cannot be read as evidence of non-inferiority, because no margin was posted for this endpoint.

Why the p-value is not the effect size. P = 0.3436 simply indicates that data of this kind would not be unusual if there were no true difference.

Cautions. As with the combination, the unadjudicated renal outcomes and the restricted diabetic-nephropathy population limit precision. With three primary endpoints in the programme, each individual test is also subject to multiplicity considerations (Section 12).

9. Primary Results: TRANSCEND

Hazard ratio, telmisartan vs placebo (TRANSCEND)

0.92

95% CI: 0.81–1.05   ·   P = 0.2192

Population: FAS of the TRANSCEND trial  ·  Hypothesis: superiority

Clinical Biostats interpretation

What the estimate means. An HR of 0.92 corresponds to an 8% lower estimated hazard of cardiovascular death, non-fatal myocardial infarction, non-fatal stroke or hospitalization for congestive heart failure with telmisartan compared with placebo.

What the interval says. The 95% CI of 0.81–1.05 spans from a 19% lower hazard to a 5% higher hazard. It is wider than the ONTARGET intervals because TRANSCEND was a smaller trial. The interval includes 1, so superiority over placebo on the primary composite was not demonstrated.

What it does not mean. Failing to reject the null hypothesis is not evidence that telmisartan has no effect. The interval leaves open a clinically meaningful reduction as well as a small increase; the trial simply could not distinguish among these possibilities with the precision available.

Why the p-value is not the effect size. P = 0.2192 does not indicate a small effect. It indicates that, given the variability in the data, an HR of 0.92 is not far enough from 1 to be declared statistically significant at the conventional level.

Cautions. TRANSCEND enrolled a distinct population (patients intolerant to ramipril), so its placebo-controlled result cannot be combined directly with the active-controlled ONTARGET comparisons. The two trials answer different questions about different patients.

10. Secondary Endpoint Results

The registry posts Cox regression analyses for many secondary endpoints, each comparing an active arm with ramipril. These analyses are supportive. The registry does not describe a hierarchical testing procedure or a multiplicity adjustment for them, so their p-values should be read as nominal. The ClinicalTrials.gov record also contains further secondary and other analyses beyond those tabulated here.

Cardiovascular components

EndpointTelmisartan + ramipril vs ramiprilTelmisartan vs ramipril
CV death, non-fatal MI, non-fatal strokeHR 1.00 (95% CI 0.93–1.09); P = 0.9086HR 0.99 (97.5% CI 0.90–1.08); P = 0.0004; non-inferiority, margin 1.13
Cardiovascular deathHR 1.04 (0.93–1.17); P = 0.4535HR 1.00 (0.89–1.12); P = 0.9421
Non-fatal myocardial infarctionHR 1.08 (0.94–1.23); P = 0.2909HR 1.08 (0.94–1.24); P = 0.2534
Non-fatal strokeHR 0.92 (0.79–1.06); P = 0.2248HR 0.91 (0.79–1.05); P = 0.1829
Hospitalization for congestive heart failureHR 0.95 (0.82–1.10); P = 0.4984HR 1.12 (0.97–1.29); P = 0.1203
Newly diagnosed congestive heart failureHR 0.94 (0.83–1.07); P = 0.3732HR 1.06 (0.94–1.20); P = 0.3633
Cardiovascular revascularization procedureHR 1.04 (0.97–1.13); P = 0.2713HR 1.03 (0.95–1.11); P = 0.5150

Intervals are 95% unless stated. The three-component composite (without heart-failure hospitalization) repeats the pattern of the primary endpoint: non-inferiority of telmisartan was tested with the same margin, with an upper 97.5% limit of 1.08, and the combination showed no evidence of superiority. None of the individual components shows a difference whose confidence interval excludes 1. Individual components have fewer events, so their intervals are wider and they are not powered for separate conclusions.

Renal outcomes in diabetic nephropathy (FAS [DN])

EndpointTelmisartan + ramipril vs ramiprilTelmisartan vs ramipril
Doubling of serum creatinineHR 1.25 (0.73–2.15, two-sided); P = 0.4221HR 0.90 (0.51–1.60); P = 0.7305
Progression to ESRDHR 0.88 (0.54–1.45); P = 0.6248HR 0.62 (0.36–1.05, two-sided); P = 0.0751
All-cause mortalityHR 0.87 (0.63–1.18); P = 0.3682HR 0.92 (0.69–1.24); P = 0.6014

These components of the renal primary composite have very wide intervals; the upper limit for doubling of serum creatinine with the combination reaches 2.15. Point estimates on either side of 1 carry little information when intervals are this wide. The telmisartan versus ramipril estimate for ESRD (HR 0.62, P = 0.0751) is the kind of result that is easily over-read: the interval reaches 1.05, and it is one of many secondary comparisons.

Renal outcomes in the full ONTARGET population

EndpointTelmisartan + ramipril vs ramiprilTelmisartan vs ramipril
Doubling of serum creatinineHR 1.14 (0.91–1.42); P = 0.2550HR 1.07 (0.86–1.34); P = 0.5297
Progression to ESRDHR 1.13 (0.82–1.56); P = 0.4593HR 0.95 (0.68–1.32); P = 0.7468
New microalbuminuriaHR 0.88 (0.80–0.97); P = 0.0133HR 0.93 (0.84–1.02); P = 0.1251
New macroalbuminuriaHR 0.81 (0.67–0.97); P = 0.0213HR 0.90 (0.75–1.07); P = 0.2396
Combined: creatinine doubling, ESRD, new micro- or macroalbuminuriaHR 0.90 (0.83–0.98); P = 0.0111HR 0.92 (0.85–1.00); P = 0.0606
Normalisation from micro- or macroalbuminuria to normoalbuminuriaHR 1.19 (1.05–1.35); P = 0.0082HR 1.08 (0.95–1.23); P = 0.2164
Surrogate markers versus hard renal outcomes. For the combination, the albuminuria-based endpoints have confidence intervals that exclude 1 in a favourable direction: fewer patients developed new micro- or macroalbuminuria, and more patients normalised (for normalisation, an HR above 1 is favourable because the event is desirable). The harder renal outcomes in the same population, doubling of serum creatinine (HR 1.14) and progression to ESRD (HR 1.13), point the other way, although neither interval excludes 1. The combined endpoint (HR 0.90) is dominated by the far more frequent albuminuria events. This is a classic illustration of why a composite that mixes a surrogate marker with hard clinical outcomes can give a different impression from the outcomes patients experience, and why nominal p-values from many secondary tests should not be treated as confirmatory.

Other secondary endpoints

EndpointPopulationTelmisartan + ramipril vs ramiprilTelmisartan vs ramipril
Newly diagnosed diabetesPatients without diabetes at baselineHR 0.92 (0.77–1.10); P = 0.3485HR 1.14 (0.96–1.36); P = 0.1235
Cognitive declinePatients with MMSE at baselineRisk ratio 0.99 (0.89–1.11); chi-squared P = 0.8690—

Newly diagnosed diabetes is analysed only among patients free of diabetes at baseline, as it must be for an incidence endpoint. Cognitive decline is analysed as a binary outcome with a risk ratio: a risk ratio of 0.99 means the proportion of patients with cognitive decline was almost the same in the two groups, and the interval of 0.89–1.11 is compatible only with small differences in either direction.

11. Safety: Serious Adverse Events

The ClinicalTrials.gov record reports the number of participants with at least one serious adverse event and the number at risk in each arm.

TrialArmAffectedAt risk
ONTARGETTelmisartan + ramipril5,2688,502
ONTARGETTelmisartan5,5068,542
ONTARGETRamipril5,3758,576
TRANSCENDTelmisartan1,7812,954
TRANSCENDPlacebo1,8302,972

The majority of participants in every arm experienced at least one serious adverse event over the long follow-up, which is expected in an older, high-risk population in which the efficacy outcomes themselves (myocardial infarction, stroke, heart-failure hospitalization) typically qualify as serious events. The registry presents these counts descriptively, without a between-arm statistical test. Because serious adverse events overlap heavily with efficacy events, overall SAE counts are a blunt instrument for comparing tolerability; specific event categories are more informative.

Adverse-event collection. The record states that only non-serious adverse events that led to discontinuation of study medication were assessed and collected, with a frequency under 5%; non-serious adverse events as such were not assessed or collected. Non-serious adverse-event tables in the registry therefore do not describe the full burden of milder side effects and cannot be compared with trials that collected all adverse events.

12. Multiplicity and Endpoint Hierarchy

The programme has three registered primary endpoints, two active comparisons within ONTARGET, and a large number of secondary analyses. Each additional test raises the chance that at least one result appears significant by chance alone.

AnalysisRoleInterpretation
Telmisartan vs ramipril, CV compositePrimary; non-inferiorityConfirmatory; judged against the 1.13 margin using a 97.5% CI
Combination vs ramipril, CV compositePrimary; superiorityConfirmatory; superiority not shown
Renal composite in diabetic nephropathyPrimary; superiorityNeither comparison shows a difference; unadjudicated outcomes in a subset
TRANSCEND, CV compositePrimary; superioritySeparate population; superiority over placebo not shown
Secondary cardiovascular, renal and other endpointsSecondarySupportive; nominal p-values; no multiplicity procedure reported in the registry

The use of a 97.5% interval for the non-inferiority comparison suggests that some form of alpha control across the primary comparisons was built into the design, but the ClinicalTrials.gov record does not describe the procedure. Among the secondary endpoints, several albuminuria outcomes have p-values between 0.0082 and 0.0213. Against the background of dozens of secondary comparisons, results of this size warrant caution unless they were protected within a prespecified hierarchy.

13. Statistical Methods Explained

Why is non-inferiority judged against the margin rather than the p-value?

The question in a non-inferiority test is not whether telmisartan differs from ramipril but whether it is worse by more than a prespecified amount. The decision therefore depends on where the confidence interval lies relative to the margin of 1.13. The upper 97.5% limit of 1.10 is below 1.13, so non-inferiority is shown. The accompanying p-value of 0.0019 is a test against the margin, not against HR = 1; reading it as "telmisartan is significantly different from ramipril" would be the opposite of what the result says.

Why can one trial test non-inferiority and superiority at the same time?

ONTARGET contains two separate contrasts with a common reference arm. Telmisartan alone is an alternative to ramipril, so the relevant question is whether it preserves ramipril's effect (non-inferiority). The combination adds to ramipril, so the relevant question is whether it improves on it (superiority). Each contrast has its own null hypothesis, and the trial's conclusions for one do not transfer to the other.

What does a hazard ratio of 0.99 with a 95% CI of 0.92–1.07 tell us?

It tells us more than "not significant". The interval is narrow enough to rule out, at the 95% level, reductions in hazard larger than 8% and increases larger than 7%. For a large, well-conducted trial, a narrow interval centred on 1 is informative evidence that any effect is small, which is different from a small trial whose interval is wide and centred on 1.

Why was Cox regression used for almost every endpoint?

The endpoints are times to a first event over a 56-month time frame, and patients were followed for different lengths of time. Cox regression uses the timing of each event and handles right-censoring for patients who had not had an event by the end of their follow-up. It gives a single relative measure, the hazard ratio, without assuming a particular shape for the baseline hazard. Kaplan-Meier curves would normally accompany these analyses to show absolute event rates over time; the registry does not report them.

Why is cognitive decline analysed with a chi-squared test and a risk ratio?

Cognitive decline was assessed by comparing Mini Mental State Examination results, which yields a yes/no outcome at assessment rather than a precisely dated event. A risk ratio compares proportions, and the chi-squared test assesses whether those proportions differ more than expected by chance. The risk ratio of 0.99 (95% CI 0.89–1.11) is directly interpretable as the ratio of the proportions of patients with decline.

How should an HR above 1 be read for "normalisation to normoalbuminuria"?

The direction of a hazard ratio always refers to the event being counted. For most endpoints in ONTARGET the event is harmful, so HR < 1 favours the experimental arm. For normalisation the event is beneficial, so the combination's HR of 1.19 (95% CI 1.05–1.35) means a 19% higher estimated rate of returning to normal albumin levels. Reading hazard ratios without checking what event is counted is a common source of error.

14. Limitations

15. Why This Trial Matters Statistically

ONTARGET is a valuable teaching case because a single, very large programme contains a non-inferiority test, a superiority test against an active control, a placebo-controlled companion trial, a subset-defined primary endpoint and an instructive conflict between surrogate and hard secondary outcomes.

ConceptHow it appears in ONTARGET
Non-inferiority designTelmisartan vs ramipril, margin 1.13 on the hazard-ratio scale
Superiority vs active controlTelmisartan + ramipril vs ramipril
Placebo-controlled companion trialTRANSCEND for patients intolerant to ramipril
Cox regressionHazard ratios for all time-to-event endpoints
Confidence-interval level97.5% intervals for non-inferiority comparisons; 95% elsewhere
Composite endpointsTime to first cardiovascular or renal event
Analysis populationsFull analysis sets and baseline-defined subsets
Risk ratio and chi-squared testBinary cognitive-decline outcome
MultiplicityThree primary endpoints and many secondary comparisons
Surrogate vs clinical outcomesAlbuminuria results diverge from creatinine doubling and ESRD
Informative null resultNarrow CI around HR 0.99 for the combination

16. Related Tutorials

Learn more about the methods used in this trial:

17. Related Calculators

18. Sources

Continue learning with Clinical Biostats

Explore the statistical methods behind this trial in more depth, or try the related calculators with your own numbers.

19. Record Summary

ONTARGET shows how much depends on the hypothesis being tested. With virtually the same point estimate near 1, telmisartan alone met its non-inferiority criterion against ramipril (HR 1.01; 97.5% CI 0.93–1.10; margin 1.13), while the combination did not show superiority (HR 0.99; 95% CI 0.92–1.07). The renal primary endpoint in diabetic nephropathy and the TRANSCEND primary endpoint were both inconclusive, with intervals that include 1. The most useful interpretation combines hazard ratios, confidence intervals relative to 1 and to the margin, the analysis population, the distinction between surrogate and clinical outcomes, and the number of comparisons performed.

Clinical Biostats methodology: A trial-results page should not merely repeat the registry entry. The goal is to set out the statistical story of the trial in a standardized format while clearly separating reported results from educational interpretation.