In 1987, the FDA approved zidovudine (AZT) for HIV based on a trial showing improved survival at 24 weeks. In 2011, the FDA revoked accelerated approval for bevacizumab in breast cancer after confirmatory trials failed to show the survival benefit the surrogate endpoint had predicted. In 2021, the approval of aducanumab for Alzheimer’s disease sparked controversy because the drug reduced amyloid plaques (the surrogate) but did not clearly improve cognition (the clinical outcome).
These cases illustrate a fundamental challenge: what should we measure to determine whether a drug works? The answer shapes trial design, determines sample size, affects regulatory strategy, and ultimately decides whether patients gain access to new therapies.
An endpoint is a defined outcome that a clinical trial is designed to measure. It is the variable that answers the trial’s central question: did the treatment work? The term “endpoint” reflects the fact that these measurements typically occur at the end of treatment or follow-up, though some endpoints are measured continuously throughout the trial.
Endpoints must be specified before the trial begins. The primary endpoint is the main outcome on which the trial will be judged: the result that determines whether the trial “succeeded” or “failed.” A Phase III trial might have survival as its primary endpoint, meaning the trial succeeds if the treatment group lives significantly longer than the control group. Secondary endpoints provide additional information but do not determine the trial’s overall success.
Choosing the right endpoint requires balancing what matters clinically (does the patient feel better? live longer?) with what is practical (can we measure it reliably? in a reasonable timeframe? with an affordable sample size?). This tension between clinical meaningfulness and operational feasibility runs through every endpoint decision.
The Endpoint Hierarchy
Table 17.1 summarizes the endpoint hierarchy with examples across therapeutic areas.
At the apex of the endpoint hierarchy sits what we ultimately care about: whether patients live longer, feel better, or function more effectively. These clinical endpoints (sometimes called hard endpoints) directly measure patient outcomes. Survival is the paradigm: either the patient is alive or they are not. There is no ambiguity, no measurement error, no room for interpretation.
Below clinical endpoints sit surrogate endpoints: measurements that are not themselves clinical outcomes but that are reasonably expected to predict clinical outcomes. Blood pressure is a surrogate for stroke and heart attack. Viral load is a surrogate for AIDS progression. Tumor shrinkage is a surrogate for cancer survival.
Further down the hierarchy are biomarkers: measurable characteristics that indicate biological processes, disease states, or responses to intervention. Some biomarkers may also be surrogates, but not all. A biomarker that reliably reflects drug activity might not predict clinical benefit.
Clinical Endpoints
The most compelling endpoint is one that directly measures what matters to patients. In oncology, overall survival (OS) is the gold standard: the most reliable way to determine whether a cancer treatment extends life. In heart failure, the composite of cardiovascular death and heart failure hospitalization captures both mortality and major morbidity.
But clinical endpoints come with practical challenges. They may require large trials because events are relatively rare. They may require long follow-up because outcomes develop slowly. A trial with survival as its primary endpoint in a slowly progressing cancer might need thousands of patients followed for years, a massive undertaking with implications for cost, time to approval, and patient access to therapy.
Patient-reported outcomes (PROs) represent another category of clinical endpoints. How does the patient feel? What is their quality of life? Can they perform daily activities? These outcomes are measured through standardized questionnaires and reflect the patient’s perspective, which may differ from what objective measurements suggest.
PROs are increasingly important to regulators and payers. A drug that improves a biomarker but does not make patients feel better is less valuable than one that produces symptomatic improvement. However, PROs introduce measurement challenges: responses can be influenced by expectations, mood, and cultural factors, and the validity of PRO instruments must be established.
Unlike a blood test, a questionnaire cannot be calibrated against a physical standard. Instead, PROs must undergo psychometric validation to prove they measure what they claim to measure. Content validity establishes that the instrument covers the relevant symptoms, typically through qualitative interviews with patients. Reliability ensures the measure is stable and reproducible. Construct validity checks whether the score correlates with other known measures of the disease, while sensitivity to change confirms the instrument can detect meaningful improvements. Regulators require this evidence to approve labeling claims based on PROs.
The proliferation of smartphones and wearables has produced a new category of endpoints: those derived from Digital Health Technologies (DHTs). DHTs offer the promise of moving from snapshot medicine to continuous, real-world monitoring. This includes digital biomarkers (such as subtle changes in gait or voice that may predict disease progression) and continuous monitoring that replaces brief clinical assessments with long-term data collection.
While promising, DHTs face significant validation hurdles. Regulators must be convinced that the device measures the concept accurately, that the measurement correlates with clinical status, and that the data is secure and attributable. The FDA has established a specific Digital Health Center of Excellence to guide the integration of these novel tools into regulatory decision-making (U.S. Food and Drug Administration 2023).
Surrogate Endpoints
Given the challenges of clinical endpoints, surrogates offer an attractive alternative. They can often be measured sooner and in fewer patients, potentially accelerating development and reducing costs.
The concept behind surrogates is straightforward: if intervention A lowers blood pressure more than intervention B, and lower blood pressure reduces cardiovascular events, then intervention A should reduce cardiovascular events more than intervention B.
But this logic can fail. The relationship between surrogate and outcome may not hold under all circumstances. An intervention might affect the surrogate through a pathway that does not influence clinical outcomes. Or it might have effects on clinical outcomes that are not mediated by the surrogate.
The history of medicine includes instructive failures. The CAST trial of the late 1980s tested antiarrhythmic drugs in post-heart attack patients. The drugs effectively suppressed the abnormal heart rhythms they were designed to treat: the surrogate looked excellent. But the trial was stopped early when it became clear that treated patients were more likely to die than those on placebo. The rhythm abnormalities were a marker of underlying heart damage, not a cause of death; suppressing them did nothing to address the underlying disease and may have introduced new cardiac risks.
Similarly, hormone replacement therapy improved lipid profiles and was widely prescribed for cardiovascular prevention based on surrogate reasoning and observational data. The Women’s Health Initiative randomized trial showed that, despite the favorable lipid effects, hormone therapy actually increased cardiovascular events. The surrogate had predicted the wrong outcome.
The FDA distinguishes between validated surrogates (where the relationship to clinical outcome is well established) and reasonably likely surrogates (where the relationship is plausible but not fully established). The latter can support accelerated approval for serious conditions, but sponsors must subsequently confirm the clinical benefit through post-marketing studies.
Primary and Secondary Endpoints
Table 17.2 defines the roles of different endpoint categories in clinical trials.
The primary endpoint is the outcome that the trial is designed and powered to assess. It is the basis for the primary analysis and typically the endpoint that regulatory decisions rest upon. A trial succeeds or fails based primarily on its primary endpoint.
Secondary endpoints provide additional information. They may address different aspects of efficacy, safety, or tolerability. They may assess outcomes in subpopulations. They may evaluate the durability of effects over time.
The relationship between primary and secondary endpoints requires careful thought. If the primary endpoint is negative (no statistically significant difference between treatment groups), positive secondary endpoints are difficult to interpret: they may represent chance findings from multiple comparisons. If the primary endpoint is positive, secondary endpoints provide useful confirmation and characterization.
Multiplicity is the statistical challenge that arises from testing multiple endpoints. With a conventional significance threshold of 0.05, one in twenty tests will be “positive” by chance alone. When multiple endpoints are tested, the overall probability of false positive findings increases. Statistical methods for multiplicity adjustment (such as hierarchical testing, gatekeeping procedures, or alpha spending) control this inflation of false positives.
Composite Endpoints
A composite endpoint combines multiple outcomes into a single measure. A participant might be counted as having an event if they experience any one of: death, heart attack, stroke, or hospitalization for unstable angina. The composite counts a single event per participant, with the most severe event taking precedence (death counts, even if the participant also had a stroke).
Composites offer statistical efficiency: because events are more common (many participants will have at least one component), smaller sample sizes may suffice. They also provide a more complete picture of disease burden, capturing multiple manifestations of benefit or harm.
But composites introduce interpretive challenges. If the composite is positive overall, which components are driving the effect? Is a treatment that reduces hospitalizations but has no effect on death meaningfully beneficial? Should a drug that reduces minor events but not major ones be approved?
The components of a composite should be clinically related and of roughly similar importance. A composite that includes both death and mild headache is problematic because the components have vastly different clinical significance.
Time-to-Event Endpoints
Many outcomes are measured not as binary events but as time-to-event: how long until a participant experiences the outcome. Time-to-event analyses can distinguish between treatments that reduce eventual event rates versus treatments that delay events.
Survival analysis methods (Kaplan-Meier curves, log-rank tests, Cox proportional hazards models) are designed for time-to-event data. They can handle participants who have not yet experienced an event (they are censored at the time of analysis) and can incorporate participants who drop out before experiencing an event.
A key assumption in many time-to-event analyses is proportional hazards: the relative risk of an event is constant over time. If a treatment reduces risk by 30% at month 6, it should also reduce risk by 30% at month 12. When this assumption is violated (as when a treatment takes time to work, or when its effects wane over time) alternative methods may be needed.
Ordinal Endpoints
Many outcomes that appear binary are actually ordered. A patient with heart failure does not simply have “a cardiovascular event” or “no event.” They may die, be hospitalized, remain symptomatic but outpatient, or be nearly asymptomatic. Collapsing this spectrum into a binary endpoint discards clinically meaningful information.
Ordinal endpoints preserve the ordering by treating the outcome as a ranked scale rather than a yes/no measurement. The modified Rankin Scale (mRS) in stroke trials is the canonical example: it runs from 0 (no symptoms) through 6 (death), with each step representing a clinically distinct level of functional disability (Saver 2007). The WHO ordinal scale used in COVID-19 trials similarly ranked patients from “uninfected” (score 0) to “dead” (score 8), with intermediate levels for different levels of oxygen support and hospitalization (WHO Working Group on the Clinical Characterisation and Management of COVID-19 infection 2020).
The standard statistical analysis for ordinal outcomes is the proportional-odds (PO) model, a generalization of logistic regression (Agresti 2010). The PO model assumes that the treatment odds ratio is the same at every cut-point of the ordinal scale (a patient on treatment is \(OR\)-times more likely to be in a better category than the reference, regardless of where the cut-point is placed). When this assumption holds, the PO model uses all the information in the ordinal scale rather than dichotomizing at an arbitrary threshold. If the PO assumption is violated, it can be checked with a score test and, if needed, replaced by a proportional-hazards model on discrete time or by non-parametric rank tests.
For sponsors, the practical benefit is statistical power: for a given sample size, an ordinal analysis is typically more powerful than a binary analysis of the same data, because the ordinal outcome contains more information. For regulators, the benefit is interpretability: an effect that shifts the entire distribution toward better categories is clinically meaningful, not just one that moves patients across a single threshold.
The Win Ratio and DOOR
Two related methods address a limitation of composite endpoints: all components are treated as equally important even when they are not. A composite of “death or hospitalization” that counts a 2-day readmission the same as death misrepresents clinical priority.
The win ratio. For each possible pair of patients (one from the treatment arm, one from the control arm), the pair is adjudicated based on a hierarchy of outcomes: check the most severe outcome first (death), and if one patient wins on that outcome, that patient’s arm wins the pair; if tied on death, check the next outcome (hospitalization), and so on (Pocock et al. 2012). The win ratio is the number of pairs where the treatment arm patient “wins” divided by the number of pairs where the control arm patient wins. Tied pairs are excluded. The clinical priority hierarchy is set by the investigators before the trial, making the clinical value judgment explicit and transparent. Win ratio analyses have been used in heart failure, oncology, and COVID-19 trials.
Desirability of Outcome Ranking (DOOR). DOOR extends the win ratio concept to continuous outcomes and patient-reported data (Evans et al. 2015). Every patient receives a score (their DOOR) based on a pre-specified combination of clinical outcomes, adverse events, and, optionally, patient preferences. Patients are then ranked from worst to best DOOR score, and the probability that a randomly selected treatment patient has a better DOOR than a randomly selected control patient (\(P[T > C]\)) is the summary statistic. A \(P[T > C] > 0.5\) indicates that, on average, treated patients have better outcomes than control patients across the entire outcome spectrum.
DOOR is particularly valuable when the goal is to summarize benefit and harm simultaneously across a broad clinical landscape, such as antibiotic trials where the endpoint should capture cure, adverse events, and resistance emergence in a single measure.
Estimands and Death as an Intercurrent Event
The ICH E9(R1) estimands framework (International Council for Harmonisation 2019) formalized a question that had long been implicit in trial design: what treatment effect, precisely, is the trial trying to estimate? An estimand is the complete specification of the treatment comparison: the population, the treatment and control, the endpoint variable, how intercurrent events are handled, and the population-level summary measure.
Intercurrent events are events that occur after treatment initiation and affect either the interpretation or the observability of the endpoint. Treatment discontinuation, rescue medication use, and crossover to the other arm are common examples. Death is the hardest case.
When the primary endpoint is a non-fatal outcome (symptom score, functional status, quality of life), death is an intercurrent event that makes the outcome unobservable: a dead patient has no quality-of-life score. The choice of how to handle death fundamentally changes the estimand:
- Treatment policy strategy: count data regardless of what happened to the patient (including deaths, with death typically assigned the worst possible score). This estimates the effect on the composite of survival and the non-fatal outcome.
- Hypothetical strategy: estimate what the outcome would have been if the patient had survived (not died and not discontinued). This is a counterfactual that requires modeling assumptions.
- While on treatment strategy: estimate the effect only during the period when patients remain on assigned treatment, excluding post-discontinuation data. This estimates treatment effect while the patient is actually taking the drug.
For clinical trial teams, the practical consequence is that endpoint selection cannot proceed without simultaneously specifying the estimand: a quality-of-life endpoint and a symptom-improvement endpoint are the same measurement but potentially different estimands, depending on how death is handled. Pre-specifying the estimand prevents post-hoc ambiguity about what the trial demonstrated, and aligns the analysis with the clinical question that motivated the trial.
Endpoint Selection in Practice
Selecting the optimal endpoint requires balancing regulatory acceptability, clinical meaningfulness, and operational feasibility. Regulators generally favor well-validated clinical outcomes, but sponsors must also confirm that the endpoint captures benefit that patients and payers recognize. On the operational side, the endpoint must be measurable with enough precision to maintain statistical power, and its timing must avoid unnecessary delays while still providing a reliable prediction of long-term benefit. For patient-reported outcomes in particular, using established and validated instruments is the most direct way to produce findings that are reproducible and credible to external stakeholders. AI-based endpoint assessment, including automated image scoring, wearable-derived digital biomarkers, and NLP extraction of outcomes from clinical notes, is covered in Chapter 26.
Agresti, Alan. 2010.
Analysis of Ordinal Categorical Data. 2nd ed. Wiley.
https://doi.org/10.1002/9780470594001.
Evans, Scott R., David Rubin, Dean Follmann, Gene Pennello, W. Charles Huskins, John H. Powers, Jerry Rhorer, et al. 2015.
“Desirability of Outcome Ranking (DOOR) and Response Adjusted for Duration of Antibiotic Risk (RADAR).” Clinical Infectious Diseases 61 (5): 800–806.
https://doi.org/10.1093/cid/civ495.
International Council for Harmonisation. 2019.
“ICH E9(R1): Addendum on Estimands and Sensitivity Analysis in Clinical Trials to the Guideline on Statistical Principles for Clinical Trials.” https://database.ich.org/sites/default/files/E9-R1_Step4_Guideline_2019_1203.pdf.
Pocock, Stuart J., Cono A. Ariti, Timothy J. Collier, and Duolao Wang. 2012.
“The Win Ratio: A New Approach to the Analysis of Composite Endpoints in Clinical Trials Based on Clinical Priorities.” European Heart Journal 33 (2): 176–82.
https://doi.org/10.1093/eurheartj/ehr352.
Saver, Jeffrey L. 2007.
“Novel End Point Analytic Techniques and Interpreting Shifts Across the Entire Range of Outcome Scales in Acute Stroke Trials.” Stroke 38 (11): 3055–62.
https://doi.org/10.1161/STROKEAHA.107.488536.
U.S. Food and Drug Administration. 2023.
“Digital Health Technologies for Remote Data Acquisition in Clinical Investigations: Guidance for Industry, Investigators, and Other Stakeholders.” https://www.fda.gov/regulatory-information/search-fda-guidance-documents/digital-health-technologies-remote-data-acquisition-clinical-investigations.
WHO Working Group on the Clinical Characterisation and Management of COVID-19 infection. 2020.
“A Minimal Common Outcome Measure Set for COVID-19 Clinical Research.” Lancet Infectious Diseases 20 (8): e192–97.
https://doi.org/10.1016/S1473-3099(20)30483-7.