In this QCast episode, co-hosts Jullia and Tom demystify estimands in clinical trials and show how a clear question, defined up front, leads to cleaner design, better data, and defensible analyses. They unpack the four elements of an estimand, explain how to handle intercurrent events without confusion, and discuss how summary measures should match the clinical decision you want to inform. The conversation also covers practical implications for protocols, case report forms, data flows, and sensitivity analyses—so teams avoid rework and keep stakeholders aligned from planning to readout.
🎧 Listen to the Episode:
Key Takeaways
What Estimands Are and Why They Matter
An estimand precisely states the clinical question by defining the population, variable, how intercurrent events are handled, and the summary measure. Distinct from the estimate (the number) and estimator (the method), it aligns objectives, data capture, and analysis so results answer the decision at hand.
Handling Intercurrent Events
Choose the strategy that fits the decision context, not habit: treatment policy for real-world use; composite when the event signals failure; hypothetical if the event is external and needs modelling; while-on-treatment for tolerability; principal stratum for the subset where the event would not occur under either arm.
Choosing Summary Measures
Match the measure to the question and communication need. Use hazard ratio for time-to-event when appropriate, consider restricted mean survival time when hazards are non-proportional or clarity is key, prefer risk difference or risk ratio for binary outcomes, and select meaningful longitudinal summaries for continuous data.
Design and Data Implications
Mirror the estimand in protocol and analysis plan, pre-specify strategies and estimators, and plan targeted sensitivity analyses. Ensure forms and external feeds capture intercurrent events with timing and reasons so the chosen strategy is implementable without guesswork.
Quick Tips and Common Pitfalls
Start the primary estimand at concept stage; justify handling of each intercurrent event in one plain sentence; select an estimator that truly targets the estimand and define two or three focused sensitivities. Avoid conflating missing data with intercurrent events, misaligned strategies and objectives, and inconsistencies across protocol, analysis plan, and data capture.
Full Transcript
Jullia
Welcome to QCast, the show where biometric expertise meets data-driven dialogue. I’m Jullia.
Tom
I’m Tom, and in each episode, we dive into the methodologies, case studies, regulatory shifts, and industry trends shaping modern drug development.
Jullia
Whether you’re in biotech, pharma or life sciences, we’re here to bring you practical insights straight from a leading biometrics CRO. Let’s get started.
Tom
Today we’re exploring estimands in clinical trials. Let’s start from the first principles. People hear the term estimand and assume it is just a fancy label for an endpoint or a statistical method. Could you explain what an estimand is, how it differs from an estimate or an estimator, and why it has become central to how we design and analyse trials today?
Jullia
So, an estimand is a precise statement of the clinical question we want to answer. It links the objective to the effect of interest in a way that is unambiguous for both clinicians and statisticians. Where an estimate is the numeric result we get from data, and an estimator is the method we use to calculate it, the estimand sits upstream and defines what that number should represent. The framework has four core elements: the population of interest, the endpoint or variable, how we handle intercurrent events, and the summary measure. Intercurrent events are things that happen after treatment starts that affect interpretation, such as treatment discontinuation, rescue medication, or death. The summary measure could be a mean difference, a proportion, a hazard ratio, or something else. When we specify these clearly, we reduce ambiguity, align teams, and make the analysis meaningful for the intended decision.
Tom
Thanks, Jullia. That gives us a good foundation. From a regulatory angle, agencies have been pushing for more clarity on objectives and handling of events that complicate interpretation. How do current expectations shape the way teams phrase their estimands, and what do regulators want to see in protocols and analysis plans?
Jullia
Current guidance expects trials to define estimands clearly and to keep them consistent from protocol through the statistical analysis plan. Regulators want to see that the estimand reflects the clinical question and that strategies for handling intercurrent events are specified before any analysis. That includes the rationale for each strategy, the data needed to implement it, and sensitivity analyses to test robustness. The aim is transparency. If a trial’s primary question is about real-world use, a treatment policy strategy that includes what happens after discontinuation may be appropriate. If the question is about the effect assuming patients could remain on treatment, a hypothetical strategy could be justified. Agencies are not prescribing one answer. They want the question stated up front, the strategy justified, and the analysis aligned to that choice.
Tom
You mentioned intercurrent events, which are often the thorniest part. Walk us through the main strategies for handling them, and when each might make sense in practice. It would help to anchor the theory with a couple of concrete examples.
Jullia
Intercurrent events are central because they change what the endpoint means. There are several recognised strategies. A treatment policy strategy includes outcomes regardless of the event, so it reflects the effect in typical use. A composite strategy folds the event into the endpoint, such as counting rescue medication use as a failure. A hypothetical strategy asks what the outcome would have been if the event had not occurred, which is useful when the event is external to treatment. A while-on-treatment strategy limits the question to outcomes before discontinuation, which suits tolerability questions. A principal stratum strategy targets the subset in whom the event would not occur under either treatment; it is conceptually appealing but often complex. Consider glycaemic control with potential rescue medication. If the decision maker cares about overall benefit in practice, treatment policy works. If rescue indicates clinical failure, a composite strategy can be more aligned. The choice should match the question, not convenience.
Tom
Let’s connect this to design. Once a team agrees on the estimand, what changes in the protocol, the data flow, and the analysis plan? Where do people underestimate the practical work involved?
Jullia
Agreement on the estimand influences everything downstream. In the protocol, objectives and endpoints should echo the estimand wording. The schedule of assessments and data collection must capture intercurrent events with enough detail to implement the chosen strategy. If a hypothetical strategy is planned, you need the covariates and timing to support modelling. If you intend a composite, case report forms should record the component events cleanly. In the analysis plan, you pre-specify the estimator that targets the estimand and document assumptions. For time-to-event questions, that may be a hazard ratio or restricted mean survival time. For longitudinal data, it could be a mixed model or a model-based imputation aligned to the strategy. You also define sensitivity analyses that probe the key assumptions, such as different missing data mechanisms or alternative intercurrent event handling. The practical work is ensuring the data and processes exist to support what you promised.
Tom
Communication is often overlooked. Stakeholders from clinical operations to senior leadership need to understand the question without statistical jargon. How do you explain estimands in a way that keeps nuance but is easy to sign off?
Jullia
Start with the decision context. Say what the number will inform and who will use it. Then express the four elements in everyday language. For instance, for a weight management study: the population is adults with obesity meeting the inclusion criteria; the endpoint is change in body weight at week fifty-two. For intercurrent events, we will include outcomes regardless of treatment discontinuation or rescue medication to reflect typical use. The summary measure is the mean difference between arms. Then add a sentence on why that choice fits the decision, such as assessing effectiveness in routine practice. If a hypothetical strategy is chosen, say why the event is external, for example an external supply interruption. Keeping the language grounded in clinical intent helps teams align and avoids debates over terminology.
Tom
Let’s talk pitfalls. Where do teams most often trip up when they try to implement estimands for the first time, and what are some common misconceptions?
Jullia
A common pitfall is treating missing data and intercurrent events as the same thing. They are different. An intercurrent event is something that changes interpretation, while missing data is a data problem. Another pitfall is choosing a strategy that does not match the objective, such as using a treatment policy approach when the real question is biological efficacy under adherence. Teams also stumble when the estimator does not target the estimand, for example applying a standard model that assumes data are missing at random while the chosen strategy implies a different mechanism. Inconsistency across documents is another trap, where the protocol, analysis plan, and data collection disagree. On misconceptions, one is that estimands remove the need for sensitivity analysis. They do not. Another is that treatment policy always equals intention-to-treat. They are related, but intention-to-treat is about analysis sets, while the strategy is about how to interpret outcomes after events.
Tom
Could you bring this to life with a therapy area example. Oncology and metabolic disease often surface in these discussions. Pick one and show how different strategies change the question and the data you need.
Jullia
Take oncology with progression-free survival. Intercurrent events include starting new anticancer therapy, discontinuation, or death before documented progression. With a treatment policy strategy, you count progression or death regardless of subsequent therapies, which reflects practice where patients may switch. With a composite strategy, you might count starting new therapy as an event, arguing it indicates failure of the randomised treatment. A hypothetical strategy could be used if switching is mandated by local policy unrelated to treatment effect. You would need models and covariates to impute the progression time absent switching. The data requirements differ. Treatment policy relies on complete follow-up and accurate event dates. Composite requires clean documentation of therapy changes and reasons. Hypothetical needs timing and prognostic factors to support modelling. The chosen summary measure might be a hazard ratio for time-to-event or restricted mean survival time when hazards are not proportional. Each choice answers a slightly different clinical question.
Tom
That naturally leads to summary measures. Teams reach for the hazard ratio by default, but it is not always the best choice. How do you select a measure that suits the estimand and remains interpretable for clinicians and payers?
Jullia
Start with interpretability and decision needs. A hazard ratio is common for time to event, but it relies on assumptions and is not a direct probability difference. Restricted means survival time gives an average time gained within a horizon and can be easier to interpret, especially when hazards cross. For binary responses, risk difference or risk ratio may be more informative for absolute effect communication. For continuous outcomes, a mean difference is intuitive but consider responder definitions if a clinically meaningful threshold is key. In longitudinal settings, model-based differences at a fixed time point or area under the curve can align with how benefits are experienced. The measure must target the estimand as written, suit the data structure, and support decisions such as labelling, reimbursement, or clinical guidelines. Pre-specifying why a measure was chosen helps downstream readers trust the result.
Tom
Before we close, a brief recap would help anchor this. Today we covered what an estimand is, why handling intercurrent events explicitly matters, and how strategy and summary measure choices change the question you answer. Do you have any final thoughts for teams heading into planning?
Jullia
Two points to keep in view. An estimand is about clarity of intention. If you can read it aloud to both a clinician and a statistician and they agree on the question, you are on the right path. Then design flows from that point. The protocol, data collection, and analysis plan must support the same question. When results arrive, the estimate will be meaningful because the team defined, implemented, and analysed to a single target. Finally, keep communication simple. The best estimands use everyday language to describe population, endpoint, intercurrent events, and measure, anchored to the decision being made. That clarity builds trust with investigators, regulators, and patients who rely on trials to answer questions that matter.
With that, we’ve come to the end of today’s episode on Estimands in Clinical Trials. If you found this discussion useful, don’t forget to subscribe to QCast so you never miss an episode and share it with a colleague. And if you’d like to learn more about how Quanticate supports data-driven solutions in clinical trials, head to our website or get in touch.
Tom
Thanks for tuning in, and we’ll see you in the next episode.
About QCast
QCast by Quanticate is the podcast for biotech, pharma, and life science leaders looking to deepen their understanding of biometrics and modern drug development. Join co-hosts Tom and Jullia as they explore methodologies, case studies, regulatory shifts, and industry trends shaping the future of clinical research. Where biometric expertise meets data-driven dialogue, QCast delivers practical insights and thought leadership to inform your next breakthrough.
Subscribe to QCast on Apple Podcasts or Spotify to never miss an episode.
