In this QCast episode, co-hosts Jullia and Tom examine missing data in clinical trials. They explain why missingness threatens unbiased estimation, how the estimand framework shapes prevention and analysis, and what robust sensitivity work and clear reporting look like from first patient in to database lock. They outline practical steps for retention, follow-up after treatment stops, and analysis choices that support inspection readiness and downstream SDTM and ADaM deliverables.
What Missing Data Is and Why It Matters
Missing data are values intended to be collected but unavailable for analysis. The concern is not only reduced sample size but biased estimates if missingness relates to outcomes. Recognising mechanisms such as missing completely at random, missing at random, and missing not at random helps teams plan credible analyses and avoid misleading conclusions.
Using Estimands to Align Design and Analysis
Define the clinical question first with an estimand. Specify how intercurrent events such as discontinuation, rescue medication or death are handled using strategies like treatment policy, hypothetical, composite or principal stratum. Align data collection with the estimand so follow-up and censoring rules support the targeted interpretation.
Prevention First in Design and Conduct
Reduce avoidable missingness with simple schedules, reasonable visit windows and flexible remote assessments. Keep participants in outcome follow-up after treatment stops when consent and safety allow. Train sites on retention steps, clarify roles for escalation, and monitor emerging gaps with targeted actions rather than broad, low-yield checks.
Choosing Analysis Methods that Fit the Question
Select methods that match the estimand and plausible missingness. For longitudinal outcomes, use mixed models for repeated measures. Apply multiple imputation with relevant auxiliary variables. For time-to-event endpoints, ensure censoring aligns with the strategy. Consider pattern-mixture or reference-based approaches to reflect alternative post-dropout behaviour. Avoid last observation carried forward and unplanned complete-case analyses.
Sensitivity Analyses and Transparent Reporting
Plan sensitivity analyses that test credible departures from primary assumptions, for example delta-adjusted or pattern-mixture scenarios and tipping-point assessments. Summarise missingness patterns, reasons by arm, and links to intercurrent events. Present primary and sensitivity results side by side with plain language rationale to support regulatory review.
Governance that Supports Delivery
Pre-specify handling of intercurrent events, visit windows and partial data in the Statistical Analysis Plan. Maintain traceable change control for rules and programs, validated systems, role-based access and active audit trails. Ensure alignment between clinical operations, data management and statistics so prevention, analysis and reporting remain consistent and inspection ready.
Practical Tips and Common Pitfalls
Design to the estimand and collect outcomes after treatment stops where appropriate. Act early on gaps with focused monitoring. Use principled methods rather than convenience choices. Do not conflate treatment discontinuation with study withdrawal. Limit sensitivity work to analyses that challenge the main assumption and explain the clinical meaning of each.
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
Jullia, could you set the scene for us? When people talk about missing data in clinical trials, what do they actually mean, and why does it matter so much for decision making?
Jullia
Thanks, Tom. So, missing data are values that were intended to be collected but are not available for analysis. They arise when a participant skips a visit, discontinues treatment, withdraws consent, or when there is an administrative lapse. The risk is not only smaller sample size. The bigger issue is bias. If data are missing in a way linked to outcomes, the treatment effect can be distorted. Missingness sits in three broad categories: missing completely at random, missing at random, and missing not at random. In practice, we rarely know which mechanism we face. That is why trials need prevention strategies in conduct and robust, pre-specified handling in the Statistical Analysis Plan, so estimates remain credible and regulators can trust the conclusions.
Tom
Let’s bring in the regulatory perspective. How do current expectations frame missing data, and where do estimands and intercurrent events fit?
Jullia
Current regulatory guidance asks sponsors to define what they want to estimate before they think about methods. That is the estimand framework. You link the clinical question to how you handle intercurrent events, such as discontinuation, rescue medication, or death. Different strategies exist. A treatment policy estimand includes post-intercurrent data as part of real-world use. A hypothetical estimand asks what would have happened if the event had not occurred. Composite strategies fold events into the endpoint, and principal stratum targets subgroups defined by the event. Once the estimand is clear, you plan data collection and analysis to align with it. This reduces avoidable missingness, avoids ad hoc choices like last observation carried forward, and supports sensitivity analyses that probe departures from the assumed missingness mechanism.
Tom
That takes us to prevention. What are the practical steps teams can take during design and conduct to minimise missing data in the first place?
Jullia
Start with protocol simplicity. Every extra visit or assessment increases drop-off risk. Use visit windows, remote or hybrid assessments where appropriate, and limit non-essential procedures. Build a retention plan with clear contact schedules, reminders, and participant-friendly logistics. Train sites on the importance of collecting outcomes even after treatment stops, if consent allows and the estimand requires it. Define rescue pathways so participants stay under follow-up. Use central and risk-based monitoring to flag sites with missing assessments early, then apply targeted support. Clarify roles for who escalates when data go off schedule. Document every attempt to obtain critical endpoints, and distinguish missed treatment from missed data. Good prevention costs less than complex imputation later and gives a stronger evidential story.
Tom
When prevention is not enough and gaps remain, what are the main analysis approaches, and how should teams choose between them?
Jullia
Choice follows the estimand and the likely missingness mechanism. For continuous outcomes, mixed models for repeated measures handle incomplete longitudinal data under the missing at random assumption and use all available visits. Multiple imputation creates several plausible datasets, analyses each, then combines results to reflect uncertainty. It is versatile and can incorporate auxiliary variables tied to missingness. For time-to-event outcomes, censoring rules must align with the estimand, and sensitivity analyses should explore informative dropout. Methods like pattern-mixture models allow explicit assumptions about different post-dropout behaviours, including reference-based imputation in comparative trials. What to avoid are simplistic methods that bias results, such as complete-case analysis when data are not random, or last observation carried forward which can misstate trajectories. Pre-specification and justification are essential, not just the software command.
Tom
You mentioned sensitivity analyses. What does good sensitivity analysis look like for missing data, and how do you communicate it?
Jullia
A good plan tests how conclusions change under credible departures from the primary assumption. Begin with the primary analysis aligned to the estimand, then add sensitivity checks that tilt assumptions towards less favourable scenarios. For example, use delta-adjusted imputation to shift post-dropout values, or implement pattern-mixture models that assume different outcomes after discontinuation. Tipping-point analysis can map the threshold of assumptions where the decision would change. Present results side by side with the primary analysis, explain the clinical meaning of each assumption, and avoid an array of minor variants that do not challenge the core premise. The goal is decision robustness: if reasonable alternative assumptions point to the same conclusion, confidence grows; if not, be transparent about uncertainty.
Tom
Many teams struggle with patient-reported outcomes and complex endpoints. Any specific advice for these cases?
Jullia
Plan the collection burden and backup routes early. For patient-reported outcomes, provide flexible completion options and reminders, and capture reasons for missingness. Ensure translations and literacy levels are appropriate to reduce skipped items. For composite or responder endpoints, clarify how partial data map to the final outcome and whether intercurrent events define non-response. If the endpoint aggregates multiple components, prioritise those that drive clinical meaning and ensure they are resilient to missed visits. Align the imputation model with the scale of measurement. Do not impute total scores if the component structure makes that inappropriate. Most importantly, keep the estimand in view: if the strategy treats treatment discontinuation as non-response, missing after discontinuation may be part of the endpoint rather than a gap to impute.
Tom
How should teams reflect missing data planning in the Statistical Analysis Plan and downstream reporting to be inspection-ready?
Jullia
The plan should tie the estimand, intercurrent event strategy, and data handling together. Define analysis populations, visit windows, handling of partial dates, and rules for out-of-window data. Describe the primary model and covariates, the imputation strategy with variable lists and number of imputations, and sensitivity analyses with their rationale. Document how protocol deviations and retention efforts will be summarised. In the clinical study report, present missingness patterns, reasons for missingness, and by-arm summaries of exposure and discontinuation. Provide clear tables that link intercurrent events to outcome availability. Include a narrative on how results held up across sensitivity analyses. This transparency helps assessors see the link from question to data to conclusion without guessing your assumptions.
Tom
Operationally, who owns what? How should statisticians, data managers, and clinical teams work together to keep missing data under control?
Jullia
It is a joint effort. Statisticians lead on the estimand, analysis choices, and sensitivity framework. Data managers design the data flow, edit checks, and dashboards that flag emerging gaps. Clinical teams and monitors drive site training, participant retention, and escalation when visits slip. Pharmacovigilance teams should align safety follow-up with the estimand to avoid avoidable gaps in key outcomes. Everyone needs shared definitions for withdrawal, discontinuation, and lost to follow-up. A short, living playbook helps: who calls the participant, who rebooks visits, who approves out-of-window data, and when to document that further follow-up is futile. When roles are clear, fewer data go missing, and analysis rests on firmer ground.
Tom
Let’s do a concise takeaways moment. What are the top practical tips and common pitfalls teams should remember?
Jullia
First, write the estimand clearly and design follow-up to match it. Collect outcomes after treatment stops when clinically appropriate and consented. Second, prevent more than you correct: simplify schedules, use remote options, and act on early missingness signals. Third, pre-specify primary and sensitivity analyses with plain-language justifications. Use methods like mixed models or multiple imputation rather than convenience choices. Regarding pitfalls to avoid: do not rely on last observation carried forward. Do not conflate treatment discontinuation with study withdrawal. Keep participants in follow-up. Do not bury the reader with unhelpful sensitivity variants. Focus on analyses that challenge your main assumption.
Tom
Thanks, Jullia. Before we close, could you recap the core message and how sponsors can apply it on their next study?
Jullia
Of course. So, the core message is simple. Define the question first using the estimand framework and let that drive data collection and analysis. Invest early in prevention through practical schedules, flexible follow-up, and proactive monitoring. When data are still missing, use principled methods that reflect plausible assumptions and test robustness with targeted sensitivity analyses. Document the plan, show the patterns, and explain assumptions in plain language. If teams do that consistently, treatment effects are more reliable, reviews move faster, and decisions are better informed.
Jullia
With that, we’ve come to the end of today’s episode on missing data 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.
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.