Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data, Analytics & AI · Missing Data in Clinical Trials: Mechanisms, Prevention and Sensitivity Analysis

Missing Data in Clinical Trials: Mechanisms, Prevention and Sensitivity Analysis

The principal consideration in handling missing data is not the quantity of data missing but whether the reason for missingness is related to the outcome. This question must be addressed before any imputation method is selected.

Pubrica AcademyWritten by a biostatistics reviewer Reviewed bya clinical data manager Published10 October 2026 Last checked10 October 2026 Reading time10 minutes Sources2 peer-reviewed

Missing data in a clinical trial comprise any planned measurement that was not obtained, such as data from a participant who withdrew, a missed visit or an unanswered item. The handling of missingness in the analysis can alter a trial’s conclusion as substantially as the treatment effect itself; for this reason, both the trial design and the statistical analysis plan must address it explicitly.

Learning objectives

  • The three missing-data mechanisms (missing completely at random [MCAR], missing at random [MAR] and missing not at random [MNAR]) and the importance of distinguishing them when selecting an analysis approach.
  • The rationale for treating the minimisation of missing data through trial design and conduct as the primary safeguard, ahead of any statistical correction.
  • The purpose of sensitivity analysis for missing data, and the reasons a single primary analysis is not considered sufficient.
  • The reasons older single-imputation methods, such as last observation carried forward, are now discouraged.

Common misconceptions

  • “A small percentage of missing data cannot meaningfully affect the result.” Whether missing data bias a result depends on the mechanism of missingness, not only on its quantity; even a modest amount of data missing for reasons related to the outcome (missing not at random) can bias a treatment-effect estimate.
  • “Imputing missing values makes the analysis as good as if the data had been observed.” Every imputation method relies on assumptions about why the data are missing; imputation resolves the practical problem of incomplete data for analysis but does not remove the underlying uncertainty about whether those assumptions are correct.
  • “Last observation carried forward is a safe, conservative default.” Carrying the last observed value forward implicitly assumes that a participant’s outcome would have remained unchanged had the participant continued, which is often neither a plausible nor a conservative assumption; the method is now generally discouraged as a primary analysis method.

Classification of missing-data mechanisms

Little and colleagues, in their widely cited 2012 New England Journal of Medicine paper on the prevention and treatment of missing data in clinical trials (produced for a U.S. National Research Council panel), describe the standard framework of three missing-data mechanisms: data missing completely at random (MCAR, where missingness is unrelated to any observed or unobserved data), missing at random (MAR, where missingness is related to observed data but not, given that, to the unobserved outcome itself), and missing not at random (MNAR, where missingness is related to the unobserved outcome itself)1. The practical consequences are greatest under MNAR: if participants with a poor response to treatment are more likely to withdraw for reasons connected to that poor outcome, the observed data for the remaining participants will overstate the true treatment effect, and no increase in sample size compensates for this bias.

Minimisation of missing data through trial design and conduct

The central recommendation of Little and colleagues is that minimising missing data through trial design and conduct should be the first priority, ahead of reliance on sophisticated statistical correction after the fact, because no analytic method can fully recover the missing information and any such correction depends on unverifiable assumptions about data that were never observed1. In practice, this entails designing outcome measures and visit schedules that participants can realistically complete, planning retention strategies before the trial begins rather than responding to dropout as it occurs, and continuing to collect outcome data from participants who discontinue their assigned treatment wherever ethically and practically possible, since discontinuation of treatment is a different event from withdrawal from the study.

Limitations of a single primary analysis

Because the appropriate handling of missing data depends on an assumption about the missingness mechanism that cannot be verified from the observed data alone, Little and colleagues recommend that the statistical analysis plan of a trial specify a primary analysis under one plausible missing-data assumption, together with sensitivity analyses under alternative, less favourable assumptions, to assess the extent to which the conclusion depends on that initial assumption1. Jakobsen and colleagues, examining missing-data handling in a cross-section of published clinical trials, found considerable variation in whether and how trials reported such sensitivity analyses, and set out explicit recommendations for the reporting and analysis of missing outcome data in practice; their work reinforces that a single reported result, without any indication of its sensitivity to the missing-data assumption, does not allow readers to judge its robustness2.

Reasons for discouraging last observation carried forward

Last observation carried forward (LOCF), in which a participant’s most recent observed value is substituted for every subsequent missing time point, was historically common because of its simplicity, but Little and colleagues explicitly identify it as relying on an implausible assumption in most clinical contexts: that the participant’s outcome would have remained unchanged from the last observed point onward1. For a progressive condition, this assumption is optimistic; for a fluctuating condition, it is arbitrary. The panel recommends that the assumptions underlying any missing-data method be stated clearly and be defensible for the specific clinical context, rather than a convention being applied by default because of its simplicity.

Key points

Key points

Missing data in a clinical trial constitute primarily a design and conduct problem and secondarily a statistical one. Minimisation through realistic outcome measures, planned retention and continued follow-up of participants who discontinue treatment is more important than any correction applied afterwards, and the chosen analysis should be reported together with sensitivity analyses under less favourable missingness assumptions rather than as a single unexamined result.

Practical application

Practical application

  1. Missing-data minimisation should be planned in the protocol, not only in the statistical analysis plan. Outcome measures and visit schedules should be designed so that participants can realistically complete them, and retention strategies should be planned before enrolment begins.
  2. Follow-up of participants who discontinue treatment should continue where possible. Discontinuation of assigned treatment and withdrawal from the study should be treated as separate events, and outcome data should continue to be collected from participants who discontinue treatment wherever ethically and practically feasible.
  3. The missing-data mechanism assumed in the primary analysis should be stated and tested. Authors should state whether the primary analysis assumes that data are missing at random (or another specific mechanism) and should report at least one sensitivity analysis under a less favourable assumption.
  4. Last observation carried forward (LOCF) should not be used as a default; any single-imputation method should be justified for the specific clinical context. Where a simple imputation method is used, the reasons its underlying assumption is plausible for the outcome and condition concerned should be stated explicitly, rather than the method being applied by convention.

Frequently asked questions

Under MAR, the probability that a value is missing can be explained by other observed data (for example, older participants withdrew more often, and age was recorded); under MNAR, the probability that a value is missing depends on the unobserved value itself (for example, participants with poorer outcomes were more likely to withdraw), which is more difficult to correct for and cannot be verified directly from the observed data.

Multiple imputation is generally preferred because it propagates uncertainty about the missing values and can incorporate more plausible assumptions than the constant-continuation assumption of LOCF; however, its validity still depends on the assumed missing-data mechanism, which is why sensitivity analyses remain necessary regardless of the imputation method.

A high retention rate reduces the amount of missing data but does not, by itself, establish that the missing data are missing at random; even a small amount of outcome-related (MNAR) missingness can bias results, so the mechanism still requires consideration.

References

  1. Little RJ, D'Agostino R, Cohen ML, et al. The prevention and treatment of missing data in clinical trials. N Engl J Med. 2012;367:1355-1360. read 10 October 2026. View on New England Journal of Medicine
  2. Jakobsen JC, Gluud C, Wetterslev J, Winkel P. When and how should multiple imputation be used for handling missing data in randomised clinical trials -- a practical guide with flowcharts. BMC Med Res Methodol. 2017;17:162. read 10 October 2026. View on BMC Medical Research Methodology

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 10). Missing Data in Clinical Trials: Mechanisms, Prevention and Sensitivity Analysis. Pubrica. https://pubrica.com/academy/concepts-definitions/missing-data-in-clinical-trials/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Missing-data analysis plan stub

For use when drafting the missing-data section of a statistical analysis plan.

Expected sources of missing data: ____
Design/conduct measures to minimise missingness: ____
Primary analysis missing-data assumption (state mechanism): ____
Primary analysis method: ____
Sensitivity analysis 1 (alternative assumption): ____
Sensitivity analysis 2 (alternative assumption): ____
Follow-up plan for participants discontinuing treatment: ____

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica’s biostatistics team incorporates missing-data minimisation into trial protocols, drafts the missing-data section of the statistical analysis plan with a stated mechanism assumption and sensitivity analyses, and reviews draft manuscripts for completeness of missing-data reporting before submission.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Designing and reporting a missing-data approach that withstands statistical review

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.