Missing Trial Data: Why Dropout Reasons Matter

By ClinicalStudyConnect.com Research Desk

Short answer: When people leave a clinical trial, their outcomes go unrecorded. Whether that gap distorts the result depends largely on why they left. If participants left for reasons unrelated to how they were doing, the remaining data may still give a fair comparison. If they left because the treatment was not working or caused side effects, the people who stayed are no longer a fair sample, and the published result can look better or worse than reality. The dropout reasons, reported separately for each study group, are how a reader judges which situation applies.

This guide explains what counts as missing data, what reporting standards require trials to disclose, how analysts fill the gap, and where the uncertainty remains. It ends with a missing-data worksheet you can use on any published trial or ClinicalTrials.gov results record. For related explainers on how trials are built and analysed, see our Study Design section.

Why Missing Data Weakens a Randomized Trial

Randomization is what makes a trial’s comparison fair: it balances known and unknown characteristics between groups at the start. The U.S. National Research Council’s panel on missing data, convened at the request of the FDA, found that missing outcome data reduces the benefit of randomization and can introduce bias into the comparison between treatment groups (National Research Council, 2010).

The reason is straightforward. Randomization balances the groups on day one. It does not guarantee that the people who finish each group are still balanced. If one group loses more participants, or loses a different kind of participant, the comparison at the end is no longer the comparison randomization set up.

Dropout, Discontinuation, and Exclusion Are Not the Same Thing

Trial reports use several overlapping terms. They mean different things for the data:

Stopped the intervention: the participant stopped taking the assigned treatment but may still have attended follow-up visits. Their outcome data may still exist.
Lost to follow-up: the study team could not collect outcome data, often because the participant stopped responding or could not be reached.
Withdrew: the participant, or sometimes the investigator, ended participation. Reports often do not say why.
Excluded from analysis: the participant was randomized but left out of the main analysis by the researchers’ own decision.

The CONSORT 2025 reporting guideline keeps these steps separate. Its flow-diagram item asks, for each group, how many people received the intervention as allocated, completed the intervention, completed follow-up as planned, and were included in the main analysis (CONSORT 2025 expanded checklist, item 22a). Someone who stops treatment but keeps attending visits is a different data problem from someone who disappears entirely.

What Reporting Standards Ask Trials to Disclose

CONSORT 2025 is the current reporting guideline for randomized trials. It was published in 2025 (Hopewell et al., BMJ 2025), and its checklists are maintained by the SPIRIT–CONSORT Group. Several of its checklist items apply directly to missing data:

Item 16a (sample size): any adjustment made for expected missing data or non-adherence when planning how many people to enroll.
Item 21b (analysis population): who was included in each analysis, including any exclusions due to missing data, and which group they were analysed in.
Item 21c (missing data handling): the assumption made about why data are missing, the method used to handle it, a justification for that method, and whether sensitivity analyses were run.
Item 22b (losses and exclusions): losses and exclusions after randomization for each group, with reasons. The checklist states that the label “protocol deviation” is not explicit enough and that exact reasons should be given.
Item 26 (numbers analysed): for every primary and secondary outcome, the number included in the analysis, the number with available data at that time point, and the reasons for missing data.
Item 27 (harms): the number of participants in each group who withdrew because of harms.

ClinicalTrials.gov uses a similar structure. Its results section includes a Participant Flow module, which reports how many participants started, completed, and did not complete each study period in each group, with counts for each “Reason Not Completed” (ClinicalTrials.gov Results Data Element Definitions).

How Dropout Reasons Change What a Result Means

Statisticians sort missing data by how closely the missingness is tied to the outcome itself. In plain terms:

Unrelated to the outcome: a participant moves away for a new job, or a scheduling problem causes a missed visit. Losses like these are less likely to tilt the comparison.
Explained by things the study measured: dropout is more common among, for example, older participants, and the study recorded age. Analysts can partly account for this. The technical term is “missing at random.”
Tied to the unrecorded outcome: participants leave because their symptoms are not improving or because of side effects. This is the most damaging case, because the missing outcomes probably differ from the observed ones. The technical term is “missing not at random.”

A 2025 article in Circulation summarized the difference. When data are missing at random, the effect on a trial’s conclusions may be small. When data are missing not at random, the integrity of the results can be compromised (Circulation, 2025).

Reported reasons are the reader’s main clue. Reasons such as “adverse event” or “lack of efficacy” point toward outcome-related dropout. A large gap between groups in those categories deserves attention even when the total dropout numbers look similar.

How Analysts Fill the Gap, and What Each Method Assumes

Every method for handling missing data rests on an assumption about the people who left. No method recovers the true missing values.

Complete-case analysis: only participants with outcome data are analysed. The NIH Pragmatic Trials Collaboratory notes that this approach relies on an untestable assumption that dropout did not select for a different kind of participant (Rethinking Clinical Trials).
Last observation carried forward (LOCF) and baseline observation carried forward (BOCF): a single value, either the last recorded or the starting measurement, stands in for the missing outcome. A 2012 New England Journal of Medicine special report explained that LOCF assumes outcomes do not change after dropout and can bias treatment effects when that assumption does not hold (Little et al., NEJM 2012).
Multiple imputation and model-based methods: these estimate missing values from observed data and build the resulting uncertainty into the results. They commonly rely on a missing-at-random assumption.
Sensitivity analyses: the analysis is repeated under different assumptions about the missing participants to see whether the conclusion holds.

The National Research Council panel recommended against using single-imputation methods such as LOCF and BOCF as the primary approach unless their underlying assumptions are scientifically justified. It also recommended that sensitivity analyses be part of the primary reporting of trial findings (National Research Council, 2010, Recommendations 10 and 15).

How Common Is Missing Outcome Data?

Missing outcome data is common even in major journals. A review of randomized trials published from July to December 2013 in BMJ, JAMA, The Lancet, and the New England Journal of Medicine found the following (Bell et al., BMC Medical Research Methodology, 2014):

73 of 77 eligible trials (95%) reported some missing outcome data.
The median share of participants with a missing outcome was 9%, with a range of 0% to 70%.
Complete-case analysis was the most common primary approach (45%), and simple imputation was used in 27% of trials.
35% reported a sensitivity analysis, but most of those did not change the missing-data assumption used in the primary analysis.

That review covers one six-month window in four journals and predates CONSORT 2025, so it should not be read as a current rate for all trials. It does show that missing data is the norm rather than the exception, and that robust handling was not.

The Missing-Data Worksheet

Use these steps with a published trial or a ClinicalTrials.gov results record. Each step matches a CONSORT 2025 item or a ClinicalTrials.gov results field, so you know where to look. Write down the answer to each step, or write “not reported.”

Record the starting numbers. How many people were randomized or started in each group? Look for the flow diagram (CONSORT item 22a) or the “Started” row in the Participant Flow module.
Record the ending numbers for the main outcome. How many in each group had data for the primary outcome at its planned time point? (CONSORT items 22a and 26.)
Calculate the missing share per group. Subtract step 2 from step 1 and divide by step 1. Do this separately for each group.
Compare the groups. Is the missing share noticeably higher in one group? Unequal losses are a signal to read more closely.
List the reasons for each group. Copy each reason and its count: adverse event, lack of efficacy, lost to follow-up, withdrawal by participant, physician decision, death, and so on. (CONSORT item 22b; ClinicalTrials.gov “Reason Not Completed.”)
Mark outcome-related reasons. Flag reasons that could be tied to how participants were doing, such as adverse events, lack of efficacy, or death. Note whether these cluster in one group.
Note vague labels. Mark reasons such as “other,” “protocol deviation,” or an unexplained “withdrawal.” CONSORT 2025 specifically says “protocol deviation” is not explicit enough.
Identify the analysis population. Were all randomized participants analysed in their assigned groups, or only those with complete data? (CONSORT item 21b.)
Identify the handling method and its assumption. Complete case, LOCF or BOCF, multiple imputation, or model-based? Did the authors state and justify an assumption about why data were missing? (CONSORT item 21c.)
Check for sensitivity analyses. Did the authors test a different missing-data assumption, and did the conclusion hold? (CONSORT items 21c and 28.)
Check withdrawals due to harms. How many in each group left because of harms? (CONSORT item 27.)
Write a one-line verdict. For example: “Low and balanced loss with specific reasons,” “Unequal or outcome-related loss; the result depends on how missing data were handled,” or “Missing-data reporting too thin to judge.”

Worked Example From a Public Results Record

ClinicalTrials.gov record NCT01964222 tested a web-based decision aid about clinical trials for people with cancer. The control group received usual care and was shown a cancer center website about clinical trials. Its posted Participant Flow gives the following counts:

Started: 101 in the decision-aid group and 100 in the control group.
Completed: 86 and 89.
Not completed: 15 (about 15%) and 11 (11%).
Lost to follow-up: 7 and 5.
Withdrawal by subject: 6 and 6.
Screening failure: 2 and 0.

Running the worksheet on these counts: losses are modest and fairly similar between groups (steps 3 and 4). No adverse-event or lack-of-efficacy reasons are listed (step 6). “Withdrawal by subject” gives no underlying reason (step 7). A reader would also ask how the two participants listed as screening failures after starting were handled in the analysis. Steps 8 through 10 require the full results record or publication. This example shows how to use the worksheet only. It is not an assessment of that study’s findings.

What Remains Uncertain

The core assumption cannot be fully tested. Because the missing outcomes were never recorded, no analysis can prove why they are missing. Methods and sensitivity analyses reduce this uncertainty but do not remove it.
Reason categories are often coarse. Labels such as “withdrawal by subject” or “lost to follow-up” record that someone left, not why. Someone lost to follow-up may have left because of side effects that were never documented.
Reporting guidelines are not enforcement. CONSORT sets out what should be reported. Individual papers vary in how completely they follow it, and older trials predate the 2025 update.
There is no universal safe dropout rate. The sources reviewed here do not set a single percentage below which missing data is harmless. The National Research Council recommended that each trial protocol set a minimum completeness rate based on what similar past trials achieved.

Your Next Practical Decision

Before relying on a trial result, run the worksheet on it. If losses are low, balanced between groups, and explained with specific reasons, and the handling method is stated and tested with sensitivity analyses, the missing data is less likely to change the conclusion. If losses are unequal, clustered in outcome-related reasons, or poorly explained, treat the headline result as less certain. Then look for other trials or systematic reviews that asked the same question. Our Research Guides cover other parts of a trial report worth checking alongside missing data.

Health and safety note: This article explains how to read research. It is not medical advice. Do not start, stop, or change any treatment based on your reading of a trial. Discuss questions about a specific treatment with a qualified clinician. If you are enrolled in a study, direct questions about your participation to the study team. For more on the limits of study evidence, see Safety & Limitations. Clinical Study Connect is an independent educational publication; learn more about how we work.