What happens to missing patient-reported outcome data

Missing patient-reported outcome data is close to a universal feature of trials that collect it, not an occasional failure. A scoping review of 118 randomised controlled breast cancer trials with PRO endpoints, published between January 2019 and February 2022, found that only 9.3% had no missing PRO data at all. For the overwhelming majority of trials, the real question isn't whether data went missing. It's what happened next, and how visible that was to anyone reading the published results.

Reporting the extent of the problem is already inconsistent

Of the 118 trials reviewed, 88, or 74.6%, actually reported the extent of their missing PRO data. That leaves a quarter of trials where readers can't tell how much data was missing at all, let alone whether it was handled appropriately. For a trial endpoint that depends entirely on participants actually completing something, that's a meaningful gap in what a reader can assess about the reliability of the reported result.

This matters more for PRO data specifically than it might for some other endpoints, because missingness in patient-reported outcomes is rarely random. A participant who stops completing a quality-of-life questionnaire partway through a study is plausibly doing so because their symptoms have changed, in either direction, which is precisely the kind of missingness that can bias a result if it isn't accounted for. Not reporting the extent of missing data removes the reader's ability to even ask that question.

The statistical handling was even less transparent than the extent

Where the review looked past whether missingness was reported and into how it was actually handled, the gap widened. Only 6 trials, 5.6% of the total, reported a sensitivity analysis to examine how much their results might change under different assumptions about the missing data. Sensitivity analysis is precisely the tool that lets a reader judge whether a trial's conclusions are robust to the missing data problem, rather than just aware that it exists, and it was essentially absent from this literature.

Among trials that did document a statistical approach to the missing data, single imputation methods were used considerably more often than multiple imputation, 57.2% versus 19.0%. That's a meaningful methodological choice with real consequences: single imputation typically understates the uncertainty introduced by the missing data, producing confidence intervals that look tighter than the underlying uncertainty actually justifies. Multiple imputation, generally considered the more defensible approach for exactly this reason, was used in barely a fifth of the cases where any method was reported at all.

Awareness campaigns haven't reached practice yet

The review's authors made a specific and slightly deflating observation: international efforts to raise awareness of best practice around missing PRO data handling are not yet reflected in the actual published literature of breast cancer trials. Guidance exists. Recommended practices, including sensitivity analysis and multiple imputation, have been argued for extensively in the methodological literature. The gap the review documents is between what's recommended and what's actually being done in a substantial sample of published trials.

What this means for a study collecting PRO data now

A few practical implications follow directly:

• Report the extent of missing PRO data as a matter of course, not as an optional detail. A quarter of the reviewed trials didn't, and that alone limits what a reader can conclude about the result.
• Sensitivity analysis is underused relative to how informative it is. At 5.6% adoption in this sample, it's close to a missed opportunity across the field, not a niche technique reserved for unusually rigorous trials.
• Multiple imputation deserves to be the default assumption, not single imputation. The methodological case for multiple imputation is well established; the review's finding that single imputation still outnumbers it nearly three to one suggests a gap between what's understood to be correct and what's actually implemented.
• The systems collecting PRO data in the first place shape how much missingness there is to handle later. A platform that makes completion easy, flags a missed entry immediately rather than at the next scheduled review, and gives coordinators visibility into who's falling behind in real time reduces the scale of the missing data problem before it ever reaches the statistical analysis stage.

None of this is a criticism specific to breast cancer trials, which simply happened to be the population this particular review examined. The pattern, PRO missingness as the norm rather than the exception, and reporting and handling of that missingness lagging behind established best practice, is very unlikely to be unique to one disease area. A trial that treats missing PRO data as an expected part of the design, planned for from the outset rather than addressed after the fact, is working from a more honest starting point than one that discovers the gap only once the analysis is already underway.