PROM Nonresponse Varied Sharply Across Radiation Oncology Clinics
In routine radiation oncology care, PROMIS Global-10 nonresponse was not a marginal data-cleaning problem. It was a structural feature of the program. Among 12,214 adults treated across five Mass General Brigham radiation oncology clinics over one year, patient-level response reached only 35.4%, and appointment-level response was 31.7%. Clinic-to-clinic variation was wide enough to alter the meaning of any aggregate quality metric, with patient-level response ranging from 12.8% to 66.2% across sites. A spread of that size is hard to ignore because it means the same questionnaire can function very differently depending on where and how it is deployed.
The study framed nonresponse as a question of representativeness rather than simple participation. In that setting, a low completion rate does not merely reduce sample size; it can distort the clinical picture if the missing respondents are systematically different from those who answer. Radiation oncology is particularly vulnerable to that problem because care is often episodic, multi-site, and dependent on portal-based workflows that require patients to engage at the right moment in the treatment course.
Patient Factors Mattered Early, Then Faded After the Clinic Was Added
The first model linked several patient characteristics to higher odds of nonresponse. Male sex, lower education, not working, and recent surgery were associated with lower completion, while a longer interval since diagnosis was associated with better response. Those findings fit a familiar pattern in patient-reported outcomes research: people facing higher treatment burden, more disrupted daily routines, or less familiarity with digital health tools are less likely to complete portal-based questionnaires.
The signal changed once provider- and clinic-level factors entered the model. Patient sex, education, and employment status no longer remained significant. Recent surgery did, with an adjusted odds ratio of 1.97, and longer time since diagnosis continued to predict lower nonresponse, with an adjusted odds ratio of 0.46 for patients more than 12 months from diagnosis. The shift is clinically important because it suggests the apparent patient-level gradient is partly explained by the care environment surrounding the patient rather than by patient traits alone.
One practical consequence follows directly from that pattern: if a clinic’s workflow suppresses completion, the dashboard may end up measuring operations more than patient experience.
Provider Patterns Helped, but Clinic Workflow Carried the Heavier Weight
Provider history mattered before the clinic was introduced into the model. A provider’s historical collection rate was protective, implying that some clinicians or care teams developed more reliable habits around PROM capture. Once clinic-level effects were added, that advantage was attenuated. The remaining signal moved decisively toward implementation structure: a later program launch was associated with markedly lower odds of nonresponse, with an adjusted odds ratio of 0.29, and a higher historical collection rate at the clinic level was also associated with lower nonresponse, with an adjusted odds ratio of 0.79.
Those findings point to a straightforward but often underappreciated mechanism. PROM collection depends on timing, staffing, patient messaging, portal usability, and the degree to which questionnaire completion is embedded into routine clinic operations. If one clinic adopts the program later, it may do so after early operational friction has been reduced, or it may benefit from more mature workflows. Either way, the program’s maturity appears to matter more than the static label of “academic” versus “community” setting, which did not show an association in the adjusted analysis.
The result is especially relevant for institutions that use PROs in quality reporting. A completion rate that looks respectable on paper can still conceal a biased respondent pool if the collection process is inconsistent across sites, providers, or appointment types.
Recent Surgery and Time Since Diagnosis Point to Real-World Burden
Two patient-level findings persisted after adjustment and deserve separate attention. Recent surgery nearly doubled the odds of nonresponse. In everyday practice, that likely reflects a patient population already coping with recovery, postoperative fatigue, symptom burden, and more fragmented care pathways. Portal completion is easy to delay when a patient is balancing pain control, wound care, transport, and multiple follow-up visits.
Longer time since diagnosis behaved in the opposite direction, with lower odds of nonresponse among patients more than 12 months from diagnosis. Patients earlier in the disease course may be navigating a more chaotic period of testing, consultation, and treatment initiation, which leaves less bandwidth for routine survey completion. As the care trajectory stabilizes, response may become more feasible. The biology is less important here than the lived logistics of cancer care: distress, scheduling pressure, and shifting contact points can all reduce the likelihood that a questionnaire is completed at the intended moment.
Response Rate Is Not the Same as Representativeness
The central message from this cohort is hard to miss. PROM nonresponse in radiation oncology was multilevel, and the strongest signals emerged from clinic implementation rather than patient characteristics alone. That has direct consequences for patient-reported outcome measures used in routine care and in performance assessment. If a program rewards sheer volume of completed surveys without checking who is missing, it risks overestimating the quality of the underlying data.
In practical terms, a clinic with a lower response rate is not automatically producing worse care, but it may be producing less representative data. A clinic with a higher response rate is not automatically solving the problem either if completion is concentrated among a narrow subset of patients. The study’s framing is a useful correction to any tendency to treat response rate as a simple proxy for data quality.
For radiation oncology programs, the more useful metric may be whether the collection process captures patients across the full range of diagnosis timing, postoperative status, and visit patterns. Representative collection, not raw volume, is the standard that matters when PRO data are used to shape care delivery or compare performance across sites.