Every clinical trial has a quiet subplot: the people who leave before it ends. In diabetic foot ulcer research, somewhere between a fifth and two-fifths of participants withdraw or are lost to follow-up. We tend to treat them as missing data — noise to be imputed and stepped past. A Report out today in Diabetes Care– the first of its kind– makes the opposite case. The people who leave are not leaving at random. Look closely and they are, disproportionately, the ones already carrying the heaviest load — and their departure is both predictable and, we believe, preventable.
The analysis draws on our SmartBoot offloading trial: 210 adults with active diabetic foot ulcers, randomized across three offloading strategies — a standard removable boot, a removable boot plus adherence education, and a sensor-enabled “smart” boot that nudges wear time in real time. We asked a deliberately simple question. Using a handful of ordinary baseline measurements, can we tell on day one who is likely to have dropped out by week twelve?
To find out, we mapped each person’s baseline profile across three domains — the wound itself (size, complexity, age, HbA1c), mobility (walking speed, balance, exhaustion, fear of falling), and demographics and mood (age, cognition, depressive symptoms) — onto a single normalized risk scale, and drew the result as a radar plot. The graphical abstract lays the two profiles side by side: the people who completed the study, and the people who did not.
What we found
Of the 210 participants, 76 (36%) withdrew or were lost to follow-up. Two signals rose above the rest. Slower walking speed and a heavier burden of depressive symptoms each independently predicted who would leave — depression carrying the larger weight (odds ratio 1.52), slow gait close behind (1.16). Combined, the integrated baseline metrics separated those who stayed from those who left with an area under the curve of 0.81 — strong discrimination for something assembled entirely from measurements you could gather in a single clinic visit.
A second pattern is worth sitting with. Withdrawal fell steadily across the three arms — 26% with the standard removable boot, 22% with added education, and 19% with the smart boot. The device that asked the least of people, and gave them the most feedback, was also the one they were most likely to stay with.
Why it matters
This reframes attrition in two ways. For the science, it is a warning. If the participants most likely to leave a trial are also the most vulnerable — slower, frailer, more depressed — then their absence is not random missingness. It is informative. Studies that quietly lose these patients will tend to overestimate how well a treatment works and underestimate the true burden of disease, because the hardest cases have walked out the door before the final visit. We have been calling this noise. It is signal.
For care, it is an opening. A brief, low-cost screen at baseline can flag the people at highest risk of falling out of the study — and, plausibly, out of their own treatment — before it happens. And the fix may be gentler than we assume. A lower-burden, feedback-enabled device held people better than education alone. Retention may be less about chasing patients down than about designing care they can actually live with.
It is a small reminder of something we keep relearning in limb preservation: the foot is never just a foot. Whether someone heals — and whether they stay with us long enough to find out — turns as much on their gait and their mood as on the wound. The radar plot simply makes it visible. Vulnerability is multidimensional, and now, at least, it is something we can see coming.
Predicting the Path to Attrition: Multidomain Risk Assessment in Diabetic Foot Ulcer Offloading Randomized Controlled Trials. Khandan A, Dehghan Rouzi M, Armstrong DG, Najafi B. Diabetes Care. 2026. doi:10.2337/dc26-0658. From the CASIT team at UCLA and SALSA at USC, supported by the National Institute of Diabetes and Digestive and Kidney Diseases (R01 DK124789).
Leave a Reply