Survey Fatigue Is a Study Design Problem, Not a Patient Commitment Problem

Rare disease natural history studies and patient registries need patients to stay engaged for years, and a real fraction drop out before the study ends. Dropout varies a lot by disease, design, and platform. Published longitudinal observational studies commonly report attrition of 20% to 50% in the first couple of years, higher over longer horizons. I won't quote a single number, because the honest answer is that it depends. The "60% dropout at year two" figure that floats around industry decks isn't something I can substantiate from the literature.

What's well-documented: ask patients why they stopped, and questionnaire burden, survey fatigue, is consistently among the top modifiable reasons.

This isn't a patient commitment problem. Rare disease communities are some of the most motivated research participants there are. When a registry loses them, the design is usually asking for something the patient can't sustain.

Two things both called survey fatigue

Within-survey fatigue is the drop in response quality inside a single questionnaire: straight-lining down a Likert column, parking on the neutral midpoint, terse open-ended answers. It's a function of instrument length and cognitive load, and survey methodologists have studied it for decades. Longer instruments produce worse back-half data.

Between-survey fatigue is the decline in completion over the life of a study. That's the retention problem. It's driven by cumulative burden weighed against whatever value the patient feels they're getting back.

The fix differs. You shorten instruments to reduce the first. You rethink the collection model to reduce the second.

What the EMA literature actually says

A few findings I'd trust to generalize:

Burden is felt momentarily, not as a weekly total. Shiffman, Stone & Hufford (2008), the foundational review on EMA, makes this point repeatedly. Many short daily prompts often feel less burdensome than one long weekly assessment, even at similar total time. Small interactions at the right moment fit into a day. A 30-minute obligation doesn't.

Recall degrades fast for episodic events. Ask a patient to reconstruct a week of seizures or migraines on Sunday night and you get a reconstruction, not a record. Near-real-time capture is closer to ground truth.

Backfilling is real and detectable. Stone et al. (2002) in BMJ showed paper-diary compliance was far lower than participants reported, with many entries filled in minutes before a clinic visit. Timestamped electronic capture makes this visible. Paper hides it.

Reciprocity helps retention. Patients who get something back stay engaged longer than those who only give and get nothing. Effect sizes vary, the direction is consistent.

What I won't claim: precise dropout percentages for "rare disease registries" as a category, or that any single intervention cuts dropout by a fixed amount. Be suspicious of industry numbers that don't come with a citation.

What seems to help

Shorter and more frequent beats longer and infrequent. A 60-to-90-second micro-assessment a few times a week is better tolerated than a 30-to-45-minute monthly battery, even at comparable annual time.

Event-triggered beats calendar-triggered for episodic conditions. A prompt right after a migraine feels relevant. A scheduled Tuesday questionnaire about last week feels like homework.

Mobile-first beats desktop portals. Smartphones are the default device for most patients under 60, and rare disease communities often skew younger. Desktop-only ePRO makes participation harder than it has to be.

Give something back. Even a simple sparkline of a patient's own symptoms changes how the study feels to be in. Write that policy alongside the data collection plan, not after IRB review.

Front-load less. Early dropout runs higher than steady-state, and the first month is where the habit forms or doesn't. Stage the instrument battery over the first few weeks instead of dumping it all at baseline.

This is where the "validated instrument plus" model earns its keep. Keep the patient's existing validated instrument on its cadence, the score the registry was built around, and add continuous real-world data around it: daily logs, wearables, automatic location-linked context, patient-submitted photos and video. That's how we run studies at Forma, including Friedreich's ataxia. The instrument stays intact. The real-world stream carries the day-to-day burden so the formal questionnaire doesn't have to fire as often.

If you're drafting a protocol

1. Audit total annual respondent time before you lock the protocol. Above three or four hours a year, expect attrition from burden alone, and ask whether each item is load-bearing for your analysis.
2. Pick your data collection platform early. It decides whether you can do event-triggered capture, verify entries with timestamps, and let patients respond from their phone. Those are measurement decisions, not IT procurement.
3. Write your reciprocity plan. "Download your own data anytime, quarterly community summary" counts.
4. Pre-specify a missing data model. No multi-year registry hits 100% completion. Saying up front how you'll handle missingness, and what completion rate your analysis needs, makes the burden-versus-completeness trade-off explicit.

Survey fatigue is the largest modifiable threat to long-term engagement, and it's mostly addressable through choices made at the protocol stage. The tools exist. The decision is to prioritize retention while there's still time to change the plan.