The Challenge
This health system had launched a free colonoscopy program for adults ages 40 to 55, but attendance was falling short. Leadership knew the program wasn't reaching people. They didn't know why.
Screening behavior involves knowledge, fear, cost, and guideline confusion, and those factors play out differently across sex, race, and age. A small qualitative sample couldn't represent that many segments, and a standard survey could size the problem but not explain it. Coding thousands of open-ended answers by hand would have taken a research team months.
The Solution
The health system fielded the study on CloudResearch Engage, using an AI-moderated conversational survey instead of a static form. When a participant gave a thin or ambiguous answer, the moderator followed up in the moment, which is what produced verbatims with real diagnostic value rather than one-line non-answers.
No one has ever brought it up. I've never thought about getting screened.
A Sample Built for Segmentation
Engage recruited into nine stratified cells by sex, race, and age band, oversampling women (1,542 vs. 511 men) as the study's priority population, so every finding could hold up under a demographic cut instead of collapsing into a single average.
AI-Coded Analysis at Research Scale
Open-ended answers to four separate questions (information sources, barriers, reasons for screening, and reasons against) were coded into consistent, testable themes across all 2,053 responses, with every classification checked against the transcript. Because those tags were structured from the start, the team ran chi-square, Kruskal-Wallis, and Benjamini-Hochberg-corrected tests directly on them, no separate analytics step required, and fused the results with survey fields into five named patient profiles: “Waiting for 50,” “Nobody Told Me,” a non-clinical information diet segment, “I Feel Fine,” and “The Family Story.”
Results
Knowledge wasn't the barrier. Participants averaged 81% accuracy on a six-item risk-factor knowledge scale, and 57.2% believed they met screening criteria. But only 39.6% had ever been screened, and even among the never-screened, 47.5% believed they qualified. The gap was between awareness and action, not information.
It is absurd to constantly search for problems. My body lets me know when something is wrong.
Four barriers emerged as co-dominant, each cited by 13 to 15% of participants: uncomfortable preparation, fear and anxiety, cost concerns, and guideline confusion. Doctor's influence was the top motivator, cited by 50% of would-be screeners. The same statistical testing showed minority participants carrying a significantly higher behavioral risk score than majority participants at every age band (1.35 vs. 1.15), and surfaced a clear priority audience: 491 participants who had never been screened and weren't planning to, carrying the highest risk score of any group.
Every one of these findings exists because Engage did something a standard survey tool couldn't: it turned 2,053 open-ended, conversational answers into structured, testable data without a separate qualitative coding phase.
What's Next
The health system is turning the five patient profiles into campaign messaging: a corrective age-45 fact for “Waiting for 50,” provider-facing EHR prompts for “Nobody Told Me,” and symptom-independent urgency messaging for “I Feel Fine.” The 491-person never-screened, not-planning subgroup is the free program's core outreach target, delivered through the physician channel patients already trust.
About This Organization
This organization is one of the world's largest health systems.