We Might Be Widening the Gap. Here Is How We Will Know
On the Structural Influence Model, intervention-generated inequalities, and why GistMD is committing to stratified reporting.
By Dan Rolls, Founder and CEO, GistMD
This post came out of a paper I found unusually interesting, one that speaks directly to GistMD’s vision. I am not here to judge the research, which is neither my profession nor my training, and I would welcome any response that deepens the discussion and advances the move to put the patient at the center.
Every morning I read one paper. Most of them help me sell. Today's did the opposite, and it is the more useful kind.
The paper is a chapter by K. Viswanath, Shoba Ramanadhan and Emily Kontos, published in 2007, which sets out what the field calls the Structural Influence Model. Its argument is uncomfortable for anyone in my business. It says that a person's structural position, their income, their education, their language, and where they live shape both the information environment they encounter and the resources they have to make use of it. Communication inequality is not a side effect of poor content. It is produced upstream of the content, and it then carries forward into health outcomes.
Read that carefully, and you will see the problem it creates for patient education. We spend our days improving the message. The model says the message is only one term in the equation.
The sharper version of the argument
A more pointed paper in the same tradition names my category directly. According to PubMed, Tiffany Veinot, Hannah Mitchell, and Jessica Ancker published "Good intentions are not enough: how informatics interventions can worsen inequality" in the Journal of the American Medical Informatics Association in 2018 (DOI). They argue that health informatics interventions risk producing what they call intervention-generated inequalities by disproportionately benefiting more advantaged people.
The paper is useful because it does not stop at the warning. It separates the ways a digital tool can widen a gap. In their words, interventions have produced inequality because they were "more accessible to, heavily used by, adhered to, or effective for those from socioeconomically advantaged groups."
Four distinct failure points. Access. Use. Adherence. Effectiveness. A tool can pass the first three and still fail the fourth. Most vendors in this space, including us until now, have not reported on any of them separately.
Where GistMD actually stands
Our context, stated plainly. GistMD delivers clinician-approved personalized video to patients preparing for and recovering from care. We are deployed across 11 hospitals in Israel; Mayo Clinic Ventures backs us, and we have a pilot at Mayo Clinic Rochester. We have a signed pilot with GI Alliance that is in onboarding, not yet deployed across their network. And we have Rhode Island, which I will come back to, because it is where this argument stops being theoretical.
Our published results are real, and they are averages: a 37 percent reduction in inadequate bowel preparation, a 68 percent reduction in cancellations, a 55 percent reduction in pre-anesthesia meetings.
Averages are exactly what the Viswanath model tells you not to trust on their own. A mean can improve while a gap widens. If our video lifts prep quality by 37 percent overall because it works very well for English-speaking patients with stable housing and somewhat for everyone else, we have improved the average and made the disparity worse. I want to be precise: we do not currently have published evidence that GistMD narrows between-group gaps. We have not measured it that way. That is a gap in our evidence, not a finding in our favor.
Where our design does engage the argument
Against Veinot's four failure points, we deliberately built our architecture to address the first two.
On access, there is no app to download and no patient portal login. The video arrives by SMS, QR code, or WhatsApp and plays on a basic smartphone. The app store and the portal password are two of the most reliable filters for sorting out less advantaged patients, and we removed both.
On use, the content is animated video in more than 20 languages, culturally adapted and literacy-adapted, rather than a PDF written at a reading level most patients cannot comfortably handle. Language and reading level are the next two filters, and we built against those as well.
On adherence and on effectiveness, I will not claim anything. Design choices aimed at a known failure mode are not evidence that we avoided it. Whether comprehension converts into completed preparation at the same rate across language and socioeconomic groups is an empirical question, and we have not yet answered it.
Why Rhode Island
We have reached our first product-market fit in the US, and it is in Rhode Island. The preliminary data is impressive.
During our visit, a local marketing expert told me something I didn't know. Rhode Island has long served as a testing ground for companies weighing a national rollout, and he gave three reasons: a diverse population, established communities, and a compact geography that lets you test and learn quickly.
I want to stay with the first of those, because it is more than a marketing fact. A diverse population is the condition under which a widening gap is detectable. In a homogeneous site, an intervention that quietly works better for advantaged patients looks like a clean success, because nothing contrasts it. Rhode Island gives us the contrast. What makes it a good commercial test bed is precisely what makes it a good equity test bed, and that coincidence is worth using rather than ignoring.
So let me be exact about what "impressive" means here and what it does not. It means the early signal points in the direction we expected. It does not mean I have a result, and I am not going to quote a number from an in-flight pilot inside a post arguing that a number without a distribution behind it is worth very little. The first thing we publish out of Rhode Island will be the breakdown, not the headline.
Rhode Island is where we learn from real patients and real clinical teams, refine the product, and build the evidence for expansion across the US. The best is yet to come.
What we are committing to
Starting with Rhode Island and carrying into every pilot after it, we will report viewing and completion rates stratified by preferred language, delivery channel, and region, with the between-group gap as a named secondary outcome rather than a subgroup footnote. Where we have a research partner, we will pre-register it that way. We will publish the result when the gap does not close, which is the only version of this commitment that means anything.
This is not only the honest position. It is the commercially correct one. The CMS ACCESS Model launches in July 2026, and health systems carrying readmission penalties and equity reporting obligations do not need another vendor with a flattering average. They need one that can show them the distribution. Measurement that can embarrass us is a feature, not a liability.
Viswanath's model is a theoretical proposition rather than causal evidence, and it maps a problem rather than predicting effect size. Veinot's paper is a perspective piece, not a trial. Neither one tells me whether GistMD works. What they do is set the standard of proof I should be held to, and it is higher than the one my industry currently meets.
So here is the question I would put to anyone building in this space: if your next pilot showed that your average improved and your gap widened, would your reporting even detect it?
References
Viswanath K, Ramanadhan S, Kontos EZ. Mass Media. In: Galea S, ed. Macrosocial Determinants of Population Health. New York: Springer; 2007:275-294. https://doi.org/10.1007/978-0-387-70812-6_13
Veinot TC, Mitchell H, Ancker JS. Good intentions are not enough: how informatics interventions can worsen inequality. Journal of the American Medical Informatics Association. 2018;25(8):1080-1088. Retrieved from PubMed. https://doi.org/10.1093/jamia/ocy052
Bekalu MA, Eggermont S. The role of communication inequality in mediating the impacts of socioecological and socioeconomic disparities on HIV/AIDS knowledge and risk perception. International Journal for Equity in Health. 2014;13:16. https://doi.org/10.1186/1475-9276-13-16
Comments