The Compliance Gap: How AI Communication Can Improve Patient Outcomes

Smallpox: not with a better vaccine

Edward Jenner introduced the smallpox vaccine in 1796, and the disease was declared eradicated in 1980. The vaccine wasn't what was missing in between. In 1967 the World Health Organization intensified the effort and changed its strategy, putting surveillance and containment ahead of mass vaccination so that every case could be tracked down. Health workers in Bangladesh in 1975 went door to door and through the markets with picture cards, asking whether anyone had seen the rash, and Somalia recorded the last naturally occurring case in October 1977. The programme cost about $300 million and, in the WHO's words, saves the world well over $1 billion every year since 1980. The delivery was worth more than the medicine.

Behavior = motivation, ability and a prompt

Why reaching people worked, in the terms of BJ Fogg's behavior model: a behavior happens when motivation, ability and a prompt occur at the same moment. Nobody wanted smallpox, so motivation was never the issue. The vaccine supplied the ability. The campaign supplied the prompt, a person at the door when it mattered. The rest of the session uses the same three parts to ask what's missing in veterinary compliance.

The prompt is the best-tested part of the medicine

Human medicine has measured the gap for decades. The World Health Organization's 2003 report on long-term therapies estimated that adherence averages about 50% in developed countries, and C. Everett Koop's line, quoted in the New England Journal of Medicine's review, still sums it up: drugs don't work in patients who don't take them. The WHO report went further, arguing that improving adherence could do more for population health than improvements in specific treatments. The prompt is also the best-tested part. Pooling 16 randomized trials, a JAMA Internal Medicine meta-analysis found that text reminders about doubled patients' odds of adherence (odds ratio 2.11), lifting it from 50% to 67.8%. Across 17 trials and 25,101 patients on cardio-metabolic drugs, automated reminders raised the odds 1.89 times, and a Cochrane review of 75 studies in ten countries found reminder and recall raised immunization rates by 28%.

Heartworm: the cleanest case

Veterinary medicine has the same problem, and heartworm is its cleanest example. A monthly preventive has been available since ivermectin reached the canine market in 1987, for a disease the American Heartworm Society describes as "almost 100 percent preventable." Even so, the Society estimates that about two in three dogs get no preventive at all. In a study of 3.5 million dogs, buyers took home an average of 7.3 of the 12 monthly doses, and in another of about 202,000 dogs, roughly one in three had a full year's coverage. At least 1.1 million dogs in the United States are positive. The cost case is settled too. Comparing 235,188 dogs on year-round prevention with 2,425 treated for heartworm disease, researchers put prevention at about $182.90 a year against about $1,457 for treatment, and found a significantly higher mortality risk in treated dogs up to nine years old. Motivation and ability are both there. The prompt is missing.

Follow-through is the medicine

AAHA's first compliance study located much of the shortfall inside the practice itself: recommendations that weren't made, and follow-through that didn't happen. Cost was rarely the reason. Later numbers look similar. Of 407 practices Covetrus surveyed in 2017, 44% sent a single appointment reminder regardless of the result, and 66% made no attempt to recall patients they hadn't seen in 24 months. Gary's reframe: the diagnosis is made in the clinic, but whether the dose gets given and the recheck gets booked is settled at home, over the eleven months and twenty-nine days of the year when the practice isn't in the room. Clients, meanwhile, already get notifications from their pharmacy, their airline and their deliveries. They aren't demanding more from their veterinarian. They're expecting what everyone else already does.

Why it isn't happening: the end-of-day list

It's 5:40, the phones have gone quiet, and there's a list: the post-op check on Tuesday's Labrador, three overdue heartworm tests, the senior panel recommended in March, a callback from two o'clock. Gary puts the same clinic's inbound week next to it. Over five weekdays at one multi-doctor general practice, roughly 160 calls a day came in and roughly 55 a day went unanswered; by Friday the missed calls totaled about 280, or 35% of inbound calls, a figure published on PupPilot's verified statistics page. A practice that can't return the calls that came in won't start making the ones that didn't. And even with the time, there's a second problem: almost every practice has software that could technically send reminders, and most still don't have it set up to work from their client list.

Why the list never existed

Veterinary record software was designed for billing, not medicine. A 2020 paper in JAMIA Open from the Association for Veterinary Informatics and the CTSA One Health Alliance notes that most dog and cat records are stored as free text because most systems are "mainly designed for efficient charging and processing of transactions throughout a patient visit." Many of those systems are old. AVImark, sold since 1988, and IDEXX Cornerstone together account for nearly half the market, at 25.4% and 19.5% on Kynetec's figures in the Companion Animal Veterinary Software Guide, and ImproMed's first product shipped in 1979. In IDEXX's Finding the Time study, 85% of veterinary professionals said their software doesn't integrate well with their practice-management system, and Instinct Science's founder, the veterinarian Caleb Frankel, wrote in Today's Veterinary Business in August that "closed proprietary ecosystems have become the norm in our industry."

The consequence is that the only reminder such a system can send is one the invoice triggers, as the Cornerstone help pages describe. What the chart knows, the reminder system can't see: "Owner declined heartworm test — revisit next visit." "Recheck ears in two weeks." "Grade II murmur. Echo recommended; owner will think about it." Human medicine faced the same wall and was pushed past it. Stage 2 of Meaningful Use, the federal incentive program finalized in 2012 and in force from 2014, required providers to use the record to find patients due for preventive or follow-up reminders before they could collect incentive payments, and office-based physicians' use of electronic records rose from 42% in 2008 to 95%. Veterinary medicine had no mandate, incentive or standard. The compliance gap was never a discipline problem. It's an infrastructure problem.

What changed: AI that reads, then reaches

Two things changed at almost the same time. First, AI can now operate software built for people. OpenAI's GPT-6 Astra launch this year showed the model working inside Blender through its menus, tools and viewports, as a person would, and within days users had it turning old locomotive drawings into full 3D models. If it can operate Blender, it can operate the practice-management system through the same screens staff use, and get the data out. Second, language models read veterinary free text well. On 250 feline records, GPT-4o extracted clinical facts with 96.9% sensitivity against human review, and pulling a measured value from 4,415 practice narratives in the SAVSNET database, a language model found every instance (100% recall), against 72.6% for the older pattern-matching method.

The reading is coming from outside the systems of record. In the past year, 31% of practices adopted or changed an AI medical-record tool, 11% a cloud practice-management system and 2% an on-premise one (Instinct Science). The new path reads invoices, notes, lab results and messages; builds the list of what's due, declined or lapsed; reaches the client by call, text or email; and routes the reply to a person. Not AI that talks: AI that reads, then reaches. That reading is what PupPilot's medical intelligence does.

The medicine was never the bottleneck

Gary closes where the smallpox story began. The medicine was never the bottleneck. Reaching people was, and for the first time, reaching them can be part of the practice.

Questions and discussion

Four questions for practice teams: how many follow-ups the practice intended to make last week, and how many happened; how compliance is measured, and whether the number tracked is doses given or reminders sent; what's in the charts right now that the reminder system can't see; and what the practice would send if the list existed tomorrow.