Chronic Disease Causes Most Caribbean Deaths. AI Underwriting Is Starting to Price That Risk.

The Bill Nobody Budgeted For

PAHO's most recent regional accounting, the "NCDs at a Glance 2025" report released on 2 July 2025, found that noncommunicable disease deaths across the Americas rose 43% since 2000, reaching 6 million of the region's 9.2 million total deaths in 2021, or 65% of everyone who died that year. Cardiovascular disease alone accounted for 2.16 million of those deaths, cancer for 1.37 million, and diabetes for more than 420,000. Nearly 40% of NCD deaths in the region occur before age 70, the years when a household is still carrying a mortgage, school fees, and a working income that a premature death takes with it.

The Caribbean sub-region specifically carries some of the highest noncommunicable disease mortality rates in the Americas. Heart disease alone accounted for 13% to 25% of deaths across the 22 territories in the 2019 study, cancer for 8% to 25%, diabetes for 4% to 21%, and cerebrovascular disease for 1% to 13%, with wide variation from country to country. That range matters, because it means an insurer pricing a health book across several Caribbean territories cannot use a single regional risk figure and expect it to hold. It is exactly the kind of variation a machine learning model, trained on enough claims and pharmacy data, is built to separate out country by country and condition by condition, rather than relying on one blended regional average.

Five countries in the Americas were on track in 2025 to hit the region's target of cutting premature NCD mortality by 25%, and three of them, Antigua and Barbuda, Barbados, and Grenada, are Caribbean states. That progress deserves equal billing with the mortality figures: this region is making real public health gains against chronic disease, even as its private insurance sector, and the AI tools increasingly sitting inside it, still lag behind the size of the problem those gains are chipping away at.

What AI Underwriting Actually Changes

Traditional health underwriting in the Caribbean still runs largely on static rules: a questionnaire, a medical report where one is required, and a rate table that treats a diagnosis code as a fixed loading regardless of how well controlled the condition is. AI-based underwriting replaces or supplements that table with a model trained on claims history, prescription refill patterns, and demographic risk factors, producing a risk score that updates as new data arrives rather than only at the point of sale. The practical difference shows up in three places: faster decisions at application, earlier flags on claims that look inconsistent or high-risk, and, where insurers use it well, more accurate pricing for a chronic condition that is genuinely well managed rather than a blanket exclusion.

None of this is unique to the Caribbean, and that is part of the point. Alternative-data credit scoring, the discipline Jamaica-based Credit Garden has applied to consumer credit in markets where thin credit files once shut people out entirely, is the closest regional analogue to what health insurers are now attempting with thin medical files: pricing risk fairly for someone whose data trail is real but incomplete, rather than declining or overcharging them by default.

The honest caveat is that most of the AI health underwriting technology currently touching Caribbean policyholders was not built for the Caribbean. Reinsurers supplying capacity to regional health and life books, and the global platforms some insurers licence rather than build, train their models mostly on data from larger, better-documented health systems. That can still beat a static rate table, but a model may still misread a Caribbean applicant's risk if the training data does not resemble the local disease pattern, the local pharmacy record-keeping, or the local rate of formal diagnosis versus undiagnosed chronic conditions.

The Clinical AI Wave Insurers Will Have to Catch

The clinical side of Caribbean health AI is moving faster than the insurance side, and the clearest evidence of that lands in San Juan, Puerto Rico this September. The Caribbean Health AI Congress runs 25-26 September 2026 at the Centro de Convenciones de Puerto Rico, expecting more than 500 attendees across 25-plus speakers and five tracks covering AI in radiology, cardiology, psychiatry, medical education, and clinical governance, organized with partners including the Society of Physician Entrepreneurs, the American Board of Artificial Intelligence in Medicine, and the University of Puerto Rico's Recinto de Ciencias Médicas.

That agenda is squarely clinical, not actuarial, and none of the public programming addresses insurance underwriting or claims directly. That gap is itself the finding worth reporting: a region building serious momentum around AI in diagnosis, triage, and chronic disease management has, so far, kept that conversation largely separate from the insurance industry that will eventually have to price around whatever these clinical tools change. The Caribbean AI Association has already mapped where AI could help across chronic disease registries and outbreak response at a public health level, work that health insurers pricing this exact risk pool would do well to read before building their own models from scratch.

The Money Isn't Going Where the Risk Is

Whatever the clinical AI wave produces, it will need an insurance product to sit underneath it, and that is where the numbers turn less encouraging. IDB Invest reported in February 2025 that Latin America and the Caribbean carry a $267 billion insurance protection gap against a $7 trillion global insurance industry, with regional insurance penetration lagging far behind global averages. Small and medium enterprises, which make up 99.5% of firms in the region and employ 60 million people, are particularly exposed: 85% of them lack adequate insurance coverage of any kind, health cover included.

Investment capital has not closed that gap. Insurtech, the category of company building exactly the AI-assisted underwriting and claims tools this article describes, attracts only about 3% of regional fintech funding, compared with 11% globally, even though regional insurtech investment did grow at a 25% compound annual rate between 2018 and 2023. Applied AI firms serving Caribbean clients, including Maestro AI Labs, form part of the small but growing base of regional technical capacity insurers could draw on, though that capacity remains thin relative to the region's uninsured health risk.

Regional AI adoption more broadly tells a related story. A StarApple AI study published in May 2026 found that only 13% of Caribbean adults aged 18 to 65 use generative AI in any form, with roughly 8.2% counted as active, regular users, though adoption among micro, small, and medium enterprises stood higher at 19%. An insurer building an AI-assisted health product into that environment is not just solving a technical modelling problem. It is trying to sell trust in a tool most of its own customer base has never used, in a market where, per the Caribbean Telecommunications Union's Caribbean AI Task Force, the region attracts only 1.12% of global AI investment against 6.6% of global GDP.

What Jamaica's NHF Already Proves

Jamaica's National Health Fund offers a useful contrast, because it shows what happens when a public system, rather than a private insurer's AI model, takes chronic disease seriously as a cost driver. In March 2026, the NHF added four chronic conditions to its subsidy list and injected $234 million in new funding, then lowered its mammogram subsidy age to 30 from 1 July, as detailed in our coverage of the NHF expansion. None of that expansion runs on AI. It runs on a government deciding, in plain policy terms, that subsidizing chronic disease management earlier is cheaper than paying for its complications later.

That is precisely the logic AI-assisted underwriting is supposed to bring to a private health insurer's book: catch the risk earlier, price it more accurately, and manage it proactively rather than waiting for an expensive claim. The NHF got there through policy and public money. Private insurers experimenting with AI are trying to get to the same place through modelling and data. Both are legitimate paths to the same goal, and a Caribbean household is generally better served asking what their actual coverage does than which method produced it. The same logic that reshaped how Caribbean life insurers price risk, covered in our earlier look at AI life underwriting, is now working its way through the health side of the same insurers' books.

Who a Bad Model Prices Out

A model trained mostly on data from wealthier, better-documented health systems can misread Caribbean risk in two directions at once. It can overprice an applicant it lacks enough comparable data to assess confidently, treating uncertainty as danger. Or it can underprice a genuine risk it cannot see in a thin file, because the region's rate of formal diagnosis for conditions like hypertension and diabetes lags the actual disease burden PAHO's data describes. Neither error is hypothetical; both are documented failure modes in health AI models built for one population and deployed on another.

A subtler risk sits inside the models themselves: treating a diagnosis code as destiny rather than checking how well controlled a condition is. A person managing type 2 diabetes with medication, diet, and regular monitoring is a materially different risk from someone with the same diagnosis and no management plan, and a crude model can flatten that distinction into a single loading. The Caribbean AI Risk Management Council works on exactly this kind of governance question: what disclosure, audit, and appeal rights a policyholder is owed when a model made the call. No CARICOM regulator has yet published binding rules on that question specific to insurance, leaving the answer, for now, up to each insurer's own internal standards.

What to Do About It Now

Most readers of this article are not going to build or buy an AI underwriting model. What the last few sections mean in practice comes down to four checks worth making before your next renewal or application.

1. Ask directly whether AI set your premium or reviewed a claim decision. Not every insurer will answer in detail, but asking establishes a record and signals that policyholders expect an answer, which is how disclosure norms eventually get built in a market with no rule requiring one yet.
2. Get any chronic condition properly documented, not just diagnosed. A model, like a human underwriter, prices what it can see. Regular monitoring records, medication adherence, and specialist follow-up notes are the evidence that separates a well-managed condition from an unmanaged one in the file an underwriter or a model actually reads.
3. Treat a sharp premium change on renewal as worth querying regardless of cause. Whether a person or a model drove the change, an insurer should be able to explain what specifically moved and why.
4. If you run a small business, check your group health cover against the region's 85% SME underinsurance figure. That statistic exists because most SMEs assume they are covered adequately until a claim proves otherwise; a short annual review with a licensed broker is cheap insurance against that specific mistake.

None of this requires treating AI in health insurance as something to resist. Used well, it is the fastest realistic path to pricing the Caribbean's chronic disease burden more accurately than a static rate table ever could, and organisations across the region, from the AI Jamaica community to regional AI associations, are actively building the literacy that makes asking good questions of these systems possible. The point is narrower: a model is a tool an insurer chose to use, and a policyholder is entitled to understand, in plain terms, what that tool decided and why.

Frequently Asked Questions