Singapore has spent two decades building one of the world’s most admired health systems: high outcomes, controlled costs, and an unsentimental willingness to learn from other countries’ mistakes before making its own. Right now, the mistake worth studying is unfolding in the United States, where the world’s largest experiment in paying insurers based on patient data has collided with the world’s largest health data audit.
The details are American, but the lesson is universal, and for a nation actively expanding value-based healthcare through Healthier SG, it arrives at a useful moment.
The experiment: paying for sickness, on paper
The US programme at the centre of the story is Medicare Advantage, under which private insurers manage government health coverage for over thirty million older Americans. The government pays each insurer a monthly amount per member, adjusted by a risk score computed from the member’s recorded diagnoses. Sicker members, higher payments. The design intent is sound: without adjustment, insurers would court the healthy and avoid the chronically ill.
The flaw emerged slowly, then all at once. Because payments follow recorded diagnoses rather than delivered care, an industry grew up around maximising what gets recorded. Insurers deployed teams and software to mine years of old medical charts for additional codable conditions. Recording became a revenue activity, increasingly detached from treatment.
By 2026 the correction arrived on three fronts. Government auditors published reviews showing that at three insurance plans, 81 to 91 percent of sampled high-risk diagnosis codes lacked supporting medical records. The Department of Justice settled with a major insurer for 117.7 million US dollars over chart-review programmes that added diagnoses while almost never removing incorrect ones. And the audit agency scaled from a few dozen reviewers to roughly two thousand certified coders running quarterly audit cycles, with error rates extrapolated across entire contracts and payments clawed back.
Independent congressional advisors estimate the cumulative excess payments at tens of billions of dollars annually. As policy experiments go, the tuition has been expensive.
The vendor ecosystem grows up
One under-reported consequence matters for any country building health data infrastructure: the shakeout in the supporting industry. A large ecosystem of technology and services firms handles diagnosis coding for American insurers, and the enforcement wave is forcibly maturing it.
The firms losing ground are those whose pitch was revenue: find more codes, lift more scores. The firms gaining ground are those built for verifiability: systems where every suggested diagnosis links to explicit evidence in the clinical note, where reviews run in both directions, removing unsupported codes as readily as adding missed ones, and where an auditor can reconstruct any decision years later. Buyers now evaluate risk adjustment coding companies primarily on audit-readiness, accuracy validation, and explainable AI rather than on promised payment uplift, a complete inversion of the criteria that ruled the market five years ago.
For health systems elsewhere, this is a preview of procurement standards worth adopting before problems force the issue: demand evidence-linking, bidirectional review, and audit trails from any data or coding partner on day one.
The Singapore angle
Singapore is not America. MediShield Life, MediSave, and heavy public provision make the incentive structure fundamentally different, and Healthier SG’s capitation payments to GPs are modest in comparison. But the direction of travel matters. Healthier SG explicitly moves money toward population health, with payments and targets attached to enrolled populations and their documented conditions. Chronic disease registries, risk stratification, and outcome-linked funding all depend on one input: recorded clinical data.
The American story is a controlled demonstration of what happens when recorded data acquires monetary value without proportional verification. Three specific lessons transfer cleanly.
First, wherever a recorded condition or measure triggers funding, build the audit before the incentive scales. America audited fifteen years after the incentive; the gap is where tens of billions leaked. Second, insist on encounter-linked data. The most defensible diagnosis is one made during a real clinical interaction, a principle US regulators now enforce by discounting diagnoses that surface only through retrospective file reviews. Third, watch the asymmetry. The tell in every US enforcement action was one-directional correction: programmes that only ever found errors that increased payment. Any honest data-quality effort finds errors in both directions, and a review that does not is a red flag any auditor can spot.
The advantage of going second
There is also a quieter opportunity. The verification technology being battle-tested in the American audit wave, AI that reads clinical notes and links every conclusion to evidence, human-in-the-loop validation, population-level anomaly detection, is maturing fast precisely because nine-figure settlements concentrate minds. Health systems adopting these tools now inherit the lessons without the lawsuits, and Singapore’s compact, digitised system with the National Electronic Health Record is unusually well placed to implement provenance-first data practices at national scale.
Singapore’s health system earned its reputation by studying others’ experiments with a clear eye. The American data reckoning is one of the most instructive experiments in decades: proof that value-based care lives or dies on the integrity of its data, and a detailed map of where integrity fails first. The tuition has been paid in full, in someone else’s currency. The syllabus is free.
