AI in Healthcare

The public system built its digital plumbing first

The Ministry of Health’s most consequential AI-adjacent work has been unglamorous: getting hospitals onto shared digital infrastructure at all. The Total Hospital Information System (THIS) is now live across 16 government hospitals, and a cloud-based clinic management system covers more than 2,400 primary healthcare facilities nationwide, alongside AI-driven diagnostics and real-time monitoring dashboards. This sits inside a RM46.52 billion health allocation in Budget 2026, with digital transformation framed as a central pillar. None of this is spectacular. All of it is prerequisite — you cannot run diagnostic AI, risk models, or DRG costing on paper records, and for most of the roadmap period, paper records were the reality in much of the system.

Diagnostic AI has a real Malaysian evidence base

The “A-eye” concept anticipated exactly the area where AI has since proven itself: image analysis. Malaysia’s own health technology assessment body, MaHTAS, evaluated AI-assisted chest X-ray reading and found detection-rate improvements of around 15.5% among radiology trainees, with AI-assisted screening sensitivity for lung nodules reported in the 56–96% range against 23–76% for unassisted readers. The MOH’s comfort with the technology traces back to the pandemic, when AI-enhanced CT analysis was deployed to screen for COVID-19 — a crisis-era shortcut that quietly normalised algorithmic reads in government hospitals. The National Institutes of Health continues to run pilot studies on AI in screening, and the ministry has been explicit that early detection of cancer and tuberculosis is where it wants the technology aimed.

Private hospitals are buying, not piloting

The private sector has moved past evaluation into procurement. In June 2026, Pantai Hospital Kuala Lumpur launched an AI-powered adaptive radiotherapy system that re-plans cancer treatment daily as a patient’s anatomy changes — officiated by the Deputy Health Minister, and paired with a public–private arrangement extending free radiotherapy to public-sector patients. It is one visible example of a broader pattern across the IHH, KPJ and Sunway networks: AI arriving embedded inside expensive imported equipment and software, rather than as standalone “AI projects”. This matters for the cost story below.

States are experimenting at the edges

Selangor began deploying 150 AI-powered fall-detection and safety-monitoring devices for senior citizens in care centres and private homes from June 2026, under its Care Economy Policy — infrared sensors plus AI, sending real-time alerts to caregivers. Small numbers, but a preview of where an ageing Malaysia will push this technology: out of the hospital and into the home.

Judged as a project plan, the healthcare chapter of the roadmap fared poorly — the flagship use cases did not ship as described, and no public accounting of them was ever issued. Judged as a forecast of where AI would matter, it holds up surprisingly well. Imaging diagnostics, proactive chronic-disease management, and logistics optimisation were the right calls; they simply arrived via MaHTAS assessments, MOH procurement and vendor roadmaps rather than through MOSTI-coordinated national projects. The honest scorecard: right themes, wrong delivery model. That verdict recurs across the whole review, but healthcare is its clearest illustration, because the counterfactual deployments are so easy to point at.

Where the roadmap’s vision remains genuinely unmet is integration. Malaysia has diagnostic AI in some hospitals, digital records in others, and risk-stratification tools in a few — but no connected pathway where an algorithm flags a diabetic patient in a klinik kesihatan and follows them through referral, treatment and monitoring. The “proactive healthcare” end-state is still a slide, not a system.

Medical records are the most sensitive data most Malaysians will ever generate, and they are now flowing through cloud clinic systems, insurer claims databases and diagnostic algorithms simultaneously. Three governance layers apply, none of them written specifically for clinical AI:

• The AIGE principles. The national AI governance and ethics guidelines are voluntary. In a hospital context, “voluntary” means the vendor’s internal standards are effectively the standard.
• The amended PDPA. The 2024 amendments — breach notification, data protection officers, processor liability — give health data real statutory teeth for the first time, but the Act still regulates data handling, not algorithmic decisions. A model that mis-triages a patient hasn’t breached anyone’s data.
• MaHTAS as accidental gatekeeper. In the absence of an AI-specific regulator, Malaysia’s health technology assessment unit has become the de facto checkpoint for clinical AI — assessing safety, efficacy and cost-effectiveness before public adoption. It is a sensible arrangement that nobody actually designed.

The open questions are the ones every jurisdiction is wrestling with: who is accountable when an AI-assisted diagnosis is wrong — the clinician, the hospital, or the vendor? What does informed consent mean when a patient cannot opt out of the software their radiologist uses? And will the anticipated AI legislation treat clinical AI as high-risk, as the EU model does? Until it answers, Malaysian clinical AI operates on professional judgment plus imported certification — workable, but thin.

AI cuts healthcare costs in theory and raises them in practice — at least at first. The theory is sound: earlier detection means cheaper treatment, and MOH’s own preliminary findings on AI-assisted radiology point that way. But the near-term reality is that AI arrives inside premium equipment — adaptive radiotherapy suites, AI-enhanced MRI, algorithm-guided pathology — and premium equipment is billed at premium rates, overwhelmingly in private hospitals, overwhelmingly paid through insurance.

The numbers are stark. Malaysian medical inflation ran at roughly 15% in 2024 and 2025 — well above the global average — and is projected to reach about 16% in 2026. Bank Negara has explicitly named advancements in medical technology as a driver, alongside the rising burden of chronic disease. Insurers’ claims costs grew faster than premiums for years, and the correction, when it came, was blunt: repricing letters that would have meant 40–70% premium jumps for some policyholders before BNM intervened.

The policy response is now reshaping the entire cover landscape:

• Interim repricing caps run out at end-2026. BNM’s December 2024 measures force insurers to stagger increases over at least three years, keeping annual adjustments below 10% for most policyholders — but only until the end of this year. What happens to premiums in 2027 depends on whether the structural reforms bite in time.
• DRG payment is coming. Private hospitals are transitioning from fee-for-service billing to diagnosis-related group pricing — fixed payments per diagnosis rather than per line item — with Act 586 amendments to regulate private hospital charges. This is the single biggest determinant of what AI-enabled treatment will actually cost insured patients.
• A government base plan launches this half. The base MHIT product — standardised medical insurance covering Malaysians to age 85, with indicative premiums from around RM80–120 a month for those in their early thirties — pilots in the second half of 2026 ahead of a 2027 rollout. Policyholders facing repricing will be able to switch into it without fresh medical underwriting.

For an individual Malaysian, the practical situation is this: medicine is getting better and dearer at the same time, your existing policy was probably priced for the medicine of five or ten years ago, and the next eighteen months bring more moving parts — expiring caps, DRG pricing, a new base plan, mandatory alternative products — than the medical insurance market has seen in decades. A plan that looked comprehensive in 2020 may quietly exclude, sub-limit or co-pay its way around exactly the newer diagnostics and treatments that AI-era care runs on.