Google and Taiwan turn 20 years of health data into Gemini-assisted diabetes screening

Google and Taiwan are moving one of the clearest public examples of national-scale preventive AI from pilot work into an operational healthcare workflow. In a March 4 post, Google said it has been working with Taiwan's National Health Insurance Administration (NHIA), under the Ministry of Health and Welfare (MOHW), to combine Gemini, Google Cloud infrastructure and population-scale records to identify diabetes risk earlier. The stated goal is not to replace clinicians, but to use historical data more effectively so doctors can focus their time on earlier intervention and the patients who need the most attention.

The first production system in that effort is AI-on-DM (Artificial Intelligence on Diabetes Mellitus), an NHIA model for diabetes-risk assessment. Google said Taiwan's health system has more than 20 years of securely aggregated data, but manual evaluation still consumes scarce clinical time. Under the previous workflow, assessing one patient's risk took about 20 minutes. Screening 20,000 people would have required 40 professionals working for three weeks without stopping. Google said the AI-on-DM model, paired with Google Cloud concurrency, reduces processing time to 25 seconds per case. That translates to a 14,400x efficiency gain and allows 20,000 evaluations to finish in under 90 minutes.

The speed increase matters because Google is positioning the system as clinician support, not automated medical decision-making. According to the company, the model flags data patterns for clinical review so doctors can prioritize intervention before complications escalate. Google also said Taiwan's NHIA will launch a Gemini-powered health assistant this month inside a government app used by 10 million people. The assistant is intended to provide personalized and secure suggestions grounded in clinical guidelines, giving patients a direct way to receive everyday care guidance inside an existing public service.

Why the rollout stands out

Google said the project builds on earlier healthcare AI deployments across Taiwan rather than starting from scratch. It pointed to China Medical University Hospital's use of MedLM for cancer care, Chang Gung Memorial Hospital's AI-enhanced ultrasound diagnostics and Taipei Medical University-affiliated hospital workflows designed to address clinician shortages. Google also said the NHIA has already used MedGemma to process more than 30,000 pathology reports.

The announcement also includes a community-health layer. Google.org awarded a $1 million USD grant to the Digital Humanitarian Association, which plans to support 300 community centers, 240,000 health check-ins and training for 200 local caregivers assisting patients with diabetes. Google said the NHIA intends to extend the same framework to hypertension and hyperlipidemia. If the deployment performs as described, Taiwan's program will offer a closely watched model for how a national health system can move from isolated AI demos to routine preventive care at scale.