SriKandi.INR

The problem

Warfarin is a blood-thinning medication used by patients with conditions that make them more likely to develop blood clots. It is also one of the most difficult to dose correctly. It requires precise, individualised adjustments based on regular blood test results, specifically a measurement called the International Normalised Ratio (INR). Getting the dose wrong can result in either inadequate protection from blood clots or an increased risk of bleeding.

A district hospital in Malaysia runs a dedicated warfarin clinic where doctors adjust patients' doses based on their latest INR result. Prescribing the correct dose involves following a detailed local guideline with numerical tables. Baseline data collected by the hospital found that up to 18% of warfarin prescriptions contained dosing errors. When an error is identified, typically caught by the pharmacist who double-checks every prescription, the patient has to return to the clinic for the prescription to be rewritten and then go back to the pharmacy. The hospital still prescribes manually, using pen and paper, and this rework adds time for both patients and staff and contributes to clinic congestion.

The team wanted to find a way to reduce these errors and make the prescribing process more efficient.

The solution

The hospital developed SriKandi.INR, a Clinical Decision Support System (CDSS) — a tool that uses Artificial Intelligence (AI) to help doctors calculate the correct warfarin dose. The name SriKandi comes from a Malay word meaning female warrior, chosen to honour the women who make up most of the team. The ".INR" refers to the blood test the tool is built around, and the naming convention is designed to allow future expansion into other clinical areas, such as diabetes or hypertension management.

How it works
SriKandi.INR is designed as a chatbot, accessible from a phone or from the clinic computer. A doctor enters a small amount of clinical information: the patient's INR value, their current warfarin dose, and any other medications the patient is taking that might interact with warfarin. The tool analyses this information against the hospital's local prescribing guideline and recommends a dose.

The tool uses a Retrieval-Augmented Generation (RAG) approach — meaning it draws its recommendations from the hospital's own local warfarin guideline rather than general medical knowledge. This ensures that the output reflects the specific prescribing protocol used at the hospital.

The chatbot does not require any patient-identifying information to function. Only clinical data is entered, meaning there is no way to identify which patient the recommendation relates to from the tool's data alone.

How it fits into the clinical workflow
SriKandi.INR is a decision support tool, not a replacement for clinical judgement. The AI provides a recommendation, but the doctor makes the final prescribing decision and writes the prescription. The pharmacist continues to double-check every prescription as before. At no point is the AI's output followed automatically.

The tool provides transparency in how it reaches its recommendation. It details its reasoning step by step — showing the calculated weekly dose and how it arrives at the daily dose — so the doctor can follow the logic before prescribing. All prescriptions are documented clearly, recording both the AI-recommended dose and the doctor's final decision, creating a clear audit trail.

Patient awareness
Patients attending the warfarin clinic are informed about the use of AI through information displayed on a screen in the waiting area. The display explains that AI is being used to help calculate medication doses, that doctors retain full decision-making authority, and that patients can raise questions with their doctor or contact the project lead directly. Patients can opt out at any time and have their dose calculated manually.

The impact

The project is structured as a six-month quality improvement programme, divided into two phases. Phase one collected baseline data, which established that up to 18% of warfarin prescriptions contained dosing errors requiring rework. Phase two — the integration of SriKandi.INR into the clinic workflow — was in its early weeks at the time of reporting.

1. Targets set for measurable improvement
The programme has set two targets: a 30% reduction in the rework process caused by prescription errors, and an increase in the rate of correct first-time prescribing. Outcome data is expected at the end of the six-month programme.

2. Validation study planned
A formal validation is planned for the third quarter of 2026. A panel of five clinical experts will review 50 scenarios covering routine cases, borderline results, and emergency situations involving bleeding complications. The same scenarios will be run through SriKandi.INR, and the AI's recommendations will be compared against the expert consensus to establish how closely they agree.

Lessons learned

• Validation against expert consensus is a necessary step for clinical AI tools. The team recognised that a tool built on a local prescribing guideline needed formal validation before it could be relied upon in practice. The planned expert panel review — comparing AI recommendations against clinical consensus across a range of scenarios — is designed to establish the level of agreement and identify where the tool performs well or needs further refinement.
• A phased quality improvement approach supports measurement. Structuring the project with a baseline data collection phase before introducing the AI tool created a clear foundation for measuring whether the intervention reduces errors. This makes it possible to track change against a known starting point.
• Building AI into an existing workflow supports adoption. The tool was designed around a process that doctors and pharmacists were already carrying out. Introducing AI to reduce friction within that workflow, rather than creating a new one, made it easier to integrate into daily practice.

The project operates within Malaysia's healthcare regulatory environment. The tool was designed to comply with the Personal Data Protection Act (PDPA) by not requiring any patient-identifying information to function. Patient consent and awareness are maintained through in-clinic information displays, and patients can opt out at any time. The system is hosted locally to ensure data sovereignty.