Automating for accuracy

Patients’ regular medicines are often inaccurately prescribed when they transfer from hospital to home; a major source of medication errors and iatrogenic harm. Medicines reconciliation (MedRec) is an evidence-based process where clinicians manually compare a patient’s regular medicines with what is prescribed in hospital then reconciling any inaccuracies. While MedRec is effective at safeguarding patients it is manually intensive; taking up a significant amount of clinician time which can be better used for direct patient care.

So what if you could automate the process to make it faster and more accurate? This is exactly what Principal Investigator Randall Britten, Dr Jerome Ng and Dr Jens Andreas, together with a team of highly experienced pharmacists and clinicians, are working on with their project Smart Medication Reconciliation. It is not a simple quest because the way that dose instructions are described varies considerably in different parts of the health system. For example, ‘Paracetamol 1g orally at 4-6 hourly for pain, max 4g per 24 hours’ is the same instruction as ‘Panadol 500 mg tablet: film-coated, 100 tablets, blister pack. Take 2 tablets 4-6 hourly, no more than 8 tablets per day’.

Fortunately, the team does not have to start from scratch. Orion Health already has medications reconciliation capability in its Medicines application, and there is academic work, such as that carried about by the University of Dundee, to build upon. The first task has been to assemble anonymised databases that can be used as training sets for machine learning applications such as Natural Language Processing (NLP). The next steps have been using de-identified data from Waitemata District Health Board which combines both primary data (general practice product orders and pharmacy dispensing) and secondary data (hospital inpatient dose-based orders).

The project team are now developing a system whereby the most relevant information can be automatically identified, analysed and categorised, so that it becomes easily recognised data, or what’s known as ‘structured data’. This project has achieved this for about 60% of the ways dose instructions are written, but the goal is achieve a much higher rate. It is anticipated that the system will become the foundation for follow-on projects that could see the development of features such as automated de-duplication within each patient’s list of medications during the reconciliation process.

The benefits of this project will be immediately realised by enhancing existing medical reconciliation software, such as Orion Health’s Medicines application, which is designed to improve patient safety and support clinical decision making. The research outcomes will also enable medication data to be used for pharmaceutical data mining to help in the development of new drug formulations, dose profiles and advanced pharmacodynamics (that is high-risk drugs with narrow therapeutic indices). The days of labour-intensive medication reconciliation are indeed numbered.