contact us

Use the form on the right to contact our team.

You can contact us to arrange a demo of PatientSource where we can show you what we can do to help your healthcare organisation.

We are always happy to hear from you and look forward to working with you.

Name *


01223 851273

PatientSource - Powered by Artificial Intelligence (AI)

PatientSource, the next-generation electronic medical records system, uses artificial intelligence (AI) to help doctors and nurses do a better job. Here are some of the AI tools we have deployed in PatientSource.

Powered by AI
Artifical Intelligence to assist diagnosis, detect deterioration and forecast resource usage.

PatientSource makes extensive use of Artificial Intelligence (AI) to make it a smarter assistant for doctors and nurses. Artificial Intelligence will never be able to replace doctors and nurses, but it will become a powerful tool for them to use as healthcare continues to grow in complexity.

We believe that AI should be used to help clinicians do a better job while still letting them have the final say. Here are some of the ways we use AI:


Extracting diagnoses, symptoms and treatments from unstructured text

We use Natural Language Processing (NLP) techniques powered by medical dictionaries, to pull out relevant terms from blocks of text. This means that a clinician can write a clinical encounter case note as they would normally, and PatientSource will automatically extract any diagnoses, symptoms, and treatments mentioned. 

PatientSource is smart enough to know the difference between when a diagnoses is being entertained (e.g. "?cauda equina"), excluded (e.g. "no neural root compression"), or affirmed (e.g. "L4/L5 posterior disc prolapse with equina compression"). It can also deal with thousands of medical abbreviations (e.g. "T2DM" for Type 2 Diabetes Mellitus) and even pick the correct term where the same abbreviation has multiple meanings.


Suggesting diagnoses from symptoms, signs, vitals, patient demographics and test results

PatientSource uses Deep Learning Neural Networks to suggest likely diagnoses based upon what it knows about a patient. It draws on information from the patient's case notes to discern symptoms and signs. It pulls in data from the Investigations module to obtain recent lab and radiology tests. It pulls data from the Observations module to look at recent vitals. It then looks at the patient demographics to see what diagnoses are most likely for age and sex of patient. This information is then weighed up, and a list of plausible diagnoses are presented to the clinician.

Our Chief Medical Officer, Dr Michael Brooks, who specialises in Emergency Medicine where diagnostics is a huge part of the job, explains:

"Diagnostic medicine is all about pulling together relevant information from the patient's story, what you observe, and test results. Most of the time there is no single data point which clinches the diagnoses and instead, you arrive at a certain diagnosis by weighing up all this information. For example, chest pain in a 30 year old is far less likely to be due to a heart attack than chest pain in a 75 year old, but it's not impossible in a 30 year old and not guaranteed in a 75 year old. You need more data, like the nature of the pain, what triggered it, and some ECGs. We've built that kind of judgement into PatientSource using neural networks.

Misdiagnosis is the number 1 cause of patient harm and medical litigation. As doctors we like to think we're good at diagnostics, but in reality, there's room for improvement. A tool that says 'have you thought about these diagnoses?'  comes in very handy to help your 4am night shift brain!"


Detecting early patient deterioration

PatientSource uses Support Vector Machines to identify patients who are deteriorating, early. This is in addition to implementing classical algorithmic systems like the National Early Warning Score.

When it comes to vitals, one of the biggest challenges is avoiding alert fatigue. When you've got a bedside monitor which is constantly alarming, your staff will begin to tune it out. The alarm has then lost its purpose. It's a classical trade-off between sensitivity and specificity. Too sensitive, and most of the time it sounds will be a false alarm, so staff will just learn to ignore it. Conversely, if it has a low sensitivity, you'll risk missing sick patients.

Support Vector Machines allow PatientSource to be much smarter about how and when it show alarms. It takes into account the recent trends in data points as well as multiple factors in a patient's set of observations. We can then identify patients who are starting to slide down the slippery slope before any single vital sign becomes abnormal.


Digital Dictation and Prescribing by Voice

PatientSource uses neural network based speech-to-text to assist with entering clinical notes and prescribing medications. Wards and clinics are busy places with lots of noise, so PatientSource employs a lot of AI based filtering to home in on the target speech. Clinicians also tend to use a very different corpus of words to the general population. 

Clinicians can walk up to a patient's bedside with a tablet, scan the patient's wristband barcode, and state "Prescribe paracetamol one gram QDS PRN". PatientSource then converts this into the right type of electronic prescription. The clinician is then asked to confirm that the prescription item PatientSource interpreted is that which they intended.