Opinion Article by Dr. Guy Wood-Gush, CEO of Deontics.
Comment: Actual technology developments in AI, machine learning, deep learning on the one side , and lowering prices for cloud services, hardware devices on the other side prepare the ground for quicker deployments as known from the past. So watch out !
A turning pointing for artificial intelligence (AI) came when the computer Deep Blue beat world champion Garry Kasparov at chess 18 years ago. Since those early days some of the best minds have been working on applying AI in medicine. Today, I'd like to see where AI could take us in clinical practice.
Here's a scenario. A man falls down in the street clutching his chest. An ambulance arrives armed with the latest tools in AI, and takes multiple streams of data to look for patterns.
Paramedics make their initial observations, speaking directly into a voice recognition system that picks up the salient data points from his natural language, such as name, GP, date of birth, symptoms and clinical signs. The system starts to look for the patient's GP record and medical history, and any clinical notes from local hospitals. The crew measure and record blood pressure, pulse, and O2 saturation. The devices they carry immediately start to show differential diagnosis, with probabilities, and recommend next steps.
Back at the hospital, the data are accessed. Clinical indicators at the scene and the GP record are pointing towards a myocardial infarction (MI). Which pathway should the patient start at hospital?
The admitting doctor agrees that MI is likely and implements an ECG and chest x-ray. A single click orders all relevant tests and records the fact that the patient is diabetic. In addition to the MI pathway, the patient is automatically entered on the diabetes comorbidity pathway and the patient's blood sugar is evaluated against historical personal data. The system immediately sends out a clinical alert that the patient's blood sugar is dangerously low and needs intervention.
On arrival, the doctor dictates a clinical assessment into a voice recognition system that recognises and records relevant data items. It confirms a suspected MI, and implements the recommended actions to address low blood sugar.
All the time, data in the clinical AI system are building. The system interprets the ECG and the chest x-ray, automatically finding a pattern that is consistent with chronic cardiac damage from hypertension. The system recommends cardiac catheterisation as the next step and the doctor confirms this. Patient-specific antihypertensive therapy is recommended.
So how is this fundamentally different to what doctors do here and now? Is this human versus machine? And is the medical Deep Blue about to beat the human clinician?
The fundamental difference is that AI adds a layer of analytics and automation to medicine that removes the need for duplication, reduces error, and drives patients towards the correct pathways while avoiding the danger of missing unusual diagnoses.
It makes sure that the right things happen to the patient at the right time, in the right place, and in the right order, and reduces unwarranted variation in clinical practice. It brings together clinical data with guidelines and allows clinicians to make the best-informed decisions for each individual.
This is human plus machine and I would argue that it equals better medicine.
With AI, the role of the doctor changes to become more like the pilot of a modern aeroplane. The computer does a lot of the flying but the plane still needs the pilot.
The doctor can be released from the paperwork and spend more time with the patient. AI enables the patient to join in shared decision making based around the same evidence as clinicians.
Early AI systems relied on coding and structured logic; but people are unique. Advanced AI can provide a smarter approach that starts from where the patient is now and changes as the patient's parameters change. Think of it as a 'sat nav' for medicine.
Fundamentally clinical logic has to be modelled accurately using a clinical logic language such as Proforma, which was explicitly designed around a patient safety and quality agenda.
So what of this scenario already exists? Algorithms that can interpret chest x-rays and ECGs are out there. Proforma-based tools can already match clinical parameters against clinical guidelines and extract the correct patient pathway. It's already being used here in the UK and the US for oncology, cardiovascular care, diabetes and other conditions. All medicine will be affected by the AI technology.
Plus the wealth of data that emerges can support audit, research and help provide the analysis for continuous improvement. But it makes most difference when it helps a doctor treat a patient.
I really do not believe that computers and AI will ever take over from doctors. But I do believe that clinical medicine can be dramatically improved - in quality, safety and efficiency terms - and AI is the tool we need to achieve this.
Comment: Actual technology developments in AI, machine learning, deep learning on the one side , and lowering prices for cloud services, hardware devices on the other side prepare the ground for quicker deployments as known from the past. So watch out !
A turning pointing for artificial intelligence (AI) came when the computer Deep Blue beat world champion Garry Kasparov at chess 18 years ago. Since those early days some of the best minds have been working on applying AI in medicine. Today, I'd like to see where AI could take us in clinical practice.
Here's a scenario. A man falls down in the street clutching his chest. An ambulance arrives armed with the latest tools in AI, and takes multiple streams of data to look for patterns.
Paramedics make their initial observations, speaking directly into a voice recognition system that picks up the salient data points from his natural language, such as name, GP, date of birth, symptoms and clinical signs. The system starts to look for the patient's GP record and medical history, and any clinical notes from local hospitals. The crew measure and record blood pressure, pulse, and O2 saturation. The devices they carry immediately start to show differential diagnosis, with probabilities, and recommend next steps.
Back at the hospital, the data are accessed. Clinical indicators at the scene and the GP record are pointing towards a myocardial infarction (MI). Which pathway should the patient start at hospital?
The admitting doctor agrees that MI is likely and implements an ECG and chest x-ray. A single click orders all relevant tests and records the fact that the patient is diabetic. In addition to the MI pathway, the patient is automatically entered on the diabetes comorbidity pathway and the patient's blood sugar is evaluated against historical personal data. The system immediately sends out a clinical alert that the patient's blood sugar is dangerously low and needs intervention.
On arrival, the doctor dictates a clinical assessment into a voice recognition system that recognises and records relevant data items. It confirms a suspected MI, and implements the recommended actions to address low blood sugar.
All the time, data in the clinical AI system are building. The system interprets the ECG and the chest x-ray, automatically finding a pattern that is consistent with chronic cardiac damage from hypertension. The system recommends cardiac catheterisation as the next step and the doctor confirms this. Patient-specific antihypertensive therapy is recommended.
So how is this fundamentally different to what doctors do here and now? Is this human versus machine? And is the medical Deep Blue about to beat the human clinician?
The fundamental difference is that AI adds a layer of analytics and automation to medicine that removes the need for duplication, reduces error, and drives patients towards the correct pathways while avoiding the danger of missing unusual diagnoses.
It makes sure that the right things happen to the patient at the right time, in the right place, and in the right order, and reduces unwarranted variation in clinical practice. It brings together clinical data with guidelines and allows clinicians to make the best-informed decisions for each individual.
This is human plus machine and I would argue that it equals better medicine.
With AI, the role of the doctor changes to become more like the pilot of a modern aeroplane. The computer does a lot of the flying but the plane still needs the pilot.
The doctor can be released from the paperwork and spend more time with the patient. AI enables the patient to join in shared decision making based around the same evidence as clinicians.
Early AI systems relied on coding and structured logic; but people are unique. Advanced AI can provide a smarter approach that starts from where the patient is now and changes as the patient's parameters change. Think of it as a 'sat nav' for medicine.
Fundamentally clinical logic has to be modelled accurately using a clinical logic language such as Proforma, which was explicitly designed around a patient safety and quality agenda.
So what of this scenario already exists? Algorithms that can interpret chest x-rays and ECGs are out there. Proforma-based tools can already match clinical parameters against clinical guidelines and extract the correct patient pathway. It's already being used here in the UK and the US for oncology, cardiovascular care, diabetes and other conditions. All medicine will be affected by the AI technology.
Plus the wealth of data that emerges can support audit, research and help provide the analysis for continuous improvement. But it makes most difference when it helps a doctor treat a patient.
I really do not believe that computers and AI will ever take over from doctors. But I do believe that clinical medicine can be dramatically improved - in quality, safety and efficiency terms - and AI is the tool we need to achieve this.
Keine Kommentare:
Kommentar veröffentlichen