This excellent report captures how we might think about healthcare in the future.
https://www.mckinsey.com/mhi/our-insights/adding-years-to-life-and-life-to-year
Healthcare includes physical, mental, social and spiritual domains and the influencing factors, such as genetics form a small part of our health outcomes. Personal behaviour, social constructs and environmental factors are the more critical part of the eco-system for health.
I love new learning so am excited by the times ahead, with the use of technology to create new ways of thinking about our health.
Our SMART devices are being rewarded for steps achieved, 23andme.com creates consumer led health information and positive words create nudges for change and this type of online presence may be far more impactful than our traditional health care models.
So my personal development plan for this year, includes Personalised Medicine (PM), Digital Therapies (DTx), Artificial Intelligence (AI), Gamification and Machine Learning (ML), Robotics, Smart Devices and understanding the power of Google search which we all use before seeking advice elsewhere.
Developments like cloud computing, blockchain, telehealth, and health apps are now routinely used. Like every change, the unknown feels uncomfortable as we grapple with clinical medicine, within a context of psychology, neurosciences, linguistics, computer science, artificial intelligence, emotion and behaviour alongside genomics, neural networks and pharmacy in a landscape of culture, history and ecology. Increasingly complex care plans, involve multiple organisations, polytherapy and polypharmacy and will require innovation to create shared understanding.
Digital recording already captures physical measurements, experiences, and social media narratives and has set the stage for a revolution in individual health and medical management, population-wide health strategies, and real-time generation of new knowledge and insight.
Chronic disease management and preventative medicine occurs in our own homes so we need to shift to understanding how to meet people in their own context to achieve better outcomes, improving quality of life, but still recognising systemic disadvantages such as accessibility including distance to be travelled to the service, access to technology, time off work, language fluency, health understanding, reading and digital literacy, numeracy and cultural beliefs all need to be addressed. Financial barriers force individuals to choose between household expenses, other priorities and can impact on healthcare and often leads to stigmatisation of individuals who use services ‘inappropriately’.
The social determinants of health (SDoH) are “the conditions in which people are born, grow, live, work and age that shape health,” with these conditions including “socioeconomic status, education, neighbourhood and physical environment, employment, and social support networks, as well as access to health care” and traditionally have been out of scope for the NHS however with Integrated Care Systems and digital technology, we have a unique opportunity to have a meaningful effect on these wider inequities.
I have listed below – from my own research what new terms as clinical practitioners we should be familiar with.
Personalised Medicine
PM can be explained by advances in clinical pharmacology, genetics and neural networks and the recognition that behaviour and the SDoH all impact on an individuals and their families.
We are shifting from being diagnosed with an illness, where the decision about which treatment you’re offered is based on average results through traditional random controlled trials and meta-analysis to PM, where we tailor medical decisions and interventions to an individual person.
We are all unique, our genetic makeup is very slightly different and the “genomics revolution”—advances our understanding of the health implications due to the human genome. There are common risk factors for many diseases, such as age, exercise, cholesterol levels, excess weight and smoking. Our genes, then affect how much difference these risk factors make to each person. This means that some people develop diseases, while others don’t and the same disease will progress differently in different people, and people react differently to treatments.
Through PM, we could minimise side effects, and improve outcomes through a precise diagnosis based on your unique situation, and even be able to predict and prevent illnesses developing in the first place.
PM isn’t just about prescribing, it about a person centred opportunity to address multiple factors including the SDoH.
Pharmacogenetics
Pharmacogenetics has described how people may metabolise drugs differently due to the genes and our unique biochemical processes which influence our drug response. This has led to the concept of a “personalised prescription” by “tailoring drugs to a patient’s genetic makeup”.
To choose the correct drug response and dose, we use genetic information in the context of environmental and personal factors; furthermore, these other factors may be more important than genetics in determining drug response.
From an industry perspective, there has been a lot of promise in terms of PM, however, the larger shift from the pharmaceutical industry is generating collaborations of organisations to increase engagement of individuals, alongside medical or surgical interventions creating support communities rather than a unique drug for an individual.
Artificial Intelligence
Utilising collections of data which arise from electronic health records and other sources, means that any aspect of medical practice such as patient characteristics, symptoms of specific diseases, diagnostic criteria, medication doses and abnormal signs on radiographs or other technology can be reviewed and aligned to our decisions on diagnosis and treatment. This data can be used to construct algorithms to create action.
AI can use computer vision to interpret visual information such as images and videos, to which it can then react based on its algorithms. Natural language processing (NLP) is how AI can understand and interpret human language, whether spoken or written.
However algorithms only work where there is a concrete answer rather than emotion, intuition, subjectivity, perception, memory and human cognition which all influence the decisions we make and currently AI and ML cannot mimic the human connection with others, our resilience and flexibility in response to experience.
Medical applications of AI have focused on diagnostic decision support, often in a specific clinical domain such as radiology and pathology, using algorithms that learn to classify. Examples include diagnosis of malignancy from photographs of skin lesions or from radiography, prediction of sight-threatening eye disease from tomography scans and prediction of impending sepsis from a set of clinical observations and test results.
Systems could autonomously triage patients or prioritise individual’s access to clinical services by screening referrals. Subcutaneous insulin pumps could be driven by information from wearable sensors, or automated ventilator control driven by physiological monitoring data, are all part of our future.
While AI-based systems are currently unable to connect with a smile, notice pain in an expression, or hear distress in a patient’s voice—skills at which humans excel—these systems offer the unique opportunity to augment our performance by creating order and transforming vast amounts of data into clinically actionable information to support optimal care. This should enable us to spend more time on explaining choices, discussing worries and anxieties and creating a shared understanding.
Digital Therapeutics and Gamification of Serious Play
Digital Therapeutics or DTx is one of the latest buzzwords in the digital health ecosystem.
DTx deliver evidence-based therapeutic interventions that are driven by high-quality software programmes to prevent, manage, or treat a medical disorder or disease. They are used independently or in concert with medications, devices, or other therapies to optimise patient care and health outcomes and build on PM utilising a variety of techniques.
Examples include:
- Daylight and Sleepio from UK based Big Health for anxiety and insomnia respectively
- Action For Happiness
- gameChange from Oxford Virtual Reality for Mental Health
- Velibra for adults with social anxiety disorder, panic disorder and general anxiety disorder
- BDD-NET for adults with body dysmorphic disorder
- Minddistrict for adults with depression
- Regul8 for adults with irritable bowel syndrome
Other areas of development are chronic pain management, oncology support programmes, substance misuse interventions, lifestyle changes including health coaching, meditation and health behaviours such as exercise and diet.
DTx solutions are typically delivered through smartphone apps however this requires digital technology so although is very accessible, can also create inequity if these resources are unavailable or not understood.
SMART devices
Healthcare professionals are very familiar with blood sugar technology and substantial improvements in diabetic management, such as normalising blood sugars through SMART technologies however we need to upskill ourselves in understanding SMART data and their link to health outcomes.
Our devices, smart watches and phones measure our heart rates, count our steps, understand our sleep patterns and are part of Microsoft teams:- Headspace to help in stress management.
This march of health sensors and wearables is expanding exponentially and now is interwoven with our clothes, can appear as digital tattoos and possibly digestibles or in our blood vessels as nanobots.
Digital health technologies can identify environmental factors, including air pollution and UV light, pollen and may lead to new ways of management of asthma or risk of malignancy and population health strategies. Food scanners could alert users to one of the constituents in their meal ensuring those allergic to certain foodstuffs are informed.
Although consumer demand for interventions that support behaviour change is high, many programmes such as weight management elicit only marginal and temporary changes in weight, with participants often experiencing weight regain. In their ideal form, evidence-based digital health tools will focus on health behaviour improving self-awareness, provide on-demand health information and education, support self-care, and promote accountability with social support networks, health coaches, and providers. They will utilise real-time data to provide personalised feedback and messaging to support change in a way that is more compelling than the traditional patient education however success of these models is still evolving.
Facilitating Off-Site Patient Management through Telemedicine
COVID-19 drove many health consultations to a virtual technology and we then have been increasingly utilising the model of hospital at home, virtual wards and virtual consultation utilising remote devices to share information and provide services in a more effective and often more acceptable manner.
Robotics
Robotics, now are part of many surgical procedures and support simulation and learning, however the technology exists for robots that could undertake phlebotomy and other technical skills, disinfect environments and even be social companions. Toy robots have been utilised as educational resources for children with ASD and other disabilities and artificial limbs have become unique and admired. We recognise how empathy, communication and the importance of human relationships underpins our therapeutic relationships but we could see robotics as partners in care.
However with every great opportunity comes unintended consequence
The bias cascade starts with the data collection process which is part of information gathering and research. Disparities are known to exist in the recruitment of subjects and health data sets, where certain populations are underrepresented, decisions made by systems that are subject to bias and fragmented services all perpetuate biased clinical performance and can be amplified by AI and ML.
We therefore need diverse and well balanced study populations, paying particular attention to racial and ethnic diversity, gender balance, socioeconomic equity, and other social, as well as social determinants of health including access. Decision making should be scrutinised for bias and wider systemic disadvantage represented. Electronic health records, and any other documents and sources of data used in AI algorithms need to use neutral and fair language to retrieve data records. Data needs to be accuracy, have identity matching capability, and privacy protections of individual data as part of the governance requirements for successful technology transformation.
Algorithm bias has already been identified in kidney function assessment (glomerular filtration rate) with age, gender, race, and levels of creatinine. The tools categorised the race of the patient as Black or not Black. When the calculations were created, black people were falsely perceived to have higher muscle mass on average, which often resulted in a higher kidney function score and triggered a delay in necessary treatment. This has been mirrored in other measures related to bone density and urinary tract infections as examples.
Recently it was noted that pulse oximeters are not accurate across different skin tones leading to altered management plans. Assessment of jaundice, pulmonary function and other parameters have been subject to incorrect data analysis in relation to ethnicity, gender and other characteristics.
The big technology giants such as Apple, Microsoft and Amazon are able to mine their health data creating more data points than traditional health services and may use AI models to create market opportunities in a lucrative lifestyle and wellbeing market place so understanding their own approach to bias is critical as part of the future of population health.
Systems can be designed to take action when explicitly rules are identified, however in healthcare, there is complexity and an ever changing environment so caution needs to be manifested before accepting digital transformation.
Our traditional model of Input – Intervention – Output should move to consider whether Experience – Conversation – Understanding is more appropriate therefore priorititising human interaction.
Other challenges in technology include:
- Distributional shift – where a system is poor at recognising a change in context resulting in the system continuing to make errors. Examples could include a change in manufacturer of a technology which alters parameters related to radiology images but the AI is unaware of the change so may misinterpret data points. Another example might be the introduction of a drug safety alert and the failure to adopt the changes needed.
- Automation complacency – Automation bias describes the phenomenon where we accept the guidance of an automated system and cease searching for confirmatory evidence, transferring responsibility for the decision to the machine. This is already seen in automation of blood pressure and oxygen saturations, believing the machine and not the presentation of the patient.
- Negative side effects – Systems can reinforces an incorrect decision and create a positive feedback loop. An example might be that an autonomous ventilator derives a ventilation strategy that successfully maintains short term oxygenation at the expense of long-term lung damage.
- Unsafe exploration – An actively learning system begins to learn new strategies by testing boundary conditions in an unsafe way, such as a continuously learning autonomous heparin infusion starts using dangerously large bolus doses to achieve rapid aPTT control.
- Interoperability – Significant progress on interoperability has occurred, but the broad interoperability of health care data platforms is still lacking including access to records, terminology standards, and concern about data management across organisations. The move needs to be towards patient owned information rather than organisation owned data.
- Cybersecurity and privacy concerns – are major obstacles to digital health adoption, continue to erode patient trust, and reinforce health systems’ reluctance to share data so leveraging the opportunity to share, aggregate, and analyse health data to improve individual health and to advance the learning health system is significant, but this has to be balanced against the risk of loss of privacy for individuals sharing their most sensitive data including through the expansion of the use of apps. Cybersecurity requires special attention to avoid intentional corruption of AI/ML training datasets (training data poisoning), use of AI by attackers, or anti-privacy designs in digital health.
- Identify Confirmation – Understanding the identity of the person is critical as many interventions have side effects alongside benefits so being correctly allocated is important alongside monetarisation capability and information sharing protocols.
- Current unclear standards and lack of regulatory guidance – has created a marketplace where promising digital health solutions that provide superior quality, impact, and value are difficult to distinguish from poor quality services.
Governance Frameworks in a Technology Centric Evidence Based Medicine
The public has routine exposure to digitally facilitated convenience based on their experience with other industries and now expects the same from the health care ecosystem. As the understanding of what creates health and well-being grows, it is important to engage patients, families, and communities in the design of new structures, processes, and solutions to support health and well-being.
An immediate priority is to ensure access to digital solutions as digital exclusion can reinforce inequity so ownership of devices, understanding of technology and broadband access is essential across all economic groups and all regions of the country and as a clinician this should now be part of our medical history.
Eliminating bias and creating fairness and equity in AI means being conscious of bias at every potential entry point but there will be the unintended consequence of being missed in design and implementation stages. We need to be alert to this and as a clinician be part of the constant review and ensuring a diverse team is present to explore any systemic disadvantages that may be created.
Quality control questions that we need to be able to ask are:
- Has the system been tested in diverse locations and populations?
- How can we be sure the training data matches what we expect to see in real life and does not contain bias?
- How can we be confident of the quality of the ‘labels’ the system is trained on?
- Do the ‘labels’ represent a concrete outcome or a clinical opinion?
- How has imbalance in the training set been addressed?
- How is the system going to be monitored and maintained over time?
- Does the system adjust its behaviour (‘err on the side of caution’) where there are high impact negative outcomes?
- Does it produce an estimate of confidence?
- How is the certainty of prediction communicated to clinicians to avoid automation bias?
- How can it accommodate changes to clinical practice?
- What aspects of existing clinical practice does this system reinforce?
Stewarding Evidence Based Medicine for Our Future
By its nature, EBM regards disease at a population level with minimal consideration to the role of the individuals. As one size does not fit all, the move towards PM, especially if we take into account the SDoH will be positive. By altering the paradigm of EBM to understand what kinds of interventions improve health outcomes, and which ones do not, there is no reason why appropriate and applicable evidence cannot be gathered about the effectiveness of interventions for everyone.
The transition from ‘outcomes that matter to the industry’ to ‘outcomes that matter to patients’ has the capacity to transform EBM and will be driven by the technology companies and our own interactions with apps, SMART devices and consumer led initiatives however it is also dependent on a highly-skilled digital health workforce, and the training challenge for leveraging digital health is our next learning journey.
So in conclusion, I will continue to enjoy learning about technology and its opportunities, alongside the responsibility to ensure that the limitations are understood. I look forward to the time saved through AI, leaving me the ability to communicate and create shared understanding with others on how to navigate the complex eco-system of healthcare.
