Thoughts on AI
Life is complex but it may look quite simple when everything is normal and is working like it ought to. So it’s not easy to determine how complex activities such as balancing, walking, talking, reading, jumping, sleeping and countless seemingly simple activities are.
However, when things are not working as they should and we begin to correct any deficits, it’s usually at this point we start to see this complexity. Attempts to return things to their normal state, are usually met with varying levels of difficulty depending on what system is out of balance.
The level of complexity observed is dependent on the condition(s) that are presented or those which can be identified. It is usually at this point that healthcare (both the health practitioners and the resources used) are stretched to their limits and costs rise astronomically.
Researchers who have been looking at complexity have noted the inherent heterogeneity/diversity in the range of tasks involved and the infinite space in which the nature of healthcare work is performed. This diversity distinguishes it from other complicated areas such as nuclear power generation and aviation.
They have also observed that in the healthcare space, it is sometimes difficult to relate cause to effect due to the interconnectedness of the domain, the often delayed feedback, and the fact that small changes can cascade and result in catastrophic events (patient death) or large changes can have little or no effect. This phenomena makes the domain unpredictable and difficult to understand.
Current digital health systems have used dominant and traditional black box views which have looked at systems like machines where distinct units can be isolated and improved individually. This is in contrast to the views expressed by complexity theory that view systems differently. These theories view the units within systems as frameworks of interrelationships that consists of diverse, robust entities that work interdependently and are capable of adapting to the context that they find themselves in.
This post is the first of many posts which will attempt to discuss the phenomena of complexity in a small way and place it in the context of time, space and ontological elements of data that are collected both when the health landscape is static(rarely) or when its in in flux( (always changing) and show why understanding this may matter.
Various researcher have discussed many of the things I blog about more in depth and references will be given for further reading in a different post.
Steve Magare is the founder of IQ Informatics . He is a health innovation enthusiast, E-health researcher and health Informatics professional interested in finding innovative solutions that can be useful in improving healthcare particularly in less developed economies.
Medical testing is central to effectively preventing a sudden and unexpected onset of disease. Testing especially when current results are compared to previous normal results or against well known normal limits can show us something more sinister like a stroke, a heart attack, heart failure, diabetes or some other adverse event could be imminent.
Knowing this information early enough could provide patients and clinicians with enough time to implement some simpler interventions that may stop such happenings or reduce their impact. Simple interventions such as prescribing a blood thinner to stem a stroke or heart attack, or a steroid infusion to reduce inflammation and soften auto-immune attacks or infusing potassium to boost low potassium levels and maintain normal heart function is what is described to here as being a relatively simple and lower cost intervention.
It could even be a non clinical intervention like making a dietary and lifestyle change. This is if testing is done early enough and if time exists for any such intervention to be successfully accomplished. These are just but a few of the cheaper interventions that can be done when clinicians know exactly what is out of balance and needs rectification.
By taking such interventions both clinicians and patients may perhaps prevent an actual disease from presenting fully or buy some time to prepare adequately if the event is inevitable
Isn’t this how preventive medicine was envisioned?
Preventive medicine
There are often a number of interventions that if implemented on time, could cost way less to the individual both in-terms of the impact to patient wellbeing over the short, medium and long term and to personal, organizational and national healthcare costs.
It usually takes a clinical specialist to listen to patient complaints, identify and isolate suspected problem areas, test and treat. A tingling sensation on the foot or a temporary loss of tasting ability which resolves with time presented during a routine clinical visit can just be seen as a clinically isolated event when it is sometimes not.
It could as well point to a neurological issue in the brain on the opposite end of where the symptom is or even point to a nervous system disorder.
Such seemingly mundane symptoms could easily be overlooked but could be early signs of potentially chronic conditions that if not handled adequately could be harder and costlier to treat in the future. Such symptoms are usually identified by specific specialists and those adept at such conditions and could easily be overlooked by non specialists.
In many cases, following best healthcare practise propels healthcare practitioners to avoid looking at symptoms in isolation. That tingling sensation when combined with a previously encountered visual disturbance for instance would point to something deeper. Experienced healthcare practitioners would want to explore further and may order a completely different set of tests and scans than they would have or refer a patient to different better suited specialists.
Such interventions are normally seen as being simple in hindsight. However, at the current time, the environment is usually filled with many other distracting elements and decisions on what to test for becomes challenging. Good clinical decisions usually beg for experienced practitioners who are able to pick out and isolate those seemingly mundane symptoms that are buried amongst a host of other more immediately urgent health complaints.
Skills to identify such symptoms are not as common and where they are available they are often over stretched and are difficult to scale.
Identifying correct clinical pathways is especially challenging when clinicians don’t usually see a certain type of condition. They may even have difficulties in knowing which specialist would be best to refer such cases to. It is not uncommon to be diagnosed with and treated for Malaria, Typhoid or Tuberculosis. Regardless of whether a patient has any of those conditions when they present to a local healthcare facility.
Would an overwhelmed clinical officer know any better?
Or spend more time than is necessary to pursue a seemingly relatively light complaint?
A fundamental shift-Digital Health
Buzz terms like Big Data, Artificial Intelligence and Machine Learning are not just that. They represent concepts which if effectively and cleverly utilized, they can begin to make massive impact to the healthcare practise both in population and individual based health. This is especially true when these technologies are used to support clinical practitioners as they go about their day to day clinical tasks. To a lay person these terms may appear disjointed and standalone. But when they act together the power of these technologies become exponential and greater than the sum of their individual parts.
These technologies have the potential of propelling even a basically trained clinician in a less developed healthcare system and/or country to higher skill levels. Levels where they could be assisted to place patients on a clinical pathway similar to what a highly trained, highly skilled health professional in specialist center anywhere in the world would.
How is that even possible?
Artificial intelligence, Machine Learning, Testing, Wearables, Big Data
Artificial Intelligence (AI) is an evolving field that could provide the technology to make specialized knowledge and expertise scalable and easier available. This is akin to having a specialist at hand consistently and always online by simply having just a power connection and perhaps access to the internet to allow health facilities tap into rare specialist expertise. This is a more achievable goal to a local health facility compared to training and retaining very highly skilled clinicians in less developed economies and often in rural settings where the need is just as great.
The recent past has seen an emergence of self driving vehicles and all types of robotics in addition to a variety of smart appliances and wearables. These technologies create opportunities for constant and continuous testing and recording of clinical results. Data from wearables meets the three Vs of big data essentially qualifying it as such.
Collecting big and real time data carries huge advantages as it can be used to feed into AI technologies and machine learning algorithms and assist clinicians in their tasks. It could also act as an alert to patients or patient carers of any impending events.
The question becomes. Why isn’t this being done yet if it appears achievable and promises to bring a lot of benefit? Don’t Clinical Decision Support Systems (CDSS) that have existed for about four decades do the same thing?
The answer is yes! CDSS have existed for sometime but the major difference now is the use of real time data. Availability of real time data means that data is constantly being churned by ever present and always online artificial intelligence systems and algorithms. Another difference is that these technologies will be based on a push of information based on existing information rather than a pull for service which has often been the case with clinical decision support systems where clinicians input data that leads them down a decision tree. However, this new approach comes with some significant caveats including privacy concerns as well as introducing certain biases and tunnel vision. It is great but it sadly is “not the be all end all”.
Additionally, certain health cases are not that simple to solve especially when cases are neither linear nor straight forward as these technologies are currently built to operate. In many instances, healthcare is not an exact science and some complex cases carry many ethical dilemmas which machines, cannot as yet solve. Choosing the best path to take involves a clinician determining what is the crucial information to consider and what information is not as essential and can be taken offline. Healthcare sometimes needs clinicians to make a judgement call based on experience, skill and clinical observation of the patient.
Decision making is based on many such assessments and it becomes complex to explain why they are making certain choices. This uncertainty adds an element of intuition and gut feeling to the decision making process which does not qualify as a decision being based on scientific principles and this makes it hard to fully automate.
Therefore AI as it currently exists is complementary to specialist decision rather than act as an alternative.
Digital health is as effective as the data it uses. Testing often and feeding the results into intelligent systems is possibly one of the easiest ways one can know what’s coming next and could provide one of the nearest things to disease prevention.
Interesting times ahead.
Test often.
Disclaimer: The article is a feature article for Health – IT educational purposes only. It should not be taken as medical advise.
Various researcher have discussed many of the things I blog about more in depth and references will be given for further reading in a different post.
Steve Magare is the founder of IQ Informatics . He is a health innovation enthusiast, E-health researcher and health Informatics professional interested in finding innovative solutions that can be useful in improving healthcare particularly in less developed economies.