Artificial brains, no fake intelligence
And yet, the surprising thing is that humanity has developed the technology to make sense of things en masse. Unfortunately, though, few vendors dare to dabble in a sector as complex and diverse as healthcare, where interoperability is a Kraken with all disparate legs, templates aplenty, and data as sensitive as can be.
But why structure data? Why take perfectly readable documents and transform them into something else? It is far less about fixing what is not broken and a lot more about opening the door to a whole other dimension of data exploitability — unleashing the full power of health data in all the ways that we can with modern technology.
What if we told you that your documents do not have to be individual entities? What if we could merge them into meaningful clinical and administrative information that would allow practitioners to grasp the whole picture at a glance? What if your clerical staff could inject all the necessary information into the corresponding patient record in just one click? Think of all the time that would save them, the gain in speed in time-sensitive information exchanges, and the percentage of human error it would avoid.
In short, what if the entire healthcare community could spend less time typing away and more time facing and listening to their actual patients?
Human logic, machine accuracy
Artificial Intelligence is a term that gets thrown around a lot, but it can mean many different things, depending on what its algorithms do. Think of an algorithm as a tiny pair of hands, connected to a big artificial brain, trained via muscle memory to do something it has seen in the past and tasked with performing it diligently - but not overly confident. This is Lifen’s chosen Machine Learning model in a nutshell.
Lifen’s Artificial Intelligence expertise heavily relies on a methodology called Machine Learning, namely Deep Learning, the former’s shiniest, newest brainchild. The former can be described as an umbrella of techniques that enable computers to learn new things from the data they process progressively.
Machine Learning is a science at the intersection of computing, statistics and good old logic: it teaches computers to recognise patterns in the data and then feed specific algorithms with those learned patterns. Once new data arrives, they run the pre-trained algorithms on it, and the computer makes predictions based on what it already knows. In short, we teach computers to make educated guesses.
As for Deep Learning, an even more recent subfield of this science, one can describe it as a highly intricate structure of algorithms that analyse data in a way that replicates the human brain’s abilities. Deep Learning algorithms are developed in a complex, layered structure called an Artificial Neural Network. The latter is, in every way, inspired by what we know about human neurons. This allows for an unprecedented boom in machine abilities.
Educated guesses
But what do these arcane and hard-to-explain algorithms do, and how do they go from a real-world document to a much more flexible data type? How do they know what to do next?
Granted, processing medical documents is no easy task. They come in a remarkable variety of formats and templates. And each facility has its own. A paper surgery report is a three-dimensional object. Its scanned PDF version is two-dimensional but still unstructured. If you want to transform it into structured data, you need to make it one-dimensional while retaining important information that can only be found on a document that contains at least two dimensions - such as where in the text certain segments occur. Getting the best of both worlds, in this case, requires highly flexible algorithms that are carefully designed to fit each customer’s use case.
To do this, Lifen works with a set of core algorithms that are trained on millions of similar documents. Each of the algorithm’s models is tasked with looking for a particular piece of information: the patient and clinician’s names, the procedure date, the date the document was created, et cetera.
Distinguishing a patient’s and clinician's names is, by the way, no easy task. The same applies to the multiple dates on a document, along with the very type of document that is being analysed - is it a referral letter? A prescription? Diagnostic findings?
Lifen uses yet another technology, Natural Language Processing applied to healthcare and on emerging use cases within facilities to address this challenge. The latter is the computerised understanding of contextual nuances in human language.
Lifen uses a method that gives a score to each algorithm's decision - or how likely each computer prediction is to be correct. Knowing that Lifen aims for a very high detection rate (as a necessary precursor to a satisfactory precision rate), algorithms can spend up to several years in training before they are deemed accurate enough to perform tasks such as detecting given names within a document and predicting whether they are a patient’s or practitioner’s name.
Predictions the computer is highly confident in are then processed by a high-capacity artificial neural network that fully uses the information they contain, allowing you to see the full picture emerge in ways you had never dreamed possible. In the healthcare sector, this often comes in the form of an Electronic Health Record that is as complete as can be, readily available, and perpetually up-to-date.
And where do humans come in?
In a sector as sensitive as healthcare, no mistakes must be made, or the consequences could be unpleasant at best. Fatal, possibly. In the context of a time-sensitive and critical diagnostic decision, they would most likely be catastrophic.
Whenever our algorithms are unsure if they predicted something correctly, they request that someone manually check it. And this happens whenever the computer is any less than completely sure it got the information right. Gladly, staff can easily carry out the necessary checks and edits in only a few clicks from the moment they receive a notification.
At the same time, at Lifen, a team of data scientists handle pseudonymised data to create new labels (a kind of new neuron containing examples of correct information that allows the AI to update itself and improve the quality of its predictions). This regular human intervention is part of our perennial commitment to continually improving the accuracy of our technology.
Human intervention, both from Lifen and healthcare staff, helps the technology get smarter and less indecisive over time. Every instance of manual input is an instance it memorises. Furthermore, as is typical of Machine Learning, the more our algorithms process your documents (with all their peculiarities and the way they evolve), the better they get, to the point where a very high percentage of information is processed without any need for human intervention.
Better care, and better everything else
The benefits of complementing human intelligence with artificial intelligence in healthcare are numerous, clinical implications being the most obvious asset. As we stand, it is believed that up to 80% of a patient’s entire existing health data capital is missing from their EHR, as it remains unstructured. Missing data reduces the validity of conclusions drawn, complicating the diagnosis process, casting shadows on prognosis, making treatment response harder to predict, and resulting in suboptimal care.
But more than providing clinicians with a reliable, up-to-date, and readily available overview of patient history, Lifen’s technology is also pivotal in assisting facilities in digitising their workflows, which leads to optimised performance, improved time efficiency and overall superior quality of service.
By eliminating the hassle of manual filing, classifying, and transferring information between multiple supports, and the chances of human error and data loss that result in significant lags, backlogs are virtually eliminated, and staff gain significant time thanks to automated EHR integration. With information now stocked electronically, by default, you can also aspire to attain 90% paperlessness, especially if you also implement Lifen’s centralised, secure sending solutions. By doing this, you can expect to see your mail budget cut by roughly 70%.
Better yet: with full digitisation comes a significant boost in the auditing process. Lifen’s extremely fine-grained analytics allow auditors to survey and filter statistics by facility, department, practitioner, and technician, among other filters. One can see the number of documents imported to EHRs by type, how many were done automatically, and how many had to be manually edited. The dashboards also show overviews of automated integration rates over time, besides quantifying how much more accurate algorithms become as they process more and more data. In real-world terms, auditors can quickly tell if a given facility is not meeting its goals, for example, and actions can be taken accordingly.
With the power of data fully unleashed, you may now gain precious insight into systemic issues. The objectiveness of figures should be an indivisible component of the compass that guides you in establishing KPIs and engaging staff to meet designated objectives and quality standards.
Brave Even-Newer World
What other doors can be opened once one attains optimal EHR integration? A complete EHR is a treasure trove of invaluable data at an individual and societal scale. When incorporated into clinical predictive models, educated guesses on diagnosis and prognosis may offer crucial guidance for clinicians on treatment. This is the natural progression expected of technologies such as Artificial Intelligence - once set up, their bewildering capacity never ceases to expand in possibilities beyond their initial intended use.
On that note, more than a technical partner, Lifen also acts as a digital health solutions facilitator. With the necessary EHR connectors already in place, Lifen’s team can quickly deploy cutting-edge technology applications in your facility by reusing the connectors it set up for your Patient Administration System. In only a few weeks, the interoperability puzzle gatekeeping your establishment from using the full potential of medical technology will have been a thing of the past. And unparalleled health management and diagnostic solutions… a thing of the present.
__
Technical contributors: Vanessa Le Roy, Florian de Sá
Technical reviewers: Pierre Rolland, Félix Le Chevallier
Copywriter: Lucy Chambel
