How a ‘Left Shift’ is Needed to Bring Healthcare Back From the Brink

By Chris Tackaberry, Co-Founder and CEO, Clinithink
Twitter: @Clinithink
Twitter: @christackaberry

Health systems are facing significant headwinds as they approach the start of a new year. Rising labor costs and acute clinical staff shortages are straining financial margins, while also threatening affordability and ease of access to care. Healthcare is grappling with an excess of demand – caused by patients deferring treatment during the pandemic, and now presenting with more advanced, complicated disease – putting additional pressure on an already-strained workforce.

Indeed, these are difficult times for anyone working within healthcare, and as we move into 2023, the combined challenges of rising workloads, insufficient resources and tighter budgets show no sign of abating.

The need for a ‘left shift’

Right now, healthcare is at a perilous juncture. Something needs to change. For the industry to navigate its way out of this difficult situation, there needs to be a ‘left shift’ – a move to increased upstream interventions, whereby instead of treating patients in hospitals once their condition has significantly progressed, we focus on earlier stage treatment in ambulatory and community settings.

This left shift requires identifying disease at an earlier stage, enabling earlier interventions. The result would almost certainly be better clinical outcomes and a reduced burden on health systems.

And for this shift to happen, technology will need to play a central role.

Covid as a catalyst for change

If there is a positive to be taken from Covid, it’s proof that healthcare – which has traditionally been far slower than other industries to embrace technology – has the ability to innovate at speed.

During Covid we saw an increase in the use of AI – from using the technology to monitor the effectiveness of treatment, through to helping to speed up vaccine development – demonstrating healthcare’s capacity for rapid technology adoption, even despite the disparate and highly-complex health ecosystem coupled with a challenging regulatory environment.

The ongoing pressures on health systems are continuing to drive a need for technology-led solutions. Interoperability and data sharing are improving, and in turn they are creating large, complicated data sets that are growing in complexity, at an exponential scale. The valuable insights from these datasets can only be unlocked using next-generation tools based on AI technology, that automate the review of unstructured data buried deep within the medical records, producing rich, structured insights.

Enabling earlier disease identification and intervention

One way in which AI is already enabling the left shift is by identifying the people most likely to be at risk of disease, thereby enabling earlier detection and diagnosis.

AI is actively being used today to spot clusters of symptoms in people that could signal either chronic diseases, which are the leading causes of illness, disability, and death in the US, and cancer. To take lung cancer as an example – most patients with the disease are put forward for chest imaging after the development of specific symptoms. But by then the disease could already be in its later stages of disease. The American Lung Association puts the five-year survival rate for lung cancer at 56 percent for cases detected when the disease is still localized, but this drops to 5 percent once patients are in the latest stage of the disease.

By using AI to help identify patients when they are feeling unwell, but before they have very specific symptoms that can be linked to an underlying cause, AI can provide physicians with the data and insights to intervene before a disease has progressed significantly.

This AI-driven left shift results in far earlier disease detection and diagnosis – increasing survival rates and driving down the length and intensity of required treatment. Finding patients with collections of characteristics that signal early disease before they present with more acute symptoms not only improves patient outcomes, but also significantly reduces the burden on health systems’ clinical resources and budgets.

Moving from identification to prediction

The next step will be to use AI to develop models that are able to accurately predict the people most likely to be at risk of specific diseases, therefore enabling highly-focused screening efforts. This predictive capability will enable much more efficient utilization of finite clinical and financial resources on the highest-risk people, but intervening to improve outcomes much sooner than is currently possible.

For example, AI could be used to predict the likelihood that a patient sitting across from a primary care physician (PCP) is at high risk of ovarian cancer. By flagging this risk to the PCP, AI enables the practitioner to decide what course of action to take – for example an ultrasound scan – increasing the chances of an early-stage diagnosis and potentially cure.

To-date, the application of AI to develop predictive models in healthcare has lagged behind other industries. The energy industry, for example, can very accurately predict surges in demand for power based on modeling from previous datasets, and can then use this information to ensure the necessary supply is available.

Healthcare data, in comparison, has traditionally been too difficult to make sense of, especially if structured data is the only source used. The challenge comes from the fact that 80% of the data held within electronic health records (EHRs) is unstructured information held in the form of physician notes, visit summaries, discharge summaries and operation reports. For analyses at scale, this information has typically gone unanalyzed, since IT systems have lacked the technological ability to process the data.

But this is starting to change. We are now beginning to see these unstructured data sources being successfully analyzed using next-generation solutions based on AI, thus enabling far more accurate predictions.

Using AI to understand human language

The unlocking of this unstructured data has been in no small part due to the capabilities of Clinical Natural Language Processing (CNLP), a highly specialized branch of AI that enables machines to ‘understand’ the clinical language. CNLP can understand billions of different word and phrase combinations (including slang and misspellings) that relate to hundreds of thousands of detailed clinical concepts, yielding valuable insights from the clinical and social data held within EHRs, which had previously been largely inaccessible.

CNLP, along with other AI technologies, is not only enabling earlier disease identification, but is also making the predictive capacity of AI a reality. This has the potential to revolutionize healthcare further still, pushing the clinical intervention point even earlier. This revolution is not something years into the future – it is happening today. But we are only at the very beginning and there is far more yet to come. It’s an exciting prospect.