There is no greater challenge for healthcare and life science organizations than ensuring that their digital transformation along with better data management will improve patient outcomes, increase operational efficiency and productivity, and better financial results.
The drivers of healthcare and life science’s transition from data rich to data driven are not new and include the race to manage cost and improve quality. Some new drivers include the growth of at risk contracting for providers, the threat of care delivery disruption by the retail industry and the impact of drug discovery in the challenge to balance speed to market with costs. Health and life science industries are data rich. IDC estimates that on average, approximately 270 GB of healthcare and life science data will be created for every person in the world in 2020.
Transformation of data into insights creates the value for health and life science organizations coupled with organizations establishing a data driven culture. Insights will be gained through advanced analytics including Artificial Intelligence (AI) and Machine Learning (ML) which are gaining acceptance in healthcare and life science and require large volumes of high-quality data.
The growth of healthcare costs in the U.S. will outstrip the growth of GDP by 1% and healthcare costs are estimated to grow to 20% of GDP by 2028. The U.S. does not deliver higher quality for that investment, the latest rate of maternal mortality in the U.S. is 26/1000 versus 9/1000 in comparable countries. The impact of health inequity was clearly demonstrated in the increase of COVID-19 positivity rates, hospitalization and death experienced in the U.S. among individuals of color and other marginalized groups. Attempts to manage costs and improve quality have shown mixed results. For example, only 5 of the 54 cost management programs sponsored by the Centers for Medicare and Medicaid Services show measurable savings. Data driven insights gained from AI are required to address and mitigate these fundamental challenges. Solutions will not be found based on anecdotal information.
AI has gained enormous importance in the life sciences industry, especially since the onset of the COVID-19 pandemic – digital resilience and sustainability developed a new meaning. As the industry searched for ways of driving continuity in clinical trials, remote, decentralized, hybrid and virtual models were implemented. One saw the development timelines being shortened and vaccines being brought to the market in less than a year. AI suddenly became integral to everything.
Applications of AI in life sciences in are diverse – a few examples range from the use of AI/ML in software in a medical device (SiMD) and software as a medical device (SaMD), acceleration of patient recruitment, enabling remote monitoring of sites and patients, the development of digital biomarkers and digital therapeutics, driving in silico drug discovery, the development of digital twins, prediction of the likelihood of a patient having an adverse event, signal detection in pharmacovigilance, as well as in sentiment analysis.
Real-world data (RWD) has also recently become more important to leverage data to identify patients and safety signals, develop synthetic control arms and leverage digital pathology and radiology data to support a precision medicine strategy. Valuable insights can also be gleaned from EHRs and claims data to support deep phenotyping and interlinking this with genomic architecture to develop a deeper understanding of mechanisms of action and clinical outcomes. One is seeing a huge convergence between pharma and healthcare as they seek to learn from each other. AI and ML serves as the enabler to be able to derive scalable insights from this data and federated learning models can power the secure use of this data and the learning models.
Challenges to AI Adoption and Implementation
Virtually all IDC AI surveys regarding AI adoption indicate that data quality, quantity and access are among their top challenges to scaling and operationalizing AI. Other top challenges are identified in Figure 1.
While other factors like lack of skilled data science professionals hold organizations back, the fact that 50% of time is spent on data preparation and deployment must be addressed before AI can be democratized and provide lasting value at scale in the health and life science industry. There are additional unique challenges in both life science and healthcare.
Challenges to AI Adoption and Implementation in Life Sciences
The journey from big data to actionable insights is fraught with challenges. As the life sciences industry rapidly pivoted to a decentralized clinical trials model, we’ve seen a huge uptick in the use of the internet of medical things (IoMT), namely wearables, sensors, etc., to capture patient data remotely and support clinical trials. RWD comes with its challenges, a lot of it is unstructured, and it comes in many shapes and sizes. However, the industry is choking on the digital exhaust.
Data science teams are not deeply integrated within business, frequently resulting in products being developed that aren’t delivering to business needs. Organizations invest in multiple proof –of-concepts but lack a cohesive digital strategy. The power of an AI solution depends on the quality of the data and RWD presents challenges. With the increased growth of digital therapeutics, ethical AI comes into the picture with concerns prevailing regarding the learning algorithms being in a black-box and concerns regarding the implications on diversity and inclusion and data bias coming in based on the data on which the algorithms were developed.
Challenges to AI Adoption and Implementation in Healthcare
The hyper heterogeneity of healthcare technology stacks makes the challenges of data fragmentation even more significant. The lack of standards and the unique configurations of healthcare applications such as the electronic health record create interoperability challenges that require highly skilled personnel and technology to accurately integrate data. Even the largest healthcare payers and providers have sufficient data volume to attain model precision.
The greatest success in AI adoption in healthcare is through embedded AI in primarily models that predict adverse events such as the onset of a chronic illness, an avoidable re-admission, avoidable emergency room visit, and missed appointments. Physicians are skeptical of the use of AI when it comes to generating treatment options or suggesting a diagnosis. For this reason, it is critical to involve physicians or other clinical/business stakeholders in the early stages of a project. As physicians begin to understand the value AI can bring to their clinical practice, particularly in reducing cognitive burden, their skepticism will diminish.
Key Take Aways/Essential Guidance
Everyone wants to dip their fingers in AI, but it is important to identify organizational champions that can define an integrated vision, the north star for the organization, outlining targeted outcomes and success criteria, as a starting point. Eighty percent of AI is really about the data, garbage in is garbage out. Nobody has all the data. Therefore, organizations should establish the right data partnerships. Once we have that data, thought needs to go into what one plans to accomplish with that data. It is important to build a data strategy and outline knowledge graphs to identify where that data is located and make it accessible. To build trust, one should make AI recommendations explainable.
This calls for moving beyond a data strategy to a knowledge strategy, tucking in the scientific rationale for the development of those algorithms as well. While focusing on pattern recognition and causal inference, it is also important to weave in the human experience into the solution as patient-centricity gains center-stage. One must never forget that while technology is key, it is really about making an impact on patient’s lives.
Finally, life sciences and healthcare organizations are at times suffering from an identity crisis. Organizations need to identify what is core to them and how much do they want to invest in building capability internally and should consider leveraging partnerships where appropriate. In a world transitioning to a more open model, driven by collaboration and co-innovation, industries need to adopt federated learning models to convert AI into true intelligence. We cover this topic in more detail in the upcoming webinar, “AI Adoption in Health and Life Science Grows as Value is being Delivered“, live on August 26th at 11 AM/ET. Click the button below to register.