On the Frontline of Prevention | Predictive Technology to Tackle Pain Crises, Drive Targeted Therapeutics, and Enrich Drug Discovery

The therapeutic landscape has seen incredible recent advancements, with treatment options for diseases such as Sickle Cell Disease (SCD) experiencing a relative boom in innovation compared to previous decades. From bench to bedside, this drug discovery process has been strengthened by advancements in genomic sequencing, protein modelling, and high-throughput technology that accelerates the identification of novel targets – applying approaches such as Artificial Intelligence (AI) and Machine Learning (ML) to the collection, analysis, and translation of key findings into therapeutic improvements for patients.

Indeed, AI/ML has seen a notable uptick in its utilisation when generating and analysing vast quantities of patient and experimental output – moving the drug discovery paradigm, as stated by CEO of Berg, Dr Niven Narain, from a place of trial-and-error to one that is driven by predictive hypotheses. Supported by more in-depth natural history studies that in turn create the foundation for target discovery and validation, this creates an ecosystem of experimental research output, retrospective and prospective patient data, and clinical trial findings that allow for unique linkage of multiple sources.

Nonetheless, identifying potential drug targets is just the first step in the journey to bringing these to patients in the real world. With several new therapies now working their way through the development pipeline, there remains a key issue frequently flagged by our clinical and life science colleagues that can sometimes form a barrier to the realisation of the potential this work produces. Past the point of target identification and into clinical trialling, it is not only the need for powerful tracking of detailed outcomes and endpoints that remains crucial, but also determining at what point in a disease cycle to provide a treatment for the greatest efficacy.

The stage at which an intervention is given can have drastically different effects on resulting patient outcomes. In the context of SCD, it has been posited that the efficacy and improvement to patient outcomes of novel vaso-occlusive crisis (VOC)-preventing treatments within ongoing trials may indeed greatly benefit from targeted intervention earlier in the cycle. While current trial endpoints may therefore see undervaluation of a treatment’s potential to improve patient outcomes, ensuring that complications such as VOC can be detected early enough, or even predicted well in advance based upon other surrounding biomarkers, to provide treatment at the right time is critical.

However, determining this exact window of benefit and subsequently guaranteeing that patients can be reached within it is no mean feat. As such, ongoing research to better understand patient physiological wellbeing, as well as the quality of life and activity factors that may impact this, remains crucial. Ensuring that new, validated, and patient-driven real-world endpoints can be captured not only during healthcare contacts but into community care and at home is vital to achieving this, supported by the remote technology and AI/ML algorithms needed to generate and utilise this data to its fullest.

Creating a Care Continuum | Insights in the Community

Many VOCs are experienced and managed by patients themselves in a home setting but can be missed in traditional medical record data. Indeed, criteria for identifying patients who may best benefit from new treatments are often driven by clinical notes around the frequency of events such as VOCs, which may not in fact reflect the true severity of each patient’s experience with their disease. Understanding players in both the physical health and day-to-day quality of life of patients is critical to providing the correct support and interventions, as well as in establishing an accurate picture of their real-world impacts on patient wellbeing.

While current care pathways involving regular clinic and check-up appointments are often based upon patient recall over long periods of time, our recent work has explored the mapping of patient comments to extract key disease-linked phrases and general indicators of lifestyle and wellbeing. This has supported the curation of patient-level timelines that allow the identification of self-reported pain events and VOCs experienced and managed at home, but which may otherwise have been missed. In particular, pain was the most commonly reported phrase, appearing in 43% of patient entries and mapped across key regions such as the back, shoulder and chest.

This proxy for traditional VOC recording has been supported by additional measures for crisis events – a basic and easily input patient pain score forming an important indicator for pain crises that supplement and go beyond traditional records. Held within a patient-centric ecosystem that integrates the tracking of usual habits such as sleep and activity, as well as physical factors such as heart rate and blood oxygen, a core part of ongoing research focuses on elucidating how these may correlate to healthcare events extracted from clinical records, and from data sources such as Hospital Episode Statistics (HES).

Combined, these metrics therefore form a critical part of ongoing research into the development of algorithms that allow for accurate prediction of upcoming VOCs and other severe complications, helping to drive better patient outcomes with earlier, targeted interventions.

Predictive Insights for Early Interventions | Data Without Borders

Establishing the underlying components of a patient’s medical history, pathology, and key characteristics is a critical part of understanding the potential efficacy of particular treatments, as well as predicting what the risk of upcoming challenges may be. Indeed, early analysis built from combined HES and medical record data has identified a number of key factors potentially linked to the rate of VOCs, with pathological markers indicating that patients below normal haemoglobin (Hb) levels and above normal haemolytic marker levels may be at greater risk of VOCs.

Beyond more traditional markers, and with consideration for the growing wealth of live data it is now possible to collect directly from patients at home and in the community, our work to date has also identified some early potential markers in the frequency of VOCs. Through early analysis, a basic comparison of VOC rates in patients below and above SCD-specific, community-derived averages demonstrated a marked difference – focusing here upon EQ-5D measurements, Eleven’s basic 0-10 pain and psychological scoring, and average patient deep sleep proportions.

Indeed, patients with above-average EQ-5D Health State and Psychological Wellbeing scores saw 5.7x and 3.7x lower annual VOC rates than those below the cohort average, respectively. In contrast, those with above-average pain and deep sleep proportions were found to have 1.7x and 6.3x higher annual VOC rates than those below the average, highlighting these metrics as key indicators of potential interest in outcomes monitoring and VOC predictions.

Monitoring how these contributing biometrics and live patient feedback measures are changing in real-time allows us to spot where a serious health challenge may be on the horizon. Supported by an in-depth history of where VOCs have occurred in the past - and what the patient’s activity, sleep, heart rate, PROs, and other key metrics looked like at that precise point in time - this combined approach forms a powerful tool in driving predictive alerts for when an upcoming challenge may be arising, and when to treat a patient before typical presentation to healthcare.

With earlier intervention, this will drive significantly improved outcomes for patients at risk of severe pain crises, while also building an evidence base for the true efficacy of novel therapeutics. No longer limited by data that only captures patients who may have been treated too late in the VOC cycle for full treatment impact, this will help to realise the true potential of new treatments and accelerate approvals – allowing interventions to reach the patients who truly need them.

Beyond an improved understanding of existing or emerging therapeutics and their real-world efficacy, this form of integrated multi-source data further holds the potential for exciting innovation in the drug discovery space – creating natural history, genomic, symptom-focused, comorbidity, and potential influencing intervention insights that support the identification of the next target for drug development.

Advanced AI technology can extract meaning and connections from such vast databases, while also deriving vital context from otherwise unstructured or unannotated data. Indeed, work in areas such as radiology and image processing through ML offers uses across not only the clinical end of the drug development spectrum but also within basic research and high throughput analysis of experimental findings. Applied from lab to analytics, to clinical validation, the evolution of drug discovery is truly set to ignite over the coming years.

How We Can Help | Making the Unpredictable Predictable

The need for a better understanding of the patient experience, be this at home or in hospital, is one that sits at the forefront of the SCD community’s minds. Remotely capturing the data needed to do so and applying the advanced AI/ML technology that in turn derives vital natural history and predictive insights from this underpins our efforts in supporting patient outcomes.

As we work to curate a truly patient-centric ecosystem that drives improvements to care pathways and treatment options, ensuring the build of a clinically powerful research base for patient community and life science insights remains our priority. Eleven’s approach to developing the knowledge base around the determinants of SCD focuses on applying the tools at hand to the creation of landscape-changing, real-world evidence. With a multi-disciplinary ethos for innovation, we welcome anyone who wishes to learn more about our ecosystem or potential opportunities to collaborate to contact us at [email protected].