š Events
AI in Healthcare - A Six Part Masterclass
TheĀ Royal Society of MedicineĀ is a world-renowned institution known for its quality programme of educational events and conferences. Donāt miss out on the annual AI in healthcare event in February 2021. Itās a virtual masterclass comprisingĀ six sessions over two afternoons. Expect amazing speakers from innovative companies like Skin Analytics, Serenus, Ultromics and many more. Donāt miss out!
Patient clusters. Disease heterogeneity. Endotypes. Precision medicine has a lot of overlapping and confusing terminology. Letās take a deeper look and see what they really meanā¦
š¤·āāļø Problem
Generic diagnostic labels like asthma or diabetes fail to capture the variety of underlying pathways that give rise to a disease.
Because treatments arenāt tailored to these different pathways, patients miss out on treatments or approaches that might be optimal for their specific āflavourā of the condition.
š” Solution
Leverage computational power, emerging biomarkers and large datasets to find distinct clusters of patients within a disease that share similar āpathobiologicalā mechanisms.
Use the resulting endotypes to identity novel biomarkers and drug targets that will make the condition management more precise and personalised.
š Terms
We mainly think about diseases in terms of the way they present (including the signs, symptoms and laboratory findings), or by the underlying change to the tissue or organs (such as cancer).
Letās think about some other ways to define them.
Endotype. A subtype of a condition that is defined by a distinct functional or pathobiological mechanism. These normally occur at the cellular or molecular level.
Phenotype. The outward clinical representation of a disease or illness. It might include the age at which a person develops the condition (young, old), their appearance (obese, thin), as well as the type of symptoms present.
Pathobiological mechanism. The underlying biological pathways that give rise to a disease. There may be many different pathways that give rise to the same ādiseaseā as we currently define it today.
Precision medicine. A model of medicine that proposes the customisation of healthcare provided to the patient. Treatments are tailored to the individual based on which disease endotype they belong to.
š History
The first use of the word endotype in relation to disease classification can be attributed to asthma researchers in the late 1990s. They established that different pathobiological mechanisms were inducing similar clusters of symptoms that would ultimately impact how the disease is treated.
As clinical research and animal models progressed during the 2000s, the cellular and molecular basis for different asthma phenotypes were further elucidated, offering up new therapeutic opportunities.
Over the last few years, the promising work in the field of asthma has spilled over into other research areas, propelled by advancements in computational analysis, molecular and cell biology.
Now there is a body of published research exploring endotypes across a wide range of different conditionsā¦
š¼ Use cases
SARS-CoV-2. Severe coronavirus disease has been associated with severe acute respiratory syndrome (ARDS). The pandemic has renewed the interest in this heterogenous condition that has a range of causes alongside a broad spectrum of severity and downstream abnormalities.
Asthma. The term asthma is now considered an umbrella diagnosis for several diseases with distinct mechanistic pathways. Examples include type 2 (T2) high or T2-low. Several treatments are now approved for T2-high asthma. Endotypes do work!
COPD. Researchers are identifying new endotypes for COPD based on the inflammatory cells present in patient āsputumā (coughed up saliva and mucus). Certain cells are associated with more frequent exacerbations and can predict a good response to steroids.
Dermatitis. Endotypes are emerging for conditions like atopic dermatitis where patient clusters have been proposed based on age, ethnicities and certain biomarkers including IgE antibodies and āfilaggrinā mutations status (a gene that can predispose to dry skin).
Chronic rhinosinusitis (CRS). Endotyping of CRS helps in understanding and predicting the natural course of the disease and in choosing the adequate treatment, from pharmacotherapy to surgery and biologics
š„Ā Players
Sensyne Health. The UK-based health technology company has applied clustering algorithms to a dataset of heart failure patients in order to find clinically useful sub-groups. Although they started with just blood pressure data, they plan to expand the work to incorporate blood tests, vital signs, imaging as well as other electronic health record data.
The Alan Turing Institute. The UKās national institute for data science and AI. They have published work applying machine learning to large cancer datasets. By capturing an individualās longitudinal clinical history they have been able to make patient specific predictions for mortality and treatments based on the disease course of similar individuals in the dataset.
š® Predictions
Cluster vs endotype. Analysis of large datasets can identify useful patient clusters that can help personalise treatments. It is the addition of a unifying pathobiological mechanism to a cluster that seems to define an endotype. Although harder to define, endotypes will likely have a higher impact on patient care.
-Omics. Advances in genomics, proteomics (and probably a few other -omics) will generate data and uncover new biomarkers that help us identify and treat new disease endotypes that are yet to be defined.
Machine learning (ML). ML techniques to identify patient clusters will become a dominant tool in establishing endotypes from patient data. Large representative datasets are required however.
Longitudinal analysis. Endotypes will be formed not just from snapshots of patient data from a single time-point, but from the way an individualās condition and treatments have changed over time. Electronic health records will be critical for this type of analysis.
A new vocabulary. As the concept of endotyping cascades to other diseases and conditions, the way we describe and label diseases is likely to fundamentally change. Expect updates to controlled terminologies like āICDā and āread codesā, but also the way we describe the disease and conditions we have.
Predictions. Endotyping will improve the predictive capability of clinicians across a range of domains including disease course, treatment responses and mortality.
š Opportunities
Better clinical trials. Knowledge of disease endotypes can improve the design and yield of clinical trials. Patient recruitment can be more informed and the likelihood of decisive results will be higher.
Better drugs. A deeper understanding of the underlying biological and molecular pathways will identify potentially novel drug targets. Whether treating a specific disease subgroup provides sufficient financial incentive for pharma companies remains to be seenā¦
Better decision making. Population level data and randomised controlled trials are relatively ābluntā tools on which to base treatment decisions. Endotyping provides an important opportunity for clinicians and patients to make more informed decisions
Better data availability. The potential for patient clustering and endotyping is huge. But only if large datasets are available to researchers. The case for a more integrated health data ecosystem has never been stronger.
š Links
This report from The Turing Institute about temporal patient clustering in cancer
This case study from Sensyne Health detailing their application of clustering and endotypes to heart failure
Thatās it for this week - catch ya next time š