Predictive Analytics – Uncovering the Hidden Value in Human Signals

Vinnie Ramesh — October 20, 2015

Predictive analytics is a hot area of research, and is increasingly gaining traction in several industries. While nothing new to healthcare, what has changed is our ability to better measure, aggregate, and make sense of previously hard-to-obtain or non-existent behavioral, psychosocial, and biometric data. Combining these new data sets with existing data (claims and EMR) allows us to gain unprecedented insight into the interaction between external factors (the environment) and human signals. In particular, understanding a patient’s environment in addition to biological and clinical factors has the ability to change how we approach clinical medicine, population health, the healthcare system, epidemiology, and pharmaceuticals. 


In addition to the existing data sources around EMR and Claims data, new digital tools allow us understand an individual’s lifestyle, preferences, and events between hospital visits as they interact with their surrounding environment. While we’re fairly good at capturing what happens to a patient in the hospital, we want to understand what’s happening to individual patients outside of the hospital and outside the healthcare system. Advances in networking, storage, and computing power allow us to start sensing human mobility patterns from mobile phones, real-time activitybiometrics, and population psychosocial wellbeing. While we’re not quite at the level of millions of tiny devices constantly measuring the ambient environment, the prevalence of smart phones, billions of data points of internet activity, and rising popularity of wearable tech are already enabling us to more accurately sense human signals. 


Human signals are messy. Our ability to make this data useful is dependent on our ability to understand it. Traditionally, predictive analytics in healthcare has relied on working with highly structured data sets, but we’re now interested in understanding unstructured data sets. Advances in natural language processing, information retrieval, and machine learning enable us to understand, structure, and infer meaning from large bodies of text, from images and videos, from time series, and from data sets with millions of data points and variables.  Current healthcare analytics operate with an extremely low sampling rate on human signals as they mainly focus on EMR and Claims data. While these contain useful information, they are updated fairly infrequently and we lose out on gaining a full picture of a patient’s daily lifestyle, their interactions with their provider, and their interactions with the healthcare system. Combining EMR and Claims data with these unstructured signals creates the more continuous picture we seek. 


Generating new data and trying to understand it is only useful if we can do so quickly and at scale. We want to make inferences over millions of patients and billions of data points – it quickly becomes impossible for humans to understand how observed and latent variables are related, which variables are important, and which ones are just noise. To do this not only requires powerful algorithms and inference techniques, but also immense amounts of cheap computing computer. Amazon, Google, and Microsoft have established themselves as leaders in cloud computing – this means lots of cheap, on-demand, computing resources for handling our data needs. Combining this computing power with appropriate computer systems allows us to process our deluge of data and provide meaningful inferences and insights. 

While these approaches have been successful in several other industries, they haven’t really seen the light of day in healthcare – the best engineers in the world typically don’t get exposed to the problems in healthcare, and thus aren’t enabled to solve them. Without giving away too much yet – a majority of our engineering work at Wellframe centers around combining the above areas: generating a more continuous stream of human signals outside of the healthcare system, combining these signals with existing clinical data, developing techniques for understanding these, and building the computer systems for doing so at scale. We believe that building the best systems for generating, understanding, and utilizing human signals requires exposing talented engineers to clinical medicine, human biology, and the inner workings of the healthcare system. If you’re interested in learning more about our approach, or curious as to what else we’re building, please reach out to me:

Vinnie Ramesh

Vinnie Ramesh

Chief Technology Officer at Wellframe

Minor Planet 24376 Ramesh was named after him in recognition of his work on mesh networks.