Horizon – A Crowdsourced Universal Health Data API

The Horizon project is a decentralized data platform and an appeal to citizens to allow de-personalized and anonymized versions of their health data currently stored in paper records to be digitized by our recurrent neural network powered EHR automation tool and further, uploaded to our blockchain based Medical Data platform. This would empower research to happen faster, administrators to target interventions better, and the healthcare industry to transition into a data driven preventive healthcare-based paradigm.

A rapidly urbanizing India with approximately 650mn people below the age of 25 and low doctor to patient ratios dreams of a preventive healthcare future to stave off its impending primary healthcare crisis. But preventive healthcare fundamentally requires that researchers, innovators and administrators spot problems before they occur. This is a tall order in a country where most healthcare data are in paper records, and doctors have a strong inertia against EHR systems, as they demonstrably add to their already gargantuan workload. India seems stuck in a double bind. 

Further, even once digitized onto an EHR or obtained from wearable devices, data is locked up in silos, and most healthcare service providers, researchers and entrepreneurs operate in a "data iceberg" scenario, with access to only the tip of the data, with most of it hidden from them, and hence only the insights that can be gained from the very limited subset. 

In order to leapfrog the "sickness care" paradigm into a wellness paradigm, India needs a privacy protected Big Data driven approach, and an open platform that provides the opportunity for organizations, individuals and public authorities to be more effective in their interventions and innovations.

In the short-run, our innovation will positively impact -

• Medical Staff - by removing overheads of digitizing prescription process to prevent errors, keep data points consistent and improve patient safety
• Patients – by providing access and control over one's medical record; inter-connected systems allow for highly effective treatment options

In the long-run, the beneficiaries of the innovation include every stakeholder in the healthcare value chain, who can use standardized and structured medical data including  

• Clinical research - to generate knowledge faster and more efficiently and accelerate innovations in health
• Healthcare Specialists & Service Providers - for building value-based healthcare, personalized medicine, and patient-centered care; dynamic infrastructure greatly reduces costs
• Administrators and policymakers - for real-time data for emergency coordination, analytics and knowledge discovery
• Drug & Pharma Companies – for clinical trials possibilities, reduces time for drug discoveries and bringing product to market

To this effect, we are currently working very closely with doctors to understand their workflows, their constraints and their support requirements through in-person meetings. Our solution approach is derived from a combination of these touch-points, learnings from best practices around the world, and drawing insights from our work in underserved markets.

The horizon PTS is a cloud based mobile patient tracking system that allows doctors to maintain patient profiles and upload medical reports, notes, prescriptions and medical imaging by simply taking a picture or scan. 

A team of medical transcriptionists then transcribe the documents and act as trainers for a Recurrent Neural Network which learns to recognize individualized doctor notes. We estimate a period of 3 months in order to train the RNN to output results with >90% accuracy. This can be sped up if the doctor is able to provide us with a large pre-existing set of prescriptions of reports. 

Horizon works on a HIPAA compliant server. Once the doctor is fully onboarded onto the platform, patients are sent a notification and a link after their visit, with a link to their medical data, and a request to share the anonymized version of their data on the horizon open data platform so that it is available for other organizations and researchers to use.

Our focus on providing the doctors with an easy to use automated EHR allows us to train a Long Short Term Memory Recurrent Neural Network to read the doctors seemingly illegible but in practice highly consistent handwriting patterns.

The blockchain based universal medical data platform provides a universal set of tools for audit and “contracts” for data access. This allows a single system to be used across geographies while conforming to varying laws, regulations etc.

Our core technology components are explained in the following - 

1. An LSTM (Long Short-Term Memory) Recurrent Neural Networks as detailed in https://www.researchgate.net/publication/311672974_Offline_Handwriting_Recognition_Using_LSTM_Recurrent_Neural_Networks

2. A blockchain based EHR like system https://www.healthit.gov/sites/default/files/5-56-onc_blockchainchallenge_mitwhitepaper.pdf

3. An FPGA based server built for LSTM RNNs that delivers both performance and energy efficiency gains detailed here for a speech recognition use case : https://arxiv.org/pdf/1612.00694.pdf

EHR systems that are not automated inevitably lead to doctor burnout. EHR automation is an inevitability. While Speech Recognition based tools exist, they usually require the doctor to dictate notes to the system either after the appointment or while the patient is present. A handwriting based system is the most seamless way to remove the burden of the EHR from the doctor’s workflow.

Still in a prototype stage, the innovation thus far has been tested by few doctors and patients across Bangalore and Delhi-NCR. The innovation aims to impact over 1 million individuals in the next 5 years.

Following are our goals - 

Pilot Phase (0 - 2 years)

On the business side, we will focus our efforts on pilots to validate the hypothesis, and gain customer validation. Through these pilots, we intend to onboard 500+ doctors and 50,000+ individuals to become regular users of Horizon beta

On the technology side, we will focus our efforts on beta release of Horizon, and the raspberry-pi based companion device "Muse". We will also create beta versions of voice profile for ML training.

Productionize Phase (2 - 5 years) 

On the business side, we intend to drive localization across diverse geographies and linguistic regions of India. We also intend to productionise "Muse".

On the product side, we intend to work on reducing the time it takes to get from 0% to 95% accuracy per doctor basis during deployment. We will be opening up API's so that other solution providers can write applications. We also intend to build integration pathways with diagnostic providers, medical reports and imaging providers. 

There are four major priority areas that we will be focusing on in the near-term for India market - 

1. Product - EHR adoption has been sporadic on account of two factors -

a. associated learning curve for doctors and medical staff, traditionally on account of legacy IT systems

b. medical staff's lack of time for data entry

2. Business Model - There is a prevalence of small-scale, single-doctor clinics across urban and rural areas. Affordability of EHR systems is a major roadblock for this segment 

3. Technology Infrastructure - Lack of adequate infrastructure and unreliable connectivity hinders deployment of universal EHR systems

4. Operations - Given the diversity of languages and regions, there is a need to contextualize solution based on these factors 

We intend to work on the above-mentioned factors in the following ways - 

1. Product - Mobile-first voice-enabled virtual assistant will seamlessly integrate with existing doctor’s workflow and capture all doctor-patient interactions, requiring no inputs or commands. This ensures that all required data is captured in the background without any learning/ active involvement of doctors or medical staff, ensuring increased efficiency for medical staff

2. Business Model - Open-source technologies will ensure drastic reduction in cost of development and remove ongoing associated costs with maintenance and upgrade; we are currently working towards providing base-features on a no-cost basis to doctors and patients, with additional functionality requiring a pay-per-use/ subscription model

3. Technology Infrastructure - Mobile-based openEHR/EMR thin-client systems are being built to run on wi-fi, cellular and offline modes. Additionally, edge computing ensures computation and automation happens on the client-side as opposed to server-side, thereby drastically reducing dependence on the network.

4. Operations - Voice-enabled virtual assistant will be initially trained through a Human-in-the-Loop (HIL) machine learning before an AI-powered automation and transcription process is activated. For this intermediate HIL step, associates proficient in vernacular languages with appropriate medical training with be on-boarded

We currently have six-full time staff members - 

1. Sidharth Sudarshan - Co-Founder & Tech Lead

2. A S Sunny - Co-Founder & Project Lead

3. Rupam Mazumdar - Co-Founder & Product Lead

4. Baisakhi Saha - Full-Stack Engineer

5. Vikram Shaw - Full-Stack Engineer

6. Kavya Katta - Full-Stack Engineer