PreSco for Neonatal Sepsis Prediction

Neonatal Sepsis is a condition that affects newborn babies and is hard to diagnose.

Causes – Delayed diagnosis, limitations of current blood and culture tests (non-specificity and non-sensitivity), low access to specialised healthcare in rural areas and low affordability are the key issues we are addressing.

Our Solution - Our integrated cloud-enabled platform uses machine learning algorithms for prediction of Neonatal Sepsis infections to address the shortfall of resources in rural areas. The platform has a referral system that connects urban health infrastructure with rural areas. It also provides a cost-effective electronic medical records platform. The entire platform is cost effective, easy to use and can be effectively used by health volunteers for making informed decisions and connect with urban health expertise with our algorithms and integrated features. It can be easily deployed by frontline healthcare volunteers to increase their efficiency and improve health parameters in rural and low resource areas.

Every year, 25 million babies are born in India and 10 million of them are at the bottom of pyramid. Six lakh babies die annually due to neonatal infections and 30% are due to Neonatal Sepsis. Neonatal mortality rate (NMR) is 25 while it is 1 or 2 in developed countries.

The current gold standard for confirmatory diagnosis of neonatal sepsis is culture test and are expensive and inaccessible. Also, test results are available only after 48 to 72 hours. During this period, empirical antibiotics are administered. Additionally, drawing at least 1 ml blood from a neonate is a challenge and not every neonate gives the sample successfully in one go. Multiple punctures increase vulnerability of a new born infant to infections. Existing methods such as blood tests (C-reactive protein or CRP) are non-specific with limited diagnostic accuracy, throw up false positives leading to antibiotic overuse. Digital health platforms that aid frontline health workers in rural areas for timely and accurate diagnosis can fill this gap.

Our solution is a digital health platform (web/mobile) that can uses machine learning techniques for clinical decision making. They can be easily used by health volunteers for an early assessment of a baby’s condition.

Our Solution – Our platform is called PreSco, Predictive Scoring. It is a cloud-based application for early detection of Neonatal Sepsis. It can be accessed from any device (mobiles/desktops). It uses data analytics and machine learning algorithms for an early assessment of a baby’s condition. This solution is ideal for low resource countries like India where 80% health specialists are concentrated in urban areas.

Technologies - Cloud, open-source technologies, data analytics and machine learning algorithms form the core of our platform.

Process - Clinical data is entered into the application and gets transmitted to a cloud server where our machine learning algorithms are housed. They process the data and return a computed predictive risk score to the device with colour coded risk assessments for ease of understanding of a baby’s condition - Red for high risk, Orange for medium risk and Green for low risk. Referral option helps to obtain expert advice from a far-off health centre or specialist. 

Primary Target – Mothers and Babies - Covid-19 crisis has severely impacted the delivery of health services and has reversed the progress made in the past few years w.r.t several health indicators listed in the UN sustainable development goals. Young mothers and infants found it hard to commute to receive quality healthcare from an urban health center during the COVID-19 pandemic. A sizeable number of frontline volunteers are working in the mother and child space in India. However, they are not equipped with the necessary tools (software) to records, identify and triage the affected babies. Moreover, lack of seamless connect with specialist doctors of which there is an 80% shortage, are concentrated in urban areas. Our application is designed for use with minimal training by a large force of frontline health volunteers (ASHAs). There are about 800,000 ASHA volunteers in India that serve the bottom pf pyramid segment.

Secondary Target -

a. Rural Clinics and healthcare professionals

b. Tertiary hospitals 

Predictive Algorithms – The uniqueness of our digital health platform is centred around three models of Predictive algorithms that are customized as per the availability of resources at a healthcare centre. They work with a combination of non-invasive (observable clinical parameters) and invasive (blood reports, X-rays etc) health data as inputs to the algorithms.

• Frontline health volunteer model is primarily a non-invasive model.
• PHC or primary health model and
• Tertiary level model uses both non-invasive and invasive clinical parameters.

The platform provides the below features –

• Upload of Laboratory reports and images
• Upload of any kind of other reports such as diagnosis notes, clinic notes etc
• Storage in cloud server and easy retrieval upon approvals
• User friendly menus for entering and storing information on vitals, baby history, episodes etc
• Transmission of all of the above reports and parameters to a referral doctor for speedy action Security and anonymisation

Child birth and child care are viewed as women’s issues. Additionally, most frontline volunteers working in the maternal and child space in India are women. Our platform works be taking in vitals and critical health information (reports or images, if available) of a baby by a health volunteer. She can then use the platform’s machine learning platform PreSco for making a quick assessment of the baby by computing a risk score. The score can be used at multiple intervals. The parameters entered can be transmitted to a far-off specialist for a second opinion, and treatment initiated, if required, to reduce time gap.

We have developed a prototype of an integrated cloud platform for predictive risk scoring of neonatal sepsis. It consists of a data collection application and machine learning algorithms for three different levels of risk scores. An ensemble of machine learning algorithms was used to build the predictive model using about 100 input parameters. Feature engineering was undertaken to identify the most critical parameters. Performance metrics were computed and the three risk ranges: Low – Green (score range 0-5), Medium – Orange (score range 5.1–8) and High – Red (score range 8.1–10).

We are currently testing our platform at two large tertiary health centres for mother and babies in India. Our platform is being used by about 20 health professionals at different levels, including nursing staff to technicians to doctors. We have collected data from two large hospitals and have developed machine learning algorithms with close to 80% accuracy. Our current Pilots are running in these hospitals. 

Data Analytics – Our platform has a wide range of data analytics dashboards that can be used by clinicians to assess and track the health of babies. These are available on both web and mobile versions or our application.

• Descriptive analytics – analyses retrospective data.
• Predictive analytics – provides a future analysis through a risk estimate using prospective data.
• Prescriptive analytics – recommends treatment actions based on data of babies.

Data Used - PreSco uses a vast range of maternal and child clinical parameters as inputs for our algorithms. Observable or non-invasive data such as colour of skin, breathing rate, abdominal distension, mother parameters etc and invasive clinical parameters such as heart rate, neurological signs or seizures, signs of pneumonia, diarrhoea etc are used for building a universal application that can be used by healthcare professionals for rapid identification of neonatal infections. Several symptoms of neonatal infections such as shortness of breath, pneumonia, fever etc overlap with COVID-19 viral infections and similar epidemics and can lead to misdiagnosis of either of the conditions and administration of wrong treatment - an antibiotic is not required to treat a viral infection but bacterial infection such as sepsis requires an antibiotic. Our platform aids in effective diagnosis and triaging of infections. It is easily scalable to other epidemics and infections as it covers more than hundred clinical parameters.

Currently – About 20 (internal testing phase)

One Year – About 100 Health Volunteers and address close to 1000 babies

Five Years – About 1000 Health Volunteers and address close to 100,000 babies

Processes – We have well defined processes in place for monitoring and evaluation such as setting Project Goals and Objectives, Defining Indicators, Data Collection Methods, Data Analysis and Reporting, Dissemination of Information through communication protocols agreed.

Data Quality Assurance Mechanisms – We follow data quality assurance plan with defined process for ensuring data validity, integrity, reliability, precision and timeliness.

Frequency of Appraisals – Daily basis for project operations, Monthly basis for comprehensive project review and quarterly for review of project goals and impact.

Results Framework – We follow a well-defined template & results framework in agreement with our funding partner. The results framework includes a comprehensive performance management plan for monitoring of key project goals and objectives as agreed with our funding partners.

Tools that project uses – We use several project tools for project execution and monitoring such as Github, Code Climate, MSP etc.

Measurable Indicators -

Percent of Target Data Collected from Collaborating Hospitals - We are targeting to collect at least 5000 data points from 3 collaborating hospitals in 24 – 36 months’ timeframe.

Percent Achievement of Key Indicators for Sepsis Score - We target to achieve 80 - 90% for key metrics such as Accuracy, Sensitivity, Specificity.

Our team consists of 6 members - 3 full time and 3 part time. We have 2 senior advisors (Neonatologist & Data Analytics) to guide us on clinical and machine learning and analytics.

Dr Hyma Goparaju, CEO & Co-Founder, Avyantra Health Technologies (Full Time)

KVKLN Rao, COO, Founder, Avyantra Health Technologies (Full Time)

Salman Sheik, Research Coordinator (Full Time)

Shreya Yadav – UI/UX Consultant

Prashanth Krishnan – Technical Architect

Dr Gothai Nachiyar S – Research Coordinator, KKCTH Hospital, Chennai

Dr M Alimelu, Head of Department, Niloufer Hospital, Hyderabad (Advisor)

Dr U Dinesh Kumar, Professor of Statistics & Head DCAL, Indian Institute of Management, Bangalore (Advisor) 

The project lead, Hyma Goparaju is Founder-CEO of Avyantra Health Technologies, a startup she founded in 2017 in India. She is a PhD and an MBA with a industry experience of close to 20 years. Under her leadership, Avyantra Health Technologies graduated from the Unicef Innovation Fund’s global cohort of 2019 focused on Data Science.

KVKLN Rao is a Fellow in Healthcare Entrepreneurship – Indian Institute of Technology, Hyderabad, 2017. He completed his one-year Business Analytics & Intelligence (BAI) certification programme from IIM Bangalore in 2016. He is also pursuing Part-Time PhD from IIM Shillong. He is a B.Tech and MBA, industry professional and Medtech & Digital Healthcare Entrepreneur with over 24 years of experience in Manufacturing and IT (Business Intelligence and Analytics). He is co-founder of the company.

Team members’ details have been shared in the above section. They have diverse backgrounds from information Technology, software development, business analytics and quality assurance methodologies.

Advisor and mentors include professionals from academics and hospitals. Subject matter and analytics professionals from academics provide necessary inputs for finetuning the platform to suit the needs of market and business. Specialist doctors guide and provide clinical inputs for building and validating the platform.

Winning the Challenge will help us overcome several barriers that we have been facing in the recent times.

Financial Barrier – The fund will help us to meet our financial needs and provide the much required budgetary boost to support our team.

Technical Barrier – The fund will help us explore the technical innovations in the area.

Legal Barrier & Policy Barrier – Fund will help us to make our product and platform complaint with domestic as well as international regulations on data privacy and protection of patient data.

Cultural Barrier – Since we are working with several partners, the Trinity Fund would be able to help us overcome cultural barriers as we walk through our partnerships.

Market Barrier – Attention from various organisations and governments across the globe to help market our solution better.

Infrastructure Barrier - The fund will help us to enhance the technical capabilities of the product as well as platform, overcome infrastructural barriers and make the application responsive in low resource settings.

Resources Available – The fund will most certainly help us get access to better resources, be it manpower, infrastructure, collaborating partners and take us towards our scale-up and commercialization goals.

Academic institutions, Hospitals, Health Organizations, Corporates etc.

Financial Barrier – We are in need of consistent funding support to enhance the features of the product, complete testing and validation and to ready our product for commercialisation and launch. We are working with several funding agencies, both domestic as well as international ones, to meet our future funding requirements.

Technical Barrier - Data availability in required digital formats for retrospective study as well as conversion of unstructured medical data to structured data for analysis is a complex task. We are planning to overcome this challenge with advanced technical architecture for big data, image analysis and data mining techniques.

Legal & Policy Barrier - Data privacy laws are becoming stringent globally and we are taking steps to adhere to data privacy laws to ensure safety and privacy of patient data.

Infrastructure Barrier - Lack of required internet bandwidth in developing countries and less developed countries could be an issue in the short term. However, since telecom infrastructure is improving in many developing countries we expect it will support usage of digital products through mobile applications.

Resources Available - We are pursuing our project goals with support from international agencies like UNICEF and government support programmes for start-ups.  

Hospitals, Health Organizations, Corporates, Teaching Hospitals, Academic institutions etc

Testing & Validation with Health Volunteers

Pilots

Precommercialization readiness

Testing & Validation with Health Volunteers

Pilots

Precommercialization readiness

Testing & Validation with Health Volunteers

Pilots

Precommercialization readiness

Testing & Validation with Health Volunteers

Pilots

Precommercialization readiness

Testing & Validation with Health Volunteers

Pilots

Precommercialization readiness