An analytical approach to the fight against COVID-19

ACT Grants is an INR 100CR grant created by India’s start-up community to give wings to ideas that could combat COVID-19 with immediate impact. Over the past 2.5 months of our existence, our aim has been to find capital-efficient, scalable solutions from NGOs and innovative start-ups which need capital to fight the spread of the pandemic and that could act as a force multiplier in today’s situation. However, with time we have realized that there is a need to prioritize solutions to target key districts which are more critically affected by COVID19. As we brainstormed the best ways to do this prioritization, we came up with the following approach which tries to find the most critical districts from a size of the problem perspective. We are sharing this on a public forum for 2 reasons:

1. We believe this could benefit other groups/governments/thought leaders/do-gooders who are focusing on solving this problem.
2. This is a first cut approach with some limitations (see below). We would love to get feedback/suggestions to overcome these limitations.

Our approach involves a 5 step analysis as given below:

1. First, we took a subset of all districts in India with more than 100 COVID-19 cases.
2. Then, we used a Bayesian implementation of the SIR (Susceptible, Infected, Recovered) model to calculate R0 i.e. effective reproduction number (how many new people does each patient infect) of the virus in each of these districts.
3. Using the daily number of new cases, and R0, we then calculated the projected number of new cases for the next 7 days assuming R0 remains constant.
4. Finally, we plotted these on a 2 by 2 chart where the X axis is the population of the district and the Y axis is the predicted number of active cases in the next 7 days.
5. We divided this chart into 4 quadrants using 45,00,000 and 500 predicted active cases after a week’s time as our key dividers. The chart we thus obtained, is attached below.

Here is how we read this chart:

1. Districts in the top right quadrant have a relatively large population (>45,00,000) as well as a relatively large number of projected active cases (>500) in the next 7 days. They thus currently are/ or likely to start seeing a strain on their healthcare system sooner than others. We call them Priority 1 districts where there is a need for immediate intervention in terms of availability of more hospitalisation facilities, ICU beds, PPEs, and home quarantine solutions to manage increasing patient load.
2. Districts in the bottom right quadrant have a relatively large population (>45,00,000) but a relatively small number of projected cases (<500) in the next 7 days These districts have so far managed to contain the spread of the virus but may see fresh outbreaks as they ease lockdown measures. They are Priority 2 districts which need to be monitored constantly. The immediate intervention needed here is large scale testing and contact tracing to keep a close tab on any spurt in cases.
3. Districts in the top left quadrant have a relatively small population (<45,00,000) but a relatively large number of projected cases (>500) in the next 7 days. These districts are seeing a rise in number of cases but given their relatively lower population we believe containing the virus is maybe easier here through lockdown as well as social distancing. We categorize them as Priority 2 districts as well. These districts could need selective intervention in terms of ICU capacity, PPEs and other solutions in case the total cases rise much higher.
4. Districts in the bottom left quadrant have a relatively small population (<45,00,000) as well as a relatively small number of projected cases (Between 100 and 300) in the next 7 days. These are Priority 3 districts as they are small in population and have contained virus spread so far. Risk of new outbreaks here is lower given the small population. We are not recommending any intervention here

There are some key assumptions which may have an impact on the accuracy of the prediction:

1. The approach assumes R0 i.e. effective reproduction number to be constant going forward. This means the sudden reduction due to efforts to reduce virus spread or outbreaks burgeoning in a district are not accounted for. We would therefore use this assumption for 7 days and would want to refresh this data every week
2. The approach does not account for external factors influencing risk/criticality like migration of workers (in or out), lockdowns being imposed or opened, new laws allowing/disallowing free movement etc. Sudden changes in policies around these factors could change the risk/criticality of a district.
3. The approach does not include the variable of existing number of hospital beds and healthcare workers. Districts with poor healthcare infrastructure could become critical much sooner.

In the light of these assumptions, we believe our approach does a reasonably good job to identify the highest priority districts in terms of scale and severity of the potential impact of COVID19. We continue to try and find improvements to this approach and welcome feedback and suggestions.

Key Contributors:

1. Shruti Deora (Senior Engagement Manager, Sattva Consulting, shruti.deora@sattva.co.in)
2. Nidhi Gupta (Senior Data Scientist, PharmEasy, nidhigupta1154@gmail.com)
3. Chaitanya Pathak (Consultant, Sattva Consulting, chaitanya.pathak@sattva.co.in)
4. Ravijot Chugh (Independent Consultant, rjchugh@gmail.com)
5. Anant Vidur Puri (Investor, Bessemer Venture Partners, avp@bvp.com)
6. Shekhar Kirani (Partner, Accel, shekhar@accel.com)

Edits — 17th August 2020

We’ve changed our criteria based on our learnings from the last month. Here is how we read this chart:

1. Districts in the top right quadrant have a relatively large population (>45,00,000) as well as a relatively large number of projected active cases (>2000) in the next 7 days. They thus currently are/ or likely to start seeing a strain on their healthcare system sooner than others. We call them Priority 1 districts where there is a need for immediate intervention in terms of availability of more hospitalisation facilities, ICU beds, PPEs, remote monitoring and home quarantine solutions to manage increasing patient load.
2. Districts in the bottom right quadrant have a relatively large population (>45,00,000) but a relatively small number of projected cases (<2000) in the next 7 days These districts have so far managed to contain the spread of the virus but may see fresh outbreaks as they ease lockdown measures. They are Priority 2 districts which need to be monitored constantly for any spurt in cases.
3. Districts in the top left quadrant have a relatively small population (<45,00,000) but a relatively large number of projected cases (>2000) in the next 7 days. These districts are seeing a rise in the number of cases but given their relatively lower population we believe containing the virus is maybe easier here through lockdown as well as social distancing. We categorize them as Priority 2 districts as well. These districts could need selective intervention in terms of ICU capacity, PPEs, remote monitoring, home quarantine and other solutions in case the total cases rise much higher.
4. Districts in the bottom left quadrant have a relatively small population (<45,00,000) as well as a relatively small number of projected cases (<2000) in the next 7 days. These are Priority 3 districts as they are small in population and have contained virus spread so far. Risk of new outbreaks here is lower given the small population. We are not recommending any intervention here.

New list of districts as per this criteria: