*Paper citing data-driven prediction of end of COVID-19 by mid-September is misleading*

This statement refers to the news articles that referenced a research article where the authors have used data analytics driven models to predict the end of COVID-19 in India by the middle of September. The said researched article has been published in Epidemiology International, Volume 5, Issue 2, 2020, Pg. No. 23-26. The authors are Dr. Anil Kumar and Rupali Roy.

Though well intentioned, the model is too oversimplified and misses many critical elements that render the model useless for any decision-making purposes. We are issuing this statement with a request that such half-baked models are avoided for any kind of use. Department of Science and Technology is working on a comprehensive all-encompassing Indian National Supermodel and we expect to have a robust model soon. Unless a model has been approved by a wide range of domain experts, it must not be used for any purpose.

We do not contest the use of data points in the model as it is clear about the limitations on the same and such use falls under reasonable assumptions.

Here are some critical components of this research paper and where they fail.

•Use of Bailey’s Mathematical Model: When Dr. Bailey created his model for epidemics with cases of removal (through recoveries and deaths), he was clear that the same has to be used only as a guiding factor, with additional of multiple local factors that were not covered in the model for any practical usage. This model itself is a simplified version of how epidemic spread can happen and does not consider specifics of the epidemic itself.

•Wrongful use of Bailey’s Relative Removal Rate: Even if a model is created using Bailey’s work, it can still be useful in certain ways. But the research in question further oversimplifies the concept of Bailey’s Relative Removal Rate and makes it equal to Total Removals upon Total Cases, where removals equal deaths plus recoveries. This is not the correct way of calculating Bailey’s Relative Removal Rate, which effectively takes into account susceptible population as well as infected, recovered and dead population, to calculate rates of infection and removal. Rate of Removal upon Rate of infection is the correct Bailey’s Relative Removal Rate.

By saying that the coefficient of 100% is reached when total number of cases equal the total number of removals, the article presents a simplified view of how to look at the situation. Conventional wisdom says when total recoveries become the same as the total number of cases, there is no epidemic as all cases have either recovered or died, unless there is a re-surge with a new base case. It does not require any modeling. The aim should always be to ensure that maximum possible cases get recovered while controlling deaths for the remaining. But this cannot follow a pattern of infection spread that does not take into account population.

•Linear Regression for Bailey’s Relative Removal Rate (BMRRR): There is no logic that explains that how linear regression is a good way of prediction of the growth of BMRRR, and yet linear regression is used. A closer analysis of the manner in which BMRRR has changed over the last few months, shows linear regression is an incorrect way of predicting its increase.

Though the research clearly mentions that it should be used with other parameters to make it effective, using it for any form of conclusion is misleading and must be avoided at all costs. So, the prediction of the epidemic ending in the middle of September is faulty and misleading, and must not be used for any decision making.