Revisiting Our Covid-19 Predictions in Florida: What We Got Right, What We

Missed and What’s Next

Authored by Ayesha Rajan, Research Analyst at Altheia Predictive Health

Introduction

In our last article we reflected on our predictions regarding the spread of Covid-19. We made observations and predictions in March using linear regression and found that our data had an error rate of less than 10% when looking at the American population as a whole. However, many states had different rules and regulations for social distancing, mask wearing and other precautions, so we also made predictions based on different levels of regulations to predict the impact Covid-19 on hospitals and ICU beds. In this article, we will explore the accuracy of those predictions for the state of Florida and tools that can help us going forward as many states begin to see their number of Covid-19 cases spiking again. 

 

Analysis and Interpretation 

In our analysis we made the following predictions for the state of Florida based on different levels of precautions taken:

Predicted Hospitalizations with All Precautions Taken

April

May

June

Mid-June

July

August

2,208

2,028

18, 251 

19, 264

10, 139

2,208

 

Predicted Hospitalizations with Some Precautions Taken 

April

May

June

July

August

2,208

10,139

35,487

10,139

2,028

 

Predicted Hospitalizations with No Precautions Taken

April

May

Mid-May

June

Mid-June

July

August

2,208

20, 278

55, 766

40, 557

10, 139

3,042

2,028

 

While looking at this data, keep in mind that the number of hospital beds available in the state of Florida is 55,727. Looking back on our data, it’s clear that our model was very conservative. On one hand, we did accurately predict peaks and the fact that Florida would be overwhelmed with Covid-19 cases, however, we predicted that even with no precautions we would begin to see this come to an end in August. As we come to mid-July, Florida is the latest state to break American records with 15,000 new Covid-19 cases in one day which indicates that the end of the battle against Covid-19 in Florida is nowhere near over (Linton). In fact, as of July 7th, 2020, “at least 56 intensive care units in Florida hospitals reached capacity” with another 35 showing less than 10% availability (Chavez). 

 

The natural question to ask when making models such as these is how to improve them. In some ways, our model was conservative – in a no precautions setting, it was almost impossible that Florida could be in back in the same place it was in April. However, some error can be attributed to the lack of knowledge regarding the spread of coronavirus in March. We created these predictions based on Imperial College London’s pandemic model, however, the pandemic model could not account for the various changes, such as when and how certain things would reopen and was also based on the fact that we had a predicted end date. It was also limited in the fact that it was an agent-based model and not a stochastic model, which accounts for randomness such as the random meeting of two people and the impact they would have on their community if they were infected with Covid-19. The most accurate prediction models take in a lot more information about a more concentrated population. For example, it would not try to make every studied population follow the same multipliers because it would consider much more specified information about that population such as the occupations of a population, their ages and activities, their political engagement and beliefs, their methods of transportation, their attendance of religious functions and even more. 

 

Conclusion

What we can take away from this interpretation of data is that there is not a one size fits all model for pandemic predictions. Different counties, cities, states and countries all follow different schedules and habits and while most hospitals in Florida are overwhelmed, there are also likely some that barely see any or limited Covid-19 related issues. For this reason, it is extremely important to consider a lot more information about a more specified population than to use broad, blanketing equations; especially when the data is used for resource allocation. As we continue to battle Covid-19, it is important to take precautions such as wearing a mask in public places, practicing social distancing and maintaining good hygiene. 

 

Prevention

Take a look at the image below to see low to high-risk situations and understand how you can limit your exposure to Covid-19.