At a time people need facts to make decisions, position you are a health analyst working for the state of your choice. You have been asked to provide a recommendation for the Governor on what third party information will best inform your Corona Management Response Team.
My write up is following:
Suggested citation: Jin, Joon Yung, "COVID-19 Dashboard: Covid Act Now", June 30, 2020. Available at http://jinjjan.blogspot.com/2020/07/mba-case-studycovid-19-dashboard-covid.html
As the health analyst of the government of Texas, I recommend Covid Act Now (CAN) as the main COVID-19 dashboard of the state. To make a decision and set an action plan, predictive and prescriptive analytics (as opposed to descriptive analytics) with solid flow data sources are critical [1]. To be a strong predictive/prescriptive analysis, substantial modeling should be placed. This means among the data analysis process of discovery, wrangling, profiling, modeling, and reporting [2], the last two steps are important to provide an insight for the future. Let’s see how CAN is doing so.
The modeling method of CAN is called SEIR, susceptible (S), exposed (E), infected (I), and recovered (R), which tracks the flow of people between four phases. The model analyzes the propagation of virus with a deductive modeling which is supported by mechanics [3, 4]. Another representative model used by other dashboard is Curve Fitting model which fits the virus propagation to the growth curves of prior regions, such as China and Italy, and this is an inductive modeling (Exhibit 1). Although both models are valid, given that every single factor affecting the propagation is specific to each region, we should adapt SEIR model to anticipate fundamental progress of infection.
Based on SEIR model, CAN provides predictions with which we will be able to set prescriptive policies accordingly. The most representative graph giving substantial insight is Future Hospitalization Projections but all the other analyses/predictions are also very important for the government. In Future Hospitalization Projections, we can see the future trajectory of hospitalization with available beds under two scenarios, “If all restrictions are lifted” and “Projected based on current trends” which is most crucial indicator for us to make a decision on reopening (Exhibit 2). Another graph showing positive test rate also delivers significant message (Exhibit 3). Most other dashboard simply show positive case numbers and death rate based on the case numbers, but this misleads the statistics. If there were many people not afford to be tested, the death rate should be exaggerated ignoring pseudo-positives. Therefore, the analysis of positive test rate is something we should focus on as it is the key insight on how we should deal with the testing capacity and reopening plan. Based on the information from the analyses described above, Texas should delay restriction lifting until the infection rate falls below at least 0.9 (“low”) which is currently 1.19 (“high”) to meet hospitalization demand. And Texas should grow the testing capacity until positive test rate decreases below 3% which is currently 12.3% as of June 27th 2020.
Another virtue of CAN is data sources on which they make analysis based. The data come from a number of sources, including The New York Times, and are updated daily. If the data were tracked once a day to produce daily figures, it would be stock data [5]. But in this case, the sources trace all the figures in real-time and just report them once a day, it can be regarded as flow data. Since it uses big data from different sources with tremendous tracking, data have volume and variety. It also has velocity with real-time flow data and appropriate analysis, and veracity as from institutions with public confidence (e.g. NYT) with strong public partners (e.g. Stanford University).
CAN, however, has some limitations of data display. Most of the data they show are rates instead of absolute numbers. Analyses from rate sometimes ignore the life cycle of a phenomenon. It would be better for them to include absolute numbers so that we could understand the situation in a timely manner. Also they lack information of other countries which gives us another approach of projections. In this case, showing projections with the Curve Fitting model with SEIR side by side will provide insights from various angles. And to see if the projections are over/under-estimated, adding margin of error in graphs would help, but decision makers would be required to follow the overestimated cases.
Exhibit 1. Representative Modellings of COVID-19 Dashboards
Exhibit 2. Future Hospitalization Projections of Texas (CovidActNow)
References
1.
Thomas
H. Davenport (2015, October 21). 5 Essential Principles for Understanding
Analytics. Retrieved June 27, 2020, from https://hbr.org/2015/10/5-essential-principles-for-understanding-analytics
2.
Kandel,
S. (2014, November 02). The Sexiest Job of the 21st Century is Tedious, and
that Needs to Change. Retrieved June 27, 2020, from https://hbr.org/2014/04/the-sexiest-job-of-the-21st-century-is-tedious-and-that-needs-to-change
3.
Compartmental
models in epidemiology. (2020, June 23). Retrieved June 27, 2020, from https://en.wikipedia.org/wiki/Compartmental_models_in_epidemiology
4.
Covid
Act Now & IHME: Why Two Models Are Better Than One. Retrieved June 27,
2020, from https://blog.covidactnow.org/covid-act-now-ihme-why-two-models-are-better-than-one/
5.
Thomas
H. Davenport, P. (2012, July 30). How 'Big Data' Is Different. Retrieved June
27, 2020, from https://sloanreview.mit.edu/article/how-big-data-is-different/
댓글 없음:
댓글 쓰기