This document describes a long-term forecasting model developed by Megaputer Intelligence to monitor and predict the spread of COVID-19 over the next 2 months in the United States. The model forecasts case numbers, the growth in severe cases, ICU bed and ventilator demand, and monitors resource usage at the state and county level. It uses machine learning algorithms and epidemiological and medical resource data to generate interactive forecasts and maps to help leaders make informed pandemic response decisions.
2. THECHALLENGE OFFORECASTINGTHEEPIDEMIOLOGICAL SITUATION
In the context of the COVID-19 pandemic, a key objective of the national
health system is to prevent its critical overload due to the exponential
increase in the number of hospitalizations of patients and, as a result, a large-
scale increase in the demand for the use of limited resources:
• intensive care unit (ICU)
• artificial ventilation of the lungs (AVL)
• testing, equipment
• medicines
• disinfectants
• staff
TASK 1
Visual presentation and monitoring of current indicators
of the spread of the COVID-19 virus.
Prediction of the volume of new cases and the
percentage of those cases that will develop into severe
(critical) cases.
TASK 2
Monitoring the actual usage of ICU beds, ventilators, and
other resources.
Prediction of the demand trend versus the amount of
available ICU beds, ventilators, and other resources.
A long-term forecast is needed to
reduce the risks of an increasing
rate of viral infection spread,
determine the optimal resource
allocation, and make
management decisions at the
federal, state, and
municipal levels.
POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
3. • In the context of the negative developments surrounding COVID-19 infection spread, it is imperative for leaders to act judiciously and make
informed decisions using the most complete and reliable information available.
• Megaputer Intelligence (www.megaputer.com) has developed a long-term forecast model with a horizon of up to 2 months for
monitoring and predicting the pattern of morbidity, the growth of critical disease cases, and the use of ICU beds, mechanical
ventilation devices, and other resources.
• The system downloads up-to-date historical information about the course of the disease, as well as the volume of ICU beds. Using
built-in machine learning and predictive analysis algorithms, the system provides an informational picture for the following
indicators:
FORECAST THE OVERALL
DISTRIBUTION OF COVID-19
INCIDENCE
FORECAST THE GROWTH IN THE
NUMBER OF CRITICAL/SEVERE
PATIENTS
FORECAST THE REQUIRED VOLUME OF
ICU HOSPITAL BEDS
FORECAST THE NEED FOR
VENTILATORS
The system presents the results of predictive modeling in the
form of a visual, interactive dashboard with the ability to display
in various tabular formats. An in-depth analysis allows the user
to view the picture nationwide and to obtain “detailed”
forecasts at the state and local level.
The system makes it possible to simultaneously view and
analyze the forecast for the number of total cases, new
cases per day / week / month, percentage of ICU bed
and mechanical ventilation capacity utilized, and the
predicted time period for the onset of a “pessimistic”
scenario involving the exhaustion of these resources.
POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
FORECASTMODEL :ASOLUTION BASEDONTHEPOLYANALYSTPLATFORM
4. The forecast model takes into account “diffusion”, or the exchange of infected carriers between territories (region / state /
municipality, etc.), and uses the following set of data for each territory obtained from available open sources:
COVID-19 incidence data for
the previous period:
• Confirmed cases
• Lethal outcomes
• Recovered
“Profile” of the analyzed
geographic region:
• Population
• Area of residence
• Coordinates of the geographical
center of a region.
Available medical resources:
ICU beds (with the presence of mechanical
ventilation).
Data that can be applied to improve the accuracy of the model:
• population distribution by age and gender
• incidence rates by age and gender
ICU beds and ventilation devices are vital
resources, but for complete control and
forecasting of the situation, it is necessary to work
with additional data on medical resources:
• number of hospital beds
• number of doctors
• number of nurses
POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
INPUTDATAANDFORECASTMODEL
5. FORECASTINGTHENUMBEROFCASES
POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
Choice of date to view
either the actual or
forecast data
Heat map of the geographical distribution
and number of cases. State /
municipality scale, selectable
The number of confirmed,
active, recovered and fatal
outcomes.
The forecast prognosis of cases
according to the required
parameters
6. POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
Choice of date to view
either the actual or
forecast data
Heat map of the actual /
forecasted share of the involved
ICU bed fund
The number of actual / predicted critical
cases of the disease and the proportion of
ICU bed capacity utilized
Time course of the actual /
forecasted share of ICU bed
capacity involved
FORECASTING THEVOLUME OFICUBEDUSAGE
7. POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
Heat map forecasting when ICU bed capacity will be reached Data table listing:
• Region
• Counter indicating
the term for ICU
bed resource
exhaustion
• Number of critical
cases (occupied
beds)
• Total number of
beds
• Share of occupied
beds
FORECASTING DURATIONUNTILEXHAUSTIONOFICUBEDRESOURCES
8. POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
Heat map of the total
number of cases
Data table listing:
• Region
• The number of
confirmed cases of
the disease
• Number of deaths
• Mortality rate
• Total number of
beds
• Share of occupied
beds
Dynamics of total confirmed/
new cases / deaths
The number and growth of sick, cured,
and fatal cases
SNAPSHOT OFTHEACTUALEPIDEMIOLOGICAL DATA
9. Interactive maps, charts,
and diagrams
Uploading forecasts for all parameters in
CSV and Excel formats
POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
CONVENIENT SIMULATION RESULTFORMATS
10. WHAT’SNEXT?
Other countries:
The model can be reconfigured for any country / territory if you have the
source data necessary for forecasting.
Test Model for the USA:
Development period: 3 days.
Sources of open data used in the model:
JHU, CDC, NYT, Wikipedia
Pilot developed
using open data
Upon request
POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
11. • Established: 1997
• Funding: Self-funded through organic growth
• Headquarters: Bloomington, Indiana
• Current size: over 100 employees
• Customer base: 300+ Customers (globally)
Megaputer – Bloomington, IN 30,000 sq ft
ABOUT MEGAPUTER INTELLIGENCE INC.
POLYANALYSTPLATFORM2020MEGAPUTERINTELLIGENCE
Identify and extract facts from text documents
We build models based on analytics and
Artificial Intelligence
We provide a cluster platform for Big Data
We digitize and automate business processes
We support a quarter of Fortune 100 companies
Headquartered in Indiana, export to the whole world
12. PRESENTED BY:
Megaputer Intelligence
Call (812) 330-0110
or email info@megaputer.com
Megaputer Intelligence, Inc.
1600 West Bloomfield Road, Suite E
Bloomington, IN 47403 USA
POLYANALYST PLATFORM 2020 MEGAPUTER INTELLIGENCE
Editor's Notes
Megaputer was first established in 1997 and was one of the first text analytics tools on the scene in the late 90s when companies were first coming out with their basic lexical analysis and bag of words approaches. Today we can better extract the meaning of documents through understanding the semantics.
Megaputer has been organically growing year-over-year and has office locations in both the US and Russia so that we can work round the clock with over 130 employees.