DELPHI - Epi Forecasting

Our Mission

Epidemiological forecasting is critically needed for decision making by public health officials, commercial and non-commercial institutions, and the general public. The Delphi group at Carnegie Mellon University focuses on developing the technological capability of epi-forecasting, and its role in decision making, both public and private. Our long term vision is to make epidemiological forecasting as universally accepted and useful as weather forecasting is today.

News

• July 2020: We are hiring! Please join us for the fall as we continue to develop data and forecasting resources for the COVID-19 pandemic. We are looking for a variety of remote-friendly roles in engineering, statistical modeling, management, and external relations. We will update this list as job postings become available:

  • Senior Project Manager - Delphi-wide operations
  • Project Manager: External Relations - communications, data partnerships, and public health
  • Project Manager: Statistical Modeling and R&D - forecasters, estimation

• April 2020: We launched our COVIDcast system, which displays indicators related to COVID-19 activity level across the U.S. These indicators are derived from a variety of anonymized, aggregated data sources made available by multiple partners, and are publicly available at the COVIDcast endpoint of our Epidata API.

  Related news articles:

  • Carnegie Mellon Unveils Five Interactive COVID-19 Maps
  • Self-reported COVID-19 Symptoms Show Promise for Disease Forecasts
  • Facebook and Carnegie Mellon Team Up to Gather COVID-19 Symptom Data

• March 2020: we are focusing our efforts at this point on COVID-19 nowcasting and forecasting. We are adapting our existing systems, and developing new ones. Some of our regular activities may be halted as a result.

• October 2019: CDC has just named us “National Center of Excellence for Influenza Forecasting” (one of two nationally.)

• October 2018: and yet again! Our forecasting systems took the top spot each of the three separate flu forecasting challenges of 2017-2018 (out of up to 30 submissions). (Results summary.)

• October 2017: We did it again! Our two systems took the top two spots in the 2016-2017 flu forecasting challenge (out of 28 submissions). (Results summary.)

• December 2016: system took the top stop in the 2015-2016 flu forecasting challenge. (Results summary.)

Operational Systems

• Real-time indicators of COVID-19 Activity

• "Wisdom-of-crowds" forecasting of ILI

  Note: This system is has been repurposed to forecast ILI during the COVID-19 pandemic.

• State-level real-time estimates (nowcasts) of ILI (influenza-like illness)

  Note: This system is designed to nowcast ILI driven by seasonal influenza and is NOT designed to nowcast ILI during the COVID-19 pandemic. We are only keeping it running as others may be relying on it. We are working on a new system for COVID-19.

• State-level weekly forecasts of ILI (influenza-like illness)

  Note: This system is designed to forecast ILI driven by seasonal influenza and is NOT designed to forecast ILI during the COVID-19 pandemic. We have temporarily shut it down to focus on COVID-19. We are working on a new system for COVID-19.

Activities

We have participated, and have done very well, in all epidemiological forecasting challenges organized by the US government to date:

• Forecasting Seasonal Influenza in the US (by CDC): 2013 – current

• Forecasting Dengue in Puerto Rico and Peru (by the White House OSTP)

• Forecasting the Chikungunya invasion of the Americas (by DARPA)

Publicly Available Tools

• Epidemiological forecasting R package: epiforecast.

• Epidemiological nowcasting via sensor fusion: nowcast.

• Epidemiological time series visualizer: EpiVis.

• API for getting up-to-date epidemiological data: epidata.

  (also available via a web interface through EpiVis)

• Visual comparison of scored submissions to CDC Flu Forecasting Challenge: FluScores.

  (provide your own score files)

• Epidemiological modeling utilities (e.g., date/epi-weeks conversions): utils.

All source code is freely available on GitHub.

Members

• Roni Rosenfeld Homepage
• Ryan Tibshirani Homepage
• Logan Brooks
• David Farrow
• Maria Jahja Homepage
• Aaron Rumack
• Jingjing Tang Homepage
• Brian Clark Homepage

We are faculty, students, and staff at Carnegie Mellon University (some alumini).

We are also part of a larger, University of Pittsburgh-based MIDAS National Center of Excellence, which includes epidemiologists, virologists, public health experts, infectious disease MDs, legal and economic experts, and other computationalists.

Members of The Delphi COVID-19 Pandemic Response Team

See the list at covidcast.cmu.edu.

Research and White Papers

• Brooks. Pancasting: forecasting epidemics from provisional data. PhD thesis, 2020

• Jahja, Farrow, Rosenfeld, Tibshirani. Kalman filter, sensor fusion, and constrained regression: equivalences and insights. Neural Information Processing Systems, 2019

• Brooks, Farrow, Hyun, Tibshirani, Rosenfeld. Nonmechanistic forecasts of seasonal influenza with iterative one-week-ahead distributions. PLOS Computational Biology, 2018

• Farrow, Brooks, Hyun, Tibshirani, Burke, Rosenfeld. A human judgment approach to epidemiological forecasting. PLOS Computational Biology, 2017

• Farrow.. Modeling the past, present, and future of influenza. PhD thesis, 2016

• Brooks, Farrow, Hyun, Tibshirani, Rosenfeld. Flexible modeling of epidemics with an empirical Bayes framework. PLOS Computational Biology, 2015

• Rosenfeld. Predicting the predictable. presentation, 2013

• Rosenfeld, Grefenstette, Burke. A proposal for standardized evaluation of epidemiological models. White paper, 2012

We are thankful for generous financial support from