What you need to know about the IHME Model…

This is the model on which the Coronavirus Task Force has most relied.


According to the WSJ and other sources:

White House coronavirus coordinator Deborah Birx said its assessment of how the pandemic would unfold closely mirrors the University of Washington’s Institute for Health Metrics and Evaluation, the so-called Murray Model.

An early-on Murray Model’s ‘most likely’ forecast was 83,967 COVID-related deaths during this epidemic cycle … with the 95% confidence interval ranging from  38,242 to 162,106.


Underlying that forecast, the Murray Model estimates that Daily New Deaths (DNDs — the number that we’ve been tracking) will  peak at about 2,200 in mid to late April.


Here are some of the key components and assumptions in the Murray Model…


The model’s primary objective is not to forecast deaths, but rather to forecast the need for hospital treatment capacity: beds, ICU beds, ventilators, healthcare professionals.

Note: The model assumed that 54% of ICU patients would be put on a ventilator.

The model’s “base case” assumption is that “aggressive social distancing measures are enacted and sustained.”  That recognizes that “individual behavioral responses and government-mandated social distancing (school closures, non-essential service closures, and shelter-in-place orders) can dramatically influence the course of the epidemic. “

Note: The models that are flashing worst case fatality numbers over 2 million make no such assumption.

But, the model understandably  does not consider the potential impacts of new therapeutic protocols (think: Hydroxychloroquine) that could materially reduce hospitalizations and deaths.


In essence, the model estimates COVID deaths based on the ratio of deaths to population (not by reported confirmed cases) by age category.  The ratios are analytically determined based largely on the experience in earlier infected countries (e.g. Italy).

Note: Appropriately, the model does not work from ‘”reported confirmed cases”, recognizing (as we have) that that number conflates infection incidence and the level of testing.

“Deaths are likely more accurately reported than cases in settings with limited testing capacity where tests are usually prioritized for the more severely ill patients.”

The model uses sequencing and time-phasing techniques to spread the fatality projections over time … and, disaggregates the national numbers down to state levels based on each state’s initial incidence (when the first cases were reported) and the age composition within each state.


The researchers are explicit about what factors they have included in the model … and which factors we not included:

The consequent main limitation of our study is that observed epidemic curves for COVID-19 deaths define the likely trajectory for US states.

In this study, we do include a covariate meant to capture the timing of social distancing measures.

And, our models explicitly take into account variation in age-structure, which is a key driver of all-age mortality.

But these efforts at quantification do not take into account many other factors that may influence the epidemic trajectory:

  • the prevalence of chronic lung disease
  • the prevalence of multi-morbidity
  • population density
  • use of public transport

… and other factors that may influence the immune response.

And again, the model understandably  does not consider the potential impacts of new therapeutic protocols (think: Hydroxychloroquine) that could materially reduce hospitalizations and deaths.


In my opinion, the model is well done — i.e. much better than some others that have been bandied around — especially since the modelers were working under tight time constraints and with highly problematic data.  It is a good first effort and a staring point for further refinements to the model and its forecasts.



For more detail, click to see the Murray Model source document .  It contains the modeling detail and discussion of the policy implications.  Great reading for quant jocks and policy people.


Follow on Twitter @KenHoma

>> Latest Posts


9 Responses to “What you need to know about the IHME Model…”

  1. Deepak Gupta Says:

    An interesting view

  2. TMK Says:

    New Orleans death rate 2X New York. I’d view the New York rates as best case scenarios – relatively healthy, wealthy citizens with a fresh hospital system. What happens to rates in NY once you hit capacity? What will death rates be in rural areas that have less wealth and more underlying chronic medical problems? How about for people in a food desert who have no choice but to ignore social distancing to get food? What about increased death rates for people being pushed out of hospitals by Covid surge? Just like the financial crisis, the timing and location of cases will change the outcomes. Our ignoramus President failed to act or take the proper posture to warn Americans. Hard to imagine we will have the “best case” outcome given the complete dis-function of this administration. You really want your life in the hands of Kushner and Pence? Good luck!

  3. “Pandemics naturally thrive most in big cities” | The Homa Files Says:

    […] See: ‘Most Likely’ Forecast: 84,000 COVID Deaths … […]

  4. Some determinants of urban viral contagion… | The Homa Files Says:

    […] See: ‘Most Likely’ Forecast: 84,000 COVID Deaths … […]

  5. Why have the model’s forecasts dropped so much? | The Homa Files Says:

    […] more detail on the IHME Model, see our previous post:‘Most Likely’ Forecast: 84,000 COVID Deaths […]

  6. April 11: COVID-19 Tracker | The Homa Files Says:

    […] ‘Most Likely’ Forecast: 84,000 COVID Deaths … […]

  7. April 12: Covid-19 UPDATE | The Homa Files Says:

    […] What you need to know about the IHME Model… […]

  8. About that forecast of 60,000 coronavirus deaths… | The Homa Files Says:

    […] See also: What you need to know about the IHME Model […]

  9. How many people are walking around infected with Covid-19? | The Homa Files Says:

    […] The model and site are the product of the University of Washington’s oft-maligned Institute for Health Metrics & Evaluation (IHME) which generated many of the early inaccurate Covid forecasts (see: What you need to know about the IHME Model) […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s