Forecasting vs. Dynamic Models

00:01 Let's talk now about the differences between forecasting and dynamic models.

00:05 Forecasting models use real data.

00:10 Actual incidence numbers.

00:12 And it predicts how an epidemic is going to look in the near term, in the next few weeks.

00:21 A dynamic model, on the other hand, uses broader indicators like the case fatality rates, observed in other populations, maybe we use some curve fitting.

00:31 We assume that the epidemic will take a certain shape and we apply these indicators to make resemble our population.

00:39 Now, dynamic model is meant to predict how an epidemic will behave in the longer term more than just weeks, perhaps months.

00:51 So, here is an example of a forecasting model for COVID-19 daily deaths in New York State, in the middle of 2020.

01:01 The red line shows us the actual deaths, and the blue line shows us what's projected to happen.

01:10 So, this would be a forecasting model or what's happening in the next few weeks.

01:15 So, the forecasting model, done in April 20 predicted they would see 100 deaths a day by May 1st.

01:23 They actually saw 300 deaths a day.

01:25 And they will see fewer than five by May 27.

01:28 They actually saw 100 by the end.

01:30 So, in this case, the forecast was off, which often happens, a model is very often wrong.

01:37 But again, sometimes it's useful in the ways in which it is wrong.

01:43 This is an example of what's called the IDEA model.

01:53 This is the IDEA model of Ontario.

01:57 IDEA is I-D-E-A and it stands for the Incidence Decay and Exponential Adjustment.

02:04 And it was a model devised by Dr. David Fisman of the University of Toronto.

02:09 And it's a forecasting model that uses actual data.

02:13 So, here we have the actual daily cases shown in pink or red, and the cumulative cases shown in blue.

02:20 In here, this has a short term projection of how the epidemic of COVID-19 was supposed to decline over the course of these weeks.

02:33 And by the way, the model is very accurately predicted very well.

02:39 Here is an example of a dynamic model for COVID-19.

02:43 This was presented by the Government of Canada in April of 2020.

02:49 And it showed you how they thought the epidemic would unfold in the nation of Canada in the coming months.

02:58 So they put forth three different scenarios.

03:03 In the blue, there's a strong epidemic control scenario, right? So, if we were to implement physical distancing, or very good contact tracing, then we can get infection rates down quite low.

03:19 The pink or red curve describes weaker controls.

03:23 So, maybe people didn't distance as well.

03:25 In which case, we'd have a higher infection peak.

03:29 But with Farr's law in place, the epidemic curve was still diminished.

03:34 And the green curve describes, what would happen if we did nothing? 70 to 80% of the population would be infected, and a certain percentage of those individuals would be hospitalized, and a certain percentage of them would die.

03:49 The implication of this model is that, we had to act.

03:54 We had to implement some population controls.

03:57 Stronger is better than weaker.

03:59 If the stronger controls are put into place, then we would have no more cases by the fall.

04:06 In reality, what happened was the first wave diminished by the end of June.

04:12 And the second wave began in spring, in rather in September.

04:16 But this model was useful in conveying the direness of the situation and the need to act fast, with strong epidemic controls to prevent the overwhelming of the healthcare system.

04:31 I've mentioned this quote several times.

04:33 "All models are wrong, but some are useful." It's an important thing to remember.

04:38 Because models have so many built in assumptions, that they do not really describe the real world.

04:47 What they do is present scenarios for planning and for understanding the likely path of a disease.

04:56 But to assume that a model can predict with the high a degree of precision, actual incidence and death rates is probably a problematic thing to do.

05:08 "Models are only really as useful as the assumptions that you put into them." This is true for all models, not just disease models anything in mathematics.

05:22 So, we have to be careful about the assumptions that we make, and interpret the curves and the data with some nuance and sensitivity.