This version is no longer current – please see<\/strong><\/span><\/h2>\nhttps:\/\/fortran.uk\/using-published-covid-19-data-to-deduce-real-infection-rates\/<\/a><\/h5>\nIntroduction<\/h2>\n
In this note, I will show that a simple model of linking testing with confirmed cases and deaths is confirmed by analysis of published data.\u00a0 The resulting equation can be extrapolated to predict true infection rates.\u00a0 It is argued that, although the model must break down for high infection rates, the way in which it breaks down can be anticipated and incorporated into a larger framework.<\/p>\n
The Model for Covid-19 Testing<\/h2>\n
If testing were random, we would expect the number of cases found (c) to be proportional to the number of tests (t), and the proportion of the population that is infected.\u00a0 The second factor is unknown, but the number of deaths (d) is probably a reasonable proxy, so we have:<\/p>\n
c \u221d d.t<\/p>\n
A better proxy for the infection rate now might be the death rate in a few days time, but initially, we will use the simpler model.<\/p>\n
In the early stages of the pandemic, testing is not random, but is designed to find as many cases as possible.\u00a0 It focuses on a small part of the population comprising the contacts of existing cases, medical staff, and those with symptoms.\u00a0 If the size of that reduced population is proportional to c, then the number of cases found will be<\/p>\n
c \u221d d.t\/c<\/p>\n
c \u221d sqrt(d.t)<\/p>\n
In effect, testing is concentrated on a bubble of contacts surrounding each case, and for small c, these bubbles are isolated.\u00a0 As c increases, the bubbles increasing overlap, and eventually encompass the entire population.\u00a0 At that point testing effectively becomes random, with c \u221d d.t.<\/p>\n
To summarize, the hypothesis is that c should vary with sqrt(d.t) for small c, and with d.t for large c.<\/p>\n
Comparison with Published Data<\/h2>\n
The data, downloaded from worldometers.info on the 14th April 2020, comprises 3 variables for each of 110 countries with more than 100 cases, and at least one death.\u00a0 China is excluded because testing data is not available.\u00a0 The variables are the number of confirmed cases (c),\u00a0 deaths (d), and tests (t) per million of population.<\/p>\n
Figure 1 shows that the data fits very well to the equation<\/p>\n
c = 1.67 * sqrt(d*t)<\/p>\n
![\"\"](\"https:\/\/fortran.uk\/wp-content\/uploads\/2020\/04\/Fig-1-new-14April.png\")
<\/a><\/p>\nThe prediction spans almost 4 orders of magnitude, but even so, almost all countries are within the tramlines, which indicate a factor of 2 either side of the predicted value.<\/p>\n
There are many reasons to expect data from different countries to differ, including:<\/p>\n
\n- Different health systems \u2013 some primitive, and others advanced<\/li>\n
- Overloading makes even advanced health systems less effective<\/li>\n
- Different demographics \u2013 Covid-19 is worse for older populations<\/li>\n
- Time lags \u2013 e.g. confirmed cases may not survive, and reports may be delayed<\/li>\n
- Different criteria for counting deaths<\/li>\n
- Political massaging of published data<\/li>\n<\/ul>\n
The conclusion is that all of these factors combine to a factor of up to about 2.<\/p>\n
UK Time Series<\/h2>\n
Historic data for the UK was downloaded from https:\/\/github.com\/tomwhite\/covid-19-uk-data\/blob\/master\/data\/covid-19-indicators-uk.csv<\/strong><\/p>\nFigure 2 shows how the model fits this data. This time-line is also superimposed, as an orange trace, on Figure 1.<\/p>\n
![\"\"](\"https:\/\/fortran.uk\/wp-content\/uploads\/2020\/04\/Fig-2-new-14April.png\")
<\/a><\/p>\nThis line is also superimposed, as an orange trace, on Figure 1.<\/p>\n
Matching Case Fatality Rates (d\/c)<\/h2>\n
Case fatality rates are an obvious but misleading measure, which is often used wrongly, when comparing\u00a0 different countries.\u00a0 In general terms, it is known that countries that test extensively tend to have lower case fatality rates, but the\u00a0 model described here allows us to quantify that relationship.<\/p>\n
Figure 3 shows that a\u00a0 good fit to the observed apparent fatality rate, d\/c, can be obtained using the equation:<\/p>\n
d\/c=0.6*sqrt(d\/t)<\/p>\n
![\"\"](\"https:\/\/fortran.uk\/wp-content\/uploads\/2020\/04\/Fig-3-new-14April.png\")
<\/a><\/p>\nOnce again, only a few countries diverge from the prediction by more than a factor of 2 (indicated by the tramlines).\u00a0 \u00a0It is possible to obtain slightly better fit metrics by multiple regression, but the difference is small, and not obvious on a scatter plot.<\/p>\n
Predictions from the Model, and How to Extend it<\/h2>\n
If we assume c = t = 1 million (i.e. the entire population is infected), the model predicts mortality as 36%.\u00a0 This is clearly wrong, and suggests that as the proportion infected increases, the curve in Figures 1 and 2 turns up so that the mortality at saturation is about 1%.\u00a0 This point is indicated by a red spot.<\/p>\n
The dashed line projected down from the red spot represents the cases that would be found with random testing if the true infection fatality rate were 1%.<\/p>\n
As the number of cases and tests increases to include a significant proportion of the population, it’s likely that testing has to be more random.\u00a0 If the true mortality rate is 1%, that suggests that the actual course of the pandemic veers away from the straight line extrapolation of the model and on to the dashed line representing random testing.<\/p>\n
The point at which the two lines cross is determined by<\/p>\n
c = 1.67*sqrt(d*t) = (d*t)\/(f*1E6)<\/p>\n
c = 2.79E6*f<\/p>\n
where f is the true infection mortality rate.\u00a0 For f = 1%, that is when the reported cases, c are 2.79% of the population.<\/p>\n
This is a gentle deviation that occurs in a region for which we have little data.<\/p>\n
The model can also be used to predict the true number of infections per million (c0<\/sub>) at any given time.\u00a0 c varies with the square root of the number of tests, so we can get an estimate of c0<\/sub>\u00a0using<\/p>\nc0<\/sub> = c . sqrt(1E6 \/ t)<\/p>\nFor the UK, on 14th April, that is about 1.8% of the population.\u00a0 However, that is probably an underestimate, as we suspect that the model may break down as the number of tests is increased towards saturation.\u00a0 In the case where c = t = 1E6, a factor of 6 increase in the extrapolated value of c was required to bring the predicted mortality down from 36% to 1%.\u00a0 In general the factor is 0.6\/sqrt(f).<\/p>\n
If the same were true for lower infection rates, we’d have to increase the computed value of c0<\/sub> by a factor of 6, meaning that around 10% of the population has been infected.\u00a0 However it is not clear whether, the model breaks primarily because of the high infection rate, or the high testing rate, or a combination.\u00a0 The true figure is most likely between 1 and 0.6\/sqrt(f).<\/p>\nFigure 4 shows how the transition from the model to a regime with c \u221d d.t might work.<\/p>\n
Introduction<\/h2>\n
In this note, I will show that a simple model of linking testing with confirmed cases and deaths is confirmed by analysis of published data.\u00a0 The resulting equation can be extrapolated to predict true infection rates.\u00a0 It is argued that, although the model must break down for high infection rates, the way in which it breaks down can be anticipated and incorporated into a larger framework.<\/p>\n
The Model for Covid-19 Testing<\/h2>\n
If testing were random, we would expect the number of cases found (c) to be proportional to the number of tests (t), and the proportion of the population that is infected.\u00a0 The second factor is unknown, but the number of deaths (d) is probably a reasonable proxy, so we have:<\/p>\n
c \u221d d.t<\/p>\n
A better proxy for the infection rate now might be the death rate in a few days time, but initially, we will use the simpler model.<\/p>\n
In the early stages of the pandemic, testing is not random, but is designed to find as many cases as possible.\u00a0 It focuses on a small part of the population comprising the contacts of existing cases, medical staff, and those with symptoms.\u00a0 If the size of that reduced population is proportional to c, then the number of cases found will be<\/p>\n
c \u221d d.t\/c<\/p>\n
c \u221d sqrt(d.t)<\/p>\n
In effect, testing is concentrated on a bubble of contacts surrounding each case, and for small c, these bubbles are isolated.\u00a0 As c increases, the bubbles increasing overlap, and eventually encompass the entire population.\u00a0 At that point testing effectively becomes random, with c \u221d d.t.<\/p>\n
To summarize, the hypothesis is that c should vary with sqrt(d.t) for small c, and with d.t for large c.<\/p>\n
Comparison with Published Data<\/h2>\n
The data, downloaded from worldometers.info on the 14th April 2020, comprises 3 variables for each of 110 countries with more than 100 cases, and at least one death.\u00a0 China is excluded because testing data is not available.\u00a0 The variables are the number of confirmed cases (c),\u00a0 deaths (d), and tests (t) per million of population.<\/p>\n
Figure 1 shows that the data fits very well to the equation<\/p>\n
c = 1.67 * sqrt(d*t)<\/p>\n
The prediction spans almost 4 orders of magnitude, but even so, almost all countries are within the tramlines, which indicate a factor of 2 either side of the predicted value.<\/p>\n There are many reasons to expect data from different countries to differ, including:<\/p>\n The conclusion is that all of these factors combine to a factor of up to about 2.<\/p>\n Historic data for the UK was downloaded from https:\/\/github.com\/tomwhite\/covid-19-uk-data\/blob\/master\/data\/covid-19-indicators-uk.csv<\/strong><\/p>\n Figure 2 shows how the model fits this data. This time-line is also superimposed, as an orange trace, on Figure 1.<\/p>\n This line is also superimposed, as an orange trace, on Figure 1.<\/p>\n Case fatality rates are an obvious but misleading measure, which is often used wrongly, when comparing\u00a0 different countries.\u00a0 In general terms, it is known that countries that test extensively tend to have lower case fatality rates, but the\u00a0 model described here allows us to quantify that relationship.<\/p>\n Figure 3 shows that a\u00a0 good fit to the observed apparent fatality rate, d\/c, can be obtained using the equation:<\/p>\n d\/c=0.6*sqrt(d\/t)<\/p>\n Once again, only a few countries diverge from the prediction by more than a factor of 2 (indicated by the tramlines).\u00a0 \u00a0It is possible to obtain slightly better fit metrics by multiple regression, but the difference is small, and not obvious on a scatter plot.<\/p>\n If we assume c = t = 1 million (i.e. the entire population is infected), the model predicts mortality as 36%.\u00a0 This is clearly wrong, and suggests that as the proportion infected increases, the curve in Figures 1 and 2 turns up so that the mortality at saturation is about 1%.\u00a0 This point is indicated by a red spot.<\/p>\n The dashed line projected down from the red spot represents the cases that would be found with random testing if the true infection fatality rate were 1%.<\/p>\n As the number of cases and tests increases to include a significant proportion of the population, it’s likely that testing has to be more random.\u00a0 If the true mortality rate is 1%, that suggests that the actual course of the pandemic veers away from the straight line extrapolation of the model and on to the dashed line representing random testing.<\/p>\n The point at which the two lines cross is determined by<\/p>\n c = 1.67*sqrt(d*t) = (d*t)\/(f*1E6)<\/p>\n c = 2.79E6*f<\/p>\n where f is the true infection mortality rate.\u00a0 For f = 1%, that is when the reported cases, c are 2.79% of the population.<\/p>\n This is a gentle deviation that occurs in a region for which we have little data.<\/p>\n The model can also be used to predict the true number of infections per million (c0<\/sub>) at any given time.\u00a0 c varies with the square root of the number of tests, so we can get an estimate of c0<\/sub>\u00a0using<\/p>\n c0<\/sub> = c . sqrt(1E6 \/ t)<\/p>\n For the UK, on 14th April, that is about 1.8% of the population.\u00a0 However, that is probably an underestimate, as we suspect that the model may break down as the number of tests is increased towards saturation.\u00a0 In the case where c = t = 1E6, a factor of 6 increase in the extrapolated value of c was required to bring the predicted mortality down from 36% to 1%.\u00a0 In general the factor is 0.6\/sqrt(f).<\/p>\n If the same were true for lower infection rates, we’d have to increase the computed value of c0<\/sub> by a factor of 6, meaning that around 10% of the population has been infected.\u00a0 However it is not clear whether, the model breaks primarily because of the high infection rate, or the high testing rate, or a combination.\u00a0 The true figure is most likely between 1 and 0.6\/sqrt(f).<\/p>\n Figure 4 shows how the transition from the model to a regime with c \u221d d.t might work.<\/p>\n<\/a><\/p>\n\n
UK Time Series<\/h2>\n
<\/a><\/p>\nMatching Case Fatality Rates (d\/c)<\/h2>\n
<\/a><\/p>\nPredictions from the Model, and How to Extend it<\/h2>\n
![\"\"](\"https:\/\/fortran.uk\/wp-content\/uploads\/2020\/04\/Fig-4-14April.jpg\")
<\/a><\/p>\n