by Mountaineerbr

#20 - Generating Graphs of Covid19 Positivity Rate


I have been making some charts with data from Johns Hopkins University CSSE and Reuters News Agency to check how the number of new cases, recovered and deaths are developing.

I update the charts almost everyday. You can check them at my github repo of covid19 graphs. Unfortunately, I only discussed my opinion about them in one unrelated forum, needed to get those references to put them in my blog sometime.

I have been very curious about another type of graphs, that of positivity. As the amount of test number per day have been variable since the pandemics start, I thought it would be helpful if we could analyse the data taking out such a parameter from the equation (in this case, graphs).

For one thing, we should not be keen to analyse the absolute numbers, as there are so many diverging opinions, such as either numbers are over or underestimated.. So they are not much good. Let's check proportions! By the way, that is a tip from John McAfee, who is currently in jail in Spain #freemcafee).

Experimental charts for Covid19 test positivity

How many percentage points of test results are positive to Covid19? I could not find positivity charts around with long time series, so I decided to make some charts with Brazilian data. Below is my try for charting data from Paraná State reports.


graph made with gnuplot
Fig 1. Positivity graph from data of the state of Paraná. Y-axis for positivity percentage of total tests analysed, X-axis relative date. That is one state with the lowest positivity rate amongst all I could see with this analysis. Data ranges between 03/feb/2020 and 06/dec/2020.

wider graph, same as fig. 1
Fig 2. Same as fig. 1, but we see absolute dates in X-axis. I am not sure how to configure gnuplot to show fewer x-ticks and make them more readable, so that is why it is so wide..

To see charts with data from all states, check my corona virus repo.

Analysing positivity rate is independent on test capacity, meaning it does not matter if there is more testing now than at the pandemics start. Everything can be levelled.

I got the data from That is not the cleanest data. I extract values for Positive cases and Negative cases with some awking.

After download the csv files, you can check the header line which contains the column keys: 

% head -1 dados-pr.csv | tr \; \\n | nl
     1	id
     2	dataNotificacao
     3	dataInicioSintomas
     4	dataNascimento
     5	sintomas
     6	profissionalSaude
     7	cbo
     8	condicoes
     9	estadoTeste
    10	dataTeste
    11	tipoTeste
    12	resultadoTeste
    13	paisOrigem
    14	sexo
    15	estado
    16	estadoIBGE
    17	municipio
    18	municipioIBGE
    19	origem
    20	cnes
    21	estadoNotificacao
    22	estadoNotificacaoIBGE
    23	municipioNotificacao
    24	municipioNotificacaoIBGE
    25	excluido
    26	validado
    27	idade
    28	dataEncerramento
    29	evolucaoCaso
    30	classificacaoFinal

The following shell function was used to calculate positivity. The shell will loop through all csv files given and group results by date.

Some awk conditionals will test values from columns $12 and $30, and will decide if that will be counted as a positive case, a negative case or be ignored. There is no guarantee this analysis is sufficient.

You can check the script with some functions to process the data at my sub repo for covid test positivity studies.

The last step is to generate graphs with gnuplot. I am having some difficulty with the x-ticks dates because gnuplot only accepts numbers by defaults.. May as it be, bear in mind the data start from about February and ends on 24th December.

Interpretation

I WILL FINISH THIS LATER ..

data for Paraná estate graph means positivity return of tests is just above 10% at 30-Nov-2020 (latest data point available). This is an EXPERIMENTAL graphs, a more robust and technical analysis will have different positivity results, but I was very curious..