Software development

The software was developed using the following programming languages:

In computer science, the terms front-end and back-end refer to the initial and final stages of the process, respectively.

Statistics and forecasts produced by the Forecast UTI application

Forecast UTI calculates average values and standard deviation for the ‘number of hospitalizations per day’ and ‘time of hospitalization’ (in days) variables. It calculates a percentage value for the ‘hospital lethality’ variable, defined as the proportion of COVID-19 cases that have died. The application provides graphic visualization for the ‘number of ICU beds occupied per day’, ‘number of new hospitalizations per day’, ‘cumulative number of hospitalizations’ and ‘number of deaths by age group’ variables, by means of bar plots. It provides boxplot graphs for ‘patients’ age’ (in years) and ‘time of hospitalization’. Finally, the ‘current patients’ situation’ variable, classified as ‘inpatient’, ‘discharged’ or ‘death’, is visualized by means of a donut chart.

Forecast UTI uses polynomial regression models⁽⁹⁾ adjusted to the fifth degree, for forecasting the ‘number of COVID-19 occupied ICU beds’. For each model, the application calculates the forecast root mean square error (RMSE) and chooses the model with the lowest RMSE. The statistical analyses consider a equal to 5%.

Forecast UTI scenario simulation: ICU bed use

Based on the ‘total number of beds’, ‘number of planning days’, ‘average number of patients per day’, ‘time of hospitalization’ and ‘number of COVID-19 occupied beds’ variables, the application simulates hypothetical care scenarios. Based on data input by the user, parameters are generated randomly for each scenario. The daily number of new patients is calculated according to a Poisson distribution, while for the calculation of hospitalization times for new patients, a Uniform distribution is considered.

The simulations generate 1,000 different scenarios for the set of variables input to the application. Each scenario is planned with the aim of allocating the new patient to the first available ICU bed. If all the beds are occupied, the patient will not be admitted to the hospital. After simulating all the scenarios, the tool shows the worst possible scenario, i.e. the scenario with the highest number of patients when cannot be cared for because of maximum ICU bed occupation.

Accessing Forecast UTI

All SUS managers interested in the application should fill in the form available at https://www.forecastuti.unifesp.br/

They will later receive a username and a password to be used to access the application free of charge.

Importing files to Forecast UTI

After accessing the application, the user should use the ‘Import Files’ tab to load the system with a file with a ‘.csv’ file extension containing the obligatory variables for their patients: ‘identification number’, ‘date of birth’, ‘inpatient bed’, ‘date of admission’ and ‘patient’s current status’. If necessary, the user can also include the following variables: patient ‘ICU type’, ‘ICD on admission’ and ‘ICD on discharge’ (Figure 1). Confidentiality of data loaded via the website is ensured as they are available only to the registered user.

Figure 1 – Forecast UTI ‘Import Files’ tab: explanation and exemplification of how to set up the database and import itICU: Intensive Care Unit. 

Examples of Forecast UTI use

For illustration purposes, we have used artificial data to provide examples of Forecast UTI features. This login (exemplo@unifesp.br) and password (dadosexemplo) can be used to access the application website, if readers would like to reproduce the following examples.

The two main monitoring tabs are available: ‘Statistics and forecasts’ and ‘ICU bed use’, which enable ICU bed occupation by COVID-19 cases to be monitored and different scenarios to be simulated, respectively.

Selecting the ‘Statistics and forecasts’ tab, one can visualize average daily ICU bed occupation, average time of hospitalization in the ICU and COVID-19 lethality. Using this tab, one can see the distribution of the daily number of occupied ICU beds and make forecasts of up to 10 days. It is also possible to visualize the daily distribution of new COVID-19 admissions, the cumulative number of admissions, monitor number of deaths by age group, length of inpatient stat and current status (Figure 2).

Figure 2 – Forecast UTI ‘Statistics and Forecasts’ tab, using artificial data: hospital indicators of interest and forecast number of occupied ICU beds per dayICU: Intensive Care Unit. 

The ‘ICU bed use’ tab (Figure 3) enables different scenarios to be simulated in order to assist with ICU bed management. This simulation enables one to check what will happen with ICU bed occupation if there is a change in the daily number of new patients, average time of hospitalization and number of beds available. The application shows each ICU bed on a line on the graph, as well as hospitalized patients, in their respective beds, along the planning horizon. The bars on the graph represent each patient’s admission and discharge date. Information is also provided about patients who were not admitted because of lack of available ICU beds on the day they arrived at the hospital. Finally, Figure 3 shows in the form of a graph the situation under the worst simulation scenario achieved by using test data. The following values were used in this example: total number of beds = 30 beds; number of planning days = 15 days; average number of patients per day = 3 patients; average length of inpatient stay = 10 days; standard deviation of length of inpatient stay = 3.6 days; and number of occupied beds = 15 beds. Using this simulation, the worst scenario produced resulted in 16 patients who could not be admitted because of lack of beds.

Figure 3 – Forecast UTI ‘ICU bed use’ tab, using artificial data: simulation of bed occupation, considering customized parameters with the largest number of patients not admitted due to maximum ICU bed occupation (worst scenario)ICU: Intensive Care Unit.