The COVID-19 pandemic has proven itself to be a formidable foe to health providers worldwide, with the ability to overwhelm and overwork facilities by stretching their capacity to breaking point. 

In recognition of this, Aginic partnered with the Queensland Government to tackle this dark cloud of uncertainty. Without knowing how many individuals the virus was likely to infect, hospitals were unable to determine whether they could handle surges of cases.

Our solution? Know the future. Or at least, predict it.

By implementing sophisticated models of COVID-19 behaviour, predictive conclusions were drawn about the nature of both when peaks would occur, and their likely severity.

At Aginic we promote a culture of solving problems differently, and to do so we focus on crafting cross-functional teams of capable individuals. By equipping problem solvers with the tools to operate at their best, we facilitate creativity and innovation in the way we deliver solutions.

Like all projects we undertake, the project approach for modelling the virus was guided by agile principles with specific adherence to the Scrum framework. Agile is important for building data solutions that have the end user in mind – in this case, Queensland Health – as it incorporates feedback into the development cycle.

Collaboration is a fundamental part of how we operate. We include clients in every discussion, and seek to provide iterative value through a collaboratively designed delivery roadmap. By adopting this approach, we can ensure that their vision matches our results. With Queensland Health, collaboration was necessary to determine how they would utilise our models, providing us with crucial use cases for UI/UX design and selection of parameters.

We developed a SIR (Susceptible, Infectious, or Recovered) model in Python, utilising research conducted by Queensland Health, ultimately allowing us to predict the spread of COVID-19 through Queensland. We  used the capture-recapture method to ensure we were accurately estimating the complete set of cases. We developed curves for all hospital districts which enabled more accurate decision making at the departmental level as well at the hospital level.

—-Image description —-

 

The prediction data fed into a Qlik Sense dashboard that included projected dates, and predicted beds occupied with COVID patients on that day, and how many of those were new admissions. Some of the relevant variables were infection rates, recovery rates, the estimated number of people that were expected to visit each region, the reported number of people in each hospital region, and the estimated length of stay in hospital. By using a series of dynamic parameters, we were able to answer vital questions, such as; 

• Do we need to purchase beds from the private sector in order to manage the current wave?
• How many ICU beds were required?
• How many virtual ward admissions were expected?

For the first wave, a curve was generated for each HHS (Hospital and Health Service) according to their specific start date. However, after the first wave, it was found that accurate predictions could be created through just proportioning the statewide curve  across the HHSs using population data from the Australian Bureau of Statistics (ABS). This allowed us to create curves for some of the smaller HHS regions. 

The dynamic parameters allowed HHSs to change parameters such as hospitalisation rates and average Length of Stay (LoS) to account for demographic factors within their communities. This work was critical for ensuring hospitals could handle their patient in-flow, making changes to their operating practices around Elective Surgery where necessary.

—- Image description —-

 

The end result was a fully functional tool that encapsulated, modelled, and displayed the Omicron variant, which was then usable in planning the allocation of hospital beds to patients and implementing safeguards to minimise the spread. The model has been maintained and updated throughout multiple waves of the virus to facilitate resource management, and also extended to include Influenza.

Compared to actual events, the model which we produced eight weeks prior was shown to have predicted the maximum number of cases with 95% accuracy.  In addition, we predicted the wave’s peak within seven days of the actual event.

If you would like to know how we can help you deliver innovative data solutions, feel free to reach out here.