As the COVID-19 pandemic continues to impact communities around the world, hospitals face the challenge of managing patient influx while ensuring optimal resource allocation. To tackle this issue, SFU Beedie professors Ian McCarthy and Michael Johnson and their co-researchers have turned to advanced statistical and machine learning techniques to forecast the demand for hospital beds at a more granular level — individual wards. By accurately predicting bed requirements, hospitals can make informed decisions that prevent overcrowding or resource shortages, ultimately saving lives. Their work is now published in the prestigious journal, Healthcare Management Science.

When it comes to COVID-19 patients, the demand for hospital beds isn't uniform. It varies across geographical regions, and even within hospitals themselves. Recognizing this variability, the research team set out to develop forecasting methods specifically tailored to the unique demands of COVID-19 wards. Their goal was to help hospital staff plan resources effectively, especially for ward-level staffing that has experienced immense strain during the pandemic.

Data-powered decision making

The research team explored the power of statistical and machine learning algorithms to forecast ward-level demand accurately. They discovered that methods commonly used for predicting population growth during pandemics were insufficient for the operational planning of hospital wards. Instead, they found that statistical and machine learning forecasting methods used to predict future values based on past data, such as ARIMA, ARIMAX, and NARX, provided valuable insights for decision-making.

ARIMA, which stands for autoregressive (AR) integrated (I) moving average (MA), and ARIMAX—an alternative, more general form of ARIMA—were found to be suitable for forecasting at the ward level, while NARX (neural net) was implemented as a more accurate machine learning approach.

How accurate forecasts can save lives

By leveraging these forecasting methods, hospitals gained the ability to anticipate bed requirements for COVID-19 wards more accurately than ever before. The significance of this accuracy cannot be understated. Over-allocating beds results in idle healthcare professionals, reduced capacity in non-COVID areas, and even impacts additional off-site wards during extreme pandemics. On the other hand, under-allocating beds leads to burnout among healthcare staff, compromised infection control, patient morbidity, and even increased mortality rates.

The researchers identified ARIMA and ARIMAX as effective forecasting methods for moderate changes in patient numbers. These techniques, known for their accuracy in various healthcare applications, proved useful in predicting ward-level demand within a 3- to 4-day range. They demonstrated that incorporating time-lagged epidemiological factors further enhanced the accuracy of predictions, particularly for longer time horizons of 7 and 14 days ahead.

A tool for effective planning

To put their findings into practice, the research team developed a user-friendly forecasting tool using statistical software, now publicly available only for purposes of research assessment and dissemination.

This tool allows hospital staff to input data and obtain real-time forecasts of bed requirements for specific wards. It even includes a capacity planning module that calculates the probability of exceeding set capacity levels, enabling informed decisions about expanding or contracting COVID-19 wards. This data-driven approach helps hospitals strike a balance between adequate bed coverage and efficient resource utilization.

Limitations and future directions

While the study focused on two urban hospitals in Vancouver, Canada, McCarthy and Johnson and their collaborators aim to validate their forecasting tool across various hospitals in the country. They also plan to explore alternative forecasting methods and examine additional variables, such as hospital discharges and emergency department admissions, to enhance prediction accuracy.

The ability to accurately predict ward-level demand for COVID-19 patients has become an invaluable asset for hospitals during the ongoing pandemic. By harnessing the power of data and employing advanced forecasting techniques, hospitals can optimize resource allocation, minimize strain on healthcare professionals, and ensure the best possible care for patients. This research marks a significant step forward in leveraging technology to combat the challenges posed by COVID-19 and lays the groundwork for future applications in healthcare resource planning beyond the pandemic.