JMIR Publications

ABSTRACT

Background:

Prior to the COVID-19 pandemic, US hospitals relied on static projections of future trends for long-term planning and were only beginning to look to forecasting methods for short-term planning of staffing and other resource needs. With the overwhelming burden imposed by COVID-19 on the healthcare system, an emergent need exists to accurately forecast hospitalization needs, including beds, intensive care units and mechanical ventilators, within an actionable timeframe of two to four weeks.

Objective:

Our goal was to leverage an existing COVID-19 case and death forecasting tool to generate expected number of concurrent hospitalizations, ICUs, and ventilators one-day to four weeks in the future for New Mexico and each of its five health regions.

Methods:

We developed a probabilistic model that took as input the number of new COVID-19 cases for New Mexico from Los Alamos National Laboratory’s COVID-19 Forecasts using Fast Evaluations and Estimation (COFFEE) tool and used it to estimate the number of new daily hospital admissions four weeks into the future based on current state-wide hospitalization rates. The model estimated how many of the new admissions would require an ICU or use of a mechanical ventilator and then projected individual lengths of hospital stays based on the resource need. By tracking the lengths of stay through time, we captured the projected simultaneous need for the inpatient beds, ICUs, and ventilators. We used a post-processing method to adjust the forecasts based on the differences between prior forecasts and the subsequent observed data. In this way, we ensured our forecasts could reflect a dynamically changing situation on the ground.

Results:

We found that our forecasts showed variability across time, healthcare resource needs, and forecast horizon. The performance of forecasting healthcare needs at the New Mexico regional level was lower than at the state level according to both accuracy and uncertainty metrics. Across geographic resolutions forecast accuracy declined with the forecast horizon, i.e., how far into the future it tried to predict. However, the stated uncertainty of the predicted improved with longer forecast horizons. Forecasts made in October, a month with steady increases in new COVID-19 cases, were the most accurate.

Conclusions:

Real-time forecasting of the burden imposed by a spreading infectious disease is a crucial component of decision support during a public health emergency. Our proposed methodology demonstrated utility in providing near-term forecasts, particularly at the state level. This tool can aid other stakeholders as they face COVID-19 population impacts now and in the future.