- During the first wave of COVID-19, imaging volume in the Massachusetts General Hospital Radiology Department declined 48%–93%, and leaders developed three novel models to predict future imaging volume
- A long-term model, created in April 2020, was based chiefly on 2003 data from the SARS epidemic and considered three possibilities for recovery of imaging volume (swift, gradual or muted); actual recovery corresponded well with the "swift" scenario
- A short-term model was highly accurate in predicting next-week imaging volume; when it was used to forecast eight weeks out, the predicted imaging volumes were within 5% of actual volumes for non-holiday weeks and 7% for holiday weeks
- A next-wave model, which assumed 38% of imaging examinations would be considered non-deferable during a subsequent wave of COVID-19, predicted an annual decline in imaging volume of 1%–3%
Subscribe to the latest updates from Radiology Advances in Motion
During the first wave of COVID-19, imaging volume in the Massachusetts General Hospital Radiology Department declined 48%–93%, depending on the modality. To help the department plan for immediate and long-term financial stability, Department Chief James Brink, MD, Oleg Pianykh, PhD, director of the Medical Analytics Group, and Steven Guitron, MSc, data scientist with the Medical Analytics Group, and colleagues developed three novel models to predict imaging volume.
In the Journal of the American College of Radiology, they describe how they formulated long-term, short-term and next-wave models, and what the results showed for their department.
The long-term model was created in April 2020. Since little was known about how recovery would proceed, the researchers chiefly relied on 2003 data from the SARS epidemic to model three potential scenarios:
- Swift recovery—Imaging volume would return to 80%–90% of normal within 1–2 months because of pent-up demand, growing desensitization to the virus, adoption of safety protocols and refined hospital safety protocols
- Gradual recovery—Volume would take 2–3 months to return to 80%–90% of normal because of lingering fear of the virus, recessionary effects on health care utilization and the constraints that new sanitation and safety protocols put on capacity
- Muted recovery—Similar to gradual recovery, but the volume would plateau at 50%–80% of normal because of lost demand, fear of contracting the virus at major hospitals and possible subsequent surges of COVID-19
Each scenario was considered when making operational decisions. As time passed, the department observed linear growth in imaging volume, largely corresponding to the swift recovery scenario.
The short-term model was designed to predict week-to-week outpatient imaging volume. The researchers estimated supply (the number of outpatient imaging studies that could be conducted by the department) using a linear regression model given in the article. To estimate demand, they used the most recent ordering data in another predictive model.
This model provided very accurate predictions for next-week imaging volume. Moreover, when it was used to forecast eight weeks out, the predicted imaging volumes were within 5% of actual volumes for non-holiday weeks and 7% for holiday weeks.
The third model forecasts how future COVID-19 surges will affect annual imaging volume. The researchers did not try to predict when subsequent waves would occur, only their shape, length and depth.
The department assumed there would be phased recovery plans and the government would roll back to an earlier phase of containment when necessary. They also assumed a subset of the most essential imaging would continue, which they estimated as 38% of the normal caseload.
Given these assumptions, the researchers estimated a subsequent wave of COVID-19 would result in a yearly decline in imaging volume of 1%–3%.
Planning for Any Uncertainty
These models can be refined as more data becomes available. Radiology departments could apply them in any situation when normal procedures for planning have been upended.
View all COVID-19 updates
Learn more about the Medical Analytics Group