Factors Contributing to Remote Radiologist Scheduling

The Challenge

The company was hiring radiologists every year (60 new radiologists per year) and transitioned from an off-hours/overflow radiology company to a 24x7x365 provider. This gave rise to the need for a cohesive strategy that was less reactive and optimized for hiring, credentialing, forecasting, and scheduling.

Current forecasting produced monthly projections with fairly high accuracy, but these projections were currently of relatively limited use when it came to schedule and credentialing. For these, volume predictions that are much more informed, specific, and accurate are critical to correctly credential and schedule radiologists to handle the increased workloads.

What They Did

The Company hired Gordian Knot Analytics to create a dynamic volume forecasting model that would more accurately predict monthly and weekly volumes as well as volumes down to the day, day-part and potentially hour (which depended on the granularity of available data)

This model was designed to handle changing conditions over time and became the core element of both the credentialing and scheduling models

We also created a dynamic state-specific credentialing model that could adapt to changing conditions and volumes over time as well as accounting for delays that are a reality of the credentialing process

We worked with the client to employ the volume predictions to help create a more effective and flexible scheduling strategy

At a high level—in this case state level—volume doesn’t look volatile; however, deeper resolutions show a volatility that must be accounted for in forecasting demand. Because of this dynamic, it is recommended volume be forecasted by modality at the hourly and facility levels so it can be matched to radiologist production capacity.

rad 1.png

Aside from holidays (which are known to have a substantial impact on volume), there are a number of factors that on a preliminary basis show strong correlations with the number of RVUs (using Georgia as a test bed. It is recommended that these factors be used as inputs to demand forecasting.

rad 2.png

Different Radiologists have different levels of capacity production by day. As an example, Rad 1 tends to produce heavily relative to his average Fri-Sun, but is well below his average the rest of the week.

rad 3.png

rad 4.png

Not all radiologists are perform equally. As key inputs into coverage planning, capacity planning, L&C, and potentially contracting, it is highly recommended that radiologist segmentations be created that can account for rad reliability, consistency, and production. Radiologists in the best segments should receive priority for expanded L&C.

rad 5.png

Given that there are very different levels of reliability, consistency and performance across different radiologists within the company, varying levels of compensation and rewards are warranted. It is recommended that these compensation tiers be heavily-based on the segmentations created for coverage and capacity planning. Due to the laws of unintended consequences and how it could potentially affect rad behaviors in a negative way (i.e. “gaming the system”), we do not currently recommend changing compensation at the study/case level.

rad 6.png

The Results

  • The project was divided into three separate phases to allow interpretation and integration of the results that would impact each subsequent project phase
  • We were able to Identify all available data including internal sources (historic volume by location, current credentials, hiring, radiologist location, prevalence of certain scan types, etc.) as well as appropriate external sources (weather, financial markets, area demographics, traffic accidents, etc) to be appended to design a comprehensive data improvement plan
  • We established which factors—and which unique combinations of factors—were driving volume and fluctuations by state, facility, and radiologist
  • Gordian Knot created and deployed an algorithm that dynamically predicts volume for states and facilities by month, week, day, day-part/work shift.