Staffing shortages have plagued US nursing homes for decades.1,2 Barriers such as lower wages compared to alternative options, staff workload, and stressful duties make hiring and keeping workers complicated for nursing homes.3, 4, 5 This problem has worsened with the COVID-19 pandemic onset.6 According to a survey, 94% of nursing homes faced staffing shortages during the COVID-19 pandemic,7 caused by nurses testing positive for COVID-19,8 staff quitting,7 lack of childcare for staff,1 and overall pandemic burnout and fatigue.1

In addition to nationwide efforts to reduce these shortages,9 many state governments have implemented various local policies for the same goal. For instance, Michigan offered bonus payments to recently hired staff, created a rapid response team, and paid for staffing services; Utah invested in recruitment efforts and volunteer programs; Washington created a new tool for requesting volunteer staff, California subsidized travel costs for volunteer staff commuting across the state, Georgia allowed nursing homes to hire temporary (not certified) nursing assistants; Delaware provided nursing aid training for unemployed workers; Indiana created a list of available staff for nursing homes facing shortages; Wisconsin gave free training programs to volunteers.10 Such a wide variety of potential solutions deployed by different states reflects not only differences in policies across states, but also our lack of understanding of how effective these solutions are.10

This paper aims to comprehensively evaluate the effectiveness of reducing staff shortages in nursing homes for one of such solutions, implemented by the Commonwealth of Massachusetts during COVID-19. The core idea was to design and run a centralized process for matching demand for and supply of nursing staff, based on prior successful implementations of similar processes in various other labor markets.11, 12, 13, 14 The solution, supervised by Commonwealth's Executive Office of Elder Affairs and executed in collaboration with local researchers (experts from the Executive Office of Health and Human Services, Northeastern University, and University of Massachusetts Chan Medical School), involved creating an online portal to enable the dynamic and real-time collection of nursing staff demand and supply data, and using this data to optimize matching staff to nursing homes in need algorithmically.15

Nursing homes looking to hire nursing staff and nurses looking for work could register and interact with the portal. Nursing homes could use the portal to report their current needs for each nursing staff position in real time; nurses could note their current availability to start working. Throughout the paper, staff positions refer to three nursing positions which are certified nursing assistant (CNA), licensed practical nurse (LPN), and registered nurse (RN).

After the Commonwealth's Executive Office of Elder Affairs set policies and criteria of matching that considered the temporal and spatial aspects of staff and nursing homes (e.g., the closeness of staff to nursing homes, staff available date and demand urgency), the centralized matching process generated matches that were then shared with the matched nursing homes. Nursing homes then could contact, interview, and hire the suggested matches to reduce their nursing staff shortage. Fig. 1 shows the roles of each stakeholder in the centralized matching process.15

The study hypothesis was that implementing the centralized matching process is strongly associated with increases in staff-to-resident ratios of nursing homes after controlling for the weekly numbers of resident deaths due to COVID-19, weekly numbers of COVID-19 positive cases among staff, and weekly numbers of resident admissions. We further hypothesized that these increases occurred faster, when compared with the setup in which nursing homes would need to seek nurses on their own.

To test these hypotheses, we examined the centralized matching process effectiveness by combining the proprietary portal data with publicly available longitudinal datasets.