Matching medical staff to long term care facilities to respond to COVID-19 outbreak

Throughout the COVID-19 outbreak, healthcare providers have suffered extended medical product and staff shortages [1]. To reduce the impact of these shortages a wide variety of measures were taken, such as bringing medical staff from all over the country to the epicenter at Hubei, China, and using students as temporary medical staff in Denmark [1, 2]. Long Term Care Facilities (LTCFs), where one of the most vulnerable population segments resides, suffered some of the worst shortages [3-5]. Based on the Centers for Medicare and Medicaid Services (Data.CMS.gov), roughly 20% of nursing homes (NHs) in the US have faced severe personal protective equipment (PPE) and staff shortages during the COVID-19 outbreak [6, 7]. As a result, the pandemic impacted LTCF residents more severely than most other population segments. Similar situations in LTCFs were reported in other countries such as Spain, England, and Italy [8-10].

The effect of LTCF staff shortages on quality of care, staff's work burden, and the likelihood of infection among residents during the COVID-19 pandemic is emphasized by prior studies [3, 4, 7, 11, 12]. In the US, the first LTCF COVID-19 infection case was reported in King County, Washington, on Feb 28th, 2020 [13]. From this date until Mar 28th, 2020, 30 LTCFs with confirmed cases were identified in King County [13]. One reason for such a quick spread of COVID-19 among LTCFs was identified as the practice of sharing medical staff, which led to LTCFs requiring staff not to work in multiple facilities [13, 14]. This restriction together with other pandemic related hardships including, staff's unwillingness or inability to work in COVID-19 positive settings and lack of childcare, resulted in LTCFs facing insurmountable challenges to satisfy their staffing needs at minimal levels. These staff shortages in turn resulted in negative outcomes even including forced evacuation of residents in some cases [3]. Moreover, several studies showed that the risk of COVID-19 infection doubled in an LTCF if the facility has fewer Registered Nurses (RNs) than the minimum recommended number [4]. Therefore, finding approaches to alleviate LTCF staff shortages has been one of the critical challenges for policymakers in responding to COVID-19 outbreak.

To relieve LTCF staff shortages, various approaches were implemented by different states within the US. For instance, in some cases, staff with suspected or confirmed COVID-19 could work in an LTCF under CDC guidelines [15, 16]. Other response initiatives included extending the temporary licensing of medical staff that do not have an active license, rapid training, engaging National Guard, using volunteer workers and students, providing a list of candidates to facilities tailored to their staffing needs, launching educational portals, and developing hiring and job posting portals [17].

Many studies point out the importance of having an adequate level of medical staff in LTCFs. Literature, covering several different fields including operations management, health systems etc., has a significant number of papers focus on matching resources to demand [18], and a branch of them is devoted to the healthcare systems, including nurse scheduling [19-21], patient allocation [22], and allocation of medical resources [23, 24]. However, these rarely suggest a central and effective way of dynamically matching medical staff to facilities in a public health emergency with severely limited resources, which is the central goal of this study.

To assist LTCFs in Massachusetts (MA), Commonwealth's Executive Office of Elder Affairs (EOEA) brought together subject matter experts from the Executive Office of Health and Human Services (EOHHS), Northeastern University, and University of Massachusetts Chan Medical School to design and operate a web portal to centrally recruit and allocate medical staff to LTCFs experiencing acute vacancies. The portal served as a central repository for collecting medical staff demand and supply data and designing a framework to match the supply to LTCFs' demand centrally and efficiently. The developed portal, a website that intakes data from medical staff and LTCFs, and the matching framework for allocating medical staff supply to LTCFs’ demand will be described in the following sections. We also developed statistical models to determine the most salient features that induced the LTCFs to submit feedback on their matches to the portal.

We analyze the outcomes of the matching process to determine whether the developed matching framework was successful in responding to facilities' demand and providing job opportunities for medical staff. We also hope to learn how the framework can be improved in terms of providing matches with higher hiring rates and determine the important characteristics that resulted in facilities providing feedback on hiring decisions.

The matching portal was operational between April 2020 and June 2021, and was updated several times based on the changing state policies and needs during its lifetime. The matching framework was used roughly 150 times during its operational time, 19,084 unique matches were made, and facilities provided feedback for 2,542 matches. Figure 5 represents the number of available staff including RCAs (blue line) and excluding them (black line) along with the eligible demand in the matching pool (green line), the number of matches (orange line), and the feedback level that the facilities provided (red line) for each round of matching over time. The figure demonstrates how the matching framework has been responsive to the demand based on supply availability on each matching date. Moreover, this figure shows that the facility feedback rate significantly increased after an improved version of the portal (portal 3.0) went live on January, 2021. This version of the portal provided a more convenient way for facilities to provide feedback on matches.

For understanding response of the framework for each position, Fig. 6 shows the distributions of the number of staff, demand quantity, number of matches, and number of feedback entries for nine different positions (excluding RCAs). This figure indicates that, except for PT and AC positions, the portal could not match the facility demand due to staff supply levels. By comparing blue and orange bars, this figure indicates the severity of staff shortages in LTCFs, specially for the three nursing positions RN, CNA, and LPN. Further, this figure shows that demand exceeds number of available staff for all positions and most of the demand entries and feedback entries are for the three nursing positions RN, CNA, and LPN.

In addition to the number of matches and facilities’ feedback entries for each position, shown in Figs. 6 and 7 indicates the percentage of demand satisfaction for facilities (Fig. 7 left), and organic and urgent demand quantities (Fig. 7 right) as other metrics. Organic demand refers to the demand that the facilities entered in the portal. On the other hand, urgent demand was created for the facilities that were prioritized for matching by the stakeholders involved in responding to the COVID-19 pandemic due to low staffing levels or expected shortages from all facility testing. Based on Fig. 7, 47.91% (19,084 out of 39,836) of the facility demand entries were matched to and 15.04% (6,085 out of 40,466) of the total demand were urgent.

Using the CMS NH dataset and the LTCF portal generated dataset, we depicted in Fig. 8 the demand, total matches, and the number of feedback entries with the number of resident and staff COVID-19 cases and the number of total resident deaths in each county in Massachusetts.⁵ From this figure, we note that counties with higher number of cases among residents and staff and number of deaths among residents entered more demand in the portal and as a result, received more matches. However, portal activity was at a higher level in Middlesex and Worcester county facilities in proportion to Covid cases and resident deaths.

On average, facilities provided feedback for 13.32% (2,542 out of 19,084) of the matched staff with 478 (2.50%) feedback showing facilities did not hire matched staff and 2,064 (10.82%) feedback showing facilities intend to hire matched staff.

We used variables’ standardized coefficients to examine the association of variables with facilities entering feedback on the matched staff. Table 3 presents the results and associations of variables with facility feedback entry. We conclude that NH and LTCFs with more demand entries were more likely to provide feedback. Moreover, staff’s years of experience, shift preferences including weekday afternoon-evening, weekend morning, and weekend overnight are associated with facilities providing feedback on the matched staff. In addition, number of times LTCFs are prioritized due to testing or low staffing levels are associated with LTCFs that did not provide feedback on the matched staff. Finally, the distance between the staff and the facility that they were matched to is associated with a higher likelihood of a facility providing feedback on matched staff. Figure 9 indicates the standardized coefficients of variables with a line showing 95% confidence interval (CI).

During the COVID-19 outbreak, LTCFs faced severe staff shortage issues. In this study, in collaboration with the EOEA, we developed a portal and a matching framework to alleviate staffing problems in LTCFs in Massachusetts. The developed matching framework facilitates staff allocation by considering restrictions of facilities and staff. The underlying matching model was adopted to the specific and changing situation on the ground and considered a variety of criteria for matching, such as the distance between facilities and staff while also incorporating restrictions due to licensing and transportation needs. We used the matching framework dynamically by updating the matching pools and adjusting daily facility prioritizations. Furthermore, the matching framework went through several updates and modifications. New features that were added to the framework overtime included improved data entry portals for both the facilities and staff and methodology for updating facility and staff pools to achieve more effective matches by removing excessive facility demand and inactive staff.

Despite continuous refinement and improvements on the matching framework, this study and the framework are subject to several limitations, including: (i) Due to work overload, LTCF staff was not always able to take action and provide feedback on the matches provided to a facility. While the portal user interface was updated several times and several follow up calls were made to facilities, the issue persisted for some. (ii) A subset of the mathematical model parameters such as matching weights and shortage penalty (i.e., a parameter that enforces the matching model to allocate available staff to LTCFs and always match an available staff to a facility if feasible) are estimated based on expert opinion and subjective; (iii) Several potentially important considerations, such as shift preferences, were not included as criteria in the matching process but only reported to the facilities as information along with their matched staff due to concerns regarding how to elicit preference information from the facilities without imposing too high of a data entry burden. We believe including such criteria will increase the quality of the matches; (iv) The matching framework only considered the current demand during each run. This led to matching all available staff to the current demand if possible, ignoring the potential for demand surges or critical demand that could arise in the future. An enhanced framework could be established by forecasting future demand based on evolving trends of the public emergency and reserving some supply for covering this future demand. (v) It was observed that a subset of the facilities inflated their demand in order to increase their matches. While the current framework tried to prevent hoarding by implementing simple limits such as those based on bed numbers, a more sophisticated process could be designed to disincentivize this behavior and recognize and adjust allocations accordingly.

In conclusion, we developed a centralized matching framework that can be an efficient tool for responding to staff shortages during public health emergencies. We further analyzed the impact of using the matching framework on staffing level of NHs in a subsequent paper [29]. Moreover, the analytical results imply that NHs that were using the portal regularly were more likely to provide feedback. Finally, staff that were matched to a close facility, had work experience, and were willing to work in the afternoons or later were likely to drive hiring feedback from the facility.