The motivation for the work stems from a grim statistical reality. Roughly 1.4 million Americans live in nursing homes, and within the first year of the pandemic, most facilities in the United States experienced outbreaks, with residents accounting for an estimated 30 to 50 percent of COVID-19 deaths. Previous modeling studies have largely concentrated on resident-focused measures such as isolation, personal protective equipment, testing, and visitor restrictions. Far fewer have asked how the structure of the workforce itself—staff shortages, low wages, multiple job holding, and weak benefits—shapes the epidemiology of infection inside these buildings.

That gap matters because the economics of nursing home work create well-documented transmission risks. Low compensation pushes many direct-care workers to hold jobs at two or even three facilities simultaneously, moving between buildings each week and potentially carrying the virus with them. The absence of paid time off means that a symptomatic worker who cannot afford unpaid leave may clock in anyway. And despite strong evidence that vaccination protects both staff and residents, skilled nursing facilities report some of the lowest vaccination rates among healthcare settings, reflecting persistent hesitancy and the lack of comprehensive institutional policies.

To capture these dynamics, the team, led by Abigail Sweet of North Carolina State University’s Biomathematics Graduate Program, constructed an agent-based model in which every individual—resident, healthcare personnel, or community member—is represented as a discrete agent moving through infection and hospitalization states. The synthetic populations were derived from 2010 U.S. Census data using the Framework for Reconstructing Epidemiological Dynamics, or FRED, which supplies demographic and spatial detail representative of the actual two million people living across Franklin, Jefferson, St. Charles, and St. Louis counties and the City of St. Louis. Zip-code-level data on social determinants of health were layered on to modulate mortality and hospitalization risk, while facility-specific characteristics such as bed capacity came from Centers for Medicare and Medicaid Services records and the LTCFocus dataset.

The technical architecture of the model is considerable. Ten submodels, each executed once per simulated day for 365 days, govern death, personnel replacement, infection transmission, hospital admission and discharge, recovery, retirement, nursing home admission and discharge, and vaccination. Transmission within facilities follows a stochastic formulation in which the infection probability per contact is combined with the prevalence of infection among residents, personnel, and the surrounding community, along with the number of daily contacts of each type. Because the model does not trace individual contacts, the probability that a given contact is infectious is approximated from the proportion of each category that is infected, recalculated at every time step for every facility. Infection status is then adjusted for each agent’s immunity, whether from prior infection or vaccination. Roughly 46 percent of simulated infections are asymptomatic, and transmission is only modeled after incubation periods of 5.47 days for asymptomatic and 2.63 days for symptomatic cases, reflecting the reduced per-contact transmission probability outside symptomatic illness.

Hospitals were included deliberately, because the movement of patients between acute care and skilled nursing facilities is a critical pathway. Up to 94 percent of nursing home residents arrive directly from hospital discharge, and many are rehospitalized as their complex medical needs evolve. Capturing these flows allowed the model to reproduce healthcare-associated exposure routes that would be invisible if facilities were simulated in isolation. The model was calibrated against empirical data collected by CMS through May 2023, focusing on discharge rates, average length of stay, and reported facility-level COVID-19 cases, and validated with 10-fold cross-validation. The experimental model achieved a root mean squared error of 0.063 against a naive baseline of 0.114 and explained approximately 70 percent of the variation in observed infection rates across facilities.

Each intervention scenario was simulated 250 times, and the results were striking in their pattern. In the baseline scenario, with no policies enforced, facilities recorded an average of 172.65 resident infections per year. The single most effective measure for protecting residents was compulsory paid sick leave, which removes symptomatic staff from the workforce pool. Combining sick leave with a full-time employment policy—which restricts each worker to a single facility at 40 hours per week instead of two or three jobs—cut resident infections by 12 percent, to 150.72, and slashed personnel infections by 45 percent, from 8.45 to 4.73 per facility. Full-time employment alone had the greatest impact on personnel infections, producing a reduction of nearly 35 percent, and on personnel-introduced outbreaks, which fell by more than half.

The vaccination mandate produced a more nuanced and somewhat sobering picture. Implementing the policy raised simulated personnel vaccination coverage from 48 percent to 81 percent, exceeding the roughly 80 percent threshold that earlier studies identified as necessary for meaningful effect. Yet the impact on resident infections remained minimal. The authors attribute this to two converging factors: the reduced efficacy of vaccines against later SARS-CoV-2 variants, and the sheer thinness of nursing home staffing, which meant that an 81 percent uptake translated to an average of only five additional vaccinated personnel per site. Larger facilities with more staff might see greater benefit, but the finding underscores that vaccination mandates alone, however well-intentioned, cannot substitute for broader workforce investment.

The outbreak findings add another dimension. Total outbreaks per facility, defined as any occurrence of two or more resident infections, changed by less than 2 percent across scenarios, but personnel-introduced outbreaks responded dramatically to staffing policies. Full-time employment alone reduced them by 51.73 percent, and combining it with sick leave or vaccination pushed the reduction to nearly 65 percent. This distinction matters because personnel-introduced outbreaks are precisely the events that facility administrators and public health agencies can most directly prevent through workforce policy, whereas outbreaks seeded by visitors fall largely outside institutional control, particularly now that federal visitor restrictions have long been lifted.

The study’s limitations are acknowledged candidly by its authors. The model assumes a simplified staffing structure without floating personnel or shift variability, excludes resident-to-personnel transmission to keep the computation tractable, omits privately funded facilities, and implements social determinants of health at a surface level that does not capture their co-occurrence. Even so, the central conclusion carries real policy weight: while no workforce intervention matched the protective power of isolation of symptomatic individuals, the combination of paid sick leave and full-time employment delivered meaningful reductions in infections among both residents and staff. Given the chronic nursing shortage, low wages, and burnout that continue to erode the long-term care workforce, the findings suggest that improving compensation and benefits is not merely a labor issue but an infection-control strategy—one that could make nursing homes safer for their residents while simultaneously making them more sustainable places to work. The modeling framework itself, publicly available on GitHub, is designed to extend beyond COVID-19 to other healthcare-associated infections in acute and post-acute settings.

Subject of Research: Agent-based modeling of workforce compensation and benefit policies on COVID-19 transmission in skilled nursing facilities

Article Title: Modeling the impact of healthcare personnel compensation and benefits on the spread of COVID-19 in skilled nursing facilities

Article References: Sweet, A., Rhea, S., Lenhart, S. M., Odoi, A., Lanzas, C., & Lloyd, A. (2026). Modeling the impact of healthcare personnel compensation and benefits on the spread of COVID-19 in skilled nursing facilities. PLOS Aging and Health, 1(3), e0000040. https://doi.org/10.1371/journal.page.0000040

Image Credits: AI Generated

DOI: 10.1371/journal.page.0000040

Keywords: COVID-19, skilled nursing facilities, agent-based model, healthcare personnel, paid sick leave, full-time employment, vaccination policy, nursing shortage, infection control, long-term care, outbreak modeling, workforce policy

The survey was conducted as a cross-sectional study between September and December 2024, targeting residents in programs accredited by the Accreditation Council for Graduate Medical Education (ACGME) across specialties. A total of 1,332 residents in postgraduate years one through four completed the survey; after excluding respondents who did not identify their gender or who completed less than half of the instrument, 1,280 responses representing 165 institutions remained for analysis. Because the survey was distributed through institutional networks rather than a single enumerable sampling frame, the authors note that a traditional response rate could not be calculated, a methodological caveat that shapes how the findings should be interpreted. Even so, the scale and geographic breadth of the sample make it one of the most comprehensive snapshots of reproductive experiences in graduate medical education assembled so far.

The demographic profile of the respondents reflects the changing face of the physician workforce. The overwhelming majority, 932 residents or 72.81 percent, identified as women, a proportion consistent with the fact that women now make up roughly half or more of entering medical school classes and are disproportionately the ones navigating pregnancy during training. Among the women respondents, about one quarter, 241 individuals or 25.86 percent, reported being pregnant at some point during residency. That figure underscores how routine pregnancy has become in the training pipeline, and yet, as the data show, the institutional machinery surrounding it remains uneven at best.

Perhaps the most clinically striking findings concern pregnancy-related health outcomes. Of the residents who were pregnant during training, 97, or 40.25 percent, reported experiencing pregnancy-related conditions, a burden of morbidity that raises questions about the physical demands of residency itself, including long call shifts, overnight duty, prolonged standing, and high-stress clinical environments. Pregnancy loss was reported by 39 residents, representing 16.18 percent of those who were pregnant during training. While the study was not designed to establish causal links between workload and adverse outcomes, the prevalence of complications and loss within this cohort provides a quantitative baseline that researchers and program directors have largely lacked, and it gives concrete weight to calls for evidence-based duty accommodations for pregnant trainees.

The survey also reached into the domain of fertility and family building more broadly. A small but notable fraction of respondents reported seeking reproductive specialist care: 87 women, or 9.33 percent of female respondents, and 23 men, or 6.61 percent of male respondents, consulted a reproductive specialist during training. These numbers touch on a growing conversation in medicine about assisted reproductive technologies (ART), delayed childbearing, and the financial and logistical barriers that residents face when pursuing fertility care on modest trainee salaries and rigid schedules. The inclusion of male residents in these figures is a reminder that family-building challenges during training are not exclusively a women’s issue, even though the burdens of pregnancy itself fall squarely on those who carry it.

When the researchers examined why pregnant residents struggled to obtain accommodations, a clear hierarchy of barriers emerged. The most frequently cited obstacle was staffing needs, reported by 137 residents, or 56.85 percent of those commenting on barriers. In other words, the most common reason pregnant trainees were denied or delayed in receiving modified duties was that their programs simply could not cover the clinical gaps their absence would create. This finding exposes a structural vulnerability in residency staffing models, which typically run with minimal slack and depend on every trainee filling a defined clinical role. When one resident needs reduced hours or lifted lifting restrictions, the burden ripples through the entire schedule.

Close behind staffing concerns were program-level and institutional-level requirements. Individual program requirements were cited by 116 residents, or 48.13 percent, while institutional requirements were cited by 108, or 44.81 percent. Taken together, these figures suggest that pregnant residents frequently encountered rules, whether written or informal, at multiple administrative levels that constrained what accommodations were available to them. The authors’ conclusion highlights three intertwined problems: unclear policies, inconsistent accommodations, and limited communication of available resources. A resident at one hospital might receive modified call schedules and ergonomic relief without difficulty, while a resident one mile away at a different institution might face opaque requirements and no clear point of contact, even when formal policies exist on paper.

The implications extend well beyond individual well-being, though that alone would justify attention. Residency is the crucible in which the physician workforce is formed, and the years of training overlap precisely with peak reproductive years. When trainees delay childbearing, experience pregnancy loss, or leave training dissatisfied because of unsupported pregnancies, the costs accumulate across the profession, contributing to burnout, attrition, and inequities that disproportionately affect women and, the authors suggest, intersect with broader concerns about underrepresented in medicine (UriM) trainees. The study was supported by the George B Quatman Foundation and The Research and Advocacy in Gender Equity Foundation, reflecting a growing philanthropic and professional investment in quantifying these workforce dynamics. The research protocol was reviewed and approved by the Advarra Institutional Review Board, participation was voluntary with informed consent obtained, and responses were anonymized to maintain confidentiality.

The authors’ recommendations center on structural fixes rather than individual resilience. Clearer policy frameworks, they argue, are needed at both the program and institutional levels, so that pregnant residents know in advance what accommodations they are entitled to and how to request them. Strengthened institutional communication, potentially through designated institutional officials (DIOs) and program directors, could ensure that existing resources actually reach the trainees who need them. The findings also implicitly support staffing innovations, such as float pools, cross-coverage systems, and flexible scheduling architectures, that would reduce the extent to which any single resident’s accommodation is treated as an unsolvable coverage crisis. Because the survey captured trainees across ACGME-accredited specialties and 165 institutions, the patterns it documents are unlikely to be idiosyncratic to a handful of programs; they point to systemic features of how graduate medical education is organized in the United States.

What makes this study resonate beyond academic medicine is the broader story it tells about institutions adapting, or failing to adapt, to a workforce that can no longer be assumed to be childless, male, or endlessly available. Nearly three-quarters of the respondents were women, a quarter of them were pregnant during training, and two in five of those pregnancies involved medical complications. Against that backdrop, barriers cited by roughly half of respondents, staffing, program rules, institutional requirements, read less like administrative details and more like a design flaw in the training system itself. The authors frame their conclusions cautiously, noting that clearer policies and better communication may improve trainee well-being and advance a more equitable and sustainable medical workforce. The data they have assembled give program directors, hospital administrators, and accreditation bodies a concrete evidence base for making those changes, and they give current and future residents something they have often lacked: confirmation that the difficulties they face during pregnancy in training are widespread, measurable, and, crucially, fixable.

Subject of Research: Pregnancy experiences and accommodation barriers among U.S. medical residents during graduate medical education

Article Title: Pregnancy and accommodations during residency training- insights from a multi-institutional survey of U.S. trainees

Article References: Bellini, L. M., Dine, C. J., Fisher, R. E., Wagner, C., Moyer, D., Weber, K., & Mladenovic, J. (2026). Pregnancy and accommodations during residency training- insights from a multi-institutional survey of U.S. trainees. BMC Medical Education. https://doi.org/10.1186/s12909-026-10211-z

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10211-z

Keywords: pregnancy, residency training, graduate medical education, ACGME, accommodations, resident well-being, gender equity, pregnancy loss, workforce policy, medical education, reproductive health, survey research