Association of migration with hospitalization and mortality among Medicare beneficiaries with end-stage renal disease in Puerto Rico after Hurricane Maria Authors: Daeho Kim, Sweta Patel, Elizabeth Fussell, Rajnish Mehrotra, Amal N. Trivedi, Maricruz Rivera-Hernandez Published: 2026 ABSTRACT BackgroundHurricane Maria, which struck Puerto Rico (PR) on September 20, 2017, was one of the worst natural disasters in modern US history. Maria disproportionately impacted chronically ill patients, such as those with end-stage renal disease (ESRD), some of whom migrated to the US mainland to receive care.ObjectiveWe aim to examine the association of migration with hospitalization and mortality among Medicare beneficiaries with ESRD.MethodsUsing Medicare claims data, we applied a difference-in-differences (DD) approach to compare changes in hospitalization and mortality after Hurricane Maria among migrants with ESRD relative to concurrent changes among non-migrant ESRD patients in PR from 2016 through 2019.ResultsThe study population of 10,630 Medicare beneficiaries with ESRD initially residing in Puerto Rico between 2016 to 2019 included 1,046 (9.8%) migrants to the mainland US and 9,584 (90.2%) non-migrants who stayed in PR. After adjusting for patient characteristics, prior hospitalizations, and calendar-quarter fixed effects, migration was associated with reductions in the number of hospital admissions per quarter by 0.10 (relative reduction of 19.2%) and hospital days per quarter by 0.81 (23.5% relative reduction), with no significant impact on mortality.ConclusionsResults show that migrants from PR to the US after Hurricane Maria experienced fewer hospitalizations and unchanged mortality. The findings suggest that migration from the disaster-affected area was associated with a reduction in the use of acute care among patients with ESRD. Introduction Hurricane Maria, which struck Puerto Rico (PR) on September 20, 2017, was one of the worst natural disasters in modern US history (FEMA, 2020). Maria led to about 3,000 deaths and $139 billion in damage to the island’s infrastructure (George Washington University, 2018; Florido, 2018). Hurricane Maria may have disproportionately impacted chronically ill patients, such as those with end-stage renal disease (ESRD) who may have been prevented from accessing dialysis and other medical care due to prolonged electricity outages following Maria (Lempert and Kopp, 2013; Bonilla-Felix and Suarez-Rivera, 2019). Our prior work in Puerto Rico before Hurricane Maria indicated that patients – including people with kidney failure – faced challenges in accessing specialty care that resulted in disparate outcomes (Rivera-Hernandez et al., 2016; Rivera-Hernandez et al., 2020; Rivera-Hernandez et al., 2021). Most patients with ESRD qualify for Medicare coverage, regardless of their age, and receive hemodialysis at outpatient dialysis facilities three times a week (Eggers, 2000). Temporary closure of dialysis facilities limited access to dialysis during Maria in some areas in Puerto Rico, and thus some patients needed to travel to other facilities on the island or to the US mainland to receive care (Bonilla-Felix and Suarez-Rivera, 2019; Rodriguez, 2018; Robles, 2017). Maria may have disrupted regular dialysis care, either in-center or home dialysis, leading to higher risk of death, hospitalization, or other adverse events (Andújar-Rivera, 2019). Previous hurricanes caused higher rates of hospitalizations, emergency department visits and mortality among patients with kidney failure (Kelman et al., 2015; Blum et al., 2022). After Hurricane Katrina, the number of missed dialysis sessions increased, which caused hypertensive emergencies, respiratory failure, and other complications that required hospitalizations (Anderson et al., 2009). Similarly, after Hurricane Sandy there was an increase in emergency department visits, hospitalizations, and mortality (Kelman et al., 2015). The importance of disaster planning for dialysis patients has grown after Hurricanes Katrina and Sandy (Kopp et al., 2007). However, there is insufficient evidence about whether migration out of the disaster zone harms or helps people with ESRD. While prior literature has described displacement and relocation patterns among kidney failure patients after a large-scale disaster (Kutner et al., 2009), little is known about the impacts of Hurricane Maria on ESRD patient outcomes (Smith et al., 2020) and the role of migration in moderating those outcomes. On one hand, patients who left PR may have missed treatments if they were not able to locate a new dialysis unit in the US. On the other hand, patients who stayed in PR may have missed treatment if the dialysis unit where they received care were damaged or they were unable to access another provider. Although evidence from past studies suggests that people who migrate might be slightly healthier than non-migrants (Abraido-Lanza et al., 1999; Rubalcava et al., 2008; Guillot et al., 2018), there are several reasons that all-cause hospitalization and mortality may be higher among migrants with ESRD. First, migrant patients may not have their medical records and prescription medications and may need to find new providers, which may disrupt the continuity of care and increase their likelihood of hospitalization or death (Arrieta et al., 2007). Second, migrant patients may have been evacuated on the recommendation of their care providers because they were very sick and could not be treated in the compromised health care system in PR and were more likely to be hospitalized or die after their arrival in the US mainland. It is important to understand the consequences of this large-scare natural disaster among this vulnerable population, given social and healthcare inequalities, and the extensive level of destruction in healthcare infrastructure caused by Maria on the island. In this study, we compared changes in hospitalization and mortality before and after Maria among migrants with ESRD (who moved from PR to the US) to concurrent changes among non-migrants with ESRD (who stayed in PR). Methods Study population and data sources The study population included all Medicare beneficiaries with ESRD (both patients undergoing dialysis and kidney transplant recipients), defined as any beneficiary who had at least one month of Medicare eligibility for ESRD benefits from January 2016 to December 2019. We used a 2-year look back period (January 2014 to December 2015) and a 2-year forward-looking period (January 2020 to December 2021) to trace the same patients over time in order to identify their migration status. We also used a 1-year look back period for prior hospitalizations. We excluded patients who did not have both Parts A and B. The final study sample included 10,630 patients. The primary data sources were 100% Centers for Medicare and Medicaid Services (CMS) Master Beneficiary Summary File (MBSF), Medicare Provider Analysis and Review (MedPAR) File, and Medicare Advantage (MA) encounter data. The MBSF was used to identify Medicare beneficiary’s ESRD status, the place of residence in each month of year, and the date of death. The MBSF also contains beneficiary’s demographic characteristics such as age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid which is primarily determined by beneficiary’s income and asset (Roberts et al., 2023), and monthly indicators for enrollment in MA. The MedPAR file and MA encounter data contain hospital inpatient claims including the dates of admission and discharge for traditional fee-for-service (FFS) Medicare enrollees and MA enrollees, respectively. We used MedPAR and MA encounter data to measure the number of hospital admissions and hospital days per person per quarter. The study was approved by the Brown University Institutional Review Board with patient informed consent waived. Analyses were conducted using Stata version 18 (StataCorp). We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. Measures We identified a migrant from PR to the mainland US (US hereafter) if a beneficiary in PR moved to the US at any point during the study period and had stayed in the US at least one year. In sensitivity analyses, we used different definitions of migration: one based on the minimum of six-month stay in the US; and the other based on the minimum of two-year stay in the US. The primary outcomes were the number of hospital admissions, hospital days, and binary indicator for death in a given quarter. If a beneficiary had multiple hospital episodes in a quarter, the number of hospital admissions and hospital days are summed over the quarter. The mortality was measured by an indicator for death in a given quarter based on the date of death. For deceased persons, the migration status was determined based on their status in a month prior to the death. The primary explanatory variable of interest was an interaction of the indicator of being a migrant and the indicator of post-Maria period (from the fourth quarter of 2017 to the last quarter of 2019). Other covariates included age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, MA enrollment, and prior hospitalizations (i.e., hospitalization in a year prior to migration for migrants and hospitalization in 2015 (as baseline hospitalization) for non-migrants). For individuals who were newly enrolled in Medicare after 2016, we assumed no prior hospitalizations. In a sensitivity analysis, we excluded those new enrollees. Statistical analysis We applied an event-study design that compares the difference in outcomes between migrants and non-migrants in each quarter, relative to the difference in the quarter before Hurricane Maria. Specifically, we estimated the following regression:(1)Yit=α+γMigranti+∑k=−7k≠−19θk(Migranti×Di,k)+μt+βXit+εitwhere Yit is the outcome (e.g., hospitalization) for individual i in time t (measured in year-quarter); Migranti is an indicator for an individual who ever migrated from PR to the US; Di,k is an event-time dummy indicating k quarters from Hurricane Maria, where the quarter before Maria (k=−1) is set as the reference period; μt is year-quarter fixed effects; Xit is a vector of covariates (age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, enrollment in MA, and prior hospitalizations); and εit is an idiosyncratic error term. The parameters of interest are θk for k≥0. The pre-Maria coefficients, θk for k<0, were assessed to test the parallel trends assumption. Standard errors were adjusted for heteroscedasticity and clustered at the person level. The unit of analysis was person-quarter. As a summary measure of an association between migration and outcomes, we also applied the following difference-in-differences (DD) approach that compares changes in the outcomes before and after Hurricane Maria among migrants with ESRD to concurrent changes among non-migrants with ESRD from 2016 through 2019:(2)Yit=α+γMigranti+Postt+θ(Migranti×Postt)+δt+βXit+εitwhere Postt is an indicator for time period after Maria; δt is quarter fixed effects to address potential seasonality or cyclicality in the outcome; and all other variables are the same as those in Eq. (1). The study period spans 1 year and 3 quarters (first quarter of 2016 to third quarter of 2017) before Hurricane Maria and 2 years and 1 quarter (fourth quarter of 2017 to the last quarter of 2019) after Maria. Sensitivity analysis We tested for parallel trends in outcomes between migrants and non-migrants during the pre-Maria period. In addition, we conducted analyses applying different ways of identifying migration from PR to the US: a Puerto Rican beneficiary was identified as a migrant if the beneficiary moved to the US and had stayed in the US at least i) 6 months or ii) 2 years. Furthermore, we applied alternative specifications for regressions – i.e., Poisson regression model for the number of hospital admissions and hospital days (count variables) and logistic regression model for mortality (binary variable). Finally, we excluded new Medicare beneficiaries who enrolled after 2015. Results The study population of 10,630 Medicare beneficiaries with ESRD initially residing in Puerto Rico between 2016 to 2019 included 1,046 (9.8%) migrants to the mainland US (mean [SD] age, 60.1 [15.4] years; 39.7% female; 75.2% dual-eligible; 15.9% enrolled in MA) and 9,584 (90.2%) non-migrants who stayed in PR (mean [SD] age, 64.6 [13.2]; 36.6% female; 2.0% dual-eligible; 54.8% enrolled in MA) (Table 1).Table 1Characteristics of Medicare beneficiaries with ESRD by migration status, 2016 to 2019.Table 1: dummy alt textMigrantsNon-migrantsP-valueNumber of person-quarter observations8,34290,919Number of patients1,0469,584Age (SD), y60.1 (15.4)64.6 (13.2)<0.001Sex, %Male60.363.4<0.001Female39.736.6<0.001Race and ethnicity, %White, non-Hispanic2.00.2<0.001Hispanic96.999.8<0.001Dual eligibility, %75.22.0<0.001MA enrollment, %15.954.8<0.001Notes: All outcomes are measured per person per quarter.ESRD = End-Stage Renal Disease, MA = Medicare Advantage. Table 1Characteristics of Medicare beneficiaries with ESRD by migration status, 2016 to 2019.Table 1: dummy alt textMigrantsNon-migrantsP-valueNumber of person-quarter observations8,34290,919Number of patients1,0469,584Age (SD), y60.1 (15.4)64.6 (13.2)<0.001Sex, %Male60.363.4<0.001Female39.736.6<0.001Race and ethnicity, %White, non-Hispanic2.00.2<0.001Hispanic96.999.8<0.001Dual eligibility, %75.22.0<0.001MA enrollment, %15.954.8<0.001Notes: All outcomes are measured per person per quarter.ESRD = End-Stage Renal Disease, MA = Medicare Advantage. Characteristics of Medicare beneficiaries with ESRD by migration status, 2016 to 2019. Notes: All outcomes are measured per person per quarter. ESRD = End-Stage Renal Disease, MA = Medicare Advantage. Fig. 1 plots the proportion of migrants from PR to the US. It shows a substantial increase in migration from about 5% before Hurricane Maria to about 10% after Maria, with a relative increase of about 100%.Fig. 1Proportion of migrants from Puerto Rico to the mainland US.Note: The vertical line indicates the occurrence of Hurricane Maria (September 2017).Fig 1: dummy alt text Proportion of migrants from Puerto Rico to the mainland US. Note: The vertical line indicates the occurrence of Hurricane Maria (September 2017). Panel A of Fig. 2 plots the event-study estimates for the differences in the number of hospital admissions per person per quarter between migrants and non-migrants. The figure shows that the number of hospital admissions were very similar between the two groups before Maria, but started to decline among migrants relative to non-migrants after Maria. The DD estimates in Table 2 confirm this pattern. The number of hospital admissions among migrants decreased from 0.52 to 0.49 (per quarter) after Maria, compared to an increase from 0.33 to 0.38 (per quarter) among non-migrants, leading to an unadjusted DD estimate of -0.08. An adjusted DD estimate indicates a reduction of 0.10 per quarter among migrants compared to non-migrants, which corresponds to a 19.2% decrease relative to the pre-Maria mean of 0.52. Panel B of Fig. 2 depicts the event-study estimates for the differences in hospital days per person per quarter, showing a reduction in the days among migrants compared to non-migrants after Maria relative to before. Table 2 presents the corresponding DD estimates. An adjusted DD estimate was -0.81 days per quarter, a 23.5% relative decrease. Panel C of Fig. 2 shows the event-study estimates for the differences in mortality rate, which reveals no differential changes between migrants and non-migrants after Maria relative to before. The adjusted DD estimate in Table 2 confirms this, with an insignificant estimate of -0.70.Fig. 2Event-study estimates.A. Number of hospital admissions.B. Hospital days.C. Mortality rate (%).Notes: Figures plot the regression coefficients with 95% confidence intervals from the event-study model adjusted for age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, being enrolled in Medicare Advantage, and hospitalization in a prior year. Heteroskedasticity-robust standard errors are clustered at the person level. The vertical line indicates the occurrence of Hurricane Maria (September 2017).Fig 2: dummy alt textTable 2Difference-in-differences estimates among ESRD patients initially residing in Puerto Rico.Table 2: dummy alt textMigrantNon-migrantDD estimateBefore MariaAfter MariaBefore MariaAfter MariaUnadjustedAdjustedNumber of hospital admissions0.520.490.330.38-0.08⁎⁎⁎-0.10⁎⁎⁎(0.02)(0.03)Hospital days3.443.153.123.50-0.66⁎⁎⁎-0.81⁎⁎⁎(0.21)(0.26)Mortality rate (%)3.493.594.144.180.06-0.70(0.46)(0.47)Notes: All outcomes are measured per quarter. Adjusted for age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, being enrolled in Medicare Advantage, hospitalization in a prior year, as well as calendar-quarter fixed effects. Heteroskedasticity-robust standard errors (in parenthesis) are clustered at the person level.ESRD = End-Stage Renal Disease.⁎⁎⁎p < 0.01; ** p < 0.05; * p < 0.1. Event-study estimates. A. Number of hospital admissions. B. Hospital days. C. Mortality rate (%). Notes: Figures plot the regression coefficients with 95% confidence intervals from the event-study model adjusted for age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, being enrolled in Medicare Advantage, and hospitalization in a prior year. Heteroskedasticity-robust standard errors are clustered at the person level. The vertical line indicates the occurrence of Hurricane Maria (September 2017). Table 2Difference-in-differences estimates among ESRD patients initially residing in Puerto Rico.Table 2: dummy alt textMigrantNon-migrantDD estimateBefore MariaAfter MariaBefore MariaAfter MariaUnadjustedAdjustedNumber of hospital admissions0.520.490.330.38-0.08⁎⁎⁎-0.10⁎⁎⁎(0.02)(0.03)Hospital days3.443.153.123.50-0.66⁎⁎⁎-0.81⁎⁎⁎(0.21)(0.26)Mortality rate (%)3.493.594.144.180.06-0.70(0.46)(0.47)Notes: All outcomes are measured per quarter. Adjusted for age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, being enrolled in Medicare Advantage, hospitalization in a prior year, as well as calendar-quarter fixed effects. Heteroskedasticity-robust standard errors (in parenthesis) are clustered at the person level.ESRD = End-Stage Renal Disease.⁎⁎⁎p < 0.01; ** p < 0.05; * p < 0.1. Difference-in-differences estimates among ESRD patients initially residing in Puerto Rico. Notes: All outcomes are measured per quarter. Adjusted for age, sex, race/ethnicity, dual eligibility for Medicare and Medicaid, being enrolled in Medicare Advantage, hospitalization in a prior year, as well as calendar-quarter fixed effects. Heteroskedasticity-robust standard errors (in parenthesis) are clustered at the person level. ESRD = End-Stage Renal Disease. p < 0.01; ** p < 0.05; * p < 0.1. The test of parallel trends assumption confirmed that there was no significant difference in trends between migrant and non-migrants during the pre-Maria period for any of the outcomes (Fig. 2 and Table A1). The main results were consistent in analyses that identify migration based on 6-month and 2-year stay in the US after move from PR (Table A2). In addition, the results were robust to different specifications in the DD estimation (i.e., Poisson and logistic regressions), as shown in Table A3. Furthermore, excluding new Medicare enrollees yielded consistent results (Table A4). Discussion In this study of Medicare beneficiaries with ESRD, we observed that there was a substantial increase in migration from PR to the US after Hurricane Maria. Applying a difference-in-differences approach, we found that the migration was associated with reductions in the number of hospital admissions and hospital days, while no impact on mortality. The results were robust to the model specifications in estimation and the way of defining migrants who stayed at least 6 months, 1 year, or 2 years in the US after moving from PR. The findings of the present study may be explained by uninterrupted and better care among migrants who moved from PR to the US after Maria, compared to a disruption in care among non-migrants staying in PR. Before Hurricane Maria, the PR healthcare system was already facing workforce and funding shortages. People with kidney failure were facing gaps in access to nephrologists and quality care in PR (Rivera-Hernandez et al., 2020). In a decade, the island lost about half of doctors (Marcos, 2024), with the exodus of medical personnel only exacerbated after Maria (Santiago-Santiago et al., 2024). In addition, previous studies on the effect of natural disasters showed that after hurricanes patients with kidney failure living in the affected area missed dialysis sessions (Anderson et al., 2009) and experienced higher rates of hospitalizations and mortality (Kelman et al., 2015). Power outages and limited water impacted the health system in PR. Hospitals and some dialysis facilities were struggling to provide care. There was a shortage of fuel to keep backup generators running, and telephone and internet services were inadequate in some areas (Alcorn, 2017). Without these essential services, healthcare facilities struggled to run machines that need clean water and energy, access health records, and preserve medications. A month after Maria, only 9 out of the 68 hospitals had power restored (FEMA, 2017). PR hospitals and dialysis facilities served a larger volume of patients as compared to pre-disaster period (Robles, 2017), and had to reduce treatment time. Care disruptions may also be related to other infrastructure challenges and lack of social support (Malik et al., 2018). Some communities in PR experienced catastrophic flooding, mudslides and road damages (FEMA, 2017). Exposure to these environmental factors may have caused people with kidney failure to seek care at a hospital, especially among vulnerable groups who may have limited access to social support. Migration dynamics among PR young adults have resulted in disrupted family structures, social support networks and informal caregivers (Acevedo, 2023). Given the high fraction of people in PR who live below poverty level and have a high rate of chronic conditions (Chandra et al., 2021), non-migrants with ESRD may have experienced exacerbations of other conditions. It is noteworthy that our findings could also be explained by systematic differences in healthcare utilization patterns between PR and the US or compositional changes in the migrant population. Although we adjusted for differences in baseline health status as well as demographic characteristics between migrants and non-migrants, there still may be time-varying unobserved differences between the two groups, which may induce selection bias in our results. Our finding on the lack of a discernible association between migration and mortality during the two years after Maria suggest that it may take a longer time for mortality to be impacted significantly. This research expands on prior literature regarding Hurricane Maria and other large-scale disasters in important ways. First, it includes data on migration behavior among people with ESRD before and after Hurricane Maria. Prior research found that after Hurricane Maria, some people with kidney failure received dialysis outside of PR (Rivera-Hernandez et al., 2022). However, it was unknown as to whether these patients moved temporarily or permanently and whether residence in Puerto Rico or the US made a difference in their outcomes. Our results elucidate differences in the associations of migration on hospitalization and mortality among people with complex needs from Puerto Rico. Second, it assesses changes across the entire Medicare population with ESRD in PR, which allows us to track individual information regarding displacement and long-term change of residence after a natural disaster. Prior research regarding large-scale disasters and people with ESRD has largely focused on non-migrants (Smith et al., 2020). We found some sociodemographic differences among migrants and non-migrants and a substantial increase in migration among people with ESRD after Hurricane Maria. With natural disasters becoming more prevalent, datasets and methodologies used in this study could be used to assess the impact of large-scale disasters among other populations in the US mainland and the territories. The study has several limitations. First, the findings only applied to Medicare beneficiaries with ESRD, which are not generalizable to the whole PR population. Second, the results were driven by migration due to Hurricane Maria and hence not apply to other natural disasters. Third, our definition of migration may exclude individuals who migrated to the US and then returned to PR or died within one year from the migration. This may retain healthier individuals in the migrant group. To address this issue, we adjusted baseline health status based on hospitalizations prior to migration and conducted a sensitivity analysis applying an alternative definition of migration employing a shorter minimum duration of stay in the US (6 months instead of 1 year). Fourth, we examined the outcomes during the 2 years after Maria, and thus could not capture longer-term effects of migration. Finally, while the approaches we used in this study (i.e., event-study and difference-in-differences designs) control for time-invariant confounding, we cannot rule out time-varying unobserved factors that may affect the decision to migrate (i.e., self-selection in migration). However, our finding of similar estimates with and without controlling for observed patient characteristics and baseline health status makes this possibility less likely. Conclusion Migration from Puerto Rico to the US after Hurricane Maria was associated with a reduction in hospitalization, but unchanged mortality among Medicare beneficiaries with ESRD. The findings suggest that migration from the disaster-affected area may be associated with lower use of acute care among patients with ESRD. Ethics declaration The author(s) declare(s) that the study does not involve humans nor animal subjects. Funding information This publication was supported by the National Institute on Minority Health and Health Disparities of the National Institutes of Health (award number: R01MD016961) as part of an award totaling $1,454,093. Disclaimer The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the Department of Veterans Affairs or the United States government. CRediT authorship contribution statement Daeho Kim: Writing – review & editing, Writing – original draft, Methodology, Formal analysis, Conceptualization. Sweta Patel: Writing – original draft, Formal analysis, Data curation. Elizabeth Fussell: Writing – review & editing, Writing – original draft, Methodology, Conceptualization. Rajnish Mehrotra: Writing – review & editing, Writing – original draft, Conceptualization. Amal N. Trivedi: Writing – review & editing, Writing – original draft, Methodology, Conceptualization. Maricruz Rivera-Hernandez: Writing – review & editing, Writing – original draft, Supervision, Methodology, Funding acquisition, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.