Long-term healthcare provider availability following large-scale hurricanes: A difference-in-differences study

Abstract

Background

Hurricanes Katrina and Sandy were two of the most significant disasters of the 21st century that critically impacted communities and the health of their residents. Despite the assumption that disasters affect access to healthcare, to our knowledge prior studies have not rigorously examined availability of healthcare providers following disasters.

Objective

The objective of this study was to examine availability of healthcare providers following large-scale hurricanes.

Methods

Using historical data on healthcare providers from the National Plan and Provider Enumeration System and county-level population characteristics, we conducted a quasi-experimental study to examine the effect of large-scale hurricanes on healthcare provider availability in the short-term and long-term. We separately examined availability of primary care physicians, medical specialists, surgeons, and nurse practitioners. A difference-in-differences analysis was used to control for time variant factors comparing county-level health care provider availability in affected and unaffected counties the year before Hurricanes Katrina and Sandy, to five years after each storm.

Results

Counties affected by Hurricane Katrina compared to unaffected locales experienced a decrease of 3.59 primary care physicians per 10,000 population (95% CI: -6.5, -0.7), medical specialists (decrease of 5.9 providers per 10,000 (95% CI: -11.3, -0.5)), and surgeons (decrease of 2.1 (95% CI: -3.8, -0.37)). However, availability of nurse practitioners did not change appreciably. Counties affected by Hurricane Sandy exhibited less pronounced changes. Changes in availability of primary care physicians, nurse practitioners, medical specialists, and surgeons were not statistically significant.

Conclusion

Large-scale hurricanes appear to affect availability of healthcare providers for up to several years following impact of the storm. Effects vary depending on the characteristics of the community. Primary care physicians and medical specialists availability was the most impacted, potentially having long-term implications for population health in the context of disaster recovery.

About the Authors:

Sue Anne Bell

Contributed equally to this work with: Sue Anne Bell, Katarzyna Klasa, Theodore J. Iwashyna, Edward C. Norton, Matthew A. Davis

Roles Conceptualization, Data curation, Formal analysis, Funding acquisition, Methodology, Project administration, Supervision, Writing – original draft, Writing – review & editing

* E-mail: [email protected]

Affiliations Department of Systems, Populations, and Leadership, University of Michigan School of Nursing, Ann Arbor, Michigan, United States of America, Institute for Healthcare Policy and Innovation, University of Michigan, Ann Arbor, Michigan, United States of America

ORCID logo https://orcid.org/0000-0002-6890-2006

Katarzyna Klasa

Contributed equally to this work with: Sue Anne Bell, Katarzyna Klasa, Theodore J. Iwashyna, Edward C. Norton, Matthew A. Davis

Roles Formal analysis, Writing – original draft, Writing – review & editing

Affiliation: Department of Health Management and Policy, University of Michigan School of Public Health, Ann Arbor, Michigan, United States of America

ORCID logo https://orcid.org/0000-0002-4468-3603

Theodore J. Iwashyna

Contributed equally to this work with: Sue Anne Bell, Katarzyna Klasa, Theodore J. Iwashyna, Edward C. Norton, Matthew A. Davis

Roles Methodology, Supervision, Writing – review & editing

Affiliations Institute for Healthcare Policy and Innovation, University of Michigan, Ann Arbor, Michigan, United States of America, Department of Internal Medicine, University of Michigan Medical School, Ann Arbor, Michigan, United States of America, University of Michigan Institute for Social Research, Ann Arbor, Michigan, United States of America

Edward C. Norton

Contributed equally to this work with: Sue Anne Bell, Katarzyna Klasa, Theodore J. Iwashyna, Edward C. Norton, Matthew A. Davis

Roles Methodology, Supervision, Writing – review & editing

Affiliations Institute for Healthcare Policy and Innovation, University of Michigan, Ann Arbor, Michigan, United States of America, Department of Health Management and Policy, University of Michigan School of Public Health, Ann Arbor, Michigan, United States of America

Matthew A. Davis

Contributed equally to this work with: Sue Anne Bell, Katarzyna Klasa, Theodore J. Iwashyna, Edward C. Norton, Matthew A. Davis

Roles Conceptualization, Methodology, Supervision, Validation, Writing – original draft, Writing – review & editing

Affiliations Department of Systems, Populations, and Leadership, University of Michigan School of Nursing, Ann Arbor, Michigan, United States of America, Institute for Healthcare Policy and Innovation, University of Michigan, Ann Arbor, Michigan, United States of America, Department of Learning Health Sciences, University of Michigan Medical School, Ann Arbor, Michigan, United States of America

ORCID logo https://orcid.org/0000-0001-6544-0451

Introduction

Hurricanes Katrina and Sandy were two of the most significant disasters of the 21st century, critically impacting communities [1–7]. During Hurricane Katrina in 2005, over 80% of New Orleans flooded, 1,577 people died, and the total sustained damages were estimated at $108 billion [8]. Widespread evacuations and long-term migration away from affected locales occurred. Orleans and Jefferson parishes, both in the New Orleans metropolitan statistical area, experienced a 40% and 19% drop in population respectively [9]. Seven years later, in 2012, Hurricane Sandy made landfall on the eastern seaboard of the United State resulting in severe damage to the New York and New Jersey area. Hurricane Sandy caused over $50 billion in damages [10] and resulted in an estimated 72 deaths [11]. Hurricanes Katrina and Sandy were two of the most damaging, and costly, in recent history, but also have now occurred with enough elapsed time in which to evaluate long-term sequelae on communities [12].

Healthcare provider availability has been associated with reduced mortality [13] and access to primary care physicians has been shown to affect health outcomes among Medicare beneficiaries [14]. Maintaining adequate access to healthcare services is critical for community recovery after disaster. As such, there is considerable interest in understanding the extent to which healthcare resources are affected by disasters. Underpinning disaster preparedness planning is the assumption that access to healthcare services diminishes for a period of time (be it short or long) as a community recovers. Furthermore, the socioeconomic characteristics of the population likely contribute to the recovery of the community, which in turn influences the degree to which healthcare provider remain in the locale, as healthcare providers tend to concentrate in affluent locales [15].

However, very few studies have empirically examined the effect of disasters on healthcare provider availability. One such study of physician characteristics and outflow after the 2011 Great Japan Earthquake and Fukishima nuclear accident found that early career physicians were more at risk for leaving the affected area in the year after the earthquake [16]. Other prior studies are descriptive in nature, examining short-term changes after Hurricane Katrina, finding that roughly 25% of the New Orleans area’s physicians had returned to practice two years after Katrina [17]. To our knowledge no prior study has examined long-term availability of health care providers after disasters. Therefore, we rigorously evaluated the impact of hurricanes on healthcare provider availability. To explore how effects may differ dependent upon community-level socioeconomic characteristics, we selected hurricanes Katrina and Sandy which affected two very different populations.

Methods

Using a combination of provider location and population data, we performed a quasi-experimental study to examine the extent to which healthcare provider availability is affected by large-scale hurricanes up to five years post hurricane. To control for time-variant factors, we compared availability before versus after hurricane landfall using difference-in-differences (DID) analysis. DID uses data from treatment groups and control groups (in our analysis, hurricane-affected counties and unaffected counties, respectively) to obtain an appropriate counterfactual to estimate a causal effect over time [18]. This study used publicly available data; and, therefore, was determined to be exempt from review by the University of Michigan Health Sciences and Behavioral Sciences Institutional Review Board.

Healthcare provider availability

Historic data from the National Plan and Provider Enumeration System (NPPES) were used to identify healthcare providers. The NPPES provides basic information about individual healthcare providers who have a National Provider Identifier (NPI). We examined availability of four types of healthcare providers: primary care physicians, nurse practitioners, medical specialists, and surgeons, following Centers for Medicare and Medicaid Services (CMS) provider taxonomy and specialty codes [19]. We operationally defined primary care physicians as those specializing in family practice, internal medicine, or general practice. Nurse practitioners provide primary care often without the supervision of a physician, and supplement primary care provision in many communities. This study excluded certain types of advanced practice nurses such as certified midwifes, nurse anesthetists, and certified clinical nurse specialists. To be included in this study, providers had to have an active license and valid primary practice address.

In 2007, the CMS changed the provider identifier used on billed services from Unique Provider Identification Number (UPIN) to NPI. The NPI is a unique 10-digit numeric identification number for covered healthcare providers, and is permanently associated with a specific individual regardless of changes in practice or location [20, 21]. Any provider, organization or individual, that is a HIPAA-covered entity must have an NPI [22]. For this study, only individual healthcare providers were included. NPPES data have been shown to provide reasonably accurate, up-to-date address information for physicians billing public and private insurers, when compared to similar databases such as the AMA Masterfile [23].

Datasets from 2001 to 2017 were merged into a single joint UPIN-NPI dataset file with total healthcare providers by Zone Improvement Plan (ZIP) codes. A ZIP code to Federal Information Processing Standards (FIPS) crosswalk file was used to create a final sum of total healthcare providers in each county. Some years of data for UPIN and/or NPI directories were missing or inaccessible: 2003, 2008, and 2014. The year 2005 was skipped as it was the year of Hurricane Katrina, as was 2012 for Hurricane Sandy. Last, the year 2006 was dropped for Hurricane Katrina because of a large temporary decrease in population across affected counties, causing a false increase in rates of providers.

Counties affected by Hurricane Katrina and Sandy were identified using Federal Emergency Management Agency disaster declarations (S1 Table) [24]. Affected counties were limited to one state per storm (Louisiana for Hurricane Katrina and New Jersey for Hurricane Sandy) in order to allow for closer matching of control counties. For each calendar year, we created county healthcare provider to population (per 10,000 adult capita) ratios for each provider type.

Selection of unaffected locales

For comparison purposes, control counties were randomly selected from U.S. counties unaffected by the respective hurricane and outside of the affected state, selecting counties based on a priori specified characteristics: population size and demographics, median county-level household income, number of hospitals, county-level population growth rate and median age. Among these randomly selected counties, counties were then further restricted to counties that (1) were not statistically different from affected counties in socio-economic and demographic characteristics, and (2) did not experience a federally declared disaster of similar scope during the study period. Counties were also excluded that had missing values for at least one of the four types of healthcare providers between 2001 to 2010. Counties that experienced a sizeable influx of Katrina migrants up to a year after the hurricane occurred were also excluded.

County population measures

We obtained population estimates and county-level sociodemographic measures from U.S. Census estimates for each of the aforementioned years, including county-level estimates of median household income, age, race/ethnicity and sex. The American Hospital Association annual survey and the Area Health Resource File were used to obtain healthcare indicators at the county-level.

Statistical analyses

For each hurricane we compared the differences pre- versus post-storm by examining the association between county-level healthcare provider supply in affected and control counties. For visual comparison, we display the change in provider to population ratios two years before and five after storms.

A difference-in-differences analysis was conducted in which we compared the difference between the mean county rate of healthcare providers (per 10,000 adult capita) before and after a hurricane for affected and unaffected counties to estimate the effect of exposure to a hurricane on provider availability [18]. We estimated effects based on the change in provider to population ratio in the short-term, by comparing the year before versus two years after the hurricanes, and in a longer period, by comparing the year before to the fifth year after the hurricanes. Difference in trends examined the difference over time between the two slopes of county ratios of providers, testing whether the change between the pre-to-post slope (using years 2001 to 2010 for Hurricane Katrina and 2008 to 2016 for Hurricane Sandy) for the control counties was statistically different from the change between the pre-to-post slope for the disaster counties. An assumption of DID analyses is that there are no significant change between groups relative to one another prior to the change being measured [25]. Parallel trends over time were accounted for in our analysis, using a visual check and a failure to reject the null hypothesis when comparing pre-period means and slopes. We assume that in the absence of the hurricane, the control and treatment groups would continue to have similar trends in provider rates into the post-period. County population growth rates were also controlled for in our analyses through control county case selection.

Selection of appropriate controls (i.e., unaffected counties) is important. Therefore, for Hurricane Katrina we also performed a sensitivity analysis by comparing affected counties to control counties in the state of Louisiana. Considering the entire state of Louisiana was affected to some degree by Hurricane Katrina, the rationale for this additional analysis was to evaluate the change in provider availability within the state of Louisiana, alongside our primary analyses (S2 Table). Analyses were conducted using Stata statistical software (version 15.1; StataCorp, College Station, TX). All models used ordinary least squares regression with robust standard errors. For all models, we ran a test of heteroskedasticity and a test of variance inflation factor. All hypothesis tests were two-sided with the critical alpha level at 0.05.

Results

Populations affected by Hurricanes Katrina and Sandy

The total population affected by Hurricane Katrina decreased after the storm (total population in affected counties decreased by 18.9 thousand adults), whereas Hurricane Sandy affected counties actually experienced population growth over time (Table 1). Concurrently, the total population among unaffected counties steadily increased over the same time frame for both hurricanes.

[Figure omitted. See PDF.]

Table 1. Characteristics of affected and unaffected counties included in study.

https://doi.org/10.1371/journal.pone.0242823.t001

County-level characteristics differed in meaningful ways among affected counties of Hurricanes Katrina versus Sandy (Table 1). For instance, the percentage of individuals that were African-American in Hurricane Katrina counties (29.5%) was close to double that of Hurricane Sandy affected populations (14.6%). The median household income in Hurricane Katrina counties was significantly lower than Sandy ($45,800 for Katina versus $65,000 for Sandy). Overall unaffected (i.e., control) counties were similar in characteristics to affected counties for Sandy but several differences were noted among Katrina unaffected counties.

Availability of healthcare providers following Hurricane Katrina

When comparing short-term (2004 vs. 2007) pre- versus post-hurricane county-level healthcare provider ratios, primary care physicians decreased by 3.59 providers per 10,000 (95% CI -6.5, -0.7) from pre-hurricane levels. Medical specialists decreased by 5.9 providers per 10,000 (95%CI -11.3, -0.5) and surgeons decreased by 2.1 providers per 10,000 (95% CI -3.8, -0.37) from pre-hurricane levels. Only nurse practitioners did not see an appreciative change (-0.45 provider per 10,000, 95% CI -1.5, 0.6). No significant associations were found when examining the difference in trends over time for difference in difference representation (Tables 2 and 3, and Fig 1).

[Figure omitted. See PDF.]

Fig 1. Provider to population ratio (No. per 10,000) before versus after Hurricane Katrina.

(A) primary care physicians, (B) nurse practitioners, (C) medical specialists, and (D) surgeons. The x-axis consists of study time periods that go from -2 to 4. In the pre-period, time -2, -1, and 0 corresponds with 2001, 2002, and 2004, respectively. In the post-period, time 0, 2, 3, and 4 corresponds with 2007, 2009, 2010, and 2011, respectively.

https://doi.org/10.1371/journal.pone.0242823.g001

[Figure omitted. See PDF.]

Table 2. Short-term changes in provider ratios by county among counties impacted by Katrina (2004 vs. 2007) and Hurricanes Sandy (2011 vs. 2013) versus matched controls.

https://doi.org/10.1371/journal.pone.0242823.t002

[Figure omitted. See PDF.]

Table 3. Long-term changes in provider ratios by county among counties impacted by Katrina (2004 vs. 2010) and Hurricanes Sandy (2011 vs. 2017) versus matched controls.

https://doi.org/10.1371/journal.pone.0242823.t003

When comparing long-term (2004 vs. 2010) pre- versus post-hurricane county-level healthcare provider ratios, primary care physicians decreased by 4.4 providers per 10,000 (95% CI -7.4, -1.4). Medical specialists decreased by 7.3 providers per 10,000 (95%CI -13, -1.7) and surgeons decreased by 2.4 providers per 10,000 (95% CI -4.2, -0.53). When we added additional control variables to control for confounding, the magnitude of the variable of interest did not change appreciably for both short-term and long-term changes.

As a sensitivity analysis we evaluated the effects of provider to population ratios among control counties within the state of Louisiana. Across all four provider types, no significant change in provider population ratios of healthcare providers were observed; primary care physicians, medical specialists, surgeons, and nurse practitioners. Similarly, no significant differences in trends were observed. (S2 Table).

Availability of healthcare providers following Hurricane Sandy

For Hurricane Sandy, little change in availability was observed. When comparing the short-term changes (2011 vs. 2013), healthcare providers per population ratios decreased. Among primary care physicians, the provider per population ratio decreased by 0.9 (95%CI -3.8, -2.0) from a pre-hurricane levels. Medical specialists also had a small decrease, at 1.21 (95%CI -6.5, 4.07), as did nurse practitioners (-0.24, 95%CI -2.6, 2.2) and surgeons (-0.21, 95%CI -1.9, 1.5). However, none of these changes were statistically significant. When comparing long-term changes (2011 vs. 2017), the difference-in-differences changes for providers were small and not statistically significant. No significant changes were found when examining the difference in trends over time. See Tables 2 and 3, and Fig 2.

[Figure omitted. See PDF.]

Fig 2. Provider to population ratio (No. per 10,000) before versus after Hurricane Sandy.

(A) primary care physicians, (B) nurse practitioners, (C) medical specialists, and (D) surgeons. The x-axis consists of study time periods that go from -2 to 4. In the pre-period, time -2, -1, and 0 corresponds with 2009, 2010, and 2011, respectively. In the post-period, time 0, 2, 3, and 4 corresponds with 2013, 2015, 2016, and 2017, respectively.

https://doi.org/10.1371/journal.pone.0242823.g002

Discussion

This study evaluated the association between changes in county-level healthcare provider supply after Hurricanes Katrina in 2005 in Louisiana and Hurricane Sandy in 2012 in New Jersey, using a difference-in-differences analysis. These two historic and devastating hurricanes are test cases to evaluate long-term population level effects on two quite different disaster-affected communities in terms of sociodemographic characteristics. Studying changes in healthcare provider availability can inform long-term community recovery after disaster.

In this study, we found evidence over time of county-level availability decreasing among several types of healthcare providers after Hurricane Katrina including primary care physicians, medical specialists and surgeons. No significant changes were seen in affected communities after Hurricane Sandy. This is likely attributable to the differences in the way the two storms impacted communities, but also in the sociodemographic differences between the two settings, including racial and economic disparities. For example, the median household income for Hurricane Katrina affected counties at $45,800 was less than the national average ($50,233) for the study time period, while Hurricane Sandy affected counties at $65,000 were well above the national average ($56,516).

An unprecedented migration from the New Orleans area occurred after Hurricane Katrina, where many thousands left the affected areas. In Louisiana, lower income populations stayed, while higher income populations, including higher-earning healthcare providers, left [17]. Counties affected by Hurricane Katrina already faced socioeconomic disparities (see Table 1); the impact of the storm and its resultant years of recovery undoubtedly exacerbated this.

Availability of medical specialists and primary care physicians were most impacted by Hurricane Katrina. Medical specialists have a broad definition in CMS, and the inclusion of many different types of providers and specialties further boosts our study results. Ratios of nurse practitioners grew considerably over both Katrina and Sandy study time periods, likely due to an increase in volume of nurse practitioners nationwide [26], resulting in difficulty assessing the impact of the hurricanes on NP supply. Primary care physicians had a significant decrease after Hurricane Katrina. This finding is especially critical given that the state of Louisiana continues to have inadequate numbers of primary care physicians to meet the current demand [27], alongside, the demand for primary care physicians outpaced the supply nationwide [28]. Attracting primary care providers (including nurse practitioners) to areas with long-term post-disaster recovery needs is a difficult prospect that policymakers have struggled to address.

While our analyses demonstrated statistically significant differences in overall availability, we did not detect meaningful differences in trends (i.e., the slopes of availability before versus after storms). Finally, in our sensitivity analysis, using controls within the state of Louisiana, no significant findings were observed in either analysis. This implies that changes in provider supply were observed state-wide in Louisiana where virtually no parishes (the term parish is used by the State of Louisiana, in place of the equivalent term, county) were spared differential impacts from the hurricane. In fact, non-affected parishes had overall lower rates of all providers compared to affected parishes both before and after the disaster.

This study was not without limitations. First, the NPPES was designed for administrative purposes, not for tracking the health professional workforce. Comparisons to other data sources suggest that this data source provides a reasonably accurate picture of the aggregate supply of providers across the geographic regions studied [23]. Second, there are limitations associated with NPI data, where certain types of providers are likely better represented in the data than those who may bill under another provider’s NPI or organization NPI, such as nurse practitioners. Providers who are not actively practicing can still have an NPI number. Third, the switch from UPIN to NPI data occurred in 2007; and, it most affected NPs and PCPs. Thus, the Hurricane Katrina DiD analysis for PCPs and NPs could have picked up this change. Finally, criteria for selection of control counties was based on literature and policy review, after considering other alternatives such as propensity matching [29].

Availability of healthcare providers has direct impacts on the health of affected communities, making this issue an important aspect of disaster recovery. Limited healthcare provider availability contributes to adverse individual and population level health effects, where these effects are compounded in communities recovering from disaster. These needs are known to be greater in communities that are sociodemographically disadvantaged, making the lack of medical specialists and primary care physicians especially concerning. Efforts to address the primary care provider shortage occurred after Hurricane Katrina, for example through a Primary Care Access and Stabilization Grant to the Louisiana Department of Health and Hospitals from the U.S. Department of Health and Human Services [30]. This program was intended to support the restoration of restore access to health care in communities affected by Hurricane Katrina.

Community recovery from disaster has no specific endpoint, and most consider both hurricane-affected regions as still in the recovery process today. Creative methods are needed to encourage primary care providers to return to, or to relocate to, communities that may not be attractive as they struggle to recover. Health workforce policy, such as that from the Health Resources and Services Administration (HRSA), could work alongside disaster and emergency management policymakers to include healthcare provider considerations in long-term recovery planning. The growing population of nurse practitioners may be one solution to address primary care needs. However, they cannot be the only solution, as medical specialists and surgeons remain critically needed.

Conclusion

The study found that effects of hurricanes on healthcare availability were largely contingent on the population-level characteristics of affected counties, where race/ethnicity and economic characteristics are key contributors. Large-scale disasters continue to have devastating effects across the United States. Populations that are most in need of healthcare, such as aging populations, those with disabilities or chronic conditions, and children, are the most impacted by the community-level disruption from these events, making the need for regular healthcare even more critical in communities that have protracted recovery periods. Long term trends in county-level healthcare provider supply after Hurricanes Sandy and Katrina suggest the need for policy level efforts to ensure stability in provision of care to disaster-affected communities.

Supporting information

[Figure omitted. See PDF.]

S1 Table. Case and control counties.

https://doi.org/10.1371/journal.pone.0242823.s001

(DOCX)

S2 Table. Changes in provider ratios by county among counties impacted by Hurricane Katrina and in-state Louisiana controls.

https://doi.org/10.1371/journal.pone.0242823.s002

(DOCX)

Citation: Bell SA, Klasa K, Iwashyna TJ, Norton EC, Davis MA (2020) Long-term healthcare provider availability following large-scale hurricanes: A difference-in-differences study. PLoS ONE 15(11): e0242823. https://doi.org/10.1371/journal.pone.0242823

References

1. Quast T, Andel R, Sadhu AR. Long-term Effects of Disasters on Seniors With Diabetes: Evidence From Hurricanes Katrina and Rita. Diabetes care. 2019;42(11):2090–7. Epub 2019/09/25. pmid:31548250

2. Raulji C, Velez MC, Prasad P, Rousseau C, Gardner RV. Impact of Hurricane Katrina on healthcare delivery for New Orleans patients, 2005–2014. Pediatr Blood Cancer. 2018;65(12):e27406. Epub 2018/09/27. pmid:30251459

3. Adams V, Kaufman SR, van Hattum T, Moody S. Aging disaster: mortality, vulnerability, and long-term recovery among Katrina survivors. Medical anthropology. 2011;30(3):247–70. Epub 2011/05/19. pmid:21590581

4. Malik S, Lee DC, Doran KM, Grudzen CR, Worthing J, Portelli I, et al. Vulnerability of Older Adults in Disasters: Emergency Department Utilization by Geriatric Patients After Hurricane Sandy. Disaster Medicine and Public Health Preparedness. 2018;12(2):184–93. pmid:28766475

5. Kelman J, Finne K, Bogdanov A, Worrall C, Margolis G, Rising K, et al. Dialysis care and death following Hurricane Sandy. American journal of kidney diseases: The official journal of the National Kidney Foundation. 2015;65(1):109–15. Epub 2014/08/27. pmid:25156306

6. Gotanda H, Fogel J, Husk G, Levine JM, Peterson M, Baumlin K, et al. Hurricane sandy: Impact on emergency department and hospital utilization by older adults in lower manhattan. Journal of the American Geriatrics Society. 2015;63:S248. pmid:26369433

7. Teperman S. Hurricane Sandy and the greater New York health care system. J Trauma Acute Care Surg. 2013;74(6):1401–10. Epub 2013/05/23. pmid:23694865

8. National Centers for Environmental Information. Billion-dollar Weather and Climate Disasters: National Oceanic and Atmospheric Administration; 2019 [April 1, 2019]. https://www.ncdc.noaa.gov/billions/.

9. Angus DC, Barnato AE, Bell D, Bellomo R, Chong CR, Coats TJ, et al. A systematic review and meta-analysis of early goal-directed therapy for septic shock: the ARISE, ProCESS and ProMISe Investigators. Intensive Care Med. 2015;41(9):1549–60. pmid:25952825

10. Federal Emergency Management Agency. Hurricane Sandy FEMA After-Action Report. Washington, D.C.: Federal Emergency Management Agency, 2013.

11. NOAA. After Sandy: Rebuilding & Recovery. Washington, DC: NOAA, 2020.

12. Bell SA, Banerjee M, Griggs JJ, Iwashyna TJ, Davis MA. The Effect of Exposure to Disaster on Cancer Survival. J Gen Intern Med. 35. United States2020. p. 380–2. pmid:31659658

13. Basu S, Berkowitz SA, Phillips RL, Bitton A, Landon BE, Phillips RS. Association of Primary Care Physician Supply With Population Mortality in the United States, 2005–2015. JAMA Intern Med. 2019;179(4):506–14. Epub 2019/02/19. pmid:30776056

14. Chang CH, Stukel TA, Flood AB, Goodman DC. Primary care physician workforce and Medicare beneficiaries’ health outcomes. Jama. 2011;305(20):2096–104. Epub 2011/05/26. pmid:21610242

15. Davis MA, Anthopolos R, Tootoo J, Titler M, Bynum JPW, Shipman SA. Supply of Healthcare Providers in Relation to County Socioeconomic and Health Status. J Gen Intern Med. 2018;33(4):412–4. pmid:29362958

16. Kashima S, Inoue K, Matsumoto M. Characteristics of Physician Outflow from Disaster Areas following the Great East Japan Earthquake. PLoS One. 2017;12(1):e0169220. Epub 2017/01/04. pmid:28046089

17. Rudowitz R, Rowland D, Shartzer A. Health care in New Orleans before and after Hurricane Katrina. Health Aff (Millwood). 2006;25(5):w393–406. Epub 2006/08/31. pmid:16940307

18. Dimick JB, Ryan AM. Methods for evaluating changes in health care policy: the difference-in-differences approach. JAMA. 2014;312(22):2401–2. Epub 2014/12/10. pmid:25490331

19. Centers for Medicare & Medicaid Services. Taxonomy Washington, DC: U.S. Department of Health and Human Services; 2019 [February 17, 2020]. https://www.cms.gov/Medicare/Provider-Enrollment-and-Certification/MedicareProviderSupEnroll/Taxonomy.

20. Bindman AB. Using the National Provider Identifier for health care workforce evaluation. Medicare Medicaid Res Rev. 2013;3(3). Epub 2013/01/01. pmid:24753977

21. Centers for Medicare & Medicaid Services. NPI: Get It. Share It. Use It. Washington, DC: 2006.

22. Andrilla CHA, Patterson DG, Garberson LA, Coulthard C, Larson EH. Geographic Variation in the Supply of Selected Behavioral Health Providers. Am J Prev Med. 2018;54(6 Suppl 3):S199–s207. Epub 2018/05/22. pmid:29779543

23. DesRoches CM, Barrett KA, Harvey BE, Kogan R, Reschovsky JD, Landon BE, et al. The Results Are Only as Good as the Sample: Assessing Three National Physician Sampling Frames. J Gen Intern Med. 2015;30 Suppl 3:S595–601. Epub 2015/06/25. pmid:26105676

24. Federal Emergency Management Agency. Disaster Declarations 2019 [January 31, 2019]. https://www.fema.gov/disasters.

25. Angrist JD, Krueger AB. Empirical strategies in labor economics. In: Card D, Ashenfelter O, editors. Handbook of Labor Economics. 3. Amsterdam, the Netherlands: Elsevier Science; 1999.

26. Burger J, Gochfeld M, Pittfield T, Jeitner C. Responses of a vulnerable Hispanic population in New Jersey to Hurricane Sandy: Access to care, medical needs, concerns and ecological ratings. Journal of toxicology and environmental health Part A. 2017;80(6):315–25. pmid:28644717

27. National Center for Health Workforce Analysis Bureau of Health Workforce Health Resources and Services Administration U.S. Department of Health and Human Services. State-Level Projections of Supply and Demand for Primary Care Practitioners: 2013–2025 Washington, D.C.: U.S. Department of Health and Human Services, 2016.

28. Resources Health and Administration Services. Projecting the Supply and Demand for Primary Care Practitioners Through 2020. Washington, DC: Health Resources and Services Administration, 2013.

29. Stuart EA, Huskamp HA, Duckworth K, Simmons J, Song Z, Chernew M, et al. Using propensity scores in difference-in-differences models to estimate the effects of a policy change. Health Serv Outcomes Res Methodol. 2014;14(4):166–82. Epub 2014/12/23. pmid:25530705

30. U.S. Government Accountability Office. Hurricane Katrina: CMS and HRSA Assistance to Sustain Primary Care Gains in the Greater New Orleans Area. Washington, D.C.: U.S. Government Accountability Office, 2010.