Introduction
Modern society is at constant risk of natural or man-made disasters, which necessitate the activation of a broad spectrum of community resources, and may overwhelm the local capacity to triage, admit, and treat patients.Reference Marcozzi, Pietrobon and Lawler1 Within this system, hospitals are the ultimate “epicenter”Reference Marcozzi, Pietrobon and Lawler1 of care delivery, where victims of disasters receive their treatment.Reference Khan2, Reference Jacobson, Inglesby and Khan3 In theory, better preparedness and higher levels of resilience will allow any system to perform better under stress and recover faster.Reference Khan2, Reference Flynn4, Reference Khan5 However, setting aside resources to counteract potential future disasters seems antithetical to the principles of lean management, resource limitations, and process optimization. Also, the more time passes between incidents, the incentive to invest resources into readiness efforts will likely diminish.Reference Siman-Tov, Davidson and Adini6
Just as metrics exist to measure readiness efforts on a Nationwide and State level,Reference Jacobson, Inglesby and Khan3 decisions at the level of the hospital or health care system should be supported by data.Reference Marcozzi, Pietrobon and Lawler1 Previously introduced methods to measure preparedness include the World Health Organization Hospital Safety Index, among others.Reference Husaini, Sugiarto, Rahmanand and Oktari7 Ideally, these assessments should be based on readily available data such that they can be repeated and assess progress over time. The Hospital Medical Surge Preparedness Index (HMSPI) is a proposed metric that characterizes hospitals with regard to their readiness capabilities.Reference Marcozzi, Pietrobon and Lawler1, Reference Marcozzi, Pietrobon and Lawler8 It is based on previous research into factors facilitating medical surge preparedness,Reference Kelen and McCarthy9 and is composed of four individual scores for system (e.g., planning, regional cooperation, supply chain distribution, continuity of operations efforts), space (e.g., medical and laboratory facilities, capacity of the hospital), staff (both in absolute numbers and by critical functions), and supplies (e.g., respirators, personal protective equipment, food and water).
It is currently unknown whether preparedness is equally distributed across the health care system, but it seems plausible that larger hospitals, and those with greater margins, would find it easier to allocate resources for emergency preparedness than smaller hospitals, which may be facing greater economic challenges at baseline. Preliminary evidence suggested that the HMSPI is higher in larger hospitals than in small hospitals, the latter defined as less than 82 beds.Reference Marcozzi, Pietrobon and Lawler8 Related investigations have shown that financial resilience quantified by the Hospital Resilience Index (HRI) is higher in top-tier, large, integrated hospitals. The HRI summarizes eleven factors influencing financial hospital buffers and hospital closures, including size, ancillary functions, staffing structure, patient volume, and operating expenses.Reference Lirk, Janjua, Rogers, Flynn, Kuo and Pietrobon10 However, there has been no further characterization of the roughly 6,000 hospitals in the United States regarding their performance on the HMSPI metric.
The current study was undertaken to investigate the relationship between the HMSPI and the type, size, location, and patient insurance for hospitals participating in the AHA database. The working hypothesis was that preparedness would be concentrated and most prominent in the upper tiers of the health care system, including large teaching hospitals.
Methods
Study Design
The primary objective was to evaluate the relationship between the Hospital Medical Surge Preparedness Index (HMSPI)Reference Marcozzi, Pietrobon and Lawler1 and characteristics of the respective hospitals, including size, type, location, performance on the Hospital Resilience Index (HRI),Reference Lirk, Janjua, Rogers, Flynn, Kuo and Pietrobon10 and Social Determinants of Health (SDoH) of the catchment area. In addition, select patient outcomes were examined.
The STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) guideline and the CHAMP (CHecklist for Statistical Assessment of Medical Papers) declaration were used to report Methods and Results.Reference von Elm, Altman and Egger11, Reference Mansournia, Collins and Nielsen12
Ethics
This study was designated as IRB exempt by the Institutional Review Board of Mass General Brigham Healthcare, Boston, MA, on March 1, 2024, under protocol number 2024-1514.
Setting
Data were obtained from the following databases:
Healthcare Cost and Utilization Project (HCUP), State Inpatient Database (SID) 13: HCUP is a group of datasets that gathers information on surgeries, emergencies, readmissions, and pediatric cases from various states or nationwide. We used SID from Florida and Maryland. The SID dataset collects longitudinal hospital care data and contains state-specific files of all inpatient records. The SID provides access to more than 100 clinical and non-clinical variables in hospital discharge abstracts, such as diagnoses and procedures, status at admission and discharge, patient demographic characteristics, expected payment source, total charges, and LOS. The SID dataset covers 97% of all hospital discharges in the United States.
American Hospital Association (AHA) survey 14, Reference Boggs, Sullivan, Espinola, Gao and Camargo15: Clinical data were then linked to the AHA survey, enhancing the hospital-level dataset derived from SID. The information generated by the AHA survey includes patient data (health service utilization, hospital service lines, and facilities), hospital characteristics (demographics, organizational structure, alternative payment models, telehealth and remote patient monitoring, social determinants of health services), and health care agency data. This database contains more than 1,000 data points on more than 6,300 hospitals, with an approximate response rate of 75%.Reference Marcozzi, Pietrobon and Lawler1
Hospital Resilience Index (HRI) Reference Lirk, Janjua, Rogers, Flynn, Kuo and Pietrobon10: The HRI was developed and validated to reflect a hospital’s financial viability and, potentially, be the basis for a risk assessment toward impending closure. The factors associated with the likelihood of closure include hospital size, ancillary functions, staffing structure, size of facilities, surgical volume, share of patients with Medicare/Medicaid coverage, operating expenses and margin, and medical school affiliation.
Distressed Communities Index (DCI) Reference Kesler16, Reference Su, Vinogradsky and Wang17: The DCI dataset depicts the geographic distribution of economic prosperity by using a 5-category classification, from prosperous to distressed communities. The classification includes seven socioeconomic indicators: no high school diploma, poverty rate, adults not working, housing vacancy rate, median household income, change in employment, and change in business establishments. Data from the DCI is calculated for four different geographic levels, such as zip codes, counties, cities, and congressional districts. The DCI covers 99% of the United States population.
RAND Corporation 18, Reference Zander, Kendall, Janjua, Kuo and Grimsley19: The RAND dataset is composed of annual expense reports from institutional providers and includes costs and statistics related to health care. This dataset contains data from health care providers regarding facility characteristics, health care service utilization, costs and charges by cost center, and financial statements.
Notably, patient data originate from HCUP (2017-2019), hospital data from RAND (1996-2021) and the AHA (2017-2023), and SDoH data from the DCI (2015-2019). The association between HCUP, RAND, and AHA is facilitated by the patient’s reported zip code in HCUP, ensuring the integration of diverse data sets for a cohesive analysis of the interactions between HRI, SDoH, and HMSPI.
Patients
Patients were included from 2017 to 2019 HCUP SID (FL and MD), who underwent major surgical procedures. Major procedures were defined as lower extremity bypass, knee arthroplasty, hip replacement, spinal fusion, colectomy, and hysterectomy. Patients were identified through the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) procedure codes (Supplementary Table 1). Exclusion criteria involved patients who were under 18 years of age, had more than 1 major surgical procedure during a single stay, and had missing zip code or mortality data.
Outcomes
The primary objective was to assess the relationship between the HMSPI and hospital characteristics such as University affiliation, patient volume, patient population served, HRI, and SDoH of patients in the hospital’s catchment area.
Predictors
To assess the association between the HMSPI and hospital size, predictive variables at a hospital level included (1) hospital volume (total facility admissions), categorized as above or below the median, (2) number of beds into small (<100 beds), medium (100-499 beds), and large (≥500 beds) categories, and (3) cost-to-charge ratios divided into those above or below the median. To examine the association between the HMSPI and hospital type and location, predictors such as (4) hospital type (Investor-owned or Not-for-profit), (5) rural hospital status, (6) Rural Referral Center designation, (7) teaching hospital (yes/no), (8) major teaching hospital (yes/no), and (9) minor teaching hospital (yes/no) were considered. For the investigation of patient and financial health, 2 types of models were employed. The first set used only hospital data, with (10) Medicare/Medicaid share and the (11) HRI as predictors.
Statistical Methods
Exploratory analysis: The exploratory analysis started with a visual exploration of all variables to evaluate the frequency, percentage, and near-zero variance for categorical variables, as well as distribution for numeric variables, and their corresponding missing value patterns.Reference Schwartz, Kolak and Pollak20 Near-zero variance occurs when a categorical variable has a low frequency of unique values over the sample size, i.e., the variable is almost constant, and this was addressed by combining different variable categorizations. Missing values were handled through imputation algorithms followed by sensitivity analyses to verify whether results were stable with and without imputation.Reference de Jager, Gunnarsson and Ho21 Statistical tests included P values for t-tests (for numeric variables) and Chi-square tests (for categorical variables). P values below 0.05 were considered statistically significant.
Modeling First, at the hospital level, crude estimates were derived to identify predictors of HMSPI. Subsequently, risk-adjusted rates for HMSPI were assessed using mixed models with a binomial distribution family. These models treated hospital size, hospital type and location, and hospital data for patient and financial health as dependent variables, accounting for multi-level models with the year grouped as layers (grouping or random effects). Mixed models were also adjusted for hospital type and rurality, hospital volume, number of beds, and cost-to-charge ratios, depending on the model.
Since these variables are likely not to have a normal distribution, variable transformations were explored, including logarithmic and Box-Cox. Alternatively, count models such as negative binomial and zero-inflated negative binomial were considered.
Once the crude and adjusted estimates were completed, a series of comparisons, still using mixed models to decrease the impact of multiple comparisons, was undertaken within a single outcome measure. A second series of models improved confounding adjustment and causal inferences through propensity scores, making use of inverse probability weighting. While it was not expected to have to calculate IPW within each of the layers (years or hospitals) of the model, use was made of extensive regression diagnostics to determine whether that strategy might have been warranted.
Results were reported as odds ratios and predicted means for binary (e.g., mortality) and continuous (e.g., LOS) outcomes, respectively, with 95% confidence intervals (CI) and P values. Odds ratio results were interpreted as significant when the confidence intervals did not cross a value of 1.0, and predicted means as significant when any 2 predicted means did not overlap between different estimates, along with P for trend <0.05.Reference Fabregas, Riley, Brant, George, Orav and Lam22
All analyses were performed using the R language.Reference Kuhn and Johnson23
Results
The first step was to characterize the hospital sample through the eyes of the patients treated. When grouping hospitals by catchment area, divided into rural–small town–suburban–urban, small but significant differences were found in HMSPI, with hospitals serving suburban patients having the highest, and hospitals serving small town patients having the lowest HMSPI. In comparison, the HRI was also lowest in the small town catchment area, and highest in urban and suburban areas. Suburban areas had the lowest prevalence of adult inhabitants without a high school diploma, the highest likelihood of inhabitants with at least a bachelor’s degree, and the lowest poverty rate (Table 1).
Hospital description by type of catchment area
The next step was to assess the relationship between hospital characteristics (patient volume, patient beds, cost-to-charge ratio) related to the HMPSI. Overall, higher patient volume, and larger hospitals as measured by bed count were associated with a higher HMSPI (Figure 1A-B and Supplementary Figure 1).
Association between the HMSPI score and hospital size metrics across 5 years (2017-2021). (A) Hospital volume (total facility admissions); (B) Number of hospital beds; (C) Cost-to-charge ratio based on HCUP data for Florida and Maryland.
Regarding hospital characteristics, investor-related hospitals had a lower HMSPI than not-for-profit hospitals (Figure 2A), teaching hospitals had a higher HMSPI than non-teaching hospitals (Figure 2B), and rural hospitals, including rural referral hospitals, had a lower HMSPI than suburban or urban centers (Figure 2C-D). There was no difference in these outcomes over time (Supplementary Figure 2).
Association between the HMSPI score and hospital type and localization hospital across 5 years among U.S. hospitals based on the AHA database, depicting (A) Hospital type (investor versus not-for-profit), (B) Teaching status, (C) Rural versus urban/suburban, and (D) Rural referral center.
Figure 3 displays unadjusted associations between the HMSPI score and Medicare and Medicaid inpatient days share, hospital resilience index (HRI), and disproportionate share hospitals across 5 years (2017-2021). A lower share of patients covered through Medicaid or Medicare was associated with a lower HMSPI (Figure 3A). In the lower range of the HRI, there was no correlation between HRI and HMSPI, but a distinctly higher HMSPI was observed in hospitals with the highest HRI scores, a surrogate marker for overall system well-being and financial resilience (Figure 3B). Disproportionate share hospitals as a group did not show a lower HMSPI than non-disproportionate share hospitals (Figure 3C). The trend over time did not show any difference in outcomes (Supplementary Figure 3).
Association between the HMSPI score and (A) share of Medicare inpatient days, (B) the hospital resilience index (HRI), and (C) disproportionate share hospital status across 5 years among U.S. hospitals based on the AHA database.
When comparing the HMSPI to the DCI for the catchment area, no apparent relationship was found Figure 3C. Table 2 presents the association between the HMSPI score and hospital type, location, and DCI, adjusted for the total number of admissions, hospital size (based on the number of beds), cost-to-charge ratio under or below 0.26, and disproportionate share hospitals. Major teaching hospitals are those with interns and residents-to-bed >0.25. Minor teaching hospitals are those with interns and residents-to-bed between 0 and 0.25 (Table 3).
Association between the HMSPI score and hospital volume, number of beds, and costs adjusted for the hospital type (teaching or non-teaching, investor owned or not-for-profit), rurality, distressed communities index (DCI), and disproportionate share hospitals (defined as hospitals with >15% Medicaid inpatient day share)
Predicted means of the HMSPI score in relation to type and localization of the hospital, and the distressed communities index
Limitations
Several limitations should be considered. The first is that the current assessment is based on data obtained from publicly available databases, such as the AHA database. Any database depends on the quality of the data entered, and in the case of the AHA, the Annual Survey has been in use since 1946 and has an average response rate of 75% among the more than 6,000 hospitals.Reference Boggs, Sullivan, Espinola, Gao and Camargo15 This necessarily means that some data is missing, but overall the AHA database is renowned to be the most comprehensive hospital database in the United States.Reference Boggs, Sullivan, Espinola, Gao and Camargo15 This measure of inaccuracy and missing data needs to be contrasted with the automated and time-tested way the data is collected, the sheer amount of data that can be used for analysis, and the fact that using the AHA database, assessments of readiness can be carried out in an automated fashion, can be analyzed across different timepoints, and can immediately be updated once new survey data become available. An alternative approach would be to perform individual surveys tailored especially to preparedness efforts,Reference Ugelvik, Thomassen, Braut, Geisner, Sjovold and Montan24 but these are very labor-intensive to conduct and collect, and missing responses are still a problem.Reference Ugelvik, Thomassen, Braut, Geisner, Sjovold and Montan24 The second limitation is that physical measures of preparedness are only part of the overall picture, and the resilience, training and ingenuity of the workforce at a specific health care institution are immeasurable contributors and modifiers of any emergency response. Lastly, and in a related matter, the metrics assessed by the HMSPI are general indicators supplied by individual hospitals to the AHA database, but they do not inform us about concrete site-specific policies and activities not captured by the database. However, the factors chosen in the original iterations of the HMSPI reflect very closely those identified as defining the capacity of hospitals to surge in times of need.
Discussion
This study is the first to investigate the relationship between the HMSPI and hospital characteristics outside the AHA data set. This study aimed to describe whether HMSPI values as a potential surrogate value for surge readiness were evenly distributed across the Medical System in the United States or were skewed toward higher-capacity and teaching institutions. Other markers of financial stability, such as the HRI, are better in larger hospitals and in teaching institutions.Reference Lirk, Janjua, Rogers, Flynn, Kuo and Pietrobon10 Along the same lines, HMSPI was higher in larger hospitals, in teaching hospitals, in hospitals serving urban and suburban areas, in not-for-profit as opposed to investor-owned hospitals, and in hospitals that had fewer patients with coverage through Medicare/Medicaid. The distressed communities index did not show a correlation with the HMSPI.
Mass casualty incidents are rising, including both man-made and natural disasters.Reference Kelley, Toscano and Gestring25 Assessing the capacity of hospitals to respond to disasters with a so-called “surge” is a crucial piece in emergency preparedness efforts. The HMSPI was based on factors identified in the landmark publication “The Science of the Surge,” and includes factors clustered into four constructs: space, staff, supplies, and system.Reference Marcozzi, Pietrobon and Lawler1, Reference Kelen and McCarthy9 The HMSPI is complementary to other metrics, such as the National Health Security Preparedness Index, which assesses preparedness efforts at the state or Territory level.Reference Jacobson, Inglesby and Khan3 One particularly appealing aspect of the HMSPI is that it is derived from the AHA database, which is populated through annual mandatory surveys which have been carried out since the 1940s, such that all the data needed for analysis is readily available, and can be analyzed over time. The HMSPI was successfully applied to quantify surge capacity during the COVID-19 pandemic.Reference Marcozzi, Pietrobon and Lawler8 The HMSPI is particularly attractive since it is grounded in a reliable dataset updated annually, and so its contemporary data can be used to continually measure and reflect on capacity, and inform policy.Reference Marcozzi, Pietrobon and Lawler1 This is especially important since preparedness efforts will necessarily reflect, at least in part, overarching political and fiscal circumstances.Reference Khan5 The HMSPI may potentially be used to assess disaster readiness on both a regional and hospital-specific level, to identify areas in need of support, and to evaluate the impact of policy interventions. As one example, funding may be directed, in part, based on need documented by the HMSPI, and this may increase preparedness potential and be reflected in an improved HMSPI over the subsequent years.
The HMSPI, as a measure of readiness, was highest in larger and teaching hospitals. This is both intended and problematic, depending on patient volume over time. For smaller incidents such as a confined terrorist attack or a multi-vehicle mass casualty event, large Level-1 trauma centers are a natural backbone of any medical response, and in that sense, it is logical that preparedness efforts are centered and emphasized on these Institutions. In the case of the Boston Marathon bombing, a large number of approximately 250 casualties originating from one single incident was relatively efficiently absorbed by the 5 trauma centers in the city, with the support of regional hospitals.26 However, when the disaster is more sustained or more overwhelming, the preparedness of the overall system, including downstream entities, becomes more and more important. This became apparent both during the COVID-19 pandemic and Superstorm Sandy. During COVID-19, patient volumes overwhelmed large hospitals, in part because smaller units of the health care system, such as community health care centers, faced enormous challenges to keep up their services.Reference Wright, Fraher, Holder, Akiyama and Toomey27 When smaller units of care are less resilient or cannot surge to meet increased demand, patients will redistribute and further overwhelm larger hospitals. During Superstorm Sandy, a large number of community care centers, dialysis centers, retail pharmacies, and nursing homes were incapacitated by the storm, without backup power, and, in many cases, no way to access patient information.Reference Flynn28 One particularly striking facet was the loss of function in several ambulatory dialysis centers in Manhattan, necessitating Beth Israel Medical Center to convert part of its Emergency Department into a dialysis center running around the clock to try to serve all the Lower Manhattan dialysis patients who had experienced loss of service.Reference Flynn28, Reference Jangi29 In conclusion, preparedness, which is concentrated at the higher tiers of the health care system, will be able to absorb mid-size discrete mass casualty events or disasters, but for a long-term disaster, preparedness across the spectrum of health care institutions, including smaller community-based offices, treatment centers, and care facilities, is paramount.
One interesting outcome of the present study was that HMSPI was higher in not-for-profit hospitals, as compared to investor-owned Institutions. This may reflect a basic struggle in surge preparedness, the balance between profitability and resilience. One balance sheet problem with preparedness efforts is that they do not produce profit; rather, they are intended to protect against a potential future threat and to mitigate its impact.Reference Herrington30 Investing in preparedness efforts means more than setting aside financial resources for supplies; it involves training, updating, and resourcing teams to facilitate readiness, and this may be hard to justify in times of economic uncertainty, especially when there are long intervals between incidents.Reference Siman-Tov, Davidson and Adini6 One area in which this is particularly palpable and concerning is the “crowding” or “overcrowding” of Emergency Rooms.Reference Puri, Tsay and Goldberg31, Reference Weiss, Rogers, Maas, Ernst and Nick32 This refers to the fact that many Emergency Rooms are operating at excess capacity in response to daily fluctuations in patient census, meaning that any additional event or disaster will be harder to absorb, as all the reserve capacity is already used up. This phenomenon is not exclusive to the United States and is found in many medical systems worldwide.Reference Anneveld, van der Linden, Grootendorst and Galli-Leslie33
This ties into the next finding, where HMSPI was inversely correlated to the share of patients covered by Medicare/Medicaid. Hospitals with higher shares of Medicare/Medicaid patients are known to often have a more precarious financial situation and narrow margins.Reference Chernew, He, Mintz and Beaulieu34 Hospitals that are already struggling financially likely find it difficult to allocate substantial resources to preparedness efforts while managing their daily operations; meanwhile, they are at risk of being the hardest hit during disasters. Interestingly, the analysis did not reveal a clear correlation between HMSPI and the DCI of the catchment area.
Conclusions
In conclusion, hospital medical surge preparedness is unevenly distributed across the U.S. health care system, with higher-tier and teaching institutions exhibiting higher preparedness scores on the HMSPI, while investor-owned status and a higher share of patients covered by Medicare/Medicaid are associated with lower scores.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/dmp.2025.10270.
Acknowledgments
This work was supported by seed grants from the Department of Anesthesiology, Brigham and Women’s Hospital, and from the Global Resilience Institute, Northeastern University, both in Boston, MA.
Author contribution
Lirk—Grant acquisition, protocol drafting, project lead, IRB and editorial correspondence, data interpretation, manuscript lead author.
Marcozzi—Protocol drafting, data interpretation, manuscript writing.
Flynn—Data interpretation, manuscript writing.
Kuo—Protocol drafting, data interpretation, manuscript writing.
Pietrobon—Protocol drafting, data analysis and interpretation, manuscript writing.
Competing interests
None.
