A comparative analysis estimating aspergillosis and histoplasmosis encounters among inpatients in the United States using multi-regional electronic health record and National Discharge Data, 2014–2020

Abstract

Fungal diseases are an emerging public health threat, yet their epidemiology remains poorly understood due to limited surveillance. Electronic health record (EHR) datasets offer opportunities for studying fungal diseases, but their national representativeness is unclear. We compared inpatient encounters for aspergillosis and histoplasmosis between Oracle EHR Real World Data (OERWD) and the Healthcare Cost and Utilization Project National Inpatient Sample (HCUP NIS), a nationally representative discharge-derived database. We analyzed inpatient encounters from both databases and applied calibration weights to OERWD to improve alignment with national benchmarks. Relative differences (RDs) in encounter counts were compared using Wald tests. Survey-weighted quasibinomial models estimated prevalence and 95% CIs; temporal trends were evaluated with Mann–Kendall tests. Dataset differences were assessed using interaction terms between a dataset indicator and demographic or geographic subgroups. OERWD had fewer total and disease-specific encounters than HCUP NIS. Weighting improved precision, and prevalence estimates did not differ significantly for most strata (31 of 42), indicating broad agreement in demographic and geographic distributions. Both datasets showed higher aspergillosis prevalence in the Pacific division and higher histoplasmosis prevalence in the East South Central division, with males and older adults consistently exhibiting higher prevalence. These findings indicate that, while HCUP NIS is optimal for nationally representative prevalence estimates, weighting EHR-derived data can reproduce key demographic and geographic prevalence patterns observed in national benchmarks. Benchmarking EHR-derived estimates against national standards represents an important step toward establishing their representativeness and supporting future studies leveraging EHR data to investigate fungal disease epidemiology in greater clinical detail.

Introduction

Fungal diseases cause severe illness and mortality worldwide, contributing to >3 million deaths annually and imposing a substantial burden on healthcare systems.¹ However, accurately measuring their impact remains challenging due to underdiagnosis, limited surveillance, and inconsistent reporting.² In the United States, the threat of fungal infections continues to rise, driven by a growing immunocompromised population, antifungal resistance, medical device use, and environmental variability influencing pathogen distribution.^3,4 National incidence and prevalence estimates are lacking for most fungal diseases, even as they account for substantial hospitalizations, disability,^5,6 and over $6 billion in yearly costs.^7,8 These challenges largely stem from incomplete diagnosis and reporting across available data sources.⁹ To improve national estimates of fungal incidence and prevalence, new approaches that integrate both clinical and epidemiologic information are needed.

Electronic health record (EHR) data provide a cost-effective means of accessing longitudinal health information on large populations,¹⁰ including for fungal diseases.^11–13 EHR-based studies have revealed that coccidioidomycosis is under-recognized and under-tested, even in highly endemic regions,¹¹ and that significant racial and ethnic differences exist in blastomycosis presentation and outcomes.¹⁴ Given the low incidence and the demographic and geographic variability of fungal diseases, comprehensive EHR databases with broad population coverage across diverse regions, hospital types, and care settings could help characterize fungal disease epidemiology. Compared with claims or discharge data, EHRs offer richer detail on laboratory results, microbiology, medication administration, and comorbidities,¹⁰ making them particularly valuable for studying diseases requiring extensive diagnostic testing or long-term management. However, several EHR databases commonly used in research have limited geographic range and incomplete demographic coverage,^11,15,16 raising questions about the generalizability of estimates derived from them.¹⁷

Electronic health record datasets are often limited to specific healthcare systems or vendors, which may not fully represent the target population and can introduce biases related to healthcare access and provider practices.^10,18 Limited geographic coverage and patient diversity can yield biased disease prevalence and incidence estimates that are not generalizable to the broader population.¹⁹ For instance, studies relying on EHR data from a single health system may overrepresent urban populations, insured patients, or those receiving specialty care.²⁰ To investigate the representativeness of EHR-derived data, the Healthcare Cost and Utilization Project (HCUP) National Inpatient Sample (NIS),²¹ a stratified, nationally representative all-payer database,²² has been used as a benchmarking tool (ie, an accepted standard to use for comparison to evaluate a metric). Previous studies have used HCUP NIS to assess the representativeness of EHR datasets by comparing disease prevalence rankings,²³ infection incidence estimates,²⁴ and surgical complication rates.²⁵ While HCUP NIS provides robust population-level benchmarks, it lacks the clinical detail needed to investigate diagnostic pathways, treatment patterns, and outcomes—areas where EHR data offer a critical advantage.

Benchmarking EHR-derived fungal disease estimates against HCUP NIS enables comparison of demographic distributions and prevalence patterns with a nationally representative sample. This approach provides insights into the inpatient care landscape, where severe fungal infections are more likely to be captured due to their high morbidity and hospitalization rates. While EHR data provide laboratory, treatment, and clinical detail that can capture disease presentation and outcomes beyond diagnosis codes, their national representativeness for fungal diseases has not been systematically assessed. Evaluating how closely EHR-derived estimates align with HCUP NIS benchmarks is an important first step in determining whether these data can be generalized to the broader inpatient population—a prerequisite for leveraging their full clinical depth to refine burden estimates and identify potentially unrecognized or underdiagnosed fungal infections. This approach enables the use of EHR data to characterize patient populations, diagnostic practices, and clinical management of fungal diseases at a national scale. Here, we analyze a large, nationwide EHR-derived database—Oracle EHR Real World Data (OERWD)—to estimate the number of inpatient encounters and prevalence of 2 fungal diseases, aspergillosis and histoplasmosis, from 2014 to 2020, and benchmark these estimates using HCUP NIS data. We examine the robustness and generalizability of the estimates derived from OERWD and evaluate whether it is representative of the national inpatient population for fungal disease encounters between 2014 and 2020.

Methods

Databases

Oracle Health Corporation maintains OERWD, 1 of the largest vendor-specific EHR databases in the United States, containing over 300 data elements from over 75 million patients. OERWD is extracted from the electronic medical records of organizations with which Oracle has a data use agreement. Encounters may include pharmacy, clinical and microbiology laboratory, admission, and billing information from affiliated patient care locations. Oracle has established Health Insurance Portability and Accountability Act-compliant operating policies to establish de-identification for OERWD. Some organizations using the Epic EHR are included in the OERWD. The Children’s Mercy Office of Research Integrity has deemed work with OERWD to be non-human subject research.

Healthcare Cost and Utilization Project, sponsored by the Agency for Healthcare Research and Quality, is the largest set of longitudinal, all-payer, encounter-level inpatient data in the United States.²¹ HCUP NIS is an annual, nationally representative sample of inpatient data representing more than 97% of all inpatient discharges from community hospitals in the United States. To construct the NIS, HCUP collects a stratified sample of approximately 20% of discharges from community hospitals in all participating states and provides sampling weights to produce national and regional estimates.²² The dataset includes patient demographics (eg, gender, age, race and ethnicity, and median household income for the patient ZIP code), visit details (eg, diagnosis codes, length of stay, primary payer, discharge status, severity and comorbidity measures, procedure group codes, and total charges), and hospital characteristics (eg, census division location of hospital, ownership).

Study design

To be included in the study, inpatient encounters for patients aged 18 years or older had to have occurred in a participating organization between 2014 and 2020. Diagnoses of aspergillosis and histoplasmosis were identified using International Classification of Diseases (ICD) codes, 9th and 10th revisions, following prior work (Table S1).⁵ Inpatient encounters with a diagnostic code listed in any position of the record were counted as cases, and patient demographics (gender, race and ethnicity, age, and the census division of the patient’s residence) were extracted. Racial and ethnic groups from OERWD were categorized to match the 6 HCUP categories (Hispanic or Latino, non-Hispanic American Indian or Alaska Native, non-Hispanic Asian or Pacific Islander, non-Hispanic Black or African American, non-Hispanic Other Race, and non-Hispanic White); age was categorized into 18–34 years, 35–64 years, and 65 years and over; and sex was categorized as self-identified male or female.

Statistical analysis

Healthcare Cost and Utilization Project National Inpatient Sample is a 20% stratified sample of all discharges from US community hospitals, excluding rehabilitation and long-term acute care hospitals; therefore, total counts and SEs of HCUP NIS all-cause and fungal disease inpatient encounters by patient characteristics were estimated using survey weights, following HCUP NIS methods.²² Accordingly, all HCUP NIS estimates reported herein reflect values calculated after applying survey weights. To make OERWD more nationally representative, we computed calibration weights as the ratio of HCUP NIS weighted encounter totals to OERWD encounter totals within each year (t)and census division (d) strata: $w_{td}=E_{td}^{\text{HCUP}}/E_{td}^{\text{OERWD}}$ (weight range: 4.0–44.1; Table S2).²⁶

The relative difference (RD) in the number of encounters for each fungal disease was compared between OERWD (unweighted and weighted) and HCUP NIS, using HCUP NIS values as the denominator,²⁶ defined as

$$\text{RD}=\frac{E_{\text{HCUP}}-E_{\text{OERWD}}}{E_{\text{HCUP}}}$$

where E_HCUP and E_OERWD are the total encounters for the disease in HCUP NIS and OERWD, respectively. To estimate prevalence and assess differences between datasets, we combined OERWD and HCUP NIS into a single dataset with a dataset indicator (HCUP NIS or OERWD) and fit separate survey-weighted quasibinomial regression models for each demographic or geographic variable using the survey package in R (version 4.4.2).²⁷

For each model, the outcome Y_i represented the number of encounters with the fungal disease in category i with denominator n_i, as the total number of encounters in that category:

$$Y_i~\text{QuasiBinomial}(n_i,p_i)$$

$$\text{logit}(p_i)={\beta }_0+{\beta }_1{(\text{Dataset})}_i+{\beta }_2{X}_i+{\beta }_3[{\text{Dataset}}_i\times X_i]$$

where i is the index strata defined by 1 level of X (eg, a census division, sex, race and ethnicity, or age group), p_i is the probability that an encounter in strata i is for the fungal disease of interest, and the interaction term tests whether prevalence differs between datasets across covariate levels. Model-based prevalence estimates, and the 95% CIs, were derived from predicted marginal probabilities. Temporal trends were evaluated using the Mann–Kendall test in the Kendall package.²⁸ P-values reported throughout the results section reflect statistical tests comparing the difference in encounters or prevalence estimates between the 2 datasets.

Results

HCUP NIS and unweighted OERWD patient characteristics

During the study period (2014–2020), 201 027 795 (SE: 7 740 262) inpatient encounters were recorded in HCUP NIS, and 14 853 499 unweighted inpatient encounters were reported in OERWD. The South Atlantic census division accounted for the highest percentage of national inpatient encounters in HCUP NIS (21.3%), whereas the Pacific census division accounted for the highest percentage in OERWD (19.1%). In both datasets, most encounters were among female patients (57.7% in HCUP NIS and 57.5% in OERWD), and non-Hispanic White was the most common race and ethnicity (67.3% in HCUP NIS and 66.1% in OERWD; Table 1). Compared to OERWD, HCUP NIS included a higher percentage of encounters for non-Hispanic Black or African American patients (15.2% in HCUP NIS and 10.8% in OERWD), while OERWD had more encounters for Hispanic or Latino patients (11.1% in HCUP NIS and 16.7% in OERWD; Table 1). The age distribution of patients between the 2 datasets was similar (Table 1).

Table 1.

Inpatient encounters in HCUP NIS and unweighted OERWD by patient characteristics—United States, 2014–2020.

Characteristic	HCUP NIS			OERWD (unweighted)
	Encounters (n)	SE	(%)	Encounters (n)	(%)
Overall	201 027 795	7 740 262	100.0	14 853 499	100.0
Census division
New England	9 730 875	1 355 818	4.8	874 987	5.9
Middle Atlantic	28 783 710	3 278 153	14.3	1 601 018	10.8
East North Central	31 234 270	4 060 724	15.5	1 241 745	8.4
West North Central	11 436 910	1 622 851	5.7	2 100 460	14.1
South Atlantic	42 848 100	4 915 174	21.3	2 169 658	14.6
East South Central	14 373 165	1 947 085	7.2	607 694	4.1
West South Central	23 835 825	2 489 492	11.9	1 021 456	6.9
Mountain	11 589 415	1 388 162	5.8	2 396 297	16.1
Pacific	27 195 525	2 742 856	13.5	2 840 184	19.1
Sex
Male	85 076 235	5 365 945	42.3	6 318 114	42.5
Female	115 951 560	6 310 401	57.7	8 535 385	57.5
Race and ethnicity
NH White	135 308 745	7 867 382	67.3	9 819 641	66.1
NH Black or African American	30 639 635	2 093 609	15.2	1 597 844	10.8
Hispanic or Latino	22 334 325	1 526 582	11.1	2 480 270	16.7
NH Asian or Pacific Islander	5 489 200	471 699	2.7	342 406	2.3
NH AI/AN	1 293 005	85 011	0.6	143 862	1.0
NH other race	5 962 885	398 645	3.0	469 476	3.2
Age group
18–34	37 866 525	2 491 575	18.8	2 854 474	19.2
35–64	76 619 590	4 834 616	38.1	5 747 551	38.7
65 and over	86 541 680	6 452 186	43.1	6 251 474	42.1

Abbreviations: AI/AN, American Indian or Alaska Native; HCUP NIS, Healthcare Cost and Utilization Project National Inpatient Sample; NH, non-Hispanic; OERWD, Oracle EHR Real World Data.

Aspergillosis encounters

A total of 86 332 (SE = 4002) weighted inpatient aspergillosis encounters (7085 unweighted, SE = 359) were recorded in OERWD, compared to 107 015 (SE = 4518) encounters in HCUP NIS (RD: 19.3% [95% CI, 9.4–29.2], P < .001; Table 2). Before weighting, aspergillosis encounter counts were substantially lower across all geographic and demographic groups in OERWD relative to HCUP NIS (Table 2). Weighting closed the gap and improved comparability across most strata; however, significant differences persisted among certain racial and ethnic, sex, and age groups. RDs in encounter counts for 3 of the 9 census divisions remained significantly different between the datasets after weighting OERWD (East North Central RD: 47.7% [95% CI, 25.1–70.2], P < .001; West North Central RD: 38.9% [95% CI, 13.1–64.7], P = .003; and West South Central RD: 31.4% [95% CI, 7.8–55.1], P = .009; Table 2).

Table 2.

Prevalence and relative difference between the unweighted and weighted OERWD inpatient dataset and HCUP NIS for aspergillosis and histoplasmosis encounters—United States, 2014–2020.

	HCUP NIS		OERWD				Relative difference (%)
	n (SE)	Prevalence/10 000 (95% CI)	n (SE)	Prevalence/10 000 (95% CI)	n, Weighted (SE)	Prevalence/10 000, weighted (95% CI)	Encounters
							Unweighted (95% CI)	Weighted (95% CI)
Aspergillosis
Total	107 015 (4518)	5.3 (5.1, 5.5)	7085 (359)	4.8a (4.5, 5.1)	86 332 (4002)	4.2a (4.0, 4.4)	93.4a (92.5, 94.2)	19.3a (9.4, 29.2)
Census division
New England	5345 (801)	5.5 (5.1, 6.0)	409 (60)	4.7 (3.8, 5.7)	4819 (741)	4.9 (3.9, 6.2)	92.3a (89.2, 95.5)	9.8 (−28.1, 47.8)
Middle Atlantic	13 610 (1637)	4.7 (4.4, 5.1)	580 (82)	3.6 (3.1, 4.2)	10 033 (1399)	3.5 (3.1, 4.1)	95.7a (94.2, 97.3)	26.3 (−0.3, 52.9)
East North Central	16 025 (2146)	5.1 (4.8, 5.5)	346 (66)	2.8a (2.3, 3.3)	8387 (1463)	2.6a (2.2, 3.1)	97.8a (96.9, 98.8)	47.7a (25.1, 70.2)
West North Central	5165 (784)	4.5 (4.0, 5.1)	484 (78)	2.3a (2.0, 2.7)	3155 (483)	2.2a (1.9, 2.6)	90.6a (86.6, 94.7)	38.9a (13.1, 64.7)
South Atlantic	20 490 (2599)	4.8 (4.5, 5.1)	996 (132)	4.6 (4.1, 5.1)	18 835 (2,353)	4.3 (3.9, 4.9)	95.1a (93.4, 96.9)	8.1 (−23.8, 39.9)
East South Central	6280 (965)	4.4 (3.9, 4.9)	254 (43)	4.2 (3.5, 5.0)	6177 (1,059)	4.3 (3.6, 5.1)	96.0a (94.1, 97.8)	1.6 (−42.7, 46.0)
West South Central	9475 (1062)	4.0 (3.7, 4.3)	275 (37)	2.7 (2.3, 3.2)	6497 (883)	2.7 (2.2, 3.2)	97.1a (96.1, 98.1)	31.4a (7.8, 55.1)
Mountain	6235 (839)	5.4 (4.9, 5.9)	1612 (211)	6.7a (6.1, 7.4)	8377 (1,078)	6.7a (6.1, 7.4)	74.1a (64.6, 83.7)	−34.3 (−83.4, 14.7)
Pacific	24 390 (2620)	9.0 (8.4, 9.6)	2129 (247)	7.5 (6.7, 8.4)	20 053 (2310)	7.4 (6.6, 8.2)	91.3a (88.6, 94.0)	17.8 (−7.6, 43.2)
Sex
Male	59 000 (3679)	6.9 (6.7, 7.2)	3757 (267)	5.9a (5.5, 6.4)	46 555 (3,087)	5.3a (4.9, 5.7)	93.6a (92.5, 94.8)	21.1a (7.0, 35.2)
Female	48 015 (3061)	4.1 (3.9, 4.4)	3328 (262)	3.9 (3.5, 4.3)	39 778 (2,850)	3.4 (3.1, 3.7)	93.1a (91.7, 94.4)	17.2a (1.6, 32.7)
Race and ethnicity
NH White	73 085 (4504)	5.4 (5.2, 5.6)	4489 (326)	4.6a (4.2, 4.9)	56 683 (3,719)	4.1a (3.8, 4.4)	93.9a (92.7, 95.0)	22.4a (8.8, 36.1)
NH Black/AA	14 695 (1157)	4.8 (4.4, 5.2)	689 (64)	4.3 (3.8, 4.9)	10 127 (1,015)	4.0 (3.4, 4.6)	95.3a (94.2, 96.4)	31.1a (13.9, 48.3)
Hispanic or Latino	11 110 (974)	5.0 (4.5, 5.5)	1368 (167)	5.5a (4.7, 6.5)	14 411 (1,626)	4.7a (4.0, 5.5)	87.7a (84.1, 91.3)	−29.7 (−66.0, 6.6)
NH Asian/PI	4590 (536)	8.4 (7.3, 9.6)	234 (39)	6.8 (5.5, 8.5)	2366 (376)	6.0 (4.8, 7.6)	94.9a (92.9, 96.9)	48.4a (28.5, 68.4)
NH AI/AN	455 (61)	3.5 (2.8, 4.5)	64 (17)	4.4 (3.0, 6.6)	382 (93)	3.8 (2.6, 5.7)	85.9a (77.7, 94.2)	16.0 (−29.6, 61.6)
NH other race	3080 (276)	5.2 (4.6, 5.8)	241 (32)	5.1 (4.2, 6.3)	2364 (307)	4.5 (3.6, 5.6)	92.2a (89.7, 94.6)	23.3 (−0.5, 47.0)
Age group
18–34	12 155 (853)	3.2 (2.9, 3.5)	924 (79)	3.2 (2.8, 3.7)	11 812 (1,042)	3.0 (2.6, 3.5)	92.4a (90.7, 94.0)	2.8 (−18.6, 24.3)
35–64	46 820 (3058)	6.1 (5.8, 6.4)	2899 (230)	5.0 (4.6, 5.6)	34 823 (2,558)	4.4a (4.0, 4.8)	93.8a (92.6, 95.1)	25.6a (11.3, 40.0)
65 and over	48 040 (3651)	5.6 (5.3, 5.8)	3262 (289)	5.2 (4.8, 5.7)	39 687 (3,222)	4.5 (4.2, 4.9)	93.2a (91.7, 94.8)	17.4 (−0.6, 35.4)
Histoplasmosis
Total	32 480 (1851)	1.6 (1.5, 1.8)	1722 (114)	1.2a (1.0, 1.3)	26 505 (1,833)	1.3a (1.2, 1.4)	94.7a (93.8, 95.6)	18.4a (4.1, 32.7)
Census division
New England	305 (53)	0.3 (0.2, 0.4)	24 (7)	0.3 (0.2, 0.4)	289 (84)	0.3 (0.2, 0.5)	92.1a (87.1, 97.3)	5.4 (−57.5, 68.3)
Middle Atlantic	1325 (176)	0.5 (0.4, 0.5)	43 (10)	0.3 (0.2, 0.4)	764 (169)	0.3 (0.2, 0.4)	96.8a (95.1, 98.4)	42.3a (13.2, 71.5)
East North Central	9250 (1380)	3.0 (2.6, 3.4)	273 (52)	2.2 (1.8, 2.7)	6292 (1,115)	2.0 (1.6, 2.4)	97.0a (95.7, 98.4)	32.0a (1.1, 62.9)
West North Central	3790 (610)	3.3 (2.9, 3.8)	393 (70)	1.9 (1.6, 2.2)	2411 (390)	1.7 (1.4, 2.1)	89.6a (84.8, 94.5)	36.4a (7.9, 64.8)
South Atlantic	3310 (465)	0.8 (0.7, 0.9)	208 (38)	1.0 (0.7, 1.2)	4048 (703)	0.9 (0.7, 1.2)	93.7a (90.9, 96.6)	−22.3 (−75.9, 31.3)
East South Central	6535 (892)	4.5 (4.1, 5.0)	206 (38)	3.4 (2.7, 4.3)	5276 (983)	3.7 (2.9, 4.5)	96.8a (95.4, 98.3)	19.3 (−17.3, 55.8)
West South Central	5765 (665)	2.4 (2.1, 2.7)	190 (32)	1.9 (1.4, 2.4)	4491 (743)	1.9 (1.4, 2.4)	96.7a (95.4, 98.0)	22.1 (−8.7, 52.9)
Mountain	670 (105)	0.6 (0.5, 0.7)	166 (30)	0.7 (0.6, 0.9)	881 (163)	0.7 (0.6, 0.9)	75.2a (63.7, 86.7)	−31.5 (−94.0, 31.0)
Pacific	1530 (180)	0.6 (0.5, 0.7)	219 (42)	0.8 (0.6, 1.1)	2053 (385)	0.8 (0.6, 1.0)	85.7a (79.4, 91.9)	−34.2 (−92.4, 24.0)
Sex
Male	18 455 (1192)	2.2 (1.9, 2.4)	1042 (94)	1.6a (1.5, 1.9)	15 995 (1,454)	1.8a (1.6, 2.1)	94.4a (93.0, 95.7)	13.3 (−7.3, 34.0)
Female	14 025 (1192)	1.2 (1.1, 1.4)	680 (69)	0.8 (0.7, 0.9)	10 510 (1,181)	0.9 (0.7, 1.1)	95.2a (93.9, 96.4)	25.1a (4.4, 45.8)
Race and ethnicity
NH White	23 580 (1832)	1.7 (1.6, 1.9)	1228 (104)	1.3a (1.1, 1.4)	19 483 (1,725)	1.4a (1.2, 1.6)	94.8a (93.6, 96.0)	17.4 (−1.7, 365)
NH Black/AA	5020 (450)	1.6 (1.4, 1.9)	201 (29)	1.3 (1.0, 1.6)	3098 (432)	1.2 (0.9, 1.7)	96.0a (94.7, 97.3)	38.3a (18.2, 58.4)
Hispanic or Latino	2685 (254)	1.2 (1.0, 1.4)	236 (42)	1.0 (0.7, 1.3)	3230 (560)	1.1 (0.8, 1.4)	91.2a (87.7, 94.7)	−20.3 (−66.8. 26.2)
NH Asian/PI	445 (63)	0.8 (0.6, 1.1)	20 (6)	0.6 (0.3, 1.0)	241 (73)	0.6 (0.3, 1.1)	95.5a (92.7, 98.3)	45.9a (10.2, 81.6)
NH AI/AN	125 (28)	1.0 (0.6, 1.5)	7 (4)	0.5 (0.2, 1.4)	66 (48)	0.7 (0.2, 2.7)	94.4a (87.9, 100.9)	47.1 (−31.4, 125.5)
NH other race	625 (70)	1.0 (0.9, 1.3)	30 (6)	0.6 (0.4, 0.9)	388 (96)	0.7 (0.5, 1.2)	95.2a (93.0, 97.4)	38.0a (4.9, 71.1)
Age group
18–34	4155 (330)	1.1 (0.9, 1.3)	233 (34)	0.8 (0.6, 1.1)	3878 (581)	1.0 (0.7, 1.3)	94.4a (92.6, 96.2)	6.7 (−24.4, 37.7)
35–64	16 675 (1480)	2.2 (1.9, 2.5)	822 (84)	1.4 (1.2, 1.7)	12 298 (1,308)	1.5 (1.3, 1.8)	95.1a (93.8, 96.4)	26.2a (6.2, 46.3)
65 and over	11 650 (1184)	1.3 (1.2, 1.5)	667 (75)	1.1 (0.9, 1.2)	10 328 (1,225)	1.2 (1.0, 1.4)	94.3a (92.6, 96.0)	11.3 (−15.8, 38.5)

Abbreviations: AI/AN, American Indian or Alaska Native; HCUP NIS, Healthcare Cost and Utilization Project National Inpatient Sample; NH, non-Hispanic; OERWD, Oracle EHR Real World Data; PI, Pacific Islander.

Apparent full overlap in some CIs is due to rounding; the underlying point estimates and SEs differ and were used for statistical comparisons.

^aA statistically significant difference between the HCUP NIS dataset and the respective OERWD dataset (both weighted and unweighted) at a 95% CI on the OERWD-derived estimate; apparent full overlap in some CIs is due to rounding; the underlying point estimates and SEs differ and were used for statistical comparisons.

As compared to weighted OERWD, HCUP NIS recorded significantly more aspergillosis encounters among patients identifying as non-Hispanic White (RD: 22.4% [95% CI, 8.8–36.1], P = .001), non-Hispanic Black or African American (RD: 31.1% [95% CI, 13.9–48.3], P < .001), and non-Hispanic Asian or Pacific Islander (RD: 48.4% [95% CI, 28.5–68.4], P < .001). While OERWD had more weighted aspergillosis encounters among Hispanic or Latino patients, the difference was not statistically significant (RD: −29.7% [95% CI, −66.0 to 6.6], P = .1; Table 2). HCUP NIS also recorded significantly higher aspergillosis encounters by sex, with the largest discrepancy after weighting remaining among male patients (RD: 21.1% [95% CI, 7.0–35.2], P = .003; Table 2). After weighting, only encounters among those aged 35–64 years remained statistically significantly different (RD: 25.6% [95% CI, 11.3–40.0], P < .001; Table 2).

Aspergillosis prevalence

Across the study period (2014–2020), the annual prevalence of aspergillosis increased in OERWD (Mann–Kendall τ = 0.7, P = .04) but remained stable in HCUP NIS (Mann–Kendall τ = 0.1, P = .8; Figure 1). National prevalence of aspergillosis diagnoses between 2014 and 2020 reported in HCUP NIS (5.3/10 000 encounters [95% CI, 5.1–5.5]) significantly exceeded that reported by OERWD (4.2/10 000 weighted encounters [95% CI, 4.0–4.4], P < .001; Table 2), with prevalence estimates becoming more aligned in 2018 and onward. Prevalence was highest in Pacific census division in both datasets (HCUP NIS: 9.0/10 000 encounters [95% CI, 8.4–9.6]; weighted OERWD: 7.4/10 000 encounters [95% CI, 6.6–8.2], P = .5; Table 2). In HCUP NIS, West South Central census division had the lowest prevalence (4.0/10 000 encounters [95% CI, 3.7–4.3]). Although prevalence was similarly low in OERWD for this region (2.7/10 000 encounters [95% CI, 2.2–3.2]), the lowest prevalence in that dataset was observed in West North Central census division (2.2/10 000 encounters [95% CI, 1.9–2.6]; Table 2). Statistically significant differences in aspergillosis prevalence were observed between OERWD and HCUP NIS for 3 of the 9 census divisions (Figure 2A).

Figure 1.

Prevalence of aspergillosis and histoplasmosis encounters (per 10 000) in Healthcare Cost and Utilization Project National Inpatient Sample (HCUP NIS) and Oracle EHR Real World Data (OERWD)—United States, 2014–2020. OERWD prevalence estimates are calculated using weighted OERWD fungal encounter and total encounter counts. Gray ribbons show 95% CIs.

Figure 2.

Prevalence of aspergillosis and histoplasmosis encounters (per 10 000) by census division (2014–2020), estimated using Healthcare Cost and Utilization Project National Inpatient Sample (HCUP NIS) and Oracle EHR Real World Data (OERWD). Panels: (A) HCUP NIS aspergillosis, (B) OERWD aspergillosis, (C) HCUP NIS histoplasmosis, (D) OERWD histoplasmosis. OERWD prevalence estimates are calculated using OERWD fungal encounter and total encounter counts that are weighted to account for non-representative facility locations. Census divisions: 1 = New England, 2 = middle Atlantic, 3 = East North Central, 4 = West North Central, 5 = South Atlantic, 6 = East South Central, 7 = West South Central, 8 = mountain, 9 = Pacific. The red outline indicates census divisions have a statistically significant difference (P < .05) in prevalence between the 2 datasets.

Aspergillosis prevalence estimates from OERWD differed significantly from those in HCUP NIS for several demographic characteristics (Table 2). However, patterns were broadly similar across datasets. The highest prevalence was observed among non-Hispanic Asian or Pacific Islander patients in both datasets (HCUP NIS: 8.4/10 000 encounters [95% CI, 7.3–9.6]; weighted OERWD: 6.0/10 000 encounters [95% CI, 4.8–7.6], P = .8; Table 2), while the lowest prevalence was among non-Hispanic American Indian or Alaska Native patients (HCUP NIS: 3.5/10 000 encounters [95% CI, 2.8–4.5]; weighted OERWD: 3.8/10 000 encounters [95% CI, 2.6–5.7], P = .1; Table 2). Prevalence estimates were similar between datasets for all races and ethnicities other than non-Hispanic White and Hispanic or Latino patients (Table 2). Aspergillosis was consistently more frequent among males in both datasets (HCUP NIS: 6.9/10 000 encounters [95% CI, 6.7–7.2]; weighted OERWD: 5.3/10 000 encounters [95% CI, 4.9–5.7], P < .001; Table 2). Patients aged 35–64 years had the highest aspergillosis prevalence in HCUP NIS (6.1/10 000 encounters [95% CI, 5.8–6.4]), while in OERWD the highest prevalence was observed in patients aged 65 years and over (4.5/10 000 encounters [95% CI, 4.2–4.9]); Table 2).

Histoplasmosis encounters

Oracle EHR Real World Data recorded 26 505 (SE = 1833) inpatient histoplasmosis diagnoses (1722 unweighted, SE = 114) between 2014 and 2020, compared to 32 480 (SE = 1851) in HCUP NIS (RD: 18.4% [95% CI, 4.1–32.7], P = .01; Table 2). Histoplasmosis encounter counts were significantly lower across all demographic and geographic characteristics in OERWD prior to weighting (Table 2). Weighting narrowed these differences, but did not eliminate them, with significant gaps persisting across 3 geographic strata and 3 racial and ethnic groups. As compared to weighted OERWD data, HCUP NIS recorded significantly more histoplasmosis encounters in the Middle Atlantic (RD: 42.3% [95% CI, 13.2–71.5], P = .004), East North Central (RD: 32.0% [95% CI, 1.1–62.9], P = .04), and West North Central (RD: 36.4% [95% CI, 7.9–64.8], P = .01) census divisions (Table 2).

Healthcare Cost and Utilization Project National Inpatient Sample recorded significantly more histoplasmosis encounters than weighted OERWD encounters for non-Hispanic Black or African American (RD: 38.3% [95% CI, 18.2–58.4], P < .001), non-Hispanic Asian or Pacific Islander (RD: 45.9% [95% CI, 10.2–81.6], P = .02), and non-Hispanic Other Race patients (RD: 38.0% [95% CI, 4.9–71.1], P = .03). Encounter numbers were comparable for non-Hispanic White, Hispanic or Latino, and non-Hispanic American Indian or Alaska Native patients (Table 2). By sex, HCUP NIS recorded significantly more encounters among females, but estimates for males were similar (female RD: 25.1% [95% CI, 4.4–45.8], P = .02; male RD: 13.3% [95% CI, −7.3 to 34.0], P = .2; Table 2). Differences were also observed across 1 age group, with HCUP NIS recording significantly more encounters among patients aged 35–64 years (RD: 26.2% [95% CI, 6.2–46.3], P = .01; Table 2).

Histoplasmosis prevalence

Annual histoplasmosis prevalence remained stable over the study period in both datasets (HCUP NIS Mann–Kendall τ < 0.01, P = 1; OERWD Mann–Kendall τ = 0.1, P = .8; Figure 1). However, overall prevalence of histoplasmosis encounters reported in HCUP NIS significantly exceeded that reported by OERWD (HCUP NIS: 1.6/10 000 encounters [95% CI, 1.5–1.8]; weighted OERWD: 1.3/10 000 encounters [95% CI, 1.2–1.4], P = .001; Table 2). In both datasets, prevalence was highest in East South Central census division (HCUP NIS: 4.5/10 000 encounters [95% CI, 4.1–5.0]; weighted OERWD: 3.7/10 000 encounters [95% CI, 2.9–4.5], P = .6), and lowest in New England (HCUP NIS: 0.3/10 000 encounters [95% CI: 0.2–0.4]; weighted OERWD: 0.3/10 000 encounters [95% CI, 0.2–0.5], P = .8) and Middle Atlantic census divisions (HCUP NIS: 0.5/10 000 encounters [95% CI, 0.24–0.5]; weighted OERWD: 0.3/10 000 encounters [95% CI, 0.2–0.4], P = .2; Table 2). No census divisions had a prevalence in OERWD that was statistically significantly different from HCUP NIS (Figure 2B).

Histoplasmosis prevalence across racial and ethnic groups was comparable between OERWD and HCUP NIS. The highest prevalence was observed among non-Hispanic White patients in both datasets (HCUP NIS: 1.7/10 000 encounters [95% CI, 1.6–1.9]; weighted OERWD: 1.4/10 000 encounters [95% CI, 1.2–1.6], P = .01; Table 2), and this was the only group with a statistically significant difference. The lowest prevalence was among non-Hispanic Asian or Pacific Islander patients (HCUP NIS: 0.8/10 000 encounters [95% CI, 0.6–1.1]; weighted OERWD: 0.6/10 000 encounters [95% CI, 0.3–1.1], P = .9; Table 2). Significant differences persisted by sex after weighting OERWD. However, histoplasmosis had a higher prevalence among males in both datasets (HCUP NIS: 2.2/10 000 encounters [95% CI, 1.9–2.4]; weighted OERWD: 1.8/10 000 encounters [95% CI, 1.6–2.1], P = .05; Table 2). No significant differences were observed among age groups after weighting (Table 2).

Discussion

This study assessed the comparability between a large, national EHR-derived dataset (OERWD) and a well-established, nationally representative administrative dataset (HCUP NIS) for monitoring inpatient encounters related to aspergillosis and histoplasmosis. Despite inherent differences in structure and sampling frames, these data sources yielded broadly similar insights into disease burden, illustrating both the potential and limitations of EHR-derived data for studying rare or underreported diseases. While HCUP NIS is well-suited for benchmarking disease prevalence at the population level to enhance the validity and generalizability of epidemiological findings,²⁹ its utility for addressing clinical outcomes is limited by its lack of detailed clinical information. For example, in a comparison with the American College of Surgeons National Surgical Quality Improvement Program, HCUP NIS underperformed in estimating postoperative complications, likely due to differences in database design and clinical granularity.²⁵ In contrast, EHR-derived datasets like OERWD offer richer clinical detail—including laboratory orders, medication administration, and patient management information—which may improve case ascertainment and support analyses of treatment patterns and outcomes.^10,30 These complementary strengths highlight the value of validating EHR data: while HCUP NIS provides a robust population-level benchmark, OERWD can offer greater clinical depth to help address longstanding gaps in disease surveillance, particularly for rare or emerging diseases, such as fungal infections, where traditional surveillance systems are fragmented or lacking.

Prevalence estimates for aspergillosis and histoplasmosis derived from OERWD generally aligned with national benchmarks from HCUP NIS, with similar patterns observed across key demographic and geographic strata. Aspergillosis prevalence was highest in the Pacific census division, and while the datasets differed in identifying the division with the lowest prevalence, both identified West North Central and West South Central as low-burden regions. Prevalence was higher among males, highest among non-Hispanic Asian or Pacific Islander patients, and lowest among non-Hispanic American Indian or Alaska Native patients in both datasets. Although HCUP NIS showed the highest prevalence among adults aged 35–64 years and OERWD showed the highest among those aged 65 and older, both datasets consistently identified these 2 age groups as having higher prevalence relative to younger age groups. For histoplasmosis, both datasets identified the highest prevalence in the East South Central census division and the lowest in New England and Middle Atlantic. Prevalence patterns were again consistent, with higher rates among males, the highest rates among non-Hispanic White patients, the lowest among non-Hispanic Asian or Pacific Islander patients, and the highest age-specific prevalence in the 35–64 year group. While not a perfect match, OERWD provides reasonable estimates of inpatient prevalence for these fungal diseases.

Encounter counts were more sensitive to differences in dataset composition, and weighting was essential for reducing RDs between OERWD and HCUP NIS. After weighting, most strata became statistically comparable, but significant differences persisted for aspergillosis encounters in the East North Central, West North Central, and West South Central divisions; by sex; among non-Hispanic White, non-Hispanic Black or African American, non-Hispanic Asian or Pacific Islander patients; and those aged 35–64 years. For histoplasmosis, differences remained after weighting in the Middle Atlantic, East North Central, and West North Central divisions; among female patients; among non-Hispanic Black or African American, non-Hispanic Asian or Pacific Islander, and non-Hispanic Other Race patients; and patients aged 35–64 years. While prevalence estimates reflect the frequency of disease within a population, encounter counts are often more directly relevant for healthcare resource planning, outbreak detection, policy evaluation, and market forecasting.^31–33 Forrare or emerging diseases, where stable rate estimates may be difficult to obtain, encounter counts may offer a practical and actionable metric for monitoring trends and informing decision-making.^34–36 Given these findings, studies focused on encounter counts in OERWD should apply appropriate weighting methods to ensure that estimates reflect national patterns.

These findings align with previous comparisons of EHR-derived datasets to HCUP NIS. For example, a comparison of Oracle Cerner’s Health Facts data to HCUP NIS found that although Health Facts had a similar relative distribution of major diagnostic categories, it had disproportionately more encounters in the Northeast and Midwest, underrepresented non-Hispanic Black or African American patients, and overrepresented non-Hispanic White patients—patterns similar to those observed in OERWD.²⁶ Comparisons of HCUP NIS with EHR data from a single-state research consortium showed strong agreement in the ranking of the most common chronic conditions,²³ and analyses using Medicare and Medicaid claims data found similar Clostridium difficile infection rates.²⁴ However, these studies focused on common conditions without applying weighting methods, limiting their applicability for rarer diseases, such as aspergillosis and histoplasmosis. To the best of our knowledge, this is the first study to benchmark fungal disease encounter counts and prevalence estimates against a nationally representative administrative dataset while explicitly applying weighting. By leveraging calibration weights to produce nationally representative demographic and geographic distributions of estimates, we demonstrated how weighting can improve the generalizability of EHR-based findings for rare conditions while preserving the clinical richness and granularity that such datasets offer.

Methodological differences between OERWD and HCUP NIS pose limitations for this study. While HCUP NIS is a nationally representative, discharge-derived dataset, OERWD captures data only from organizations that have agreed to contribute to an EHR vendor-led initiative and is not a random sample. Our application of calibration weighting and benchmarking against HCUP NIS aimed to mitigate concerns about bias and representativeness that have historically limited the utility of EHR data for population-level estimates. By combining the strengths of both datasets, we improve the reliability of fungal encounter and prevalence estimates. However, our weighting approach is less detailed than the HCUP NIS design and may introduce error, particularly in regions of the United States with fewer contributing hospitals. While calibration weighting improved overall alignment with HCUP NIS, encounter counts and prevalence estimates for some strata remained statistically significantly different between the datasets, which may reflect unmeasured structural differences between datasets that remained unaddressed in our weighting scheme. Future analyses of OERWD would benefit from incorporating additional weighting variables (eg, hospital type or size), calibrating to a census (eg, American Hospital Association’s Annual Survey)³⁷ rather than a representative sample, validating case definitions across institutions, and expanding comparisons to outpatient settings. Finally, since this study focuses exclusively on inpatient encounters, findings may not be generalizable to outpatient encounters, where coding and documentation practices may differ. Despite these limitations, OERWD provides fungal disease estimates comparable to HCUP NIS when properly weighted, supporting its use in inpatient fungal disease research.

Conclusions

In this study, we compared a large, national EHR database with a well-established, nationally representative sample of inpatient encounters to assess the utility of EHR-derived data for estimating inpatient fungal disease encounters and prevalence. Despite methodological differences, our findings demonstrate that geographic and demographic distributions of prevalence were broadly consistent between datasets with no statistically significant differences for most subgroup comparisons (31 of 42). Both datasets identified higher aspergillosis prevalence in the Pacific division and higher histoplasmosis prevalence in the East South Central division, with males and older adults consistently showing higher prevalence. By showing that weighted EHR-derived prevalence patterns across demographic and geographic strata generally parallels national benchmarks, and that weighting estimates improves precision, this study offers an initial framework for using large EHR databases to investigate fungal disease epidemiology in greater clinical detail. Benchmarking representativeness enables future work to leverage EHR data to examine diagnostic practices, treatment patterns, and patient outcomes across diverse populations and care settings, and ultimately to refine prevalence estimates and detect infections that may otherwise go unrecognized in traditional surveillance systems. Future research should evaluate the representativeness of EHR databases across different care settings, refine weighting strategies to further improve national representativeness, and expand analyses to additional diseases. Broadening the use of EHR data can enhance fungal disease research, inform clinical decision-making, and support public health efforts to address the growing burden of fungal infections.

Supplementary Material

Supplementary material is available at AJE Advances: Research in Epidemiology online.

Data availability

Due to proprietary restrictions, the datasets analyzed in this study are not publicly available. Researchers from academia, healthcare systems, or life sciences organizations can access OERWD if their institution contributes de-identified data to the dataset or partners with Oracle Health through the Learning Health Network to use Healthe-DataLab, a cloud-based, distributed learning platform for approved research projects. Other researchers or organizations may apply for access to OERWD, subject to approval. The code for this analysis is available at https://github.com/b-lmb/OERWD-HCUP-Benchmarking.