Social Determinants of Health and Surgical Desirability of Outcome Ranking in Older Veterans

Key Points

Question

What are the associations between social determinants of health, care fragmentation, and veteran surgical outcomes?

Findings

This cohort study of US veterans found worse surgical outcomes among veterans who identified as Black and with greater proportions of non–Veterans Affairs (VA) care, similar outcomes in veterans living in highly resource-deprived neighborhoods, and better outcomes among veterans living in rural areas.

Meaning

To address these disparities, programs should direct resources to reduce presentation acuity among Black veterans, incentivize veterans to receive care within the VA where possible, and create support programs to better coordinate veterans’ treatment and records between health care professionals.

Abstract

Importance

Evaluating how social determinants of health (SDOH) influence veteran outcomes is crucial, particularly for quality improvement.

Objective

To measure associations between SDOH, care fragmentation, and surgical outcomes using a Desirability of Outcome Ranking (DOOR).

Design, Setting, And Participants

This was a cohort study of US veterans using data from the Veterans Affairs (VA) Surgical Quality Improvement Program (VASQIP; 2013-2019) limited to patients aged 65 years or older with inpatient stays between 2 and 30 days, merged with multiple data sources, including Medicare. Race and ethnicity data were retrieved from VASQIP, Medicare and Medicaid beneficiary summary files, the Veterans Health Administration Corporate Data Warehouse, and the United States Veterans Eligibility Trends and Statistics file. Data were analyzed between September 2023 and February 2024.

Exposure

Living in a highly deprived neighborhood (Area Deprivation Index >85), race and ethnicity used as a social construct, rurality, and care fragmentation (percentage of non-VA care days).

Main Outcomes and Measures

DOOR is a composite, patient-centered ranking of 26 outcomes ranging from no complication (1, best) to 90-day mortality or near-death complications (6, worst). A series of proportional odds regressions was used to assess the impact of SDOH and care fragmentation adjusted for clinical risk factors, including presentation acuity (presenting with preoperative acute serious conditions and urgent or emergent surgical procedures).

Results

The cohort had 93 644 patients (mean [SD] age, 72.3 [6.2] years; 91 443 [97.6%] male; 74 624 [79.7%] White). Veterans who identified as Black (adjusted odds ratio [aOR], 1.06; 95% CI, 1.02-1.10; P = .048) vs White and veterans with higher care fragmentation (per 20% increase in VA care days relative to all care days: aOR, 1.01; 95% CI, 1.01-1.02; P < .001) were associated with worse (higher) DOOR scores until adjusting for presentation acuity. Living in rural geographic areas was associated with better DOOR scores than living in urban areas (aOR, 0.93; 95% CI, 0.91-0.96; P < .001), and rurality was associated with lower presentation acuity (preoperative acute serious conditions: aOR, 0.88; 95% CI, 0.81-0.95; P = .001). Presentation acuity was higher in veterans identifying as Black, living in deprived neighborhoods, and with increased care fragmentation.

Conclusions and Relevance

Veterans identifying as Black and veterans with greater proportions of non-VA care had worse surgical outcomes. VA programs should direct resources to reduce presentation acuity among Black veterans, incentivize veterans to receive care within the VA where possible, and better coordinate veterans’ treatment and records between care sources.

This cohort study of US veterans examines the associations between social determinants of health, care fragmentation, and surgical outcomes.

Introduction

Social determinants of health (SDOH), the conditions in which people are born, grow, learn, work, and age,¹ drive disparities in health outcomes.^2,3 These disparities, studied across various health care domains, are critically important in surgical care,⁴ as even minor surgery can sometimes include serious risk.^5,6 Moreover, SDOH are incredibly varied: Black patients can have 1.2 to 1.5 times higher surgical mortality vs White patients,⁴ patients living in areas labeled by the US Government Home Owners Loan Corporation as historically “hazardous” neighborhoods can have 1.19 times higher mortality than those living in areas labeled the “best” neighborhoods,⁷ and insurance type (means-tested insurance like Medicaid or no insurance) drives greater urgent and emergent surgical procedures,⁸ complications,⁹ readmissions,^8,10 and lengths of hospital stay.¹¹ All this prior work suggests that multiple SDOH need to be evaluated simultaneously, not in isolation, because SDOH can have independent effects¹² or be heavily confounded with each other.¹³

While previous, private-sector work has evaluated multiple SDOH effects simultaneously,^12,14,15 these studies are lacking in veteran populations. Veterans have both unique opportunities (comprehensive medical service coverage and transportation to health care appointments) and unique challenges (less education¹⁶ and higher rates of disability,¹⁶ mental illness,^17,18 and suicide¹⁹). Furthermore, while veterans are more prone to homelessness,²⁰ veteran access to Veteran Affairs (VA) mortgage loans may increase their ability to live in more affluent neighborhoods.²¹ All these SDOH differences make it crucial to determine how SDOH negatively impact veteran outcomes, especially to develop care pathways that alleviate SDOH disparities.

A prior study of veterans examined multiple SDOH on surgical mortality.²² This study suggested a survival benefit among Black veterans vs White veterans and found similar mortality among veterans living in resource-deprived vs nondeprived neighborhoods.²² However, while mortality is an important quality indicator, it is insufficient to establish the absence of SDOH-related disparities. A composite outcome, one with greater statistical power than individual outcomes, is needed. We aimed to develop such an outcome, a Desirability of Outcome Ranking (DOOR)²³ with a single, ordinal score combining various surgical outcomes, for use in VA surgical data. DOOR was originally developed to evaluate antibiotic performance in clinical trials,²³ and these methods were later adapted for surgical outcomes using National Surgical Quality Improvement Program (NSQIP) data²⁴ and then adding electronic health record (EHR)–based outcomes.¹⁴ DOOR has more statistical power than analyzing outcomes separately^25,26 or binary composites,^27,28,29 like textbook outcomes. At every iteration, surgical DOORs used a modified Delphi process,³⁰ a valid, established method to substantiate consensus among experts (and previously used to determine research priorities for various surgical specialties^{31,32,33,34,35} and a surgical safety checklist³⁶). Surgical DOORs demonstrated improved detection of complex SDOH relationships within private-sector cohorts.^14,24 We aimed to use a VA Surgical Quality Improvement Program (VASQIP)–specific DOOR to examine outcome disparities from a variety of SDOH simultaneously: (1) minoritized race; (2) living in highly resource-deprived areas; (3) living in rural areas; and (4) care fragmented between the VA and private-sector sources.

Methods

Design and Setting

VASQIP contains nurse-abstracted data on VA surgical procedures.³⁷ This retrospective cohort study used a national sample of noncardiac VASQIP patients (2013-2019) with inpatient stays of at least 2 days to select for higher-risk procedures that have greater complication risk.^38,39 We excluded patients with inpatient stays longer than 30 days to prevent extreme outliers from driving our SDOH disparity estimates, and we limited patients to those aged 65 years or older at the time of surgery, ensuring Medicare eligibility. Veterans younger than 65 years have relatively little non-VA data, so patients younger than 65 years were excluded to ensure coverage of non–Veterans Health Administration (VHA) health care days. Although we could have analyzed other SDOH using a larger sample that included younger patients, that approach would have undermined our goal of examining multiple SDOH associations in the same model. The study was determined exempt by the VA Central Institutional Review Board owing to the use of deidentified data and is reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.⁴⁰

Data Sources

The VASQIP cohort was linked to the following: (1) race and diagnosis codes from VHA Corporate Data Warehouse (CDW) records; (2) Medicare and Medicaid claims data from Centers for Medicare and Medicaid Services (CMS); (3) private-sector claims data for VA-covered care from the VA Fee-Based and Program Integrity Tool files; and (4) veteran geographic residence from the VA Planning Systems Support Group (PSSG). Mortality was determined using the Vital Status File, which contains mortality data from multiple VA and non-VA data sources, such as the Social Security Administration Death Master File and the Integrated Benefits System Death File.

To ensure that each case represented a single, unique patient, cases with scrambled Social Security numbers associated with more than 1 unique patient identifier across data sources were excluded. This approach accounts for rare data entry errors that produce 2 distinct patients sharing the same scrambled Social Security number. VASQIP cases within 30 days of the index case were excluded. Subsequent cases occurring more than 30 days from a previous procedure were considered independent. We excluded patients with addresses reported as (1) missing; (2) Post Office Box numbers; (3) outside the United States (including Puerto Rico); or (4) incomplete address data. We also excluded patients (1) with no record of VA care in the 2 years prior to surgery; (2) missing patient- or procedure-specific data; (3) missing SDOH data; (4) enrolled in Medicare health maintenance organization plans because claims for managed care enrollees are not routinely available in CMS inpatient, outpatient, or carrier files (we lacked information on services or diagnoses incurred by managed care enrollees); or (5) enrolled in Medicare for less than 1 year to ensure adequate diagnoses for Gagne score calculation and to accrue adequate time to detect care fragmentation.

Preoperative Comorbidities

Frailty was measured using the recalibrated Risk Analysis Index (RAI)⁴¹ with preoperative VASQIP variables. The RAI provides a composite measure of patient-related risk that avoids problems with model fit when models include each RAI component separately^42,43,44,45 and was used as a continuous risk factor (0-81). Other comorbidities were measured using the Gagne score⁴⁶ with VA diagnoses and Medicare or Medicaid claims data from the year prior to surgery. Age at the time of surgery was used as a separate risk factor.

Presentation Acuity

Patients with preoperative acute serious conditions (PASC)⁴² were defined as patients presenting with any of several VASQIP variables (eTable 1 in Supplement 1). Case status was classified as elective, urgent, or emergent. Emergency cases were flagged as emergent in the VASQIP database. Cases were considered urgent if VA surgical scheduling data indicated they were scheduled as (1) urgent; (2) add-on; (3) standby; or (4) emergent but with a nonemergent VASQIP flag. All other cases were considered elective unless scheduling type was missing; these cases were excluded.

Operative Variables

Operative variables included surgery-related physiological stress measured by the expanded Operative Stress Score (OSS).^5,45,47 The OSS assigns ratings between 1 (very low stress) and 5 (very high stress) to 2343 Current Procedural Terminology (CPT) codes⁴² and is an effective estimate of procedure-related variability.^42,43,47 OSS captures physiological stress of specific surgical procedures to account for procedure-related variation in risk while (1) preserving the statistical power of the entire sample and (2) accounting for rare, but potentially meaningful, poor outcomes in low-risk procedures, as suggested by Chen et al.⁴⁸ The expanded OSS is comprehensive, with 98% coverage in NSQIP data.⁴⁷ Cases with multiple CPT codes were assigned the highest OSS value (eTable 2 in Supplement 1). Cases with unscorable principal procedures were excluded. The primary surgeon’s specialty from VASQIP was used to adjust for specialty-specific variation.

SDOH

Race and ethnicity data were retrieved from, in order, (1) VASQIP, (2) Medicare or Medicaid beneficiary summary files using codes from the Research Triangle Institute,⁴⁹ (3) CDW, and (4) the United States Veterans Eligibility Trends and Statistics file. Race and ethnicity, used as a social construct, were categorized as White (non-Hispanic), Black (non-Hispanic), Hispanic (any race), or other (non-Hispanic). Other (non-Hispanic) included Native American and Asian or Pacific Islander, which had too few cases as separate groups for statistically reliable comparisons. Rurality came from the PSSG, which uses the rurality definition from the Office of Management and Budget using rural-urban commuting area codes. The Area Deprivation Index (ADI)⁵⁰ ranks neighborhoods by disadvantage using a composite measure of 17 education, employment, housing quality, and poverty measures from the American Community Survey at the census block group level.⁵⁰ Coordinates associated with each patient’s address were mapped to a census block group to determine ADI. We dichotomized ADI to identify veterans living in the 15% most deprived areas, as the ADI is more effective at identifying deprived neighborhoods than wealthy ones.⁵⁰

Preoperative Care Fragmentation

Care fragmentation was defined by adapting the Usual Provider Continuity index⁵¹ to measure the percentage of days receiving non-VA care during the 12 months before the surgical admission relative to all preoperative care days in CDW, VA fee-based care, and CMS claims data. Although care fragmentation can be conceptualized as either an SDOH or the consequence of other SDOH, we follow the Centers for Disease Control and Prevention in defining it as a social determinant of health care access and quality.⁵²

DOOR

A previous surgical DOOR study derived DOOR from NSQIP and EHR variables.¹⁴ Although NSQIP and VASQIP share many variables, VASQIP contains 4 postoperative complications not in NSQIP. Therefore, the same surgeons who created the previous ranking system¹⁴ (P.K.S., K.B.S., L.S.K., and D.E.H.) reconvened to rank the 4 VASQIP-unique variables using a modified Delphi process.^30,39 These 4 outcomes, combined with 22 previously ranked outcomes, were used to create a VASQIP-specific DOOR (Table 1). DOOR rankings were assigned 1, representing no complications (the most desirable outcome), or 2 to 6, representing progressively less desirable (worse) outcomes. For each round, previous rankings were summarized and shared, disagreements were discussed to build consensus, and discussions were continued until a final, unanimous consensus was reached.

Table 1. Variables Included in Desirability of Outcome Ranking (DOOR)^a.

Variable	Definition	Score
NA	Most desirable outcome—absence of all outcomes below	1
URNINFEC	Infection in the kidneys, ureters, bladder, or urethra, occurring within 30 d after surgery	2
SUPINFEC	Superficial incisional surgical site infection that occurs within 30 d after surgery	2
30-d EDOS	A single EDOS within 30 d of discharge from the index hospitalization, derived from the CDW and CMS	2
WNDINFD	Deep incision surgical site infection that occurs within 30 d after surgery	3
OUPNEUMO	Pneumonia that occurs within 30 d after surgery	3
RENAINSF	Reduced kidney capacity (without requirement for dialysis) within 30 d after surgery	3
CDIFCOLITIS	Clostridioides difficile colitis within 30 d after surgery	3
OTHBLEEDb	Use of ≥5 units of packed or whole red blood cells within 72 h after surgery	3
CNSCOMAb	Significantly impaired level of consciousness (excluding transient disorientation or psychosis) for >24 h during the postoperative hospitalization	4
NEURODEFb	Peripheral nerve damage may result from damage to the nerve fibers, cell body, or myelin sheath during surgery; peripheral nerve injuries (eg, motor, sensory, and mixed motor and sensory injury) to the cervical plexus, brachial plexus, ulnar plexus, lumbosacral plexus (sciatic nerve), perineal nerve, and/or femoral nerve should be included	4
OTHDVT	New diagnosis of blood clot or thrombus within the venous system, occurring within 30 d after surgery	4
30-d Readmission	A readmission within 30 d of discharge from the index hospitalization, derived from the CDW and CMS	4
ORGSPCSSI	Organ or space surgical site infection that occurs within 30 d after surgery	4
DEHIS	Wound separation that compromises integrity of closure, occurring within 30 d after surgery	4
OTHSYSEP	Sepsis (short of septic shock) within 30 d after surgery	4
REINTUB	Intubation intraoperatively or within 30 d after surgery	4
FAILWEAN	Requirement of a ventilator for >48 h cumulatively within 30 d after surgery	4
CDMI	Acute myocardial infarction occurring intraoperatively or within 30 d after surgery	4
Multiple EDOSs or readmissions	Multiple EDOSs or readmissions within 30 d of discharge from the index hospitalization	5
OTHGRAFLb	Mechanical failure of an extracardiac vascular graft or prosthesis, including myocutaneous flaps and skin grafts, requiring return to the operating room or a balloon angioplasty within 30 d after surgery	5
PULEMBOL	New diagnosis of a pulmonary embolism within 30 d after surgery	5
OTHSESHOCK	Sepsis associated with organ and/or circulatory dysfunction within 30 d after surgery	5
OPRENAFL	Kidney failure requiring dialysis within 30 d after surgery	6
CNSCVA	Cerebral vascular accident or stroke with motor, sensory, or cognitive dysfunction for ≥24 h within 30 d after surgery	6
CDARREST	Chaotic or absent cardiac rhythm requiring CPR within 30 d after surgery	6
90-d Mortality	Death within 90 d of surgery, derived from the Vital Status File	6

Abbreviations: CDW, Corporate Data Warehouse; CMS, Centers for Medicare and Medicaid Services; CPR, cardiopulmonary resuscitation; EDOS, emergency department visit or observation stay; NA, not applicable.

^a All DOOR rankings are as previously established¹⁴ unless otherwise noted. See Methods for more details. Surgeons rated complications based on (1) their clinical significance or risk to the patient, (2) the expected duration of the complication and needed treatment, and (3) their desirability (or not) for the typical patient.

^b Variable is unique to the Veterans Affairs Surgical Quality Improvement Program (VASQIP) or has different definitions between VASQIP and the American College of Surgeons National Surgical Quality Improvement Program; its DOOR ranking was established for this study using procedures mirroring the approach used in a previous DOOR version.¹⁴

Statistical Analysis

Our main study outcome was DOOR, with the main analysis being the association between DOOR and various SDOH, adjusted for RAI, OSS, PASC, and case status. We used a series of nested ordinal logistic (proportional odds) models to assess these associations, with the coefficient for patient age allowed to vary across rank comparisons because age violated the proportional odds assumption. Three models were used: (1) race and clinical variables; (2) adding other SDOH to evaluate their associations and establish the robustness of the race associations from model 1; and (3) adding PASC and case status to evaluate what SDOH are associated with surgical outcomes through increasing acuity because previous work demonstrates that many SDOH are associated with surgical outcomes by driving increased acuity.^12,15,24 Considering this prior work, we performed post hoc analyses examining a series of logistic regression models with PASC and case status as outcomes. All models incorporated random facility intercepts to control for the clustering of patients within VA facilities. Analyses were performed using SAS version 9.4 statistical software (SAS Institute Inc). Two-tailed P < .05 was considered statistically significant, and confidence intervals are reported at the 95% level. For descriptive statistics, categorical data were summarized using count and percentage, with continuous data using mean and SD. χ² tests and F tests were used to test for differences between groups for categorical and continuous variables, respectively.

Results

Demographic Characteristics

The cohort had 93 644 cases (eFigure in Supplement 1), with a mean (SD) age of 72.3 (6.2) years (Table 2). Most identified as White (74 624 [79.7%]), followed by Black (13 625 [14.5%]), Hispanic (4066 [4.3%]), and other (1329 [1.4%]); 91 443 (97.6%) were male. Most had normal RAI frailty scores (60 661 [64.8%]) and underwent moderate-stress procedures (OSS of 3) (48 350 [51.6%]). A total of 14 856 cases (15.9%) lived in highly deprived neighborhoods (ADI >85), and 33 531 (35.8%) lived in rural or highly rural areas. Few cases presented with PASC (4115 [4.4%]), and most cases were elective (68 558 [73.2%]). Most had optimal surgical outcomes (DOOR of 1) (62 392 [66.6%]). Demographic characteristics stratified by various SDOH (ADI, race and ethnicity, and rurality) are provided in eTables 3-5 in Supplement 1.

Table 2. Descriptive Statistics Stratified by DOOR.

Characteristic	Overall (N = 93 644)	DOOR						P value
		1 (n = 62 392)	2 (n = 740)	3 (n = 9082)	4 (n = 4636)	5 (n = 11 917)	6 (n = 4877)
Age, mean (SD), y	72.3 (6.2)	72.0 (6.0)	73.0 (6.6)	72.1 (6.0)	72.7 (6.4)	72.6 (6.3)	76.0 (8.1)	<.001
Sex, No. (%)
Female	2201 (2.4)	1575 (2.5)	18 (2.4)	201 (2.2)	85 (1.8)	234 (2.0)	88 (1.8)	<.001
Male	91 443 (97.6)	60 817 (97.5)	722 (97.6)	8881 (97.8)	4551 (98.2)	11 683 (98.0)	4789 (98.2)
Race and ethnicity, No. (%)a
Black, non-Hispanic	13 625 (14.5)	8541 (13.7)	109 (14.7)	1439 (15.8)	698 (15.1)	1972 (16.5)	866 (17.8)	<.001
Hispanic, any race	4066 (4.3)	2723 (4.4)	28 (3.8)	381 (4.2)	198 (4.3)	548 (4.6)	188 (3.9)
White, non-Hispanic	74 624 (79.7)	50 219 (80.5)	597 (80.7)	7138 (78.6)	3663 (79.0)	9244 (77.6)	3763 (77.2)
Other, non-Hispanicb	1329 (1.4)	909 (1.5)	6 (0.8)	124 (1.4)	77 (1.7)	153 (1.3)	60 (1.2)
RAI
Mean (SD)	28.7 (6.6)	27.8 (5.8)	28.6 (5.9)	28.2 (5.9)	30.1 (7.0)	29.7 (6.9)	36.7 (9.7)	<.001
Category, No. (%)
Robust, ≤20	818 (0.9)	618 (1.0)	4 (0.5)	77 (0.8)	24 (0.5)	82 (0.7)	13 (0.3)	<.001
Normal, 21-29	60 661 (64.8)	43 473 (69.7)	476 (64.3)	6099 (67.2)	2540 (54.8)	6737 (56.5)	1336 (27.4)
Frail, 30-39	24 871 (26.6)	15 030 (24.1)	215 (29.1)	2385 (26.3)	1573 (33.9)	3941 (33.1)	1727 (35.4)
Very frail, ≥40	7294 (7.8)	3271 (5.2)	45 (6.1)	521 (5.7)	499 (10.8)	1157 (9.7)	1801 (36.9)
Gagne score, mean (SD)	3.4 (3.4)	2.9 (3.1)	3.7 (3.2)	3.4 (3.2)	4.5 (3.5)	4.4 (3.6)	6.6 (3.9)	<.001
Care fragmentation, mean (SD), %c	22.3 (33.0)	21.5 (32.8)	23.8 (34.9)	21.7 (32.2)	24.5 (34.0)	23.1 (32.7)	28.8 (36.2)	<.001
ADI, No. (%)
≤85	78 788 (84.1)	52 694 (84.5)	598 (80.8)	7680 (84.6)	3845 (82.9)	10 007 (84.0)	3964 (81.3)	<.001
>85	14 856 (15.9)	9698 (15.5)	142 (19.2)	1402 (15.4)	791 (17.1)	1910 (16.0)	913 (18.7)
Rurality, No. (%)
Urban	60 113 (64.2)	39 507 (63.3)	450 (60.8)	5864 (64.6)	3108 (67.0)	7916 (66.4)	3268 (67.0)	<.001
Rural or highly rural	33 531 (35.8)	22 885 (36.7)	290 (39.2)	3218 (35.4)	1528 (33.0)	4001 (33.6)	1609 (33.0)
Expanded OSS, No. (%)
1, Very low	1465 (1.6)	887 (1.4)	7 (0.9)	145 (1.6)	69 (1.5)	226 (1.9)	131 (2.7)	<.001
2, Low	30 196 (32.2)	21 334 (34.2)	148 (20.0)	2914 (32.1)	1257 (27.1)	3483 (29.2)	1060 (21.7)
3, Moderate	48 350 (51.6)	32 249 (51.7)	393 (53.1)	4543 (50.0)	2397 (51.7)	6037 (50.7)	2731 (56.0)
4, High	12 181 (13.0)	7207 (11.6)	169 (22.8)	1311 (14.4)	776 (16.7)	1891 (15.9)	827 (17.0)
5, Very high	1452 (1.6)	715 (1.1)	23 (3.1)	169 (1.9)	137 (3.0)	280 (2.3)	128 (2.6)
Surgeon specialty, No. (%)
General	22 417 (23.9)	13 989 (22.4)	330 (44.6)	2160 (23.8)	1229 (26.5)	3203 (26.9)	1506 (30.9)	<.001
Neurosurgery	5696 (6.1)	3994 (6.4)	32 (4.3)	497 (5.5)	224 (4.8)	677 (5.7)	272 (5.6)
Orthopedic	31 286 (33.4)	23 260 (37.3)	131 (17.7)	2904 (32.0)	1089 (23.5)	2797 (23.5)	1105 (22.7)
Otolaryngology	1741 (1.9)	1087 (1.7)	15 (2.0)	195 (2.1)	130 (2.8)	248 (2.1)	66 (1.4)
Thoracic	5608 (6.0)	3677 (5.9)	34 (4.6)	633 (7.0)	276 (6.0)	696 (5.8)	292 (6.0)
Urologic	8201 (8.8)	5241 (8.4)	58 (7.8)	955 (10.5)	293 (6.3)	1233 (10.3)	421 (8.6)
Podiatry	1060 (1.1)	659 (1.1)	9 (1.2)	82 (0.9)	97 (2.1)	153 (1.3)	60 (1.2)
Peripheral vascular	16 475 (17.6)	9805 (15.7)	124 (16.8)	1530 (16.8)	1227 (26.5)	2685 (22.5)	1104 (22.6)
Other	1160 (1.2)	680 (1.1)	7 (0.9)	126 (1.4)	71 (1.5)	225 (1.9)	51 (1.0)
PASC, No. (%)	4115 (4.4)	1769 (2.8)	18 (2.4)	343 (3.8)	362 (7.8)	709 (5.9)	914 (18.7)	<.001
Case status
Elective	68 558 (73.2)	48 092 (77.1)	519 (70.1)	6860 (75.5)	2972 (64.1)	7993 (67.1)	2122 (43.5)	<.001
Urgent	17 519 (18.7)	10 349 (16.6)	147 (19.9)	1552 (17.1)	1126 (24.3)	2717 (22.8)	1628 (33.4)
Emergent	7567 (8.1)	3951 (6.3)	74 (10.0)	670 (7.4)	538 (11.6)	1207 (10.1)	1127 (23.1)

Abbreviations: ADI, Area Deprivation Index; DOOR, Desirability of Outcome Ranking; OSS, Operative Stress Score; PASC, preoperative acute serious conditions; RAI, Risk Analysis Index.

^a Race and ethnicity data were retrieved from the Veterans Affairs Surgical Quality Improvement Program, Medicare and Medicaid beneficiary summary files, the Veterans Health Administration Corporate Data Warehouse, and the United States Veterans Eligibility Trends and Statistics file.

^b Includes Native American and Asian or Pacific Islander.

^c Care fragmentation measures the percentage of days patients received non–Veterans Affairs care during the 12 months before the index surgery admission.

Black Race and Care Fragmentation Associated With Worse Outcomes

Patients identified as Black had higher (worse) DOORs than White patients (adjusted odds ratio [aOR], 1.06; 95% CI, 1.02-1.10; P = .048) (Table 3). Worse outcomes persisted after adjusting for ADI, care fragmentation, and rurality (aOR, 1.04; 95% CI, 1.00-1.08; P = .0497), but not after adjusting for PASC and case status. Hispanic and other non-Hispanic patients had similar DOORs vs White patients. Patients with higher proportions of non-VA care in the year before surgery had higher (worse) DOORs (per 20% increase in VA care days relative to all care days: aOR, 1.01; 95% CI, 1.01-1.02; P < .001), until adjusting for PASC and case status.

Table 3. Association of DOOR With Clinical, Social Determinant of Health, and Care Acuity Variables^a.

Variable	Model 1		Model 2		Model 3
	aOR (95% CI)	P value	aOR (95% CI)	P value	aOR (95% CI)	P value
Age in 10-y increments by DOOR levelb
1	1 [Reference]	NA	1 [Reference]	NA	1 [Reference]	NA
2	0.95 (0.93-0.98)	<.001	0.95 (0.92-0.97)	<.001	0.96 (0.94-0.99)	.004
3	0.95 (0.93-0.98)	<.001	0.95 (0.92-0.97)	<.001	0.96 (0.93-0.98)	.002
4	1.03 (1.01-1.06)	.04	1.03 (1.00-1.05)	.04	1.04 (1.01-1.07)	.003
5	1.06 (1.03-1.09)	<.001	1.06 (1.03-1.08)	<.001	1.07 (1.04-1.10)	<.001
6	1.61 (1.55-1.68)	<.001	1.60 (1.54-1.66)	<.001	1.63 (1.57-1.69)	<.001
RAI	1.05 (1.05-1.05)	<.001	1.05 (1.05-1.05)	<.001	1.04 (1.04-1.04)	<.001
Gagne score	1.10 (1.09-1.10)	<.001	1.10 (1.09-1.10)	<.001	1.09 (1.09-1.10)	<.001
OSS
1-2, Very low to low	0.99 (0.96-1.02)	.55	0.99 (0.96-1.02)	.55	0.98 (0.95-1.01)	.25
3	1 [Reference]	NA	1 [Reference]	NA	1 [Reference]	NA
4, High	1.36 (1.30-1.42)	<.001	1.36 (1.30-1.42)	<.001	1.42 (1.36-1.49)	<.001
5, Very high	1.77 (1.61-1.96)	<.001	1.78 (1.62-1.97)	<.001	1.94 (1.76-2.15)	<.001
Surgeon specialty
General	1 [Reference]	NA	1 [Reference]	NA	1 [Reference]	NA
Neurosurgery	0.90 (0.84-0.96)	.001	0.90 (0.84-0.96)	.001	1.02 (0.96-1.09)	.66
Orthopedic	0.75 (0.72-0.78)	<.001	0.75 (0.72-0.78)	<.001	0.83 (0.80-0.87)	<.001
Otolaryngology	0.86 (0.78-0.95)	.004	0.86 (0.78-0.95)	.004	1.01 (0.91-1.12)	.86
Thoracic	0.76 (0.71-0.81)	<.001	0.76 (0.71-0.81)	<.001	0.85 (0.79-0.90)	<.001
Urology	1.03 (0.98-1.09)	.20	1.04 (0.98-1.09)	.20	1.17 (1.11-1.23)	<.001
Podiatry	0.90 (0.79-1.03)	.10	0.90 (0.79-1.02)	.10	0.86 (0.75-0.98)	.02
Peripheral vascular	1.09 (1.04-1.14)	<.001	1.09 (1.05-1.14)	<.001	1.19 (1.14-1.23)	<.001
Other	0.97 (0.87-1.10)	.66	0.97 (0.87-1.10)	.66	0.89 (0.79-1.01)	.06
Race and ethnicityc
Black, non-Hispanic	1.06 (1.02-1.10)	.048	1.04 (1.00-1.08)	.0497	1.04 (1.00-1.08)	.08
Hispanic, any race	0.99 (0.93-1.06)	.51	0.98 (0.91-1.05)	.51	0.96 (0.90-1.03)	.28
White, non-Hispanic	1 [Reference]	NA	1 [Reference]	NA	1 [Reference]	NA
Other, non-Hispanicd	0.91 (0.81-1.03)	.10	0.91 (0.81-1.02)	.10	0.91 (0.81-1.02)	.10
ADI
≤85	NA	NA	1 [Reference]	.72	1 [Reference]	.89
>85	NA	NA	1.01 (0.97-1.05)		1.00 (0.96-1.04)
Rurality
Urban	NA	NA	1 [Reference]	<.001	1 [Reference]	<.001
Rural or highly rural	NA	NA	0.93 (0.91-0.96)		0.94 (0.92-0.97)
Care fragmentatione	NA	NA	1.01 (1.01-1.02)	<.001	1.00 (1.00-1.01)	.41
PASC	NA	NA	NA	NA	1.76 (1.65-1.88)	<.001
Case status
Elective	NA	NA	NA	NA	1 [Reference]	NA
Urgent	NA	NA	NA	NA	1.30 (1.25-1.34)	<.001
Emergent	NA	NA	NA	NA	1.66 (1.58-1.75)	<.001

Abbreviations: ADI, Area Deprivation Index; aOR, adjusted odds ratio; DOOR, Desirability of Outcome Ranking; NA, not applicable; OSS, Operative Stress Score; PASC, preoperative acute serious conditions; RAI, Risk Analysis Index.

^a Model 1 includes age, clinical variables, and race and ethnicity; model 2 adds other social determinants of health (ADI, rurality, and care fragmentation); and model 3 further adds PASC and case status.

^b Age was used as a continuous variable (in 10-year increments), but the coefficient was allowed to vary across different levels of DOOR because age violated the proportional odds assumption.

^c Race and ethnicity data were retrieved from the Veterans Affairs Surgical Quality Improvement Program, Medicare and Medicaid beneficiary summary files, the Veterans Health Administration Corporate Data Warehouse, and the United States Veterans Eligibility Trends and Statistics file.

^d Includes Native American and Asian or Pacific Islander.

^e Care fragmentation measures the percentage of days patients received non–Veterans Affairs care during the 12 months before the index surgery admission.

Better Outcomes for Rurality and Similar Outcomes for Area Deprivation

Patients living in rural or highly rural areas had lower (better) DOORs than those living in urban areas (aOR, 0.93; 95% CI, 0.91-0.96; P < .001), and this persisted after adjusting for PASC and case status (aOR, 0.94; 95% CI, 0.92-0.97; P < .001) (Table 3). Patients living in highly deprived neighborhoods had similar DOORs vs those living in less deprived neighborhoods.

Area Deprivation, Black Race, and Care Fragmentation Associated With Higher Presentation Acuity

The odds of PASC are shown in Table 4, and the odds of urgent or emergent surgery are shown in Table 5. Patients identified as Black had higher odds of PASC (aOR, 1.18; 95% CI, 1.08-1.30; P < .001) and urgent (aOR, 1.16; 95% CI, 1.10-1.22; P < .001) or emergent (aOR, 1.28; 95% CI, 1.19-1.38; P < .001) surgery. Patients living in highly deprived neighborhoods had greater odds of PASC (aOR, 1.11; 95% CI, 1.02-1.21; P = .02) and urgent (aOR, 1.08; 95% CI, 1.03-1.14; P = .002) or emergent (aOR, 1.12; 95% CI, 1.05-1.20; P = .001) surgery vs elective surgery. After adjusting for rurality and care fragmentation, Black race and living in highly deprived neighborhoods were associated with higher odds of PASC and urgent or emergent surgery. Patients with higher proportions of non-VA care had higher odds of PASC (aOR, 1.08; 95% CI, 1.06-1.10; P < .001) and urgent (aOR, 1.17; 95% CI, 1.16-1.18; P < .001) or emergent (aOR, 1.19; 95% CI, 1.17-1.21; P < .001) surgery.

Table 4. Association of Preoperative Acute Serious Conditions With Clinical, Social Determinant of Health, and Care Acuity Variables^a.

Variable	Model 1		Model 2		Model 3
	aOR (95% CI)	P value	aOR (95% CI)	P value	aOR (95% CI)	P value
Ageb	0.64 (0.61-0.68)	<.001	0.64 (0.61-0.67)	<.001	0.63 (0.60-0.66)	<.001
RAI	1.13 (1.12-1.13)	<.001	1.13 (1.12-1.13)	<.001	1.13 (1.12-1.13)	<.001
Gagne score	1.04 (1.03-1.05)	<.001	1.04 (1.03-1.05)	<.001	1.04 (1.03-1.05)	<.001
Race and ethnicityc
Black, non-Hispanic	1.18 (1.08-1.30)	<.001	NA	NA	1.14 (1.04-1.25)	.007
Hispanic, any race	1.11 (0.94-1.30)	.21	NA	NA	1.09 (0.93-1.29)	.28
White, non-Hispanic	1 [Reference]	NA	NA	NA	1 [Reference]	NA
Other, non-Hispanicd	1.07 (0.81-1.40)	.64	NA	NA	1.06 (0.81-1.39)	.70
ADI
≤85	NA	NA	1 [Reference]	.02	1 [Reference]	.05
>85	NA	NA	1.11 (1.02-1.21)		1.09 (1.00-1.20)
Rurality
Urban	NA	NA	NA	NA	1 [Reference]	<.001
Rural or highly rural	NA	NA	NA	NA	0.88 (0.81-0.95)
Care fragmentatione	NA	NA	NA	NA	1.08 (1.06-1.10)	<.001

Abbreviations: ADI, Area Deprivation Index; aOR, adjusted odds ratio; NA, not applicable; RAI, Risk Analysis Index.

^a Model 1 includes age, clinical variables, and race and ethnicity; model 2 includes age, clinical variables, and ADI; and model 3 includes age, clinical variables, race and ethnicity, ADI, rurality, and care fragmentation.

^b Age was used as a continuous variable in 10-year increments.

^c Race and ethnicity data were retrieved from the Veterans Affairs Surgical Quality Improvement Program, Medicare and Medicaid beneficiary summary files, the Veterans Health Administration Corporate Data Warehouse, and the United States Veterans Eligibility Trends and Statistics file.

^d Includes Native American and Asian or Pacific Islander.

^e Care fragmentation measures the percentage of days patients received non–Veterans Affairs care during the 12 months before the index surgery admission.

Table 5. Association of Case Status With Clinical, Social Determinant of Health, and Care Acuity Variables^a.

Variable	Urgent						Emergent
	Model 1		Model 2		Model 3		Model 1		Model 2		Model 3
	aOR (95% CI)	P value	aOR (95% CI)	P value	aOR (95% CI)	P value	aOR (95% CI)	P value	aOR (95% CI)	P value	aOR (95% CI)	P value
Ageb	1.21 (1.16-1.24)	<.001	1.20 (1.15-1.23)	<.001	1.16 (1.13-1.20)	<.001	1.15 (1.10-1.20)	<.001	1.14 (1.09-1.18)	<.001	1.09 (1.05-1.14)	<.001
RAI	1.08 (1.08-1.09)	<.001	1.08 (1.08-1.09)	<.001	1.08 (1.08-1.09)	<.001	1.10 (1.09-1.10)	<.001	1.10 (1.09-1.10)	<.001	1.10 (1.09-1.10)	<.001
Gagne score	1.05 (1.05-1.06)	<.001	1.05 (1.05-1.06)	<.001	1.05 (1.04-1.06)	<.001	1.02 (1.01-1.03)	<.001	1.02 (1.01-1.03)	<.001	1.02 (1.01-1.02)	<.001
Race and ethnicityc
Black, non-Hispanic	1.16 (1.10-1.22)	<.001	NA	NA	1.12 (1.06-1.18)	<.001	1.28 (1.19-1.38)	<.001	NA	NA	1.20 (1.11-1.30)	<.001
Hispanic, any race	1.16 (1.06-1.27)	.001	NA	NA	1.15 (1.05-1.25)	.003	1.12 (1.06-1.35)	.004	NA	NA	1.17 (1.03-1.32)	.01
White, non-Hispanic	1 [Reference]	NA	NA	NA	1 [Reference]	NA	1 [Reference]	NA	NA	NA	1 [Reference]	NA
Other, non-Hispanicd	1.09 (0.94-1.26)	.28	NA	NA	1.07 (0.92-1.24)	.40	0.97 (0.78-1.19)	.74	NA	NA	0.93 (0.75-1.15)	.49
ADI
≤85	NA	NA	1 [Reference]	.002	1 [Reference]	.003	NA	NA	1 [Reference]	.001	1 [Reference]	.002
>85	NA	NA	1.08 (1.03-1.14)		1.08 (1.03-1.14)		NA	NA	1.12 (1.05-1.20)		1.12 (1.04-1.20)
Rurality
Urban	NA	NA	NA	NA	1 [Reference]	<.001	NA	NA	NA	NA	1 [Reference]	<.001
Rural or highly rural	NA	NA	NA	NA	0.83 (0.80-0.87)		NA	NA	NA	NA	0.74 (0.69-0.78)
Care fragmentatione	NA	NA	NA	NA	1.17 (1.16-1.18)	<.001	NA	NA	NA	NA	1.19 (1.17-1.21)	<.001

Abbreviations: ADI, Area Deprivation Index; aOR, adjusted odds ratio; NA, not applicable; RAI, Risk Analysis Index.

^a Model 1 includes age, clinical variables, and race and ethnicity; model 2 includes age, clinical variables, and ADI; and model 3 includes age, clinical variables, race and ethnicity, ADI, rurality, and care fragmentation.

^b Age was used as a continuous variable in 10-year increments.

^c Race and ethnicity data were retrieved from the Veterans Affairs Surgical Quality Improvement Program, Medicare and Medicaid beneficiary summary files, the Veterans Health Administration Corporate Data Warehouse, and the United States Veterans Eligibility Trends and Statistics file.

^d Includes Native American and Asian or Pacific Islander.

^e Care fragmentation measures the percentage of days patients received non–Veterans Affairs care during the 12 months before the index surgery admission.

Rurality Associated With Reduced Presentation Acuity

Patients living in rural or highly rural areas had lower odds of PASC (aOR, 0.88; 95% CI, 0.81-0.95; P < .001) (Table 4). They also had lower odds of urgent (aOR, 0.83; 95% CI, 0.80-0.87; P < .001) or emergent (aOR, 0.74; 95% CI, 0.69-0.78; P < .001) surgery (Table 5).

Discussion

We linked VASQIP, a reliable, nurse-adjudicated source of clinical and surgical data, to multiple SDOH variables, obtaining more complete data for each case. Each data source was scrutinized to be the VA’s most reliable data source for every exposure and outcome. In contrast to prior work that suggested a survival benefit among veterans identified as Black,²² we found that the DOOR composite outcome measure demonstrated increased odds of worse outcomes among veterans identified as Black.

These findings suggest that DOOR, as a more comprehensive, granular measure of surgical outcomes, more accurately captures complex SDOH associations. Other studies have also found an increase in statistical power with DOOR compared with other outcome measures.¹⁴ We recommend using DOOR in site-level and national-level analyses of surgical data.

The contrast between previous 30-day mortality and DOOR findings suggests 2 potential causes: (1) Black veterans could have a greater comorbidity burden^53,54 that is adequately controlled for when predicting risk of 30-day mortality, but not complications and postoperative health care encounters, or (2) mortality for Black veterans could occur after 30 days vs White veterans, as DOOR uses 90-day mortality.⁵⁵ Explanation 1 suggests a potential conflict between care-quality metrics and patient-centered outcomes, which may have different goals. For quality-of-care metrics, the higher comorbidity burden of Black patients is relevant and needs statistical adjustment to avoid unfairly penalizing facilities serving higher proportions of patients of racial and ethnic minority groups.^56,57,58 However, alleviating these disparities requires metrics to target reducing this greater comorbidity burden. Explanation 2 requires analyzing longer-term outcomes than typically captured by surgical studies.^8,10,59,60 While the discrepancy between racial disparities in 30-day mortality vs DOOR requires further investigation, we nonetheless suggest focusing resources on preventive care to reduce presentation acuity among veterans identified as Black, as acuity seems to be a key driver in outcome disparities for Black veterans.

Veterans with larger proportions of non-VA care had worse outcomes, and private-sector surgery has worse mortality than the VA.⁴³ Patients and policymakers should reconsider whether the Maintaining Internal Systems and Strengthening Integrated Outside Networks (MISSION) Act,⁶¹ which aims to facilitate veteran access to private-sector health care, may unintentionally drive care fragmentation and worsen outcomes. Veterans might seek non-VA care for 2 reasons: (1) personal preference, or (2) lack of adequate care resources at their local VA (ie, long wait times or referrals for specialty services not available within the VA). Quality improvement initiatives might consider examining causes of greater non-VA care among veterans, incentivizing veterans to receive care within the VA where possible, and creating support programs to better coordinate veterans’ treatment and EHR data to and from the private sector. When adjustment for presentation acuity (PASC and case status) attenuates the association of care fragmentation, we consider this suggestive evidence that preventive care to reduce presentation acuity might partially alleviate care fragmentation–associated disparities.

We also found reduced risk among rural veterans and evidence to suggest that reduced presentation acuity is critically associated with this risk reduction. As those in rural areas tend to be older and have fewer resources⁶² and as private-sector studies have found higher surgical risk among people in rural areas,^63,64 these findings are likely limited to VA settings. Improved outcomes for veterans living in rural areas may reflect the tendency for veterans’ urgent or emergent surgical procedures to occur outside the VA or the effectiveness of the VA’s Office of Rural Health in alleviating rurality-based disparities.⁶⁵ For the private sector, health care systems may want to adapt rurality-focused programs from the VA to alleviate regional disparities in health care access^66,67 and coverage.⁶⁸ For the VA, data collection on veteran urgent and emergent surgical procedures in the private sector may be necessary to accurately capture outcome disparities. As rural veterans are more likely to seek private-sector care,⁶⁹ there may be trade-offs between getting rural veterans the care they need and preventing worsened outcomes from fragmented care. To suggest potential solutions, we speculate that telehealth appointments may help rural veterans to access needed care while maintaining care continuity through the VA.

Veterans living in highly deprived neighborhoods had similar DOOR outcomes vs less deprived neighborhoods. This contrasts with data from the private sector, where higher ADI is associated with worse outcomes,^12,14,15 and suggests that the VA’s care model provides more resources (eg, appointment transportation, comprehensive pharmacy management) for care access than the private sector. However, the association of deprived neighborhoods and acute presentation suggests that the VA should target reducing presentation acuity among veterans from highly deprived neighborhoods and collect data on veteran urgent or emergent surgical procedures in the private sector. Private-sector systems and insurers may wish to consider implementing some of the VA’s strategies to mitigate the association between area deprivation and presentation acuity.

Limitations

While these findings yield critical insight within the VA context, they are limited to VA surgical procedures. We excluded patients younger than 65 years to ensure adequate capture of non-VA health care encounters. Findings may be different in nonveteran cohorts, younger cohorts, or largely female cohorts. Rural or highly rural veterans with urgent or emergent surgical needs might be largely treated in nearby private-sector hospitals, therefore not appearing in VASQIP. Furthermore, our sample had relatively few Hispanic patients, so we do not wish to overinterpret the similar DOOR outcomes between White and Hispanic patients. This is a retrospective study, limited to measuring associations rather than causal effects. The DOOR outcome has not been externally validated among large groups of surgeons or patients; this could be the focus of future work. However, the rankings here are based on decades of experience caring for and communicating with patients. Some complications are heavily specific to certain surgical procedures, such as complications in procedure-targeted versions of American College of Surgeons NSQIP data, and are not included in VASQIP; these complications could be added into alternative definitions of DOOR.

Conclusions

We adapted a surgical DOOR method for use in VA surgical data. Combining VASQIP data with various other data sources demonstrated worse surgical outcomes among veterans identified as Black and with greater proportions of non-VA care. To address these disparities, we recommend directing resources to reduce presentation acuity among Black veterans, incentivizing veterans to receive care within the VA where possible, and creating support programs to better coordinate veterans’ treatment and records between health care professionals. As rural veterans are more likely to seek care outside the VA, there may be trade-offs between helping rural veterans access the care they need and preventing fragmented care from worsening outcomes. To suggest potential solutions, we speculate that telehealth appointments may succeed at allowing rural veterans to access needed care while maintaining care continuity through the VA.

Supplement 1.

eTable 1. Veterans Affairs Surgical Quality Improvement Program Variables Used for Preoperative Acute Serious Conditions (PASC)

eTable 2. Most Common Procedures for Each Expanded Operative Stress Score

eTable 3. Descriptive Statistics Stratified by Area Deprivation Index Groups

eTable 4. Descriptive Statistics Stratified by Race/Ethnicity Groups

eTable 5. Descriptive Statistics Stratified by Rurality Groups

eFigure. Flow Diagram With Cohort Exclusions

Supplement 2.

Data Sharing Statement