Discussion

We show that higher SVI and SDI levels are associated with lower guideline concordance in US EDs among older adults, driven by increased antibiotics underuse, partially offset by modest reductions in overuse. Our findings suggest that structural disadvantage thus shifts the balance between indicated and discretionary use, increasing undertreatment while tempering overuse.^{Reference Al Mohajer, Slusky, Nix and Nicodemo11} While the descriptive quintile trends showed only modest differences, the IV estimates clarified that deprivation causally increased underuse while slightly reducing overuse, underscoring how adjusted models reveal a stronger inequity signal than raw patterns alone.

In contrast, temporal trends demonstrated that concordance rose between 2015 and 2020 and remained high through 2024, but this pattern reflected declining underuse alongside rising overuse. Much of the latter was linked to empiric antibiotic use during the early COVID-19 pandemic and the frequent absence of a plausible coded indication.^{Reference Ladines-Lim, Fischer, Linder and Chua4} COVID-19 accounted for a large share of overuse, while UTI diagnoses contributed most to underuse; however, diagnosis codes may include presentations later determined to be asymptomatic bacteriuria or noninfectious syndromes, which could inflate apparent underuse. These trends mirror prior national reports of rising empiric use and persistent diagnostic uncertainty in older adults, especially for urinary and respiratory syndromes, where appropriateness is often difficult to adjudicate.^{Reference Ladines-Lim, Fischer, Linder and Chua4}

Consistent with prior literature, disadvantaged populations are more likely to receive non-concordant antibiotic prescribing.^{Reference Al Mohajer, Slusky, Nix and Nicodemo11–Reference Young, Strey and Lee13,Reference Kissler, Klevens, Barnett and Grad35} Deprivation-related constraints—such as delayed presentation and reduced outpatient continuity and follow-up capacity—may increase diagnostic uncertainty and shape clinicians’ thresholds for initiating indicated therapy.^{Reference Butler, Petterson, Phillips and Bazemore17,Reference Mullachery, Li and Melly19,Reference Kissler, Klevens, Barnett and Grad35} The same constraints may limit discretionary prescribing when antibiotics are not indicated, reducing overuse.^{Reference Linder10,Reference Al Mohajer, Slusky, Nix and Nicodemo11,Reference Kissler, Klevens, Barnett and Grad35} For instance, nonemployment, household overcrowding, and limited infrastructure showed the strongest associations with lower concordance. This points to mechanisms tied to transportation, household constraints, and instability of work and caregiving arrangements. Prior work links area deprivation to differences in access and utilization; our results extend that literature by relating specific components to tier-defined prescribing outcomes in the ED.^{14,Reference Butler, Petterson, Phillips and Bazemore17,Reference Morenz, Liao, Au and Hayes20,Reference Kissler, Klevens, Barnett and Grad35}

After adjusting for deprivation and facility characteristics, antibiotic overuse was more common among older adults who were non-white, unmarried, and insured through non-Medicaid plans, with higher rates also observed in larger, academic, and rural EDs. This pattern is consistent with lower treatment thresholds in settings or patient groups where clinicians perceive greater risk of clinical deterioration and weaker safety nets for follow-up or continuity of care.^{Reference Ladines-Lim, Fischer, Linder and Chua4,Reference Linder10,Reference Palms, Hicks and Bartoces36} For race and ethnicity, higher concordance was observed among Black and other racial/ethnic groups, reflecting markedly lower underuse despite modestly higher overuse. Several ambulatory and ED studies^{Reference Seibert, Hersh and Patel12,Reference Young, Strey and Lee13,Reference Goyal, Johnson and Chamberlain37} have reported lower overall antibiotic use among Black and Hispanic patients, although limited in incorporating causal inference for area-level socioeconomic conditions. In an older ED population with atypical presentations, clinicians anchored on physiologic instability or anticipated barriers to return care, lowering the threshold to “cover” with antibiotics.^{Reference Ladines-Lim, Fischer, Linder and Chua4,Reference Linder10} Rural EDs, compared with metropolitan EDs, may face constrained access to imaging, laboratory turnaround, and next-day clinics, which can push toward empiric treatment in low-probability scenarios and help explain the rural-to-urban overuse gradient we observed. In contrast, longer patient travel time was associated with higher underuse and lower overuse, suggesting that patient-level access constraints may operate differently than facility rurality and may reflect referral patterns or delays in early antibiotic initiation.^{Reference Mullachery, Li and Melly19,Reference Kissler, Klevens, Barnett and Grad35,Reference Palms, Hicks and Bartoces36}

Predicted SVI and SDI were spatially clustered, as was concordance, with high-disadvantage/low-concordance clusters in the South and low-disadvantage/high-concordance clusters in the Midwest. These patterns are consistent with broader regional differences, where the South has higher poverty, more rural hospital closures, and weaker public health investment, while the Midwest has greater primary care availability and stronger continuity of services.^{Reference Butler, Petterson, Phillips and Bazemore17,Reference Mullachery, Li and Melly19,Reference Van Dyke21,Reference Kissler, Klevens, Barnett and Grad35} Such contrasts help explain why prescribing diverges regionally and highlight the importance of targeting stewardship to the local health system context. The IV results are consistent with a potential causal interpretation under IV assumptions. Maximum state EITC exhibited strong first-stage associations with both SVI and SDI, and the IV estimates for underuse and overuse were directionally consistent with OLS but larger in magnitude, which is expected if unmeasured features of access and practice culture bias conventional estimates toward the null. As a policy-determined exposure that varies across states and time and does not plausibly affect a clinician’s antibiotic choice except through community deprivation when state and temporal effects are modeled, EITC meets standard relevance and exclusion criteria used in the literature.^{Reference Al Mohajer, Slusky, Nix and Nicodemo11,Reference Angrist and Pischke22,Reference Anselin32,Reference Larrimore38,Reference Hoynes and Patel39} Conversely, the high F-statistics, agreement across specifications, and coherence with the component analysis increase confidence in the direction and structure of the associations. These findings suggest ED stewardship could monitor underuse and overuse separately and stratify performance by patient and neighborhood factors to identify inequities. Because tiering relies on diagnosis codes, any metric should be paired with audit safeguards against coding drift. The tier-by-prescribing framework lends itself to automated dashboards and iterative decision support linked to orders/symptoms, and spatial clustering can guide where to prioritize interventions.

The study has limitations, including the potential misclassification from diagnosis coding and tier assignment, using a single infectious disease specialist, unobserved clinical considerations not captured in the EHR, and measurement error in geocoding and travel time. SVI was originally developed for disaster-management vulnerability^{Reference Flanagan, Gregory, Hallisey, Heitgerd and Lewis16} and may not fully capture deprivation constructs most relevant to healthcare delivery, introducing measurement error when applied to prescribing outcomes. Complete-case IV estimates were generally less precise and sometimes attenuated relative to imputed models, consistent with reduced sample size and potential selection bias due to missing covariates. Outcomes reflect antibiotics administered within 24 hours of ED presentation; for admitted patients, some early antibiotic decisions may reflect inpatient rather than ED teams. ICD-coded UTIs may include asymptomatic bacteriuria or noninfectious syndromes, potentially overstating underuse. Because underuse and overuse were uncommon (∼7–10%), IV estimates may appear large and should be interpreted cautiously. Moreover, the study setting may limit generalizability beyond participating systems and regions.

Despite these limitations, the study draws on a large, multi-state, encounter-level EHR dataset; applies a tier-based framework that separates underuse from overuse; examines both SVI and SDI, including SDI components; uses an IV design with strong first-stage performance; and incorporates spatial methods that move beyond individual-level associations to identify tract-level clustering relevant for implementation. In summary, area-level disadvantage was linked to greater underuse and lower overuse, with clear geographic clustering, while patient- and facility-level factors showed the opposite tendency, with non-White, non-married, and higher-acuity patients, and larger, academic, and rural EDs more often receiving empiric therapy. Stewardship that addresses both sides of error, targets disadvantaged communities, and builds safeguards in settings prone to overuse is most likely to improve outcomes and advance equity.