Hasil untuk "Insurance"

Caroline Hillairet, Olivier Lopez, Lionel Sopgoui

This work evaluates the impact of contagious cyber-events, over a finite horizon, on firms' financial health and on a cyber insurance portfolio. Our approach builds on key empirical findings from economics and cybersecurity. In economics, firm size and growth-rate distributions are non-Gaussian and exhibit heavy tails. In cybersecurity, contagion dynamics strongly depend on firm size and environmental conditions. To capture these features, we propose a stochastic multi-group SIR model coupled with a granular model of firm growth. This framework allows us to quantify the financial impact of cyber-attacks on firms' revenues and on the insurer's portfolio. In the model, the arrival time and duration of cyber-attacks are driven by a combination of a Cox process and a Bernoulli random variable. The Cox process represents external contagion, with an intensity given by the force of infection derived from the stochastic SIR dynamics. The Bernoulli component captures contagion originating from an infected sister or subsidiary firm. Environmental variability enables stochastic scenario generation and the computation of aggregate exceedance probabilities, a standard metric in catastrophe modeling that provides insurers with immediate insight into the financial severity of an event. We apply the framework to the LockBit ransomware attacks observed between May and July 2024. For a portfolio of 2,929 firms located in Ile-de-France, the model predicts that, with 50% probability, the insurer will need to compensate losses equivalent to up to two days of revenue over a 100-day cyber incident.

Teerapong Panboonyuen

Automated property risk detection is a high-impact yet underexplored frontier in computer vision with direct implications for real estate, underwriting, and insurance operations. We introduce HOMEY (Heuristic Object Masking with Enhanced YOLO), a novel detection framework that combines YOLO with a domain-specific masking mechanism and a custom-designed loss function. HOMEY is trained to detect 17 risk-related property classes, including structural damages (e.g., cracked foundations, roof issues), maintenance neglect (e.g., dead yards, overgrown bushes), and liability hazards (e.g., falling gutters, garbage, hazard signs). Our approach introduces heuristic object masking to amplify weak signals in cluttered backgrounds and risk-aware loss calibration to balance class skew and severity weighting. Experiments on real-world property imagery demonstrate that HOMEY achieves superior detection accuracy and reliability compared to baseline YOLO models, while retaining fast inference. Beyond detection, HOMEY enables interpretable and cost-efficient risk analysis, laying the foundation for scalable AI-driven property insurance workflows.

Jodie A. Dionne, Ashutosh Tamhane, Taylor Golden et al.

Abstract Background Congenital syphilis rates in the United States have increased significantly over the past decade. Syphilis is a curable infection with the potential for lifelong sequelae in the absence of timely diagnosis and treatment. Routine serologic syphilis screening is universally recommended during prenatal care with the traditional or the reverse diagnostic algorithm. False positive syphilis serologic testing in pregnancy can occur and comparative performance data for recommended algorithms in pregnancy are limited. Primary objective To compare the performance of the traditional algorithm and the reverse algorithm for the diagnosis of syphilis in pregnancy. Study design This retrospective analysis included pregnant women who delivered at our tertiary care center in the Southeastern United States during a period of increasing syphilis rates with testing performed between November 1, 2012 and December 31, 2019. We evaluated results according to the diagnostic algorithm used by facility laboratories at the time of syphilis screening (traditional 2012–2014 and reverse 2015–2019). Screen positivity, false positive test results, confirmed infection, and pregnancy outcomes were compared between the two periods. For secondary outcomes, multivariable logistic regression models were conducted to identify factors associated with false positive screening results and confirmed infection including maternal age, race, insurance status, test timing and location, and human immunodeficiency virus/sexually transmitted infection coinfection. Results Of 26,519 pregnant women tested for syphilis during the study period, 8781 were evaluated using the traditional algorithm and 17,738 were evaluated using the reverse algorithm. Mean age was 27.9 years, 85.3% of women were initially screened in the first trimester, and the mean number of syphilis testing episodes in pregnancy was 2.3. Screen positivity was 0.6% among women screened using the traditional algorithm compared to 1.6% for those tested with the reverse algorithm (p < 0.001). The proportion diagnosed with confirmed infection was similar in both algorithms: 0.2% traditional algorithm versus 0.3% reverse algorithm. Among those who screened positive with follow‐up testing performed, 52.4% and 60.4% were classified as falsely positive with negative confirmatory testing with the traditional algorithm and the reverse algorithms, respectively. In an adjusted model, delayed testing in pregnancy (odds ratio [OR], 22.8; 95% confidence interval [CI], 14.1–36.8 for ≥28 weeks compared to <14 weeks), inpatient or ER screening location (OR, 22.7; 95% CI, 13.4–38.4 vs. clinic), Black race (OR, 5.4; 95% CI, 3.2–9.2) compared to White, other sexually transmitted infection in pregnancy (OR, 2.2; 95% CI, 1.2–4.1]), and lack of private insurance (OR, 1.8; 95% CI, 1.2–2.9) were associated with false positive syphilis screening. The same factors were associated with confirmed syphilis in pregnancy except for STI coinfection. The reverse algorithm was only associated with false positive screening in the crude model (OR, 2.2; 95% CI, 1.4–3.5). Conclusion Syphilis screen positivity rates in pregnancy were nearly twice as high with the reverse algorithm compared to the traditional algorithm. Since false positive screening tests were common, improved diagnostic testing for active infection in pregnancy is needed.

Ziheng Geng, Jiachen Liu, Ran Cao et al.

Flood insurance is an effective strategy for individuals to mitigate disaster-related losses. However, participation rates among at-risk populations in the United States remain strikingly low. This gap underscores the need to understand and model the behavioral mechanisms underlying insurance decisions. Large language models (LLMs) have recently exhibited human-like intelligence across wide-ranging tasks, offering promising tools for simulating human decision-making. This study constructs a benchmark dataset to capture insurance purchase probabilities across factors. Using this dataset, the capacity of LLMs is evaluated: while LLMs exhibit a qualitative understanding of factors, they fall short in estimating quantitative probabilities. To address this limitation, InsurAgent, an LLM-empowered agent comprising five modules including perception, retrieval, reasoning, action, and memory, is proposed. The retrieval module leverages retrieval-augmented generation (RAG) to ground decisions in empirical survey data, achieving accurate estimation of marginal and bivariate probabilities. The reasoning module leverages LLM common sense to extrapolate beyond survey data, capturing contextual information that is intractable for traditional models. The memory module supports the simulation of temporal decision evolutions, illustrated through a roller coaster life trajectory. Overall, InsurAgent provides a valuable tool for behavioral modeling and policy analysis.

Tim J. Boonen, Xinyue Fan, Zixiao Quan

Machine learning improves predictive accuracy in insurance pricing but exacerbates trade-offs between competing fairness criteria across different discrimination measures, challenging regulators and insurers to reconcile profitability with equitable outcomes. While existing fairness-aware models offer partial solutions under GLM and XGBoost estimation methods, they remain constrained by single-objective optimization, failing to holistically navigate a conflicting landscape of accuracy, group fairness, individual fairness, and counterfactual fairness. To address this, we propose a novel multi-objective optimization framework that jointly optimizes all four criteria via the Non-dominated Sorting Genetic Algorithm II (NSGA-II), generating a diverse Pareto front of trade-off solutions. We use a specific selection mechanism to extract a premium on this front. Our results show that XGBoost outperforms GLM in accuracy but amplifies fairness disparities; the Orthogonal model excels in group fairness, while Synthetic Control leads in individual and counterfactual fairness. Our method consistently achieves a balanced compromise, outperforming single-model approaches.

Adam Bradley, John Hastings, Khandaker Mamun Ahmed

The insurance industry is undergoing a paradigm shift through the adoption of artificial intelligence (AI) technologies, particularly in the realm of intelligent conversational agents. Chatbots have evolved into sophisticated AI-driven systems capable of automating complex workflows, including policy recommendation and claims triage, while simultaneously enabling dynamic, context-aware user engagement. This paper presents the design, implementation, and empirical evaluation of Axlerod, an AI-powered conversational interface designed to improve the operational efficiency of independent insurance agents. Leveraging natural language processing (NLP), retrieval-augmented generation (RAG), and domain-specific knowledge integration, Axlerod demonstrates robust capabilities in parsing user intent, accessing structured policy databases, and delivering real-time, contextually relevant responses. Experimental results underscore Axlerod's effectiveness, achieving an overall accuracy of 93.18% in policy retrieval tasks while reducing the average search time by 2.42 seconds. This work contributes to the growing body of research on enterprise-grade AI applications in insurtech, with a particular focus on agent-assistive rather than consumer-facing architectures.

Nur Azizah, Rida Perwita Sari

The insurance industry has developed rapidly enough to cause intense competition between companies. In competition, it is necessary to have good work prospects and public trust, one of which is having a healthy financial performance. Financial performance can be influenced by several factors, namely risk-based capital and claim ratio. The study analyzed whether risk-based capital and claim ratios affect financial performance. The population of this study is insurance companies listed on the Indonesia Stock Exchange (IDX) in 2019-2023. This research uses quantitative methods with purposive sampling methods with a total sample of 15 companies over 5 5-year period. The data used in this study are secondary data obtained through the company’s financial statements and annual reports. The results of this study indicate that risk-based capital affects financial performance. Meanwhile, the claim ratio does not affect financial performance.

Vladyslav Shevchuk

Мета роботи. Розробити оптимізаційну математичну модель максимізації значення інтегрального індексу безпеки державного кордону. Метод дослідження. В основу моделі покладено інтегральний індекс безпеки державного кордону (ІІБ) як цільову функцію, яку необхідно максимізувати. Змінними в моделі виступають не самі показники, а керуючі впливи або ресурси, що спрямовуються на покращення цих показників. Результати дослідження. Розроблено математичну модель максимізації значення інтегрального індексу безпеки державного кордону та реалізовано її у середовищі Matlab. Теоретична цінність дослідження. Теоретично методика збагачує область математичного моделювання в системах безпеки, демонструючи, як оціночні індекси можна трансформувати в оптимізуючі задачі. Оригінальність/Цінність дослідження. Новизна полягає у інтеграції оціночної моделі та агрегації цільової функції до специфічного індексу IIБ з елементами математичного програмування для оптимального розподілу ресурсів у контексті безпеки державного кордону. Майбутні дослідження. У ході подальших досліджень доцільно опрацювати вхідні дані та розробити програмний продукт для комп’ютерної підтримки. Тип статті. Теоретична.

Jared W Magnani, Sachin Yende, Seyed Mehdi Nouraie et al.

Objectives Racial disparities in critical illness outcomes are well-described, with social determinants of health as likely contributors. We sought to identify inequalities in readmissions and mortality between black and white patients among survivors of critical illness with sepsis and assess whether these disparities were explained by neighbourhood characteristics, health insurance and hospital quality.Design Retrospective cohort study examining 90-day and 9-month readmissions and survival as coprimary outcomes. Models included age, sex, race and area deprivation index (ADI), Medicaid status or hospital Centers for Medicare &amp; Medicaid Star rating. Accelerated failure time and Cox proportional hazards models with subgroup analyses by age and surgical status were employed.Setting 14 community and tertiary hospitals in Western Pennsylvania.Participants 48 027 survivors of sepsis with critical illness; 20 952 (50.4%) male; 6489 (13.5%) identified as black.Results Black patients were younger (mean age 59.0 years vs 65.8 years), more likely to have higher ADI, Medicaid insurance and receive care at lower-quality hospitals. Black patients had higher readmission risk: (90-day subdistribution HR (SDHR) 1.13 (95% CI 1.04 to 1.23); p=0.003); 9-month SDHR: 1.11 (95% CI 1.03 to 1.20); p=0.005). Adjusting for age and sex, we found no difference in 90-day and 9-month mortality (90-day acceleration factor (AF): 1.04 (95% CI 0.91 to 1.19); p=0.556; 9-month: 1.08 (95% CI 0.96 to 1.22); p=0.196), which remained consistent when including ADI, Medicaid status or hospital quality. Mortality among black patients was increased relative to white patients among patients ≥60 years (9-month AF 1.23 (95% CI 1.07 to 1.42; p=0.004)) and among surgical patients (90-day AF: 1.23 (95% CI 1.01 to 1.50; p=0.04); 9-month AF: 1.28 (95% CI 1.07 to 1.53; p=0.006)). Medicaid status, but not ADI or hospital quality, attenuated racial differences in subgroup mortality.Conclusions In a retrospective analysis of intensive care unit (ICU) survivors with sepsis, black patients had higher readmission rates but comparable mortality to white patients, except among older and surgical subgroups. Medicaid status influenced racial inequalities in mortality, highlighting a need for targeted post-ICU interventions.

Vahé Heboyan, Phillip Coule, Davide Mariotti et al.

Background: The emergency department (ED) provides a significant portion of health care services in the United States, and its utilization has increased over the past decade. ED overcrowding remains a considerable challenge to many EDs. The objectives of this study were (1) to evaluate the knowledge of telehealth and readiness to use it among patients who visit EDs in a nonurgent triage category and (2) to estimate their willingness-to-use and willingness-to-pay for telehealth consultations. Methods: A structured questionnaire was administered using a tablet to adult patients who visited the ED of a large medical center and who were triaged into a nonurgent category. Respondents were asked about their sociodemographic and ED visit characteristics and health and telehealth utilization history. Then, we presented them with a hypothetical scenario for visiting a board-certified ED doctor through telehealth instead of in-person visits, and, using a double-bound dichotomous choice iterative bidding algorithm, we solicited their willingness-to-pay for such a telehealth visit. Results: A total of 171 patients agreed to participate in the study. More than half of the respondents (n = 107; 62.6%) said they have health insurance. Almost half of the respondents (n = 71; 41.5%) reported the main reason for going to the ED was an ongoing condition or concern. More than two-thirds of the respondents identified themselves as being very proficient with using a smartphone or tablet (n = 116; 67.8%), and only a few (n = 21; 12.3%) reported not having any internet-capable device. Most respondents (n = 148; 86.5%) had never heard about telehealth. However, after a brief description of telehealth, we found that approximately two-thirds of the patients would be willing to use or consider using telehealth (n = 107; 62.6%), and one-third (n = 64; 37.4%) would not be interested. We did not observe any statistically significant differences in willingness-to-use. However, we observed statistically significant differences in the willingness-to-pay $50 by gender (p < 0.01), by currently having a regular doctor/clinic (p < 0.05), and by health insurance status. Conclusions: Hospitals should consider investigating telehealth services that can be provided to their communities as an option instead of visiting their EDs. While technology does not seem to be a barrier to telehealth, more educational initiatives to inform the public about telehealth are desirable. A targeted advertisement campaign to recommend telehealth for nonlife-threatening ED visits could be developed once more user characteristics are collected.

Serguei Maliar, Bernard Salanie

The positive correlation test for asymmetric information developed by Chiappori and Salanie (2000) has been applied in many insurance markets. Most of the literature focuses on the special case of constant correlation; it also relies on restrictive parametric specifications for the choice of coverage and the occurrence of claims. We relax these restrictions by estimating conditional covariances and correlations using deep learning methods. We test the positive correlation property by using the intersection test of Chernozhukov, Lee, and Rosen (2013) and the "sorted groups" test of Chernozhukov, Demirer, Duflo, and Fernandez-Val (2023). Our results confirm earlier findings that the correlation between risk and coverage is small. Random forests and gradient boosting trees produce similar results to neural networks.

Felix Fießinger, Mitja Stadje

This paper studies the equity holders' mean-variance optimal portfolio choice problem for (non-)protected participating life insurance contracts. We derive explicit formulas for the optimal terminal wealth and the optimal strategy in the multi-dimensional Black-Scholes model, showing the existence of all necessary parameters. In incomplete markets, we state Hamilton-Jacobi-Bellman equations for the value function. Moreover, we provide a numerical analysis of the Black-Scholes market. The equity holders on average increase their investment into the risky asset in bad economic states and decrease their investment over time.

Alex Zarifis, Peter Kawalek, Aida Azadegan

Trust and privacy have emerged as significant concerns in online transactions. Sharing information on health is especially sensitive but it is necessary for purchasing and utilizing health insurance. Evidence shows that consumers are increasingly comfortable with technology in place of humans, but the expanding use of AI potentially changes this. This research explores whether trust and privacy concern are barriers to the adoption of AI in health insurance. Two scenarios are compared: The first scenario has limited AI that is not in the interface and its presence is not explicitly revealed to the consumer. In the second scenario there is an AI interface and AI evaluation, and this is explicitly revealed to the consumer. The two scenarios were modeled and compared using SEM PLS-MGA. The findings show that trust is significantly lower in the second scenario where AI is visible. Privacy concerns are higher with AI but the difference is not statistically significant within the model.

Qi Zhang, Ruili Wang, Lei Chen et al.

Background: Launched in March 2019, the National Centralized Drug Procurement (NCDP) initiative aimed to optimize the drug utilization framework in public healthcare facilities. Following the integration of Non-Vitamin K Antagonist Oral Anticoagulants (NOACs) into the procurement catalog, healthcare establishments in Suining swiftly transitioned to the widespread adoption of NOACs, beginning 1 March 2020.Objective: This study aims to comprehensively assess the impact of the NCDP policy on the efficacy of anticoagulation therapy, patient medication adherence, and the incidence of hemorrhagic events in individuals with non-valvular atrial fibrillation (NVAF) residing in Suining. The analysis seeks to elucidate the broader impacts of the NCDP policy on this patient demographic.Methods: This study analyzed patient hospitalization records from the Department of Cardiology at Suining County People’s Hospital, spanning 1 January 2017, to 30 June 2022. The dataset included demographic details (age, sex), type of health insurance, year of admission, hospitalization expenses, and comprehensive information on anticoagulant therapy utilization. The CHA2DS2-VASc scoring system, an established risk assessment tool, was used to evaluate stroke risk in NVAF patients. Patients with a CHA2DS2-VASc score of 2 or higher were categorized as high-risk, while those with scores below 2 were considered medium or low-risk.Results: 1. Treatment Cost Analysis: The study included 3,986 patients diagnosed with NVAF. Following the implementation of the NCDP policy, a significant increase in the average treatment cost for hospitalized patients was observed, rising from 8,900.57 ± 9,023.02 CNY to 9,829.99 ± 10,886.87 CNY (p &lt; 0.001). 2. Oral Anticoagulant Utilization: Overall, oral anticoagulant use increased from 40.02% to 61.33% post-NCDP (p &lt; 0.001). Specifically, NOAC utilization among patients dramatically rose from 15.41% to 90.99% (p &lt; 0.001). 3. Hemorrhagic Events: There was a significant decrease in hemorrhagic events following the NCDP policy, from 1.88% to 0.66% (p = 0.01). Hypertension [OR = 1.979, 95% CI (1.132, 3.462), p = 0.017], history of stroke [OR = 1.375, 95% CI (1.023, 1.847), p = 0.035], age ≥65 years [OR = 0.339, 95% CI (0.188, 0.612), p &lt; 0.001], combination therapy of anticoagulants and antiplatelets [OR = 3.620, 95% CI (1.752, 7.480), p &lt; 0.001], hepatic and renal insufficiency [OR = 4.294, 95% CI (2.28, 8.084), p &lt; 0.001], and the NCDP policy [OR = 0.295, 95% CI (0.115, 0.753), p = 0.011] are significant risk factors for bleeding in patients with atrial fibrillation. 4. Re-hospitalization and Anticoagulant Use: Among the 219 patients requiring re-hospitalization, there was a notable increase in anticoagulant usage post-NCDP, from 36.07% to 59.82% (p &lt; 0.001). NOACs, in particular, saw a substantial rise in usage among these patients, from 11.39% to 80.92% (p &lt; 0.001). 5. Anticoagulant Type Change: The NCDP policy [OR = 28.223, 95% CI (13.148, 60.585), p &lt; 0.001] and bleeding events [OR = 27.772, 95% CI (3.213, 240.026), p = 0.003] were significant factors influencing the alteration of anticoagulant medications in patients.Conclusion: The NCDP policy has markedly improved anticoagulation management in patients with AF. This policy has played a crucial role in enhancing medication adherence and significantly reducing the incidence of hemorrhagic events among these patients. Additionally, the NCDP policy has proven to be a key factor in guiding the selection and modification of anticoagulant therapies in the AF patient population.

Oytun Haçarız, Torsten Kleinow, Angus S. Macdonald

We revisit surplus on general life insurance contracts, represented by Markov models. We classify technical bases in terms of boundary conditions in Thiele's equation(s), allowing more general regulations than Scandinavian-style `first-order/second-order' regimes, and replacing the traditional retrospective policy value. We propose a `canonical' model with three technical bases (premium, valuation, accumulation) and show how each pair of bases defines premium loadings and surplus. Along with a `true' or `real-world' experience basis, this expands fundamental results of Ramlau-Hansen (1988a). We conclude with two applications: lapse-supported business; and the retrospectively-oriented regime proposed by Møller & Steffensen (2007).

Freek Holvoet, Katrien Antonio, Roel Henckaerts

Insurers usually turn to generalized linear models for modeling claim frequency and severity data. Due to their success in other fields, machine learning techniques are gaining popularity within the actuarial toolbox. Our paper contributes to the literature on frequency-severity insurance pricing with machine learning via deep learning structures. We present a benchmark study on four insurance data sets with frequency and severity targets in the presence of multiple types of input features. We compare in detail the performance of: a generalized linear model on binned input data, a gradient-boosted tree model, a feed-forward neural network (FFNN), and the combined actuarial neural network (CANN). The CANNs combine a baseline prediction established with a GLM and GBM, respectively, with a neural network correction. We explain the data preprocessing steps with specific focus on the multiple types of input features typically present in tabular insurance data sets, such as postal codes, numeric and categorical covariates. Autoencoders are used to embed the categorical variables into the neural network, and we explore their potential advantages in a frequency-severity setting. Model performance is evaluated not only on out-of-sample deviance but also using statistical and calibration performance criteria and managerial tools to get more nuanced insights. Finally, we construct global surrogate models for the neural nets' frequency and severity models. These surrogates enable the translation of the essential insights captured by the FFNNs or CANNs to GLMs. As such, a technical tariff table results that can easily be deployed in practice.

Ayse Arik, Andrew J. G. Cairns, Erengul Dodd et al.

In this paper we consider pricing of insurance contracts for breast cancer risk based on three multiple state models. Using population data in England and data from the medical literature, we calibrate a collection of semi-Markov and Markov models. Considering an industry-based Markov model as a baseline model, we demonstrate the strengths of a more detailed model while showing the importance of accounting for duration dependence in transition rates. We quantify age-specific cancer incidence and cancer survival by stage along with type-specific mortality rates based on the semi-Markov model which accounts for unobserved breast cancer cases and progression through breast cancer stages. Using the developed models, we obtain actuarial net premiums for a specialised critical illness and life insurance product. Our analysis shows that the semi-Markov model leads to results aligned with empirical evidence. Our findings point out the importance of accounting for the time spent with diagnosed or undiagnosed pre-metastatic breast cancer in actuarial applications.

Lei Liu, Zheng Pei, Peng Chen et al.

Abstract Deep learning has achieved tremendous success in various applications owing to its robust feature representations of complex high-dimensional nonlinear data. Financial time-series prediction is no exception. Hence, the volatility trend prediction in financial time series (FTS) has been an active topic for several decades. Inspired by generative adversarial networks (GAN), which have been studied extensively in image processing and have achieved excellent results, we present the ordinal regression GAN for financial volatility trends (ORGAN-FVT) method for the end-to-end multi-classification task of FTS. An improved generative model based on convolutional long short-term memory (ConvLSTM) and multilayer perceptron (MLP) is proposed to capture temporal features effectively and mine the data distribution of volatility trends (short, neutral, and long) from given FTS data. Meanwhile, ordinal regression is leveraged for the discriminator to improve the multi-classification performance, making the model more practical. Finally, we empirically compare ORGAN-FVT with several state-of-the-art approaches on three real-world stock datasets: MICROSOFT(MSFT), Tesla(TSLA), and The People’s Insurance Company of China(PAICC). ORGAN-FVT demonstrated significantly better AUC and F1 scores, at most 20.81% higher than its competitors.

Chakravarthy Varadarajan, Klaus R. Schenk-Hoppé

We explain the evolution of the volatility market and present the infamous day of ‘Volmageddon’ as an insightful case study. Our survey focuses on the pricing and trading of volatility-linked assets, highlighting the impact of mechanical hedging in markets for futures and higher-order derivatives. We supplement the vast statistical analysis of volatility derivatives with a financial economist’s perspective.

Shara Fuller, MD, Molly Kuenstler, MD, Marie Snipes, PhD et al.

BACKGROUND: Racial and ethnic disparities in health care exist and are rooted in long-standing systemic inequities. These disparities result in significant excess health care expenditures and are due to complex interactions between patients, health care providers and systems, and social and environmental factors. In perinatal care, these inequities also exist, with Black patients being 3 to 4 times more likely to die of childbirth compared with White patients. Similar health care inequities may also exist in the Military Health System despite universal health care coverage, stable employment, and social programs that benefit military families. OBJECTIVE: This study aimed to evaluate racial disparities in obstetrical outcomes in the Military Health System. STUDY DESIGN: This is a retrospective cohort study of deliveries from 2019 to 2021 in the Military Health System, which provides obstetrical care for approximately 35,000 annual deliveries. The study was conducted using National Perinatal Information Center data on cesarean delivery, postpartum hemorrhage, and severe maternal morbidity by race and ethnicity from direct-care military hospitals representing tertiary care medical centers and community hospitals in the United States and abroad. Chi-square analyses and binary logistic regression were used to compare groups. RESULTS: The cohort included 68,918 deliveries. Of these, 32,358 (47%) were White, 9594 (13.9%) Black, 3120 (4.5%) Asian Pacific Islander, 456 (0.7%) American Indian/Alaska Native, 19,543 (28.4%) other, 3976 (5.8%) unknown, 7096 (10.3%) Hispanic, 58,009 (84.2%) non-Hispanic, and 4399 (6.4%) other ethnicity. Rates of cesarean delivery were significantly higher for Black (30%; odds ratio, 1.44; 95% confidence interval, 1.37–1.52), Asian Pacific Islander (27%; odds ratio, 1.24; 95% confidence interval, 1.14–1.35), and other (26%; odds ratio, 1.20; 95% confidence interval, 1.15–1.25) compared with White race (23%) (P<.001). Postpartum hemorrhage rates were higher for Black (5.9%; odds ratio, 1.11; 95% confidence interval, 1.00–1.24) and Asian Pacific Islander (7.7%; odds ratio, 1.49; 95% confidence interval, 1.29–1.72) compared with White race (5.3%) (P<.001). Severe maternal morbidity was higher for Black (2.9%; odds ratio, 1.44; 95% confidence interval, 1.24–1.67), Asian Pacific Islander (2.9%; odds ratio, 1.45; 95% confidence interval, 1.15–1.82), and other (2.8%; odds ratio, 1.36; 95% confidence interval, 1.21–1.54) compared with White race (2.1%) (P<.001). For severe maternal morbidity excluding blood transfusions, rates were also significantly higher for Black (1%; odds ratio, 1.68; 95% confidence interval, 1.30–2.17) than for White race (0.6%) (P<.002). Hispanic ethnicity was associated with a lower rate of severe maternal morbidity excluding transfusions (0.5%; odds ratio, 0.68; 95% confidence interval, 0.48–0.98) compared with non-Hispanic ethnicity (0.7%) (P=.04). CONCLUSION: Racial disparities in obstetrical outcomes exist in the Military Health System despite universal health care coverage, with significantly higher rates of cesarean delivery and severe maternal morbidity in Black, Asian Pacific Islander, and other races compared with White race. These findings suggest that these disparities are likely related to other factors or social determinants of health rather than availability of health care and insurance coverage. Further work should include investigation into such social determinants of health to address their causes, including systemic and structural barriers.