The COVID-19 pandemic had disparate impacts on Black and Hispanic Americans, as observed in mortality and infection rates. The mental health impact of the pandemic is less clear, with little research exploring mental health disparities and differences by state and region. Using the Behavioral Risk Factor Surveillance System (BRFSS), state-level differences in unhealthy mental health days were examined between 2019 and 2021 for White, Black, and Hispanic respondents. Nonparametric tests assessed group differences in unhealthy mental health days, while state-level within- and between-group difference scores highlighted areas with the greatest disparities. Both White and Black respondents experienced significantly more unhealthy mental health days in 2021 than in 2019; though Hispanic respondents reported more unhealthy days in 2021 than in 2019, this finding was not statistically significant. Black respondents, but not Hispanic, reported significantly more unhealthy mental health days than White respondents in 2021. Missouri had the worst outcomes for Black respondents, with the greatest increase in unhealthy mental health days between 2019 and 2021 and the greatest difference between White and Black groups in 2021. Regionally, the Northeast had the best outcomes for Black respondents and the Midwest had the worst. These findings can help identify population groups and geographic areas most in need of disaster-preparedness efforts and policy interventions for future public health emergencies. Practitioners and state health officials can use these findings to identify potentially impactful community interventions, or to develop infrastructure for addressing community mental health.

Parental emotional socialization (PES) has been recognized as a critical mechanism in parenting programs to enhance children's well-being, especially following adversity. However, few studies have examined the potential moderating effect of baseline PES levels. This study aimed to examine whether supportive and unsupportive PES can mediate the intervention effects of a parenting program on child adjustment (Aim 1), and whether baseline PES can moderate the effect (Aim 2). This study utilized data from two randomized controlled trials for post-deployed military families (N = 335, Mage = 8.25, 54% girls). Families were either assigned to in-person intervention condition (n = 226) or a treatment-as-usual condition (n = 109). Baseline-targeted moderation mediation (BTMM) models were conducted for mothers and fathers separately, with child age, child sex, child minority status, family household income, and deployment length included as covariates. The parenting program had an indirect effect on child internalizing and externalizing problems 1-year post-baseline through reduced maternal unsupportive PES at post-intervention, while no indirect effect was found through supportive PES. Additionally, baseline PES moderated the impact of the ADAPT program on maternal supportive and unsupportive PES post-intervention, such that mothers who reported less supportive PES or more unsupportive PES at baseline benefited more. No intervention effect was found through fathers' PES. The findings underscore the crucial role of baseline PES in shaping behavioral parenting intervention effectiveness. The study highlights that one size does not fit all and future research and practice should consider the diverse needs and responses of families, emphasizing the delivery of personalized interventions to best meet parents' needs and maximize support.

Adverse childhood experiences (ACEs) are preventable, potentially traumatic events that can predict a wide range of short- and long-term negative health outcomes. More research is needed on policies that may reduce ACEs. Paid Family Leave (PFL) is a state-level policy that provides economic support for employees to care for a newborn or a recently adopted child. This study evaluated effects of PFL on reduction in household challenge ACEs. To investigate whether PFL implementation reduced the occurrence of household challenge ACEs among young children in Rhode Island and New York relative to other states without PFL implementation. This study used 2011-2019 population-based National Survey of Children's Health (NSCH) data to compare the proportion of household challenge ACEs experienced in intervention versus control states. Bayesian Additive Regression Tree (BART) and two-way fixed effects (TWFE) methods estimated post-intervention average treatment effects. United States (U.S.) Census covariates controlled for variation between states. Placebo checks bolstered the robustness of inference. Rhode Island and New York (intervention states) implemented PFL in 2014 and 2018, respectively. Control states included states that did not implement PFL as of 2019. The NSCH surveys included a 9-item ACEs questionnaire in which caregivers reported on whether their children ages 0-2 experienced various forms of abuse, and potentially traumatic household challenges (i.e., caregiver divorce, domestic violence, caregiver incarceration, caregiver mental health problems, caregiver substance use). This study focused on five household challenge ACEs and the occurrence of any household challenge ACE as primary outcomes. A treatment indicator identified intervention and control states. State-level covariates were median household income, unemployment rate, and percent uninsured.  Results suggest that PFL implementation was associated with significant 4% reduction in caregiver separation and 2% reduction in caregiver incarceration ACEs in the intervention states post-PFL adoption and a significant 4% reduction in the proportion of young children experiencing any household challenge ACE relative to no PFL implementation in control states. This study adds rigorous scientific support for the public health benefits of PFL for young children and their families. Existing evidence indicates caregiver incarceration and separation may affect children's short- and long-term mental health, substance use, and school related outcomes. Support through PFL may help reduce the occurrence of household challenge ACEs and promote children's well-being.

In 2020, firearm violence became the leading cause of death for American children and teens, a critical datapoint informing the 2024 U.S. Surgeon General's advisory on firearm violence. However, firearm violence has been a leading cause of death for Black youth-particularly, Black males-for decades, disproportionately impacting their morbidity and mortality. As the rights of Black youth to experience safety from firearm violence converge with the interests of white youth now increasingly impacted by it, it is imperative that prevention scientists critically interrogate what contributes to the national willfulness to see Black youth as perpetrators of violence deserving of punishment and a national reluctance to see Black youth as victims of violence deserving of healing and prevention? Guided by the Cycle of Dehumanization framework, we contend this pattern is symptomatic of racial dehumanization, a root cause of structural racism and violence. We argue that to successfully prevent youth firearm violence using a structural approach, we must disrupt dehumanizing narratives about Black male criminality and offer a viable solution through our visual storytelling and narrative change campaign, In All Ways Human. Using an adapted community engaged participatory action research approach and qualitative interviewing, our multimodal narrative change project captured 50 strategically disseminated (mural, billboards, kiosks, exhibits, digital galleries) portraits and stories that construct a counter-narrative with the power to prevent youth violence by transforming the ways in which Black males are seen, see one another, and see themselves. The impact and future directions of our narrative change effort are discussed.

Addressing socio-demographic differences that affect mental health service encounters is crucial for ethical practice and enhancing therapeutic alliances. Yet discussing personal perceptions of socio-demographic influences within a large healthcare system can be challenging due to discomfort among staff and the need for engagement at both interpersonal and organizational levels. The SITE project sought to intervene on one healthcare system's mental health providers' willingness to discuss patient background during care coordination. An internal workgroup used intervention mapping enhanced with frameworks from socio-demographic-focused literature and implementation science. Data collection included surveys, interviews, and a participatory consensus process. The results were two multi-component intervention packages designed to address interpersonal and organizational barriers, each targeting providers' willingness and psychological safety in addressing patient-specific background factors with colleagues. The interventions were adopted by the setting and then later dissolved due to administrative shifts. While the resultant interventions are unique to this setting, we demonstrate a repeatable process for adapting a well-known intervention development method (intervention mapping (IM)) informed by theory and implementation science. This process can be applied in other healthcare systems for discerning multi-level interventions appropriate to different contexts.

Image-based sexual abuse is an increasingly prevalent form of technology-facilitated harm, yet disclosure rates remain low. Understanding why youth do not disclose image-based sexual abuse is critical for developing effective prevention and intervention strategies. This paper examined the reasons youth do not disclose image-based sexual abuse incidents and identified incident- and person-level characteristics associated with different nondisclosure motives. Participants were recruited online to this US-based cross-sectional study between June 28, 2023, and April 1, 2024, using social media advertisements targeting individuals aged 18-28. A total of 6204 individuals completed the survey; 2854 (46.0%) reported experiencing image-based sexual abuse before age 18. The analytic sample included 2522 incidents reported by 1551 participants that were not disclosed. The most frequently cited reasons for nondisclosure were fear of getting in trouble with family (53.9%), embarrassment (52.9%), and the belief that they could handle the incident alone (45.2%). Reasons varied by image-based sexual abuse subtype. Longer incident duration and explicit content were related to fear of getting into trouble with their family or the police, and having multiple people responsible was related to many reasons for nondisclosure (i.e., fear of getting in trouble, fear the person would find out, embarrassment, and feeling like no one could help) ,. Female participants and sexual/gender minority youth were more likely to report barriers specific to fear and shame. Prior victimization was associated with a greater belief that no one could help and fear of getting in trouble or the person finding out. Prevention should address common fears, challenge stigma and self-blame, and ensure youth have access to trusted adults and non-punitive disclosure options. These findings support clinical efforts to reduce barriers and promote safe disclosure pathways for these survivors.

Asian American and Pacific Islanders (AAPI) youth report a high prevalence rate of youth violence. Logic models that guide violence prevention programming are often not culturally specific and may not address the unique structural/social drivers of youth violence in minoritized communities. Monsoon has received Rape Prevention and Education (RPE) funding from the Iowa Department of Health and Human Services (Iowa HHS) since 2007 through a competitive application process. The specific aim of Monsoon's Youth Sexual Violence Prevention Program (YVPP) is to "prevent gender-based violence by educating and empowering peers through outreach and workshops." In response to growing awareness at Iowa HHS about the limitations of the State RPE programmatic logic model in addressing AAPI-specific context for youth violence, Monsoon received additional RPE funds to establish the first culturally specific RPE logic model in Iowa. This initiative aims to fill a gap in culturally relevant evaluation resources. This paper describes the academic-community partnership initiated in 2023 that guided the participatory development of the culturally specific logic model and evaluation plan, for youth violence prevention in AAPI communities. Monsoon partnered with staff and faculty from the Center for Disease Control and Prevention (CDC)-funded Prevention Research Center for Rural Health (PRC-RH) in this work. PRC-RH faculty and staff reviewed documents and interviewed Monsoon leaders and youth staff. An AAPI-specific youth staff logic model and evaluation research questions were developed using a community-based research approach. Based on these, a mixed-methods evaluation plan was outlined. This case study highlights the critical importance of community-engaged research and practice to develop culturally specific logic models and evaluation designs for youth sexual violence prevention in the AAPI community.

Suicide has become the second leading cause of death among U.S. college students, exacerbated by COVID-19. A more comprehensive understanding of its risk factors can guide the development of effective prevention strategies tailored to this population. We employed a time-series cross-sectional approach and used the national survey data from the American College Health Association to examine the effects of institutional characteristics (school locale, region, size, and type) and behavioral factors (physical and social activities) on suicide risks among U.S. college students across different pandemic phases (before, early phase, and late phase). We also tested whether behavioral factors moderated the association between the pandemic period and suicide risk. After adjusting for confounders, multilevel regression results showed that (1) suicide risk increased during the pandemic, particularly in the late phase; (2) institutional characteristics, including geographic region and religious affiliation, as well as student behaviors such as physical and social activities, were significant predictors of suicide risks; and (3) behavioral factors moderated the pandemic's effect on suicide risk, as shown by the role of social activities (e.g., team sports, socializing with friends) amplifying the effect, whereas spending time with family mitigated it. Because risk rose fastest where certain social activities intensified pandemic effects, campuses should prioritize structured, lower-risk social engagement, while pairing selective strategies in high-activity settings (e.g., targeted screening, gatekeeper training) with universal measures (e.g., campus-wide suicide prevention program). Overall, this study offers evidence-informed guidance to shape campus environments that mitigate suicide risk and support student well-being, considering institutional context, behavioral factors, and public health crises.

Students of color are disciplined for behavioral infractions at higher rates than white students in K-12 schools in the USA. The consequences of racism in K-12 schools include mental health problems, school dropout, and disproportionate disciplinary practices, leading to the school-to-prison pipeline. Many school personnel implement interventions to address student behavior rather than racism and other implicit biases. Furthermore, culturally relevant practices are imperative to address the root causes of racial disparities in student discipline. For these reasons, a systematic and comprehensive review of the published literature on school-based interventions in the USA (including public and private K-12 schools) was conducted to identify interventions to remedy racial disparities in school discipline, as well as the research designs used to test their efficacy. The final sample includes 48 studies that directly or indirectly attempt to address the race discipline gap. There were only three studies that reduced race disparities in school discipline with a culturally relevant intervention. Future researchers may consider the importance of the school's cultural context and intervention audience when developing and testing efforts to reduce racial disproportionality.

Mediation analysis is essential for understanding causal mechanisms and designing effective policy interventions. This paper explains a comprehensive framework for implementing VanderWeele's (2009) three-way decomposition using natural effect models (NEM), focusing on its application in health equity research. While three-way decomposition helps social epidemiologists understand how social exposures influence health outcomes through differential mediator effects, its implementation through NEM has not been fully utilized. Using the R package medflex to implement NEM, we demonstrate the methodology through two case analyses: healthcare expenditure disparities between Medicaid beneficiaries and health insurance enrollees using continuous outcomes, and racial/ethnic disparities in preterm birth using binary outcomes. We provide detailed interpretations of NEM estimates in terms of three-way decomposition and discuss methodological considerations for researchers designing health policies to reduce disparities.

Preventive parenting programs effectively promote positive parenting behaviors and prevent negative child behavioral health outcomes, but engagement is often low. Self-directed eHealth interventions are increasingly available as an option that offers flexibility and increased access for parents. Yet, reports of and operationalizations of engagement within self-directed programs vary and correlates of eHealth engagement are not well understood. Thus, this study systematically reviewed reports to identify common operationalizations of engagement in self-directed preventive eHealth parenting programs, predictors of engagement, and associations between engagement and child behavioral health outcomes. Of 835 records screened, 31 unique studies had caregivers that participated in a self-directed eHealth preventive parenting programs that defined engagement or program adherence. Most studies reported the number of modules completed as a measure of parent engagement. About one-third of programs also reported frequency of parent program access. There was great variability in variables examined as correlates of eHealth program engagement, limiting the ability to draw consistent patterns of predictors of engagement. Evidence linking engagement to youth mental and behavioral health outcomes was similarly limited. Although module or session completion was the most common metric, studies varied in how engagement was further defined and applied. This heterogeneity poses challenges for comparing engagement rates across studies, but individual studies provide emerging evidence of potential ways to increase engagement within self-directed programs (e.g., phone coaching, ordering of content). Overall, this review underscores the need for further research as eHealth interventions continue to be offered to parents.

Depression and substance use are often comorbid, but less is known about these associations in non-clinical community samples. Moreover, existing research is primarily cross-sectional or focuses on long-term trajectories of depressive symptoms and substance use. The present study examines within-person associations to estimate the extent to which monthly fluctuations in depressive symptoms relate to alcohol and cannabis use. A community sample of 778 young adults (M = 21.1 at baseline, 56% female) completed up to 33 monthly surveys (82.7% total response rate) scattered across a 6-year period (2015-2021). Zero-truncated regression was used to stratify associations with any use and amount of use on months that alcohol and cannabis were reported, respectively. Pertaining to alcohol use, depressive symptoms predicted lower odds of drinking alcohol on a given month; however, if young adults did drink, then depressive symptoms predicted heavier drinking (i.e., more drinks per week). For cannabis, depressive symptoms predicted both greater odds of using cannabis and heavier use in that month (i.e., greater hours high per week). Associations between depressive symptoms and alcohol/cannabis quantity indices were mediated by coping motives. Findings highlight the risks of elevated depressive symptoms for young adults, in terms of alcohol and cannabis use.

Evidence-based program repositories have been designed to help practitioners in their decision-making. Most repositories supplement summaries of effectiveness evidence with information intended to assist with implementation of each included program (e.g., training costs). It is less common for guidance to be included to support translation of findings for a broader range of purposes, such as enhancing related programs already in place. To help address this gap within the area of youth mentoring, the National Mentoring Resource Center has appended "Insights for Mentoring Practitioners" to profiles of 47 mentoring programs included in the CrimeSolutions.gov repository of the National Institute of Justice. We qualitatively analyzed these commentaries to elucidate themes across them that can inform the development and improvement of mentoring programs. Themes included (1) ensuring alignment across program goals, design, implementation, and evaluation; (2) connecting the intervention to mentees' home, parents, and larger environment; (3) tailoring mentor engagement and support to effectively serve youth; and (4) optimizing the role of mentoring within multi-component programs. Discussion focuses on how findings inform the improvement of mentoring programs, and how content geared toward the translation of evidence to practice could enhance and improve evidence repositories.

Program adaptations or modifications are often necessary to suit local contexts, populations, and resources available. Despite the frequency with which program modifications are made in practice, they are rarely systematically recorded and reported comprehensively, particularly in the context of scale-up delivery led by implementers and in low- and middle-income countries. We use the FRAME framework to document the modifications of a parenting program called Parenting for Lifelong Health for Parents and Adolescents, locally known as Furaha Teens, which was delivered to over 30,000 families in Tanzania in 2020-2021. We draw on thematic analysis of 12 focus groups and 67 semi-structured interviews with program facilitators, coaches, coordinators, and managers (164 participants). Both proactive and reactive modifications were made to the program context and content. Proactive modifications included delivering the program as part of a wider package of services for families with adolescent girls, focused on HIV prevention, and adding HIV-related content. Both proactive and reactive modifications were made to make the material more acceptable to participants, such as by translating into local languages. Modifications to condense the number and frequency of sessions were reactively made by implementers to meet delivery timelines, particularly due to COVID-related closures. Study findings suggest that a range of program modifications may be required to scale programs to large cohorts as well as new contexts. To ensure successful delivery at scale, funders can support implementers in learning from the modifications and encouraging reflection on whether and how modifications affect program fidelity.

The purpose of this study was to estimate the costs of starting up diverse interventions for preventing opioid misuse in young people aged 15 to 30. Interventions were to be delivered in the context of research trials funded under the National Institutes of Health HEAL (Helping to End Addiction Long-term®) Prevention Initiative. Start-up activities were conducted in systems and settings across the United States from 2019 to 2023. Start-up costs were estimated using activity-based costing from a provider perspective and common methods and data collection procedures across projects (n = 8). Descriptive statistics were used to understand the magnitude of and variability in start-up costs, cost drivers, and staff time. Factors explaining variability were identified from project activity descriptions. Start-up activities cost $37,541 on average, and $33,492 at the median (2020 USD). Labor drove costs (89% on average). Training, project management, and partner engagement accounted for 78% of start-up costs on average. There was considerable variability in total and activity costs. Start-up activities provide an essential foundation for successful intervention delivery and sustainability, yet the resources needed and associated costs for this phase of intervention implementation do not receive much attention. We found that interventions faced unique start-up challenges and leverage points, resulting in differences in total cost and activity cost burden. ClinicalTrials.gov Identifier: NCT01088542, NCT04901312, NCT04678960, NCT04617938, NCT04550715, and NCT04135703.

The objective of this study is to examine cross-over effects of the Raising Healthy Children universal childhood preventive intervention on adult suicide behaviors and related mental health. A nonrandomized controlled trial was conducted in elementary schools serving higher-risk areas in Seattle, Washington (during ages 6-13, grades 1-6) and followed up at ages 21-39. The panel originated in Seattle but was followed in and out of state. This study examines participants who had been in the intervention (n = 156) vs. control (n = 220) conditions in grades 1-6. The intervention provided teachers with methods of classroom management, parents with family management skills, and children with social-emotional skills training. Outcomes examined were ever suicide ideation, attempt, or completion, and DSM-IV-based criterion counts for depression, generalized anxiety disorder, PTSD, and social phobia across 6 waves, ages 21-39. At follow-up, the intervention group showed significantly lower suicide ideation and behavior, depression, generalized anxiety disorder, PTSD, and social phobia than the control group. Universal childhood preventive intervention can reduce suicide ideation and behaviors and related mental health problems in adulthood. Clinical Trials.gov ID: NCT04075019.

Social spending refers to public or private funds aimed at helping vulnerable groups. Evidence-based decision-making supports social spending by applying existing evidence to inform funding and involves integrating relevant information from various sources, such as experimental studies and community feedback. Online clearinghouses support evidence-based decision-making by synthesizing evidence on what works, though manually updating literature is incomplete. In addition, passively summarizing evaluations is insufficient for end-users to implement preventive solutions that achieve population impacts. A responsive platform and a dissemination plan are needed to encourage the uptake of equitable and culturally relevant preventive interventions grounded in transparency and rigorous evidence of effectiveness. A two-part conceptual framework is proposed for a clearinghouse platform that integrates stepwise, interactive, AI-driven capabilities safeguarded by human touchpoints. First, clearinghouses should adopt a "living" review to automatically update evaluations-an approach embraced globally by organizations such as the World Health Organization, Cochrane Collaboration, and the U.N.'s Pan American Health Organization. The second component involves adding a chatbot to support assessment and implementation guidance and make recommendations on (1) the provision of all evidence-based preventive interventions (EBPIs) and their key activities, (2) EBPIs shown to achieve equitably distributed outcomes, (3) culturally relevant EBPIs that align with the values and needs of a target population, (4) implementation support, such as materials, training, and fidelity measures, and (5) delivery costs. The resulting platform will ethically expedite the translational process of identifying and scaling EBPIs, leading to a more complete, comprehensive, and accessible body of evidence on effective preventive strategies.

A good understanding of the associations of COVID-19 infection and mortality with contextual factors when vaccines were not widely available is necessary for human societies to be better prepared for future outbreaks of infectious diseases. This retrospective ecological study aimed to explore the spatially varying associations of COVID-19 incidence, death, and case fatality rates with contextual socioeconomic, health, and environmental factors during the period of partial population coverage of vaccination at county level in the state of Georgia, USA. The associations of COVID-19 rates and contextual factors were analyzed using geographically weighted regression (GWR), compared with ordinary least squares regression (OLS) analysis. OLS results showed that most factors were significantly associated with COVID-19 death rate and case fatality rate, but not incidence rate. GWR results demonstrated that the associations of all three COVID-19 rates with factors varied across space: A factor might have a significant positive, significant negative, or nonsignificant association with each rate in certain counties. Most factors for poor health outcomes were significantly associated with higher risks of COVID-19 infection and mortality in more counties compared to non-significant or inverse associations. The spatially varying associations for some contextual factors were related to the socioeconomic and urbanization characteristics of counties. Some factors also affected COVID-19 infection and mortality differently. For example, persons aged 65 and older percentage was not a significant risk factor of COVID-19 infection in most counties, but it was the most spatially consistent risk factor of COVID-19 death in Georgia; fully vaccinated percentage was a more significant indicator of reducing COVID-19 infection in rural counties compared to urban and suburban areas. This study provides useful information for public health agencies and professionals to make and implement more specific and targeted local health policies.

The growing momentum to combat racial and ethnic disparities in youth violence victimization and exposure by developing structural interventions is an important shift toward addressing the root causes of disparities in youth violence. However, discussions of structural interventions sometimes overlook existing programs such as Summer Youth Employment Programs (SYEPs). We conduct an integrative review of evaluation studies on SYEPs and use the findings to re-examine the evidence and arguments for SYEPs, advancing this issue's goal of highlighting viable solutions to address youth violence. We identified 13 evaluation studies of SYEPs in Boston, Chicago, New York, Philadelphia, Cuyahoga County, OH, and Washtenaw County, MI. Findings from these evaluations suggest that SYEPs impact behavioral and sociocultural environmental determinants of youth violence across the individual, interpersonal, and community levels. No evaluations reported on societal-level determinants of youth violence. Yet, well-designed SYEPs align with key components of interventions to address disparities in youth violence: (1) they are sustainable, having been implemented in several cities and states for decades despite precarious funding; (2) as evidenced by evaluation findings, they address behavioral and sociocultural environmental domains across multiple levels of influence; and (3) they address racial and economic disparities by serving youth disproportionately affected by community violence, Black and Latine youth in economically disadvantaged communities. We argue that when these programs are made universal, they are structural interventions that address youths' lack of employment opportunities at the societal level.

Youth violence is a significant public health concern with differential impacts on youth who reside in high-burden communities. Structural factors, including institutional practices and policies, contribute to the concentration of youth within neighborhoods where violence is more likely to occur. Across disciplines, scholars have identified structural factors that affect not only the prevalence of violence but also environmental conditions that make violence more likely. Consequently, interventions may fall short of addressing the root causes of violence, perpetuating cycles of harm rather than fostering long-term solutions. It is crucial to advance community and societal level approaches to address macro-level factors influencing youth violence. The Institute of Medicine's Framework for Collaborative Public Health Action suggests that community coalitions may contribute to preventing youth violence by facilitating multisectoral engagement across socioecological levels to promote sustainable change. Guided by this framework, the ThrYve Coalition in Kansas City convened over 40 community partners across 15 sectors to address youth violence (Watson-Thompson et al., 2020). Through a collaborative process, 87 change lever strategies (i.e., strategies to implement program, policy, practice changes) were identified to support community and systems-level improvements aligned with risk and protective factors related to youth violence. Between 2018 and 2022, the ThrYve Coalition implemented more than 199 community changes, addressing social and structural determinants related to youth violence. This study examines the collaborative process for facilitating change as part of a comprehensive community intervention to prevent youth violence.

Hospitalization and death following COVID-19 infection continue to pose a major public health concern and place strain on health system resources. Outpatient antiviral medication can reduce the risk of COVID-19 hospitalization and death for those at risk of poor outcomes, but identifying high-risk populations who may benefit most from treatment is challenging. The objective of this study was to develop and validate a prediction model for the composite outcome of hospitalization or death in the 14 days following COVID-19 infection. Our sample included 67,530 COVID-19 infections documented in outpatient care and occurring between April 1, 2020, and November 1, 2022, for 64,529 Kaiser Permanente Washington patients who did not receive outpatient antiviral treatment; 1378 (2.0%) of these infections resulted in hospitalization or death. Our prediction model, estimated using logistic regression with LASSO variable selection and ridge penalization, included 19 risk factors and showed high performance, including an area under the curve of 0.825 (95% confidence interval 0.813-0.836). Among the 10% of infections with the highest risk predictions, the true positive rate was 48% (46-51%) and the positive predictive value was 9.9% (9.2-10.6%). Supplemental analyses confirmed strong model performance across racial and ethnic subgroups and over time. We also present our process for selecting a risk threshold above which to recommend antiviral treatment and discuss considerations for prospective clinical implementation. This project demonstrates that machine learning tools can be used by health systems to deliver timely, targeted secondary prevention to reduce the risk of serious illness or death.