Access to technology, internet usage, and online health information-seeking behaviors in a racially diverse, lower-income population.

Background: This study examined access to technology, internet usage, and online health information-seeking behaviors, in a racially diverse, lower-income population. Methods: Data were obtained via a cross-sectional survey of low-income communities in Houston, Los Angeles, and New York between April and August 2023. Binary responses to the following online health information-seeking behaviors, internet and technology access, were examined: using the internet to (i) understand a medical diagnosis, (ii) fill a prescription, (iii) schedule a healthcare appointment, (iv) email communication with a healthcare provider, and (v) access electronic health records and medical notes. Results: 41% of survey respondents identified as non-Hispanic Black individuals, 33% as non-Hispanic White individuals, and 22% as Hispanic individuals. 69% reported a pre-tax annual household income of less than $35,000. 97% reported ownership/access to a smart device; 97% reported access to reliable internet. In the past year, only 59% reported using the internet to better understand their medical diagnosis, 36% reported filling a prescription online, 47% scheduled a medical appointment online, 47% viewed electronic health records online, and 56% emailed healthcare providers. Female sex, higher incomes, and having at least a bachelor's degree were significantly associated with all five online health information-seeking attributes. Conclusion: Despite high technology adoption rates, we observed suboptimal online health information-seeking behaviors. This underutilization has potential adverse implications for healthcare access and use given the documented advantage of HIT. Efforts to increase health information-seeking behaviors should explore the identification of HIT barriers, and patient education to increase familiarity and usage in this population.

Prevalence of youth experiencing homelessness and its association with suicidal thoughts and behaviors: Findings from a population-based study.

Although various studies have examined factors associated with suicidal behaviors among youth, few studies have investigated the association between youth experiencing homelessness (YEH) and suicidal thoughts and behaviors (STBs) using a large nationally representative sample. The objectives of this study were to investigate prevalence of YEH and its association with STBs. Data for this study came from the 2021 Youth Risk Behavior Survey. An analytic sample of 17,033 youth aged 14-18 (51.7 % male) was analyzed using binary logistic regression. Of the 17,033 youth examined, 3 % experienced homelessness during the past 30 days, 21.3 % experienced suicidal ideation, 17.3 % made a suicide plan, and 10.9 % attempted suicide during the past 12 months. Controlling for demographic characteristics and feeling sad or hopeless, YEH was associated with 2.48 times higher odds of experiencing suicidal ideation (AOR=2.48, p<.001), 2.46 times higher odds of making a suicide plan (AOR=2.46, p<.001), and 4.38 times higher odds of making a suicide attempt (AOR=4.38, p<.001). The findings of this study highlight the importance of identifying youth who are at risk of experiencing homelessness to ensure early interventions are put in place to prevent suicidal behaviors.

Racial and ethnic differences in predictors of participation in an intergenerational social connectedness intervention for older adults.

BACKGROUND: Social connectedness is a key determinant of health and interventions have been developed to prevent social isolation in older adults. However, these interventions have historically had a low participation rate amongst minority populations. Given the sustained isolation caused by the COVID-19 pandemic, it is even more important to understand what factors are associated with an individual's decision to participate in a social intervention. To achieve this, we used machine learning techniques to model the racial and ethnic differences in participation in social connectedness interventions. METHODS: Data were obtained from a social connectedness intervention that paired college students with Houston-area community-dwelling older adults (> 65 yo) enrolled in Medicare Advantage plans. Eligible participants were contacted telephonically and asked to complete the 3-item UCLA Loneliness Scale. We used the following machine-learning methods to identify significant predictors of participation in the program: k-nearest neighbors, logistic regression, decision tree, gradient-boosted decision tree, and random forest. RESULTS: The gradient-boosted decision tree models yielded the best parameters for all race/ethnicity groups (96.1% test accuracy, 0.739 AUROC). Among non-Hispanic White older adults, key features of the predictive model included Functional Comorbidity Index (FCI) score, Medicare prescription risk score, Medicare risk score, and depression and anxiety indicators within the FCI. Among non-Hispanic Black older adults, key features included disability, Medicare prescription risk score, FCI and Medicare risk scores. Among Hispanic older adults, key features included depression, FCI and Medicare risk scores. CONCLUSIONS: These findings offer a substantial opportunity for the design of interventions that maximize engagement among minority groups at greater risk for adverse health outcomes.

Population-based assessment of the burden of COVID-19 infection in African countries: a first-year report card and public health implications.

Background: The COVID-19 pandemic constitutes a global health threat and poses a major burden on the African continent. We assessed the real-world burden of COVID-19 infection in African Union (AU) member states to determine the distributional patterns of epidemiological measures during the first 1 year of the pandemic. Methods: This retrospective cross-sectional study utilized COVID-19 data from publicly available data repositories of the African Center for Disease Control and Prevention and Our World in Data for the period February 2020 to January 2021. AU member states were classified into low, medium, and high burdens based on COVID-19 morbidity. We conducted descriptive and inferential analyses of COVID-19-reported cases, deaths, recoveries, active cases, COVID-19 tests, and epidemiological measures that included morbidity and mortality rates, case fatality rate (CFR), and case ratios. Results: A total of 3.21 million cases were reported during the 1-year period, with 2.6 million recoveries, 536,784 cases remaining active, and 77,486 deaths. Most countries (49.1%, n = 26) in AU experienced a low burden of COVID-19 infection compared to 28.3% (n = 15) with medium burden and 22.6% (n = 12) with high burden. AU nations with a high burden of the disease were mainly in the northern and southern regions. South Africa recorded the highest number of cases (1.31 million), followed by Morocco with 457,625 and Tunisia with 175,065 cases. Correspondently, death tolls for these countries were 36,467, 7888, and 5528 deaths, respectively. Of the total COVID-19 tests performed (83.8 million) during the first 1 year, 62.43% were from high-burden countries. The least testing occurred in the medium-burden (18.42%) countries. The overall CFR of AU was 2.21%. A morbidity rate of 327.52/105 population and mortality rate of 5.96/105 population were recorded during the first 1-year period with significant variations (p < 0.0001) across burden levels. Continental morbidity and mortality rates of 17,359/105 and 315.933/105 populations were recorded with significant correlation (r = 0.863, p < 0.0001) between them and variations across selected epidemiological measures by COVID-19 burden levels. Conclusion: Understanding the true burden of the disease in AU countries is important for establishing the impact of the pandemic in the African continent and for intervention planning, preparedness, and deployment of resources during COVID-19 surges and future pandemics.

Evaluating Predictors of Participation in Telephone-Based Social-Connectedness Interventions for Older Adults: A Dual Machine-Learning and Regression Approach.

Social isolation is a well-documented contributor to poor mental and physical health, and interventions promoting social connectedness have been associated with various health benefits. This study examined predictors of participation in a telephone-based social connectedness intervention for socially isolated older adults. Data were obtained from a social-connectedness intervention that paired college students with Houston-area, community-dwelling adults aged 65 years and older and enrolled in Medicare Advantage plans. We combined machine learning and regression techniques to identify significant predictors of program participation. The following machine-learning methods were implemented: (1) k-nearest neighbors, (2) decision tree and ensembles of decision trees, (3) gradient-boosted decision tree, and (4) random forest. The primary outcome was a binary flag indicating participation in the telephone-based social-connectedness intervention. The most predictive variables in the ML models, with scores corresponding to the 90th percentile or greater, were included in the regression analysis. The predictive ability of each model showed high discriminative power, with test accuracies greater than 95%. Our findings suggest that telephone-based social-connectedness interventions appeal to individuals with disabilities, depression, arthritis, and higher risk scores. scores. Recognizing features that predict participation in social-connectedness programs is the first step to increasing reach and fostering patient engagement.

Rethinking access to care: A spatial-economic analysis of the potential impact of pharmacy closures in the United States.

Data chronicling the geo-locations of all 61,589 pharmacies in the U.S. (from the Homeland Infrastructure Foundation-Level Data (HIFLD) Open Data interface, updated on April 2018) across 215,836 census block groups were combined with Medically Underserved Areas (MUAs) information, and the Centers for Disease Control and Prevention's Social Vulnerability Index (CDC-SVI). Geospatial techniques were applied to calculate the distance between the center of each census block and the nearest pharmacy. We then modeled the expected additional travel distance if the nearest pharmacy to the center of a census block closed and estimated additional travel costs, CO2 emissions, and lost labor productivity costs associated with the additional travel. Our findings revealed that MUA residents have almost two times greater travel distances to pharmacies than non-MUAs (4,269 m (2.65 mi) vs. 2,388 m (1.48 mi)), and this disparity is exaggerated with pharmacy closures (107% increase in travel distance in MUAs vs. 75% increase in travel distance in non-MUAs). Similarly, individuals living in MUAs experience significantly greater average annual economic costs than non-MUAs ($34,834 ± $668 vs. $22,720 ± $326). Our findings suggest the need for additional regulations to ensure populations are not disproportionately affected by these closures and that there is a significant throughput with community stakeholders before any pharmacy decides to close.

Comparing In-Person Only, Telemedicine Only, and Hybrid Health Care Visits Among Older Adults in Safety-Net Clinics.

Introduction: Older adults face challenges in seeking health care. This study examined factors associated with in-person only versus telemedicine only versus hybrid health care visits among adults 65+ in safety-net clinics. Methods: Data were obtained from a large Texas-based Federally Qualified Health Center (FQHC) network. The dataset included 12,279 appointments for 3914 unique older adults between March and November 2020. The outcome of interest was a 3-level indicator of telemedicine visits: in-person visits only, telemedicine visits only, and hybrid (in person + telemedicine) visits during the study period. We used a multinomial logit model adjusting for patient level characteristics to assess the strength of the relationships. Results: Compared to their white counterparts, black and Hispanic older adults were significantly likely to have telemedicine only visits versus in-person only visits (black RRR: 0.59, 95% confidence interval [CI]: 0.41-0.86; Hispanic RRR: 0.46, 95% CI: 0.36-0.60). However, there were no significant racial and ethnic differences in hybrid utilization (black RRR: 0.91, 95% CI: 0.67-1.23; Hispanic RRR: 0.86, 95% CI: 0.70-1.07). Discussion: Our findings suggest that hybrid opportunities may bridge racial and ethnic disparities in access to care. Clinics should consider building capacity for both in-person and telemedicine opportunities as complementary strategies.

Assessment of Unmet Health-Related Social Needs Among Patients With Mental Illness Enrolled in Medicare Advantage.

Importance: A large body of literature has found associations between unmet health-related social needs (HRSNs) and adverse mental health outcomes. A comparative analysis of the risks associated with HRSNs among patients with varying severity of mental illness and an assessment of how these risks compare with those of individuals without mental illness are needed. Objective: To examine the prevalence and risks of HRSNs among patients with serious and persistent mental illness (SPMI), patients with mental health diagnoses but no serious and persistent mental illness (non-SPMI), and patients with both SPMI and non-SPMI compared with individuals without mental illness. Design, Setting, and Participants: This retrospective cohort study used data from the Accountable Health Communities HRSN Screening Tool surveys, which target a nationally representative sample of Medicare Advantage members of a large payer (Humana Inc). The surveys were conducted between October 16, 2019, and February 29, 2020. Of the initial 329 008 eligible Medicare Advantage enrollees, 70 273 responded to the survey (21.4% response rate). Of those, 56 081 respondents (79.8%) had complete survey responses and were included in the final analytic sample. Main Outcomes and Measures: Outcomes of interest included 7 HRSNs (financial strain, food insecurity, housing instability, housing quality, severe loneliness, transportation problems, and utility affordability) based on responses to the survey. The major independent variable was the presence of mental illness up to 12 months preceding the date of survey completion. Codes indicating mental illness listed as the primary, principal, or secondary diagnoses of a patient's inpatient or outpatient medical claims data were identified, and participants were grouped into 4 cohorts: SPMI, non-SPMI, SPMI plus non-SPMI, and no mental illness. Results: Among 56 081 older adults, the mean (SD) age was 71.31 (8.59) years; 32 717 participants (58.3%) were female, and 43 498 (77.6%) were White. A total of 21 644 participants (38.6%) had at least 1 mental illness diagnosis in the past year, 30 262 (54.0%) had an HRSN, and 14 163 (25.3%) had both mental illness and an HRSN. Across all specific HRSNs, the odds of experiencing the respective HRSN was most substantial for those with SPMI plus non-SPMI vs those with only non-SPMI or SPMI. The HRSN with the largest risk differences among the study cohorts was severe loneliness; compared with the cohort without mental illness, the non-SPMI cohort had 2.07 times higher odds (95% CI, 1.84-2.32; P < .001), the SPMI cohort had 3.35 times higher odds (95% CI, 3.03-3.71; P < .001), and the SPMI plus non-SPMI cohort had 5.13 times higher odds (95% CI, 4.68-5.61; P < .001) of severe loneliness. Conclusions and Relevance: In this study, the increased risk of having HRSNs associated with SPMI, alone or in combination with non-SPMI, emphasizes the need for more targeted interventions to address social needs in this vulnerable population.

Hospital Length of Stay in Patients with and without Serious and Persistent Mental Illness: Evidence of Racial and Ethnic Differences.

BACKGROUND: Prior studies have documented racial and ethnic differences in mental healthcare utilization, and extensively in outpatient treatment and prescription medication usage for mental health disorders. However, limited studies have investigated racial and ethnic differences in length of inpatient stay (LOS) in patients with and without Serious and Persistent Mental Illness. Understanding racial and ethnic differences in LOS is necessary given that longer stays in hospital are associated with adverse health outcomes, which in turn contribute to health inequities. OBJECTIVE: To examine racial and ethnic differences in length of stay among patients with and without serious and persistent mental illness (SPMI) and how these differences vary in two age cohorts: patients aged 18 to 64 and patients aged 65+. METHODS: This study employed a retrospective cohort design to address the research objective, using the 2018 Healthcare Cost and Utilization Project (HCUP) National Inpatient Sample. After merging the 2018 National Inpatient Sample's Core and Hospital files, Generalized Linear Model (GLM), adjusting for covariates, was applied to examine associations between race and ethnicity, and length of stay for patients with and without SPMI. RESULTS: Overall, patients from racialized groups were likely to stay longer than White patients regardless of severe mental health status. Of all races and ethnicities examined, Asian patients had the most extended stays in both age cohorts: 8.69 days for patients with SPMI and 5.73 days for patients without SPMI in patients aged 18 to 64 years and 8.89 days for patients with SPMI and 6.05 days for patients without SPMI in the 65+ cohort. For individuals aged 18 to 64, differences in length of stay were significantly pronounced in Asian patients (1.6 days), Black patients (0.27 days), and Native American patients/patients from other races (0.76 days) if they had SPMI. For individuals aged 65 and older, Asian patients (1.09 days) and Native American patients/patients from other races (0.45 days) had longer inpatient stays if they had SPMI. CONCLUSION: Racial and ethnic differences in inpatient length of stay were most pronounced in Asian patients with and without SPMI. Further studies are needed to understand the mechanism(s) for these differences.

Optimizing Disaster Preparedness Planning for Minority Older Adults: One Size Does Not Fit All.

By 2050, one in five Americans will be 65 years and older. The growing proportion of older adults in the U.S. population has implications for many aspects of health including disaster preparedness. This study assessed correlates of disaster preparedness among community-dwelling minority older adults and explored unique differences for African American and Hispanic older adults. An electronic survey was disseminated to older minority adults 55+, between November 2020 and January 2021 (n = 522). An empirical framework was used to contextualize 12 disaster-related activities into survival an0000000d planning actions. Multivariate logistic regression models were stratified by race/ethnicity to examine the correlates of survival and planning actions in African American and Hispanic older adults, separately. We found that approximately 6 in 10 older minority adults did not perceive themselves to be disaster prepared. Medicare coverage was positively associated with survival and planning actions. Income level and prior experience with disaster were related to survival actions in the African American population. In conclusion, recognizing the gaps in disaster-preparedness in elderly minority communities can inform culturally sensitive interventions to improve disaster preparedness and recovery.

County-Level Determinants of COVID-19 Testing and Cases: Are There Racial/Ethnic Disparities in Texas?

Reports suggest that COVID-19 affects people of color disproportionately. Texas ranks second in the count of confirmed US cases. This study examined the relationship between county-level racial/ethnic composition and COVID-19 testing/cases in Texas, adjusting for population-level demographic characteristics, health factors, and health care access measures. County-level testing and case data, obtained from the Texas Department of State Health Services, were combined with the 2020 Robert Wood Johnson Foundation County Health Rankings data. Outcome variables were tests per 100,000 population and cases per 100,000 population. The independent variable of interest was percent of racial and ethnic composition. Multivariable linear regression analyses were used. There was a statistically significant increase in COVID-19 testing/100,000 population with every 1% increase in the proportion of African Americans/Blacks (ß = 2065.4; P = 0.009), Asians (ß = 2056.2; P = 0.015), and Hispanics (ß = 1641.1; P = 764.7). After controlling for county characteristics and cases/100,000 population, these relationships were no longer significant. However, primary care physician rate was significantly associated with testing/100,000 population (ß = 64.0; P = 0.027), as was the percent of uninsured (ß = -469.9; P = 0.024). An analysis of case data showed that African Americans had the largest number of cases/100,000 (ß = 432.2; P = 0.001), followed by Hispanics (ß = 422.8; P < 0.001) and Asians (ß = 415.4; P = 0.004). As in other parts of the United States, African Americans and Hispanics are most affected by COVID-19 in Texas. Community-based strategies to improve access to testing or reduce community spread outside clinical settings should target counties with low primary care physician rates or a high proportion of uninsured residents.

Factors Associated with COVID-Related Mortality: the Case of Texas.

BACKGROUND: Texas ranks 2nd in the count of COVID cases. Pre-existing disparities in healthcare may be intersecting with COVID-19 outcomes. OBJECTIVES: To explore the relationship between county-level race/ethnic composition and COVID-19 mortality in the state of Texas and determine whether county-level health factors, healthcare access measures, and other demographic characteristics explain this relationship. METHODS: This retrospective study uses county-level case and fatality data obtained from the Texas Department of State Health Services and merged with the 2020 Robert Wood Johnson foundation (RWJF) county health rankings data. The outcome variables were fatalities per 100,000 population. A two-part/hurdle model examined (1) the probability of having a COVID-19 fatality and (2) fatalities per 100,000 population in counties with 1+ fatalities. For both parts of the hurdle model, we examined the impacts of racial and ethnic composition, adjusting for county characteristics and health factors. RESULTS: The odds of having a COVID-19 fatality decreased with a unit increase in the rate of primary care physicians in a county (OR = 0.93; 95% CI = 0.89, 0.99). In the second part of the model, there was a statistically significant increase in COVID-19 fatalities/100,000 population with every 1 % increase in the proportion of Hispanics (ß = 5.41; p = 0.03) and African Americans (ß = 5.08; p value = 0.04). CONCLUSION: Counties with higher rates of minorities, specifically Hispanics and African Americans, have a higher COVID-19 fatality burden. Targeted interventions are needed to raise awareness of preventive measures in these communities.

Relationship Between Accountable Care Organization Status and 30-Day Hospital-wide Readmissions: Are All Accountable Care Organizations Created Equal?

OBJECTIVE: We compare hospital readmission rates by accountable care organization (ACO) status with national readmission averages, to determine whether ACO affiliation influences 30-day hospital-wide readmission rates. METHODS: Data from the 2015 American Hospital Association Survey of Care Systems and Payment database were merged with Centers for Medicare and Medicaid's 2015 Hospital Compare Deaths and Readmissions data set. A multinomial logistic regression model is used to examine readmission rates, categorized as better, no different, or worse, in comparison to national averages, by ACO status. RESULTS: Compared with Non-ACO hospitals and holding the covariates constant, the relative risk of having better than national average readmissions was 1.85 in Medicare ACO hospitals (p = .36). Compared with facilities in the Northeast region, the relative risk of having better than national average readmissions was 2.21 for facilities in the West (p = .10). Facilities in the Midwest and Southern regions had a lower risk of having better than national average rates (Relative Risk: 0.90 and 0.23, respectively; p = .83 and .06, respectively). As hospital beds increase, facilities have significantly lower risks of having worse than national average readmissions. CONCLUSIONS: Overall, the ACO status did not significantly improve readmissions. However, Medicare ACOs performed better than non-Medicare ACOs and those hospitals without any reported ACO arrangements.

Health Care Disparities in the Post-Affordable Care Act Era.

Disparities in health care have been targeted for elimination by federal agencies and professional organizations, including the American Public Health Association. Although the Affordable Care Act (ACA) provides a valuable first step in reducing the disparities gap, progress is contingent upon whether opportunities in the ACA help or hinder populations at risk for impaired health and limited access to medical care.

Modeling the determinants of Medicaid home care payments for children with special health care needs: A structural equation model approach.

BACKGROUND: The management of children with special needs can be very challenging and expensive. OBJECTIVE: To examine direct and indirect cost drivers of home care expenditures for this vulnerable and expensive population. METHODS: We retrospectively assessed secondary data on children, ages 4-20, receiving Medicaid Personal Care Services (PCS) (n = 2760). A structural equation model assessed direct and indirect effects of several child characteristics, clinical conditions and functional measures on Medicaid home care payments. RESULTS: The mean age of children was 12.1 years and approximately 60% were female. Almost half of all subjects reported mild, moderate or severe ID diagnosis. The mean ADL score was 5.27 and about 60% of subjects received some type of rehabilitation services. Caseworkers authorized an average of 25.5 h of PCS support per week. The SEM revealed three groups of costs drivers: indirect, direct and direct + indirect. Cognitive problems, health impairments, and age affect expenditures, but they operate completely through other variables. Other elements accumulate effects (externalizing behaviors, PCS hours, and rehabilitation) and send them on a single path to the dependent variable. A few elements exhibit a relatively complex position in the model by having both significant direct and indirect effects on home care expenditures - medical conditions, intellectual disability, region, and ADL function. CONCLUSIONS: The most important drivers of home care expenditures are variables that have both meaningful direct and indirect effects. The only one of these factors that may be within the sphere of policy change is the difference among costs in different regions.

Effects of diabetes self-management programs on time-to-hospitalization among patients with type 2 diabetes: a survival analysis model.

OBJECTIVE: This study compared time-to-hospitalization among subjects enrolled in different diabetes self-management programs (DSMP). We sought to determine whether the interventions delayed the occurrence of any acute event necessitating hospitalization. METHODS: Electronic medical records (EMR) were obtained for 376 adults enrolled in a randomized controlled trial (RCT) of Type 2 diabetes (T2DM) self-management programs. All study participants had uncontrolled diabetes and were randomized into either: personal digital assistant (PDA), Chronic Disease Self-Management Program (CDSMP), combined PDA and CDSMP (COM), or usual care (UC) groups. Subjects were followed for a maximum of two years. Time-to-hospitalization was measured as the interval between study enrollment and the occurrence of a diabetes-related hospitalization. RESULTS: Subjects enrolled in the CDSMP-only arm had significantly prolonged time-to-hospitalization (Hazard ratio: 0.10; p=0.002) when compared to subjects in the control arm. Subjects in the PDA-only and combined PDA and CDSMP arms showed no improvements in comparison to the control arm. CONCLUSION: CDSMP can be effective in delaying time-to-hospitalization among patients with T2DM. PRACTICE IMPLICATIONS: Reducing unnecessary healthcare utilization, particularly inpatient hospitalization is a key strategy to improving the quality of health care and lowering associated health care costs. The CDSMP offers the potential to reduce time-to-hospitalization among T2DM patients.

Can chronic disease management programs for patients with type 2 diabetes reduce productivity-related indirect costs of the disease? Evidence from a randomized controlled trial.

The objective was to assess the impacts of diabetes self-management programs on productivity-related indirect costs of the disease. Using an employer's perspective, this study estimated the productivity losses associated with: (1) employee absence on the job, (2) diabetes-related disability, (3) employee presence on the job, and (4) early mortality. Data were obtained from electronic medical records and survey responses of 376 adults aged ≥18 years who were enrolled in a randomized controlled trial of type 2 diabetes self-management programs. All study participants had uncontrolled diabetes and were randomized into one of 4 study arms: personal digital assistant (PDA), chronic disease self-management program (CDSMP), combined PDA and CDSMP, and usual care (UC). The human-capital approach was used to estimate lost productivity resulting from 1, 2, 3, and 4 above, which are summed to obtain total productivity loss. Using robust regression, total productivity loss was modeled as a function of the diabetes self-management programs and other identified demographic and clinical characteristics. Compared to subjects in the UC arm, there were no statistically significant differences in productivity losses among persons undergoing any of the 3 diabetes management interventions. Males were associated with higher productivity losses (+$708/year; P<0.001) and persons with greater than high school education were associated with additional productivity losses (+$758/year; P<0.001). Persons with more than 1 comorbid condition were marginally associated with lower productivity losses (-$326/year; P=0.055). No evidence was found that the chronic disease management programs examined in this trial affect indirect productivity losses.