Faculty Bibliography

Anticipating food crisis outbreaks is crucial to efficiently allocate emergency relief and reduce human suffering. However, existing predictive models rely on risk measures that are often delayed, outdated, or incomplete. Using the text of 11.2 million news articles focused on food-insecure countries and published between 1980 and 2020, we leverage recent advances in deep learning to extract high-frequency precursors to food crises that are both interpretable and validated by traditional risk indicators. We demonstrate that over the period from July 2009 to July 2020 and across 21 food-insecure countries, news indicators substantially improve the district-level predictions of food insecurity up to 12 months ahead relative to baseline models that do not include text information. These results could have profound implications on how humanitarian aid gets allocated and open previously unexplored avenues for machine learning to improve decision-making in data-scarce environments.

Background: Healthy lifestyle behaviors that can prevent adverse health outcomes, including obesity, are formed in early childhood. This study describes feeding, television, and sleep behaviors among one-year-old infants and examines differences by sociodemographic factors. Methods: Caregivers of one-year-olds presenting for well care at two clinics, control sites for the Greenlight Study, were queried about feeding, television time, and sleep. Adjusted associations between sociodemographic factors and behaviors were performed by modified Poisson (binary), multinomial logistic (multi-category), or linear (continuous) regression models. Results: Of 235 one-year-olds enrolled, 81% had Medicaid, and 45% were Hispanic, 36% non-Hispanic Black, 19% non-Hispanic White. Common behaviors included 20% exclusive bottle use, 32% put to bed with bottle, mean daily juice intake of 4.1 ± 4.6 ounces, and active television time 45 ± 73 min. In adjusted analyses compared to Hispanic caregivers, non-Hispanic Black caregivers were less likely to report exclusive bottle use (odds ratio: 0.11, 95% confidence interval [CI] 0.03"“0.39), reported 2.4 ounces more juice (95% CI 1.0"“3.9), 124 min more passive television time (95% CI 60"“188), and 37 min more active television time (95% CI 10"“64). Increased caregiver education and higher income were associated with 0.4 (95% CI 0.13"“0.66) and 0.3 (95% CI 0.06"“0.55) more servings of fruits and vegetables per day, respectively. Conclusion: In a diverse sample of one-year-olds, caregivers reported few protective behaviors that reduce the risk for adverse health outcomes including obesity. Differences in behavior by race/ethnicity, income, and education can inform future interventions and policies. Future interventions should strive to create culturally effective messaging to address common adverse health behaviors.

IMPORTANCE:Active surveillance (AS) is endorsed by clinical guidelines as the preferred management strategy for low-risk prostate cancer, but its use in contemporary clinical practice remains incompletely defined. OBJECTIVE:To characterize trends over time and practice- and practitioner-level variation in the use of AS in a large, national disease registry. DESIGN, SETTING, AND PARTICIPANTS:This retrospective analysis of a prospective cohort study included men with low-risk prostate cancer, defined as prostate-specific antigen (PSA) less than 10 ng/mL, Gleason grade group 1, and clinical stage T1c or T2a, newly diagnosed between January 1, 2014, and June 1, 2021. Patients were identified in the American Urological Association (AUA) Quality (AQUA) Registry, a large quality reporting registry including data from 1945 urology practitioners at 349 practices across 48 US states and territories, comprising more than 8.5 million unique patients. Data are collected automatically from electronic health record systems at participating practices. EXPOSURES:Exposures of interest included patient age, race, and PSA level, as well as urology practice and individual urology practitioners. MAIN OUTCOMES AND MEASURES:The outcome of interest was the use of AS as primary treatment. Treatment was determined through analysis of electronic health record structured and unstructured clinical data and determination of surveillance based on follow-up testing with at least 1 PSA level remaining greater than 1.0 ng/mL. RESULTS:A total of 20 809 patients in AQUA were diagnosed with low-risk prostate cancer and had known primary treatment. The median age was 65 (IQR, 59-70) years; 31 (0.1%) were American Indian or Alaska Native; 148 (0.7%) were Asian or Pacific Islander; 1855 (8.9%) were Black; 8351 (40.1%) were White; 169 (0.8%) were of other race or ethnicity; and 10 255 (49.3%) were missing information on race or ethnicity. Rates of AS increased sharply and consistently from 26.5% in 2014 to 59.6% in 2021. However, use of AS varied from 4.0% to 78.0% at the urology practice level and from 0% to 100% at the practitioner level. On multivariable analysis, year of diagnosis was the variable most strongly associated with AS; age, race, and PSA value at diagnosis were all also associated with odds of surveillance. CONCLUSIONS AND RELEVANCE:This cohort study of AS rates in the AQUA Registry found that national, community-based rates of AS have increased but remain suboptimal, and wide variation persists across practices and practitioners. Continued progress on this critical quality indicator is essential to minimize overtreatment of low-risk prostate cancer and by extension to improve the benefit-to-harm ratio of national prostate cancer early detection efforts.

Knowledge on the role in cancer etiology of environmental exposures as pesticides is a pre-requisite for primary prevention. We review 62 epidemiological studies on exposure to pesticides and cancer risk in humans published from 2017 to 2021, with emphasis on new findings, methodological approaches, and gaps in the existing literature. While much of the recent evidence suggests causal relationships between pesticide exposure and cancer, the strongest evidence exists for acute myeloid leukemia (AML) and colorectal cancer (CRC), diseases in which the observed associations were consistent across several studies, including high quality prospective studies and those using biomarkers for exposure assessment, with some observing dose-response relationships. Though high-quality studies have been published since the IARC monograph on organophosphate insecticides in 2017, there are still gaps in the literature on carcinogenic evidence in humans for a large number of pesticides. To further knowledge, we suggest leveraging new techniques and methods to increase sensitivity and precision of exposure assessment, incorporate multi-omics data, and investigate more thoroughly exposure to chemical mixtures. There is also a strong need for better and larger population-based cohort studies that include younger and non-occupationally exposed individuals, particularly during developmental periods of susceptibility. Though the existing evidence has limitations, as always in science, there is sufficient evidence to implement policies and regulatory action that limit pesticide exposure in humans and, hence, further prevent a significant burden of cancers.

SIGNIFICANCE STATEMENT:Low muscle mass is related to frailty and increased mortality in older adults. However, muscle mass is not easily assessed in routine clinical practice. This paper describes a novel creatinine muscle index (CMI) on the basis of serum creatinine and cystatin C. CMI was moderately associated with frailty among older adults. A significantly higher proportion of individuals with weak grip strength were in the lowest tertile of CMI. The index was also associated with mortality. These results are consistent with the hypothesis that creatinine filtration may be an index of muscle mass, which may have utility in clinical practice. BACKGROUND:Low muscle mass is related to frailty and increased mortality in older adults. However, muscle mass is not easily assessed in routine clinical practice. METHODS:This study describes a novel creatinine muscle index (CMI) on the basis of serum creatinine and cystatin C in a community-based sample of older adults from the Atherosclerosis Risk in Communities Study. Analyses included 4639 participants who attended visit 5 (2011-2013) and 12,786 participants who attended visit 2 (1990-1992). CMI was defined as creatinine filtration (the product of serum creatinine times eGFR on the basis of cystatin C) and was analyzed in sex-specific tertiles. Cross-sectional associations of CMI with a frailty trichotomy, defined by the number (robust [0]/prefrail [1-2]/frail [3-5]) of five frailty components (weight loss, slowness, exhaustion, weakness, and low physical activity), were studied using polychotomous logistic regression and binary logistic regression with each frailty component. Cox regression was used to estimate associations of CMI at visit 5 and visit 2 with mortality. Models were adjusted for demographics, clinical variables, and comorbid conditions. RESULTS:CMI (tertile 1 versus 3) was moderately associated with frailty (visit 5: adjusted odds ratio 4.23 [95% confidence interval (CI), 2.02 to 8.87] in men and 2.34 [95% CI, 1.41 to 3.89] in women) and with mortality (visit 5: adjusted hazard ratio 1.45 [95% CI, 1.08 to 1.94] in men and 1.55 [95% CI, 1.13 to 2.12] in women; similar results were seen at visit 2). CONCLUSION:Lower CMI was associated with frailty and increased mortality, two clinical outcomes known to be associated with decreased muscle mass. Creatinine filtration may be an index of muscle mass and have utility in clinical practice, particularly at low levels.

SIGNIFICANCE STATEMENT:The kidney failure risk equation (KFRE) uses age, sex, GFR, and urine albumin-to-creatinine ratio (ACR) to predict 2- and 5-year risk of kidney failure in populations with eGFR <60 ml/min per 1.73 m 2 . However, the CKD-EPI 2021 creatinine equation for eGFR is now recommended for use but has not been fully tested in the context of KFRE. In 59 cohorts comprising 312,424 patients with CKD, the authors assessed the predictive performance and calibration associated with the use of the CKD-EPI 2021 equation and whether additional variables and accounting for the competing risk of death improves the KFRE's performance. The KFRE generally performed well using the CKD-EPI 2021 eGFR in populations with eGFR <45 ml/min per 1.73 m 2 and was not improved by adding the 2-year prior eGFR slope and cardiovascular comorbidities. BACKGROUND:The kidney failure risk equation (KFRE) uses age, sex, GFR, and urine albumin-to-creatinine ratio (ACR) to predict kidney failure risk in people with GFR <60 ml/min per 1.73 m 2 . METHODS:Using 59 cohorts with 312,424 patients with CKD, we tested several modifications to the KFRE for their potential to improve the KFRE: using the CKD-EPI 2021 creatinine equation for eGFR, substituting 1-year average ACR for single-measure ACR and 1-year average eGFR in participants with high eGFR variability, and adding 2-year prior eGFR slope and cardiovascular comorbidities. We also assessed calibration of the KFRE in subgroups of eGFR and age before and after accounting for the competing risk of death. RESULTS:The KFRE remained accurate and well calibrated overall using the CKD-EPI 2021 eGFR equation. The other modifications did not improve KFRE performance. In subgroups of eGFR 45-59 ml/min per 1.73 m 2 and in older adults using the 5-year time horizon, the KFRE demonstrated systematic underprediction and overprediction, respectively. We developed and tested a new model with a spline term in eGFR and incorporating the competing risk of mortality, resulting in more accurate calibration in those specific subgroups but not overall. CONCLUSIONS:The original KFRE is generally accurate for eGFR <45 ml/min per 1.73 m 2 when using the CKD-EPI 2021 equation. Incorporating competing risk methodology and splines for eGFR may improve calibration in low-risk settings with longer time horizons. Including historical averages, eGFR slopes, or a competing risk design did not meaningfully alter KFRE performance in most circumstances.

BACKGROUND:Maternal antenatal depression experienced around conception or during pregnancy may adversely affect child development. This study explores three potential mechanisms of the effects of antenatal depression on children's developmental delays at 2-3 years: gestational age of the child, continued depressive symptoms postnatally, and interrupted breastfeeding practices. METHODS:Mothers (N = 2888) of 3450 children, including 2303 singletons and 1147 multiples from the Upstate KIDS cohort provided data. Linked hospital discharge data was combined with mothers' reports to identify women with moderate to severe antenatal depression. Gestational age was extracted from birth certificates. Mothers completed a depression screener at 4 months postpartum, reported about their breastfeeding practices from 4 to 12 months postpartum, and completed a developmental delay screener when children were 24, 30, and 36 months. RESULTS:In unadjusted path analysis models, mothers with antenatal depression had more postnatal depressive symptoms and breastfed fewer months, which translated into children being more likely to have developmental delays. Gestational age was not a mediator. Effects were similar across girls and boys and singletons and twins, and largely held when adjusting for covariates. LIMITATIONS/CONCLUSIONS:Main limitations were the relatively advantaged sample and reliance on maternal report. CONCLUSIONS:Maternal antenatal depression may impact child development through continued depressive symptoms in the postpartum period and through reduced breastfeeding duration suggesting additional targets for intervention.

IMPORTANCE:Head injury is associated with significant short-term morbidity and mortality. Research regarding the implications of head injury for long-term survival in community-dwelling adults remains limited. OBJECTIVE:To evaluate the association of head injury with long-term all-cause mortality risk among community-dwelling adults, with consideration of head injury frequency and severity. DESIGN, SETTING, AND PARTICIPANTS:This cohort study included participants with and without head injury in the Atherosclerosis Risk in Communities (ARIC) study, an ongoing prospective cohort study with follow-up from 1987 through 2019 in 4 US communities in Minnesota, Maryland, North Carolina, and Mississippi. Of 15 792 ARIC participants initially enrolled, 1957 were ineligible due to self-reported head injury at baseline; 103 participants not of Black or White race and Black participants at the Minnesota and Maryland field centers were excluded due to race-site aliasing; and an additional 695 participants with missing head injury date or covariate data were excluded, resulting in 13 037 eligible participants. EXPOSURES:Head injury frequency and severity, as defined via self-report in response to interview questions and via hospital-based International Classification of Diseases diagnostic codes (with head injury severity defined in the subset of head injury cases identified using these codes). Head injury was analyzed as a time-varying exposure. MAIN OUTCOMES AND MEASURES:All-cause mortality was ascertained via linkage to the National Death Index. Data were analyzed between August 5, 2021, and October 23, 2022. RESULTS:More than one-half of participants were female (57.7%; 42.3% men), 27.9% were Black (72.1% White), and the median age at baseline was 54 years (IQR, 49-59 years). Median follow-up time was 27.0 years (IQR, 17.6-30.5 years). Head injuries occurred among 2402 participants (18.4%), most of which were classified as mild. The hazard ratio (HR) for all-cause mortality among individuals with head injury was 1.99 (95% CI, 1.88-2.11) compared with those with no head injury, with evidence of a dose-dependent association with head injury frequency (1 head injury: HR, 1.66 [95% CI, 1.56-1.77]; 2 or more head injuries: HR, 2.11 [95% CI, 1.89-2.37]) and severity (mild: HR, 2.16 [95% CI, 2.01-2.31]; moderate, severe, or penetrating: HR, 2.87 [95% CI, 2.55-3.22]). Estimates were similar by sex and race, with attenuated associations among individuals aged 54 years or older at baseline. CONCLUSIONS AND RELEVANCE:In this community-based cohort with more than 3 decades of longitudinal follow-up, head injury was associated with decreased long-term survival time in a dose-dependent manner, underscoring the importance of measures aimed at prevention and clinical interventions to reduce morbidity and mortality due to head injury.