Detecting intimate partner violence circumstance for suicide: development and validation of a tool using natural language processing and supervised machine learning in the National Violent Death Reporting System.

BACKGROUND: Intimate partner violence (IPV) victims and perpetrators often report suicidal ideation, yet there is no comprehensive national dataset that allows for an assessment of the connection between IPV and suicide. The National Violent Death Reporting System (NVDRS) captures IPV circumstances for homicide-suicides (<2% of suicides), but not single suicides (suicide unconnected to other violent deaths; >98% of suicides). OBJECTIVE: To facilitate a more comprehensive understanding of the co-occurrence of IPV and suicide, we developed and validated a tool that detects mentions of IPV circumstances (yes/no) for single suicides in NVDRS death narratives. METHODS: We used 10 000 hand-labelled single suicide cases from NVDRS (2010-2018) to train (n=8500) and validate (n=1500) a classification model using supervised machine learning. We used natural language processing to extract relevant information from the death narratives within a concept normalisation framework. We tested numerous models and present performance metrics for the best approach. RESULTS: Our final model had robust sensitivity (0.70), specificity (0.98), precision (0.72) and kappa values (0.69). False positives mostly described other family violence. False negatives used vague and heterogeneous language to describe IPV, and often included abusive suicide threats. IMPLICATIONS: It is possible to detect IPV circumstances among singles suicides in NVDRS, although vague language in death narratives limited our tool's sensitivity. More attention to the role of IPV in suicide is merited both during the initial death investigation processes and subsequent NVDRS reporting. This tool can support future research to inform targeted prevention.

Patient Characteristics Associated with Access to Minimally Invasive Gynecologic Surgery: Changes during the COVID-19 Pandemic.

STUDY OBJECTIVE: To evaluate patient characteristics that affect access to minimally invasive gynecologic surgery (MIGS) subspecialty care and identify changes during the coronavirus disease 2019 pandemic. DESIGN: Retrospective cohort study of patients referred to MIGS from 2014 to 2016 (historic cohort) compared with those referred to MIGS in 2020 (pandemic cohort). Primary outcome was the interval between referral and first appointment. SETTING: Single-institution academic MIGS division. PATIENTS: Historic cohort (n = 1082) and pandemic cohort (n = 770). INTERVENTIONS: Not applicable. MEASUREMENTS AND MAIN RESULTS: Demographics and socioeconomic variables (race, ethnicity, language, insurance, employment, and socioeconomic factors by census tract) and distance from hospital were compared between historic and pandemic cohorts with respect to referral interval using the chi-square, Fisher exact tests, and logistic regression. After adjusting for referral indication, being unemployed and living in an area with less population density, less education, and higher percentage of poverty were associated with a referral interval >30 days in the historic cohort. In the pandemic cohort, only unemployment persisted as a covariate associated with prolonged referral interval and new associated variables were primary language other than English (odds ratio, 3.20; 95% confidence interval [CI], 1.60-6.40) and "other" race (odds ratio, 2.22; 95% CI, 1.34-3.68). The odds of waiting >30 days increased by 6% with the addition of 1 demographic risk factor (95% CI, 1.01-1.10) and by 17% for 3 risk factors (95% CI, 1.03-1.34) in the historic cohort whereas no significant intersectionality was identified in the pandemic cohort. Average referral intervals were significantly shorter during the pandemic (31 vs 50 days, p <.01). Telemedicine appointments had a significantly shorter referral interval than in-person appointments (27 vs 47 days, p <.01). Of patients using telemedicine, a greater proportion were non-Hispanic, English speaking, employed, privately insured, and lived further from the hospital (p <.05). CONCLUSION: Time from referral to first appointment at a tertiary-care MIGS practice during the coronavirus disease 2019 pandemic was shorter than that before the pandemic, likely owing to the adoption of telemedicine. Differences in socioeconomic and demographic factors suggest that telemedicine improved access to care and decreased access disparities for many populations, but not for non-English-speaking patients.

How do we assess a racial disparity in health? Distribution, interaction, and interpretation in epidemiological studies.

Identifying the exposures or interventions that exacerbate or ameliorate racial health disparities is one of the fundamental goals of social epidemiology. Introducing an interaction term between race and an exposure into a statistical model is commonly used in the epidemiologic literature to assess racial health disparities and the potential viability of a targeted health intervention. However, researchers may attribute too much authority to the interaction term and inadvertently ignore other salient information regarding the health disparity. In this article, we highlight empirical examples from the literature demonstrating limitations of overreliance on interaction terms in health disparities research; we further suggest approaches for moving beyond interaction terms when assessing these disparities. We promote a comprehensive framework of three guiding questions for disparity investigation, suggesting examination of the group-specific differences in (1) outcome prevalence, (2) exposure prevalence, and (3) effect size. Our framework allows for better assessment of meaningful differences in population health and the resulting implications for interventions, demonstrating that interaction terms alone do not provide sufficient means for determining how disparities arise. The widespread adoption of this more comprehensive approach has the potential to dramatically enhance understanding of the patterning of health and disease and the drivers of health disparities.

Child Maltreatment Surveillance Improvement Opportunities: A Wake County, North Carolina Pilot Project.

BACKGROUND As child maltreatment often occurs in private, child welfare numbers underestimate its true prevalence. Child maltreatment surveillance systems have been used to ascertain more accurate counts of children who experience maltreatment. This manuscript describes the results from a pilot child maltreatment surveillance system in Wake County, North Carolina.METHODS We linked 2010 and 2011 data from 3 sources (Child Protective Services, Raleigh Police Department, and Office of the Chief Medical Examiner) to obtain rates of definite and possible child maltreatment. We separately analyzed emergency department visits from 2010 and 2011 to obtain counts of definite and possible child maltreatment. We then compared the results from the surveillance systems to those obtained from Child Protective Services (CPS) data alone.RESULTS In 2010 and 2011, rates of definite child maltreatment were 11.7 and 11.3 per 1,000 children, respectively, when using the linked data, compared to 10.0 and 9.5 per 1,000 children using CPS data alone. The rates of possible maltreatment were 25.3 and 23.8 per 1,000, respectively. In the 2010 and 2011 emergency department data, there were 68 visits and 84 visits, respectively, that met the case definition for maltreatment.LIMITATIONS While 4 data sources were analyzed, only 3 were linked in the current surveillance system. It is likely that we would have identified more cases of maltreatment had more sources been included.CONCLUSION While the surveillance system identified more children who met the case definition of maltreatment than CPS data alone, the rates of definite child maltreatment were not considerably higher than official reports. Rates of possible child maltreatment were much higher than both the definite case definition and child welfare records. Tracking both definite and possible case definitions and using a variety of data sources provides a more complete picture of child maltreatment in North Carolina.