Publications

2022

Goldman-Mellor SJ, Olfson M, Schoenbaum M. Acute injury mortality and all-cause mortality following emergency department presentation for alcohol use disorder. Drug and Alcohol Dependence. 2022;236:109472.
Background. Alcohol-related morbidity and mortality have increased substantially in the U.S. Understanding the population health implications of these concerning trends, including by identifying clinical subgroups of alcohol users at increased risk for potentially preventable acute causes of mortality, is of critical importance. Methods. This retrospective cohort study used statewide, all-payer, longitudinally-linked ED patient record and mortality data from California. Participants comprised all residents presenting to a licensed ED at least once in 2009-2011 with a diagnosis of alcohol use disorder (AUD). Participants were followed for one year after index ED visit to assess acute injury (unintentional poisoning, suicide, homicide, motor vehicle crash, and fall- or fire-related injury) and all-cause mortality rates per 100,000 person-years. Age-, sex-, race/ethnicity-adjusted standardized mortality rates (SMRs) for acute injury causes of death were determined using statewide mortality data. Results. Among 437,855 patients with index non-fatal ED visits for AUD, the 12-month acute injury mortality rate was 608.6 per 100,000 (SMR=8.0; 95% CI=7.7, 8.3), and all-cause mortality was 5,700.7 per 100,000 (SMR=6.5; 95% CI=6.4, 6.6). Unintentional poisoning accounted for 46.5%, and suicide for 19.7%, of acute-injury deaths. Acute injury deaths comprised 71.7% of all-cause mortality among patients aged 10-24 years, but much lower proportions among older patients. Female AUD patients had lower rates for all mortality outcomes. Conclusions. Emergency department patients with a recognized AUD comprise a population at persistently elevated risk for mortality. Age-related AUD patient differences in common causes of death, including drug overdose and suicide, can inform the structure of future clinical interventions.
Reeves M, Bhat HS, Goldman-Mellor S. Resampling to address inequities in predictive modeling of suicide deaths. BMJ Health & Care Informatics. 2022;29(e100456). doi:doi:10.1136/ bmjhci-2021-100456 ►
Objective Improve methodology for equitable suicide death prediction when using sensitive predictors, such as race/ethnicity, for machine learning and statistical methods. Methods Train predictive models, logistic regression, naive Bayes, gradient boosting (XGBoost) and random forests, using three resampling techniques (Blind, Separate, Equity) on emergency department (ED) administrative patient records. The Blind method resamples without considering racial/ethnic group. Comparatively, the Separate method trains disjoint models for each group and the Equity method builds a training set that is balanced both by racial/ethnic group and by class. Results Using the Blind method, performance range of the models sensitivity for predicting suicide death between racial/ethnic groups (a measure of prediction inequity) was 0.47 for logistic regression, 0.37 for naive Bayes, 0.56 for XGBoost and 0.58 for random forest. By building separate models for different racial/ethnic groups or using the equity method on the training set, we decreased the range in performance to 0.16, 0.13, 0.19, 0.20 with Separate method, and 0.14, 0.12, 0.24, 0.13 for Equity method, respectively. XGBoost had the highest overall area under the curve (AUC), ranging from 0.69 to 0.79. Discussion We increased performance equity between different racial/ethnic groups and show that imbalanced training sets lead to models with poor predictive equity. These methods have comparable AUC scores to other work in the field, using only single ED administrative record data. Conclusion. We propose two methods to improve equity of suicide death prediction among different racial/ethnic groups. These methods may be applied to other sensitive characteristics to improve equity in machine learning with healthcare applications.

2021

Goldman-Mellor SJ, Hall C, Cerda M, Bhat H. Firearm suicide mortality among emergency department patients with physical health problems. Annals of Epidemiology. 2021;54:38–44. doi:10.1016/j.annepidem.2020.09.007
Purpose: Individuals with poor physical and mental health may face elevated risk for suicide, particularly suicide by firearm. Methods: This retrospective cohort study used statewide, longitudinally linked emergency department (ED) patient record and mortality data to examine 12-month incidence of firearm suicide among ED patients presenting with a range of physical health problems. Participants included all residents pre- senting to a California ED in 2009e2013 with nonfatal visits for somatic diagnoses hypothesized to in- crease suicide risk, including myocardial infarction, congestive heart failure, cerebrovascular disease, chronic obstructive pulmonary disease, diabetes, cancer, back pain, headache, joint disorder, and injuries. For each patient diagnostic group, we calculated rates of firearm suicide per 100,000 person-years and standardized mortality ratios (SMRs) relative to the demographically matched California population. Results: Firearm suicide rates per 100,000 person-years ranged from 9.6 (among patients presenting with unintentional injury) to 55.1 (patients with cancer diagnoses), with SMRs from 1.48 to 7.45 (all p
Hall C, Ha S, Yen IH, Goldman-Mellor S. Risk factors for hyperthermia mortality among emergency department patients. Annals of Epidemiology. 2021;64:90–95. doi:10.1016/j.annepidem.2021.09.009
PURPOSE: This study examines risk factors for heat-related mortality due to hyperthermia in emergency department patients, a vulnerable population. METHODS: This matched case-control study used statewide, longitudinally linked emergency department (ED) data and death records from California. Cases comprised California residents (≥18 years) who presented to a state-licensed ED and died of hyperthermia during the study period (2009-2012). For each case, up to five ED patients were randomly selected as live controls and matched on sex and age. Patients demographic characteristics and history of ED utilization for alcohol use, drug use, psychiatric disorders, heart-related conditions, chronic respiratory disease, neurodegenerative disorders, and cerebrovascular disease were assessed in relationship to hyperthermia mortality. RESULTS: Using multivariate conditional logistic regression models, hyperthermia mortality cases had higher odds of prior ED utilization for alcohol use (OR=11.16, 95% CI=3.87, 32.17) compared to controls. Cases were also more likely than controls to have Medicare insurance (OR=5.80, 95% CI=1.70, 15.15) or self-pay (OR=5.39, 95% CI=1.73, 16.79), at their most recent ED visit. CONCLUSIONS: ED patients presenting with alcohol problems may face increased risk of hyperthermia mortality. To help reduce heat-related mortality, EDs should consider interventions that target patients vulnerable to heat exposure.
Catalano R, Goldman-Mellor S, Bruckner TA, Hartig T. Sildenafil and suicide in Sweden. European Journal of Epidemiology. 2021;36(5):531–537. doi:10.1007/s10654-021-00738-4
Much theory asserts that sexual intimacy sustains mental health. Experimental tests of such theory remain rare and have not provided compelling evidence because ethical, practical, and cultural constraints bias samples and results. An epidemiologic approach would, therefore, seem indicated given the rigor the discipline brings to quasi-experimental research. For reasons that remain unclear, however, epidemiologist have largely ignored such theory despite the plausibility of the processes implicated, which engender, for example, happiness, feelings of belonging and self-worth, and protection against depression. We use an intent-to-treat design, implemented via interrupted time-series methods, to test the hypothesis that the monthly incidence of suicide, a societally important distal measure of mental health in a population, decreased among Swedish men aged 50–59 after July 2013 when patent rights to sildenafil (i.e., Viagra) ceased, prices fell, and its use increased dramatically. The test uses 102 pre, and 18 post, price-drop months. 65 fewer suicides than expected occurred among men aged 50–59 over test months following the lowering of sildenafil prices. Our findings could not arise from shared trends or seasonality, biased samples, or reverse causation. Our results would appear by chance fewer than once in 10,000 experiments. Our findings align with theory indicating that sexual intimacy reinforces mental health. Using suicide as our distal measure of mental health further implies that public health programming intended to address the drivers of self-destructive behavior should reduce barriers to intimacy in the middle-aged populations.
Margerison CE, Hettinger K, Kaestner R, Goldman-Mellor S, Gartner D. Medicaid expansion associated with some improvements in perinatal mental health. Health Affairs. 2021;40(10):1605–1611. doi:10.1377/hlthaff.2021.00776
Poor perinatal mental health is a common pregnancy-related morbidity with potentially serious impacts that extend beyond the individual to their family. A possible contributing factor to poor perin...

2020

Goldman-Mellor S, Phillips D, Brown P, Gruenewald P, a C, Wiebe D. Emergency department use and inpatient admissions and costs among adolescents with deliberate self-harm: A five-year follow-up study. Psychiatric Services. 2020;71(2). doi:10.1176/appi.ps.201900153
© 2020 American Psychiatric Association. All rights reserved. Objective: Self-harm rates among U.S. adolescents have risen substantially. Health and social outcomes among contemporary self-harming youths are infrequently tracked and poorly understood. This study investigated long-term health service utilization (emergency department [ED] visits and inpatient admissions) and inpatient costs among a recent cohort of adolescents with deliberate self-harm. Methods: This retrospective cohort study used statewide, all-payer, longitudinally linked discharge data from California. All residents ages 10-19 presenting to EDs in 2010 with deliberate self-harm (N=5,396) were compared with two control groups: a random sample of adolescent ED patients with other complaints, matched on sex, age, residential zip code, and month of index visit (general control patients, N=14,921), and matched ED patients with psychiatric complaints but no self-harm (psychiatric control patients, N=15,835). Outcomes included 5-year rates of ED visits, inpatient admissions, and inpatient costs, overall and for psychiatric and nonpsychiatric complaints separately. Results: Self-harm patients ED use, inpatient admissions, and inpatient costs were significantly higher than those of general control patients (by 39%, 81%, and 21%, respectively), when the analysis controlled for confounding demographic and utilization characteristics. Associations mostly persisted, although smaller in magnitude, in comparisons between self-harm and psychiatric control patients. Psychiatric and nonpsychiatric complaints contributed to self-harming adolescents excess health service utilization and costs. Conclusions: Deliberate self-harm among adolescents was found to be associated with long-lasting and costly patterns of health service utilization, often but not exclusively for psychiatric complaints. Future research should investigate the pathways underlying these associations and incorporate service utilization as a key patient outcome.
Olfson M, Schoenbaum M, Goldman-Mellor S. Risks of mortality following nonfatal intentional and unintentional opioid overdoses. JAMA Psychiatry. 2020;[epub ahea. doi:10.1001/jamapsychiatry.2020.1045
Following nonfatal opioid overdose, patients were at high risk of mortality from several causes. Suicide risks were greater for patients with nonfatal intentional overdoses compared with unintentional overdoses, while risks of unintentional overdose death were greater for patients with nonfatal unintentional overdoses than inten- tional overdoses. Shared increased risks for all external- cause mortality across groups support a unified self-injury conceptualization that emphasizes common underlying determinants, while differential mortality risks for suicide and unintentional overdose supports the clinical utility of distinguishing nonfatal overdoses by intent.
Phillips D, on-Moyano CL, a MC, Gruenewald P, Goldman-Mellor S. Association between unintentional injuries and self-harm among adolescent emergency department patients. General Hospital Psychiatry. 2020;64:87–92. doi:https://doi.org/10.1016/j.genhosppsych.2020.03.008
Background Unintentional injury, a leading cause of morbidity among adolescents, may also be a risk factor for deliberate self-harm. To inform clinical and public health prevention efforts in adolescent populations, we examined whether distinct subtypes of unintentional injury were differentially associated with deliberate self-harm. Methods Statewide, all-payer, individually linkable administrative data on adolescent patients presenting to any California emergency department (ED) in 2010 (n = 490,071) were used to investigate longitudinal associations between subtypes of unintentional injury and deliberate self-harm. Adolescents aged 10–19 years presenting with unintentional drug poisoning, other poisoning, fall, suffocation, or cutting/piercing injuries formed the exposure groups; adolescents presenting with unintentional strike injuries formed the primary referent group. Study patients were followed back in time (2006–2009) to compare the groups odds of a prior ED visit for deliberate self-harm, as well as forwards in time (2010–2015) to compare their risks of subsequent self-harm. Results Unintentional drug-poisoning injury was strongly associated with increased likelihood of ED visits for deliberate self-harm, assessed both retrospectively (adjusted OR = 4.52; 95% confidence interval [CI] = 3.08, 6.64) and prospectively (adjusted RR = 3.74; 95% CI = 3.03, 4.60). Positive associations with odds of prior self-harm and/or risk of subsequent self-harm were also observed for patients with unintentional non-drug poisoning, suffocation, and cutting/piercing injuries. Conclusions Certain subtypes of unintentional injury, particularly drug poisoning, are strongly associated with risk for deliberate self-harm among adolescents, a finding with implications for targeting clinical assessment and intervention in emergency department settings. More research is needed to understand the mechanisms underlying these associations.
Goldman-Mellor S, Olfson M, Lidon-Moyano C, Schoenbaum M. Mortality following nonfatal opioid and sedative/hypnotic drug overdose. American Journal of Preventive Medicine. 2020;59(1):59–67. doi:10.1016/j.amepre.2020.02.012
Introduction: Opioid and sedative/hypnotic drug overdoses are major causes of morbidity in the U.S. This study compares 12-month incidence of fatal unintentional drug overdose, suicide, and other mortality among emergency department patients presenting with nonfatal opioid or sedative/hypnotic overdose. Methods: This is a retrospective cohort study using statewide, longitudinally linked emergency department patient record and mortality data from California. Participants comprised all residents presenting to a licensed emergency department at least once in 2009–2011 with nonfatal unintentional opioid overdose, sedative/hypnotic overdose, or neither (a 5% random sample). Participants were followed for 1 year after index emergency department presentation to assess death from unintentional overdose, suicide, or other causes, ascertained using ICD-10 codes. Absolute death rates per 100,000 person years and standardized mortality ratios relative to the general population were calculated. Data were analyzed February–August 2019. Results: Following the index emergency department visit, unintentional overdose death rates per 100,000 person years were 1,863 following opioid overdose, 342 following sedative/hypnotic overdose, and 31 for reference patients without an index overdose (respective standardized mortality ratios of 106.1, 95% CI=95.2, 116.9; 24.5, 95% CI=21.3, 27.6; and 2.6, 95% CI=2.2, 3.0). Suicide mortality rates per 100,000 were 319, 174, and 32 following opioid overdose, sedative/hypnotic overdose, and reference visits, respectively. Natural causes mortality rates per 100,000 were 8,058 (opioid overdose patients), 17,301 (sedative/hypnotic overdose patients), and 3,097 (reference patients). Conclusions: Emergency department patients with nonfatal opioid or sedative/hypnotic drug overdose have exceptionally high risks of death from unintentional overdose, suicide, and other causes. Emergency department–based interventions offer potential for reducing these patients overdose and other mortality risks.