Targeted Maximum Likelihood Estimation Research Articles

Machine learning in causal inference for epidemiology.

In causal inference, parametric models are usually employed to address causal questions estimating the effect of interest. However, parametric models rely on the correct model specification assumption that, if not met, leads to biased effect estimates. Correct model specification is challenging, especially in high-dimensional settings. Incorporating Machine Learning (ML) into causal analyses may reduce the bias arising from model misspecification, since ML methods do not require the specification of a functional form of the relationship between variables. However, when ML predictions are directly plugged in a predefined formula of the effect of interest, there is the risk of introducing a "plug-in bias" in the effect measure. To overcome this problem and to achieve useful asymptotic properties, new estimators that combine the predictive potential of ML and the ability of traditional statistical methods to make inference about population parameters have been proposed. For epidemiologists interested in taking advantage of ML for causal inference investigations, we provide an overview of three estimators that represent the current state-of-art, namely Targeted Maximum Likelihood Estimation (TMLE), Augmented Inverse Probability Weighting (AIPW) and Double/Debiased Machine Learning (DML).

The effects of hypothetical psychological interventions on alcohol use in European young adults

Abstract Background Low psychosocial well-being (PWB) and high emotion-driven impulsiveness (EDI) are associated with alcohol. Yet, it is unclear whether a hypothetical intervention targeting one or the other in adolescence might be more effective in reducing alcohol consumption (AC) in young adulthood. Therefore, we aimed to compare the separate causal effects of PWB and EDI in adolescence on AC in young adulthood. Methods We included 505 European young adults from the IDEFICS/I.Family cohort (mean age: 20.2 years; age range: 18.2-23.5 years) who did not drink alcohol at study entry. AC was operationalized as the amount of weekly consumed alcoholic beverages (mean: 4.2 drinks per week; range: 0.3-70 drinks per week). EDI was assessed using the negative urgency subscale from the UPPS-P Impulsive Behaviour Scale. PWB was assessed using the KINDLR Health-Related Quality of Life Questionnaire. Following the principles of target trial emulation, we estimated, separately, the average causal effects of PWB and EDI on AC accounting for relevant confounders and applying a semi-parametric doubly robust method (targeted maximum likelihood estimation). We stratified the results by sex and parental education. Results If all adolescents, hypothetically, had high levels of PWB, compared to low levels, we estimated a decrease in the average amount of alcoholic beverages in young adulthood by 0.1 drinks per week [95%-confidence interval: -2.3; 2.1]. Furthermore, if all adolescents had low levels of EDI, compared to high levels, we estimated an increase in alcoholic beverages in young adulthood by 1.5 drinks per week [0.1; 2.9]. Different effects for sex and parental education groups were found. Conclusions Hypothetical interventions targeting PWB in adolescents were not found to have effects on reducing AC in young adulthood. Interventions targeting EDI, however, would lead to an increase in AC. This may be due to unmeasured confounding and the missing distinction between drinking motives. Key messages • We demonstrate that causal inference methods, compared to traditional ones, improve the robustness of estimated effect measures and address important sources of bias in European cohort data. • To inform public health interventions on reducing alcohol consumption, future research should investigate different drinking motives, e.g. alleviating negative emotions or enhancing positive emotions.

HIGHLY ADAPTIVE LASSO: MACHINE LEARNING THAT PROVIDES VALID NONPARAMETRIC INFERENCE IN REALISTIC MODELS.

Understanding treatment effects on health-related outcomes using real-world data requires defining a causal parameter and imposing relevant identification assumptions to translate it into a statistical estimand. Semiparametric methods, like the targeted maximum likelihood estimator (TMLE), have been developed to construct asymptotically linear estimators of these parameters. To further establish the asymptotic efficiency of these estimators, two conditions must be met: 1) the relevant components of the data likelihood must fall within a Donsker class, and 2) the estimates of nuisance parameters must converge to their true values at a rate faster than . The Highly Adaptive LASSO (HAL) satisfies these criteria by acting as an empirical risk minimizer within a class of càdlàg functions with a bounded sectional variation norm, which is known to be Donsker. HAL achieves the desired rate of convergence, thereby guaranteeing the estimators' asymptotic efficiency. The function class over which HAL minimizes its risk is flexible enough to capture realistic functions while maintaining the conditions for establishing efficiency. Additionally, HAL enables robust inference for non-pathwise differentiable parameters, such as the conditional average treatment effect (CATE) and causal dose-response curve, which are important in precision health. While these parameters are often considered in machine learning literature, these applications typically lack proper statistical inference. HAL addresses this gap by providing reliable statistical uncertainty quantification that is essential for informed decision-making in health research.

Association of crop commercialization and rural households' multidimensional poverty using targeted maximum likelihood estimation

Recent evidence shows the highest prevalence of multidimensional poverty in the world. To alleviate poverty, Ethiopia is investing in stimulating agricultural commercialization for poverty reduction, driving sustainable economic growth, and creating jobs. Therefore, this study investigated the causal relationship between crop commercialization and multidimensional poverty across Ethiopian rural households using the Ethiopia Socioeconomic Survey 2018/19 data, which included 1129 crop commercialized and 1585 completely subsistence rural households. Moreover, this study also examined the causal effect of crop commercialization on the multidimensional poverty mediated by mobile phone ownership status. The study used the double robust method of the augmented inverse probability of treatment weighted (AIPTW) and targeted maximum likelihood estimation (TMLE) methods. The result showed that crop commercialization has an association with multidimensional poverty. Specifically, the TMLE with a default super learner model estimates were the most precise and suggest that the prevalence of risk of being multidimensionally poor among rural households who were crop commercialized is 12.2 % lower than for those who were completely subsistence households. Thus, it is important to have more robust policies like crop commercialization to target rural multidimensional poverty reduction. Moreover, the controlled direct effect estimation analyses showed that crop-commercialized rural households reduced the adjusted risk of multidimensional poverty by; 12 % for those who owned mobile phones and 7.8 % for those who did not. To ensure sustainable livelihoods for rural multidimensional poor households, it is necessary to have a multidisciplinary policy intervention to promote crop commercialization through the spread of mobile phones in rural areas.

SSRI use during acute COVID-19 and risk of long COVID among patients with depression

BackgroundLong COVID, also known as post-acute sequelae of COVID-19 (PASC), is a poorly understood condition with symptoms across a range of biological domains that often have debilitating consequences. Some have recently suggested that lingering SARS-CoV-2 virus particles in the gut may impede serotonin production and that low serotonin may drive many Long COVID symptoms across a range of biological systems. Therefore, selective serotonin reuptake inhibitors (SSRIs), which increase synaptic serotonin availability, may be used to prevent or treat Long COVID. SSRIs are commonly prescribed for depression, therefore restricting a study sample to only include patients with depression can reduce the concern of confounding by indication.MethodsIn an observational sample of electronic health records from patients in the National COVID Cohort Collaborative (N3C) with a COVID-19 diagnosis between September 1, 2021, and December 1, 2022, and a comorbid depressive disorder, the leading indication for SSRI use, we evaluated the relationship between SSRI use during acute COVID-19 and subsequent 12-month risk of Long COVID (defined by ICD-10 code U09.9). We defined SSRI use as a prescription for SSRI medication beginning at least 30 days before acute COVID-19 and not ending before SARS-CoV-2 infection. To minimize bias, we estimated relationships using nonparametric targeted maximum likelihood estimation to aggressively adjust for high-dimensional covariates.ResultsWe analyzed a sample (n = 302,626) of patients with a diagnosis of a depressive condition before COVID-19 diagnosis, where 100,803 (33%) were using an SSRI. We found that SSRI users had a significantly lower risk of Long COVID compared to nonusers (adjusted causal relative risk 0.92, 95% CI (0.86, 0.99)) and we found a similar relationship comparing new SSRI users (first SSRI prescription 1 to 4 months before acute COVID-19 with no prior history of SSRI use) to nonusers (adjusted causal relative risk 0.89, 95% CI (0.80, 0.98)).ConclusionsThese findings suggest that SSRI use during acute COVID-19 may be protective against Long COVID, supporting the hypothesis that serotonin may be a key mechanistic biomarker of Long COVID.

The Conditional Autoregressive F-Riesz Model for Realized Covariance Matrices

Abstract We introduce a new model for the dynamics of fat-tailed (realized) covariance-matrix-valued time-series using the F-Riesz distribution. The model allows for heterogeneous tail behavior across the coordinates of the covariance matrix via two vector-valued degrees of freedom parameters, thus generalizing the familiar Wishart and matrix-F distributions. We show that the filter implied by the new model is invertible and that a two-step targeted maximum likelihood estimator is consistent. Applying the new F-Riesz model to U.S. stocks, both tail heterogeneity and tail fatness turn out to be empirically relevant: they produce significant in-sample and out-of-sample likelihood increases, ex-post portfolio standard deviations that are in the global minimum variance model confidence set, and economic differences that are either in favor of the new model or competitive with a range of benchmark models.

Long-Term Dynamic Effect of Body Mass Index on Adverse Cardiovascular Outcomes with the Targeted Maximum Likelihood Estimation Method: Results from the KNOW-CKD Study

Long-Term Dynamic Effect of Body Mass Index on Adverse Cardiovascular Outcomes with the Targeted Maximum Likelihood Estimation Method: Results from the KNOW-CKD Study

Comparing methods for estimating causal treatment effects of administrative health data: A plasmode simulation study.

Estimating the causal effects of health policy interventions is crucial for policymaking but is challenging when using real-world administrative health care data due to a lack of methodological guidance. To help fill this gap, we conducted a plasmode simulation using such data from a recent policy initiative launched in a deprived urban area in Germany. Our aim was to evaluate and compare the following methods for estimating causal effects: propensity score matching, inverse probability of treatment weighting, and entropy balancing, all combined with difference-in-differences analysis, augmented inverse probability weighting, and targeted maximum likelihood estimation. Additionally, we estimated nuisance parameters using regression models and an ensemble learner called superlearner. We focused on treatment effects related to the number of physician visits, total health care cost, and hospitalization. While each approach has its strengths and weaknesses, our results demonstrate that the superlearner generally worked well for handling nuisance terms in large covariate sets when combined with doubly robust estimation methods to estimate the causal contrast of interest. In contrast, regression-based nuisance parameter estimation worked best in small covariate sets when combined with singly robust methods.

Identifying Racial Disparities in Utilization and Clinical Outcomes of Ambulatory Hip Arthroscopy: Analysis of Temporal Trends and Causal Inference via Machine Learning.

Arthroscopic diagnosis and treatment of femoroacetabular pathology has experienced significant growth in the last 30 years; nevertheless, reduced utilization of orthopaedic procedures has been observed among the underrepresented population. The purpose of this study was to examine racial differences in case incidence rates, outcomes, and complications in patients undergoing hip arthroscopy. It was hypothesized that racial and ethnic minority patients would undergo hip arthroscopy at a decreased rate compared with their White counterparts but that there would be no differences in clinical outcomes. Cross-sectional study. The State Ambulatory Surgery and Services Database and the State Emergency Department Database of New York were queried for patients undergoing hip arthroscopy between 2011 and 2017. Patients were stratified into White and racial and ethnic minority races, and intergroup comparisons were performed for utilization over time, total charges billed per encounter, 90-day emergency department (ED) visits, and revision hip arthroscopy. Temporal trends in the utilization of hip arthroscopy were identified, and racial differences in secondary outcomes were analyzed with a semiparametric method known as targeted maximum likelihood estimation (TMLE) backed by a library of machine learning algorithms. A total of 9745 patients underwent hip arthroscopy during the study period, with 1081 patients of minority race (11.1%). White patients underwent hip arthroscopy at 5.68 (95% CI, 4.98-6.48) times the incidence rate of racial and ethnic minority patients; these incidence rates grew annually at a ratio of 1.11 in White patients compared with 1.03 in racial and ethnic minority patients (P < .001). Based on the TMLE, racial and ethnic minority patients were significantly more likely to incur higher costs (P < .001) and visit the ED within 90 days (P = .049) but had negligible differences in reoperation rates at a 2-year follow-up (P = .53). Subgroup analysis identified that higher likelihood for 90-day ED admissions among racial and ethnic minority patients compared with White patients was associated with Medicare insurance (P = .002), median income in the lowest quartile (P = .012), and residence in low-income neighborhoods (P = .006). Irrespective of insurance status, racial and ethnic minority patients undergo hip arthroscopy at a lower incidence and incur higher costs per surgical encounter.

History-restricted marginal structural model and latent class growth analysis of treatment trajectories for a time-dependent outcome.

In previous work, we introduced a framework that combines latent class growth analysis (LCGA) with marginal structural models (LCGA-MSM). LCGA-MSM first summarizes the numerous time-varying treatment patterns into a few trajectory groups and then allows for a population-level causal interpretation of the group differences. However, the LCGA-MSM framework is not suitable when the outcome is time-dependent. In this study, we propose combining a nonparametric history-restricted marginal structural model (HRMSM) with LCGA. HRMSMs can be seen as an application of standard MSMs on multiple time intervals. To the best of our knowledge, we also present the first application of HRMSMs with a time-to-event outcome. It was previously noted that HRMSMs could pose interpretation problems in survival analysis when either targeting a hazard ratio or a survival curve. We propose a causal parameter that bypasses these interpretation challenges. We consider three different estimators of the parameters: inverse probability of treatment weighting (IPTW), g-computation, and a pooled longitudinal targeted maximum likelihood estimator (pooled LTMLE). We conduct simulation studies to measure the performance of the proposed LCGA-HRMSM. For all scenarios, we obtain unbiased estimates when using either g-computation or pooled LTMLE. IPTW produced estimates with slightly larger bias in some scenarios. Overall, all approaches have good coverage of the 95 % confidence interval. We applied our approach to a population of older Quebecers composed of 57,211 statin initiators and found that a greater adherence to statins was associated with a lower combined risk of cardiovascular disease or all-cause mortality.

How Does CEO Duality Influence ESG Scores in Hospitality and Tourism Companies? Confounding Roles of Governance Mechanisms and Financial Indicators

Previous studies have yielded inconsistent results about the impact of CEO duality on corporate performance in the hospitality and tourism (H&amp;T) industry. To further delve into this relationship, we investigated the causal relationship between CEO duality and environmental, social, and governance (ESG) performance under various board characteristics and financial indicators. The data from the Thomson Reuters Eikon database were evaluated using a machine learning technique that included targeted maximum likelihood estimation (TMLE), augmented inverse probability weighting (AIPW), and neural network analysis, all of which are doubly robust estimators with cross-fitting. The findings suggest that CEO duality negatively impacts environmental pillar scores but not other outcomes (i.e., governance and social pillar scores). Among the governance practices and financial indicators, policy executive compensation performance, policy executive compensation ESG performance, and return on invested capital (ROIC) have positive relations with total ESG scores. The results have important ramifications for helping H&amp;T companies develop effective boards of directors and governance systems, as well as achieve targeted ESG performance objectives.

Machine learning for causal inference: An application to ECLS-K data

This paper explores the use of machine learning for causal inference to estimate the average treatment effect of special education services on fifth-grade math scores. Causal inference is the study of the relationship between cause and effect when changes in one variable directly affect another variable. The use of machine learning techniques in causal inference problems has been growing rapidly, offering advantages over traditional methods such as propensity score matching. such as propensity score matching. This paper compares the performance of four machine learning methods: Ordinary Least Squares (OLS), Multi-Layer Perception (MLP), Targeted Maximum Likelihood Estimation (TMLE), and Bayesian Additive Regression Trees (BART) in estimating the average treatment effect of special education services on fifth-grade math scores. This study utilizes the Early Childhood Longitudinal Study, Kindergarten Class of 1998-1999 (ECLS-K) dataset. A factor analysis is conducted to identify the key variables that influence math performance, paving the way for examining their causal effects. Our results show that BART outperforms the other methods in accuracy and robustness and that receiving special education services does not have a causal effect on math scores. This paper discusses the implications and limitations of our findings and suggests directions for future study.

P-660 Finding the optimal starting dose of gonadotropins for an ovarian stimulation treatment through Machine Learning

Abstract Study question Can machine learning algorithms provide an individualized starting dose of gonadotropins to maximize the number of oocytes retrieved in patients undergoing ovarian stimulation ? Summary answer Methods of Policy Learning based on Causal Inference and Machine Learning optimize starting doses of gonadotropins with a substantial gain of oocytes under appropriate treatment. What is known already Despite attempts toward standardization, there has been no consensus about establishing the optimal starting dose of gonadotropins to be used in ovarian stimulation. In the real world, medical practice differs considerably across countries and even among fertility clinics nationally. The machine learning community has proposed a variety of models for doing this, but with limited comparisons, causal inference can provide the statistical tools to enable the best model to be chosen and allow its explicability. Study design, size, duration In a retrospective single-center study, we reviewed the data of 11,436 cycles of ovarian stimulation for IVF between 2012 and 2023. In order to estimate the causal effect of the starting dose of gonadotropins on the number of oocytes retrieved, we selected a relevant subset of confounding covariates – including age, Antral Follicle Count (AFC), Anti-Müllerian Hormone (AMH) and oestradiol (E2) rates. Participants/materials, setting, methods We used five different Machine Learning models : Linear Regression, Regression Forest, Super Learner (SL), Multi-Arm Causal Forest (MACF) and a model based on the Targeted Maximum Likelihood Estimation (TMLE). We employed these models for Policy Learning through an Outcome Regression Modeling approach. We compared the gain of oocytes obtained through the optimal policy of our different models and analyzed the variables which played a crucial role in determining the optimal dose. Main results and the role of chance We use two approaches to compare our models, the first one gives the expected average gain of oocytes under the optimal policy : in this case, the Linear Regression is the best with an average gain of 0.90 oocytes, followed by the MACF (0.80), the SL (0.58), the TMLE (0.44) and the Regression Forest (0.18). Considering the lack of heterogeneity of our database we use a second approach, an Augmented Inverse Propensity Score based value, to compare with more robustness the performance of the optimal policy of each model. In this case the MACF (an explainable model specifically designed for causal inference) is the best with a gain of 0.50 oocytes, followed by the SL (0.42), the Linear Regression (0.37), the TMLE (0.23) and the Regression Forest (0.18). In general, our models recommend lower doses, for instance 20% of the patients are recommended by the MACF to reduce their initial doses of gonadotropins by on average 100 IU. The most significant covariates in the models are the AFC, the BMI, and the rate of AMH, with p-values &amp;lt; 0.001 in the Linear Regression that has a R-squared of 0.418. Limitations, reasons for caution Full IVF history was not included in patients who had previous IVF attempts, but only data from the last previous attempt was considered. Also, the study is monocentric and more heterogeneity will be present considering other centers. Wider implications of the findings This study needs to be completed by data from different centers in France and around the world to bring diversity to the patients’ data and to the treatment prescriptions in order to improve external validity. Algorithm recommendations and their adherence need to be tested prospectively. Trial registration number not applicable

A targeted likelihood estimation comparing cefepime and piperacillin/tazobactam in critically ill patients with community-acquired pneumonia (CAP)

Cefepime and piperacillin/tazobactam are antimicrobials recommended by IDSA/ATS guidelines for the empirical management of patients admitted to the intensive care unit (ICU) with community-acquired pneumonia (CAP). Concerns have been raised about which should be used in clinical practice. This study aims to compare the effect of cefepime and piperacillin/tazobactam in critically ill CAP patients through a targeted maximum likelihood estimation (TMLE). A total of 2026 ICU-admitted patients with CAP were included. Among them, (47%) presented respiratory failure, and (27%) developed septic shock. A total of (68%) received cefepime and (32%) piperacillin/tazobactam-based treatment. After running the TMLE, we found that cefepime and piperacillin/tazobactam-based treatments have comparable 28-day, hospital, and ICU mortality. Additionally, age, PTT, serum potassium and temperature were associated with preferring cefepime over piperacillin/tazobactam (OR 1.14 95% CI [1.01–1.27], p = 0.03), (OR 1.14 95% CI [1.03–1.26], p = 0.009), (OR 1.1 95% CI [1.01–1.22], p = 0.039) and (OR 1.13 95% CI [1.03–1.24], p = 0.014)]. Our study found a similar mortality rate among ICU-admitted CAP patients treated with cefepime and piperacillin/tazobactam. Clinicians may consider factors such as availability and safety profiles when making treatment decisions.

MSR48 Utilization of High-Dimensional Propensity Score and Targeted Maximum Likelihood Estimation with Machine Learning to Improve Causal Effect Estimation in Patients with Nonvalvular Atrial Fibrillation and Hypertension

MSR48 Utilization of High-Dimensional Propensity Score and Targeted Maximum Likelihood Estimation with Machine Learning to Improve Causal Effect Estimation in Patients with Nonvalvular Atrial Fibrillation and Hypertension

Cognitive-Behavioral-Based Physical Therapy for Improving Recovery After a Traumatic Lower-Extremity Injury: The Results of a Randomized Controlled Trial.

Lower-extremity injuries can result in severe impairment and substantial years lived with a disability. Persistent pain and psychological distress are risk factors for poor long-term outcomes and negatively influence the recovery process following a traumatic injury. Cognitive-behavioral therapy (CBT) interventions have the potential to address these risk factors and subsequently improve outcomes. This study aimed to evaluate the effect of a telephone-delivered cognitive-behavioral-based physical therapy (CBPT) program on physical function, pain, and general health at 12 months after hospital discharge following lower-extremity trauma. The CBPT program was hypothesized to improve outcomes compared with an education program. A multicenter, randomized controlled trial was conducted involving 325 patients who were 18 to 60 years of age and had at least 1 acute orthopaedic injury to the lower extremity or to the pelvis or acetabulum requiring operative fixation. Patients were recruited from 6 Level-I trauma centers and were screened and randomized to the CBPT program or the education program early after hospital discharge. The primary outcome was the Patient-Reported Outcomes Measurement Information System (PROMIS) Physical Function (PF) scale. The secondary outcomes were objective physical function tests (4-square step test, timed stair ascent test, sit-to-stand test, and self-selected walking speed test), PROMIS Pain Intensity and Pain Interference, and the Veterans RAND 12-Item Health Survey. Treatment effects were calculated using targeted maximum likelihood estimation, a robust analytical approach appropriate for causal inference with longitudinal data. The mean treatment effect on the 12-month baseline change in PROMIS PF was 0.94 (95% confidence interval, -0.68 to 2.64; p = 0.23). There were also no observed differences in secondary outcomes between the intervention group and the control group. The telephone-delivered CBPT did not appear to yield any benefits for patients with traumatic lower-extremity injuries in terms of physical function, pain intensity, pain interference, or general health. Improvements were observed in both groups, which questions the utility of telephone-delivered cognitive-behavioral strategies over educational programs. Therapeutic Level I . See Instructions for Authors for a complete description of levels of evidence.

Comparison of outcomes between off-pump and on-pump coronary artery bypass graft surgery using collaborative targeted maximum likelihood estimation

There are some discrepancies about the superiority of the off-pump coronary artery bypass grafting (CABG) surgery over the conventional cardiopulmonary bypass (on-pump). The aim of this study was estimating risk ratio of mortality in the off-pump coronary bypass compared with the on-pump using a causal model known as collaborative targeted maximum likelihood estimation (C-TMLE). The data of the Tehran Heart Cohort study from 2007 to 2020 was used. A collaborative targeted maximum likelihood estimation and targeted maximum likelihood estimation, and propensity score (PS) adjustment methods were used to estimate causal risk ratio adjusting for the minimum sufficient set of confounders, and the results were compared. Among 24,883 participants (73.6% male), 5566 patients died during an average of 8.2 years of follow-up. The risk ratio estimates (95% confidence intervals) by unadjusted log-binomial regression model, PS adjustment, TMLE, and C-TMLE methods were 0.86 (0.78–0.95), 0.88 (0.80–0.97), 0.88 (0.80–0.97), and 0.87(0.85–0.89), respectively. This study provides evidence for a protective effect of off-pump surgery on mortality risk for up to 8 years in diabetic and non-diabetic patients.

Statistical methods to control for confounders in rare disease settings that use external control

ABSTRACT In the drug development for rare disease, the number of treated subjects in the clinical trial is often very small, whereas the number of external controls can be relatively large. There is no clear guidance on choosing an appropriate statistical method to control baseline confounding in this situation. To fill this gap, we conduct extensive simulations to evaluate the performance of commonly used matching and weighting methods as well as the more recently developed targeted maximum likelihood estimation (TMLE) and cardinality matching in small sample settings, mimicking the motivating data from a pediatric rare disease. Among the methods examined, the performance of coarsened exact matching (CEM) and TMLE are relatively robust under various model specifications. CEM is only feasible when the number of controls far exceeds the number of treated, whereas TMLE has better performance with less extreme treatment allocation ratios. Our simulations suggest bootstrap is useful for variance estimation in small samples after matching.

Powerless in the storm: Severe weather-driven power outages in New York State, 2017–2020

The vulnerability of the power grid to severe weather events is a critical issue as climate change is expected to increase extreme events, which can damage components of the power grid and/or lessen electrical power supply, resulting in power outages. However, largely due to an absence of granular spatiotemporal outage data, we lack a robust understanding of how severe weather-driven outages, their community impacts, and their durations distribute across space and socioeconomic vulnerability. Here, we pair hourly power outage data in electrical power operating localities (n = 1865) throughout NYS with urbanicity, CDC Social Vulnerability Index, and hourly weather (temperature, precipitation, wind speed, lightning strike, snowfall) data. We used these data to characterize the impact of extreme weather events on power outages from 2017–2020, while considering neighborhood vulnerability factors. Specifically, we assess (a) the lagged effect of severe weather on power outages, (b) common combinations of severe weather types contributing to outages, (c) the spatial distribution of the severe weather-driven outages, and (d) disparities in severe weather-driven outages by degree of community social vulnerability. We found that across NYS, 39.9% of all outages co-occurred with severe weather. However, certain regions, including eastern Queens, upper Manhattan and the Bronx of NYC, the Hudson Valley, and Adirondack regions were more burdened with severe weather-driven outages. Using targeted maximum likelihood estimation, we found that the frequency of heat-, precipitation-, and wind-driven outages disproportionately impacted vulnerable communities in NYC. When comparing durations of outages, we found that in rural regions, precipitation- and snow-driven outages lasted the longest in vulnerable communities. Under a shifting climate, anticipated increases in power outages will differentially burden communities due to regional heterogeneity in severe weather event severity, grid preparedness, and population socioeconomic profiles/vulnerabilities. As such, policymakers must consider these characteristics to inform equitable grid management and improvements.

Bridging Differences in Cohort Analyses of the Relationship between Secondhand Smoke Exposure during Pregnancy and Birth Weight: The Transportability Framework in the ECHO Program.

Estimates for the effects of environmental exposures on health outcomes, including secondhand smoke (SHS) exposure, often present considerable variability across studies. Knowledge of the reasons behind these differences can aid our understanding of effects in specific populations as well as inform practices of combining data from multiple studies. This study aimed to assess the presence of effect modification by measured sociodemographic characteristics on the effect of SHS exposure during pregnancy on birth weights that may drive differences observed across cohorts. We also aimed to quantify the extent to which differences in the cohort mean effects observed across cohorts in the Environmental influences on Child Health Outcomes (ECHO) consortium are due to differing distributions of these characteristics. We assessed the presence of effect modification and transportability of effect estimates across five ECHO cohorts in a total of 6,771 mother-offspring dyads. We assessed the presence of effect modification via gradient boosting of regression trees based on the H-statistic. We estimated individual cohort effects using linear models and targeted maximum likelihood estimation (TMLE). We then estimated transported effects from one cohort to each of the remaining cohorts using a robust nonparametric estimation approach relying on TMLE estimators and compared them to the original effect estimates for these cohorts. Observed effect estimates varied across the five cohorts, ranging from significantly lower birth weight associated with exposure [; 95% confidence interval (CI): , ] to higher birth weight with wide CIs, including the null (; 95% CI: , 99.8). Transported effect estimates only minimally explained differences in the point estimates for two out of the four cohort pairs. Our findings of weak to moderate evidence of effect modification and transportability indicate that unmeasured individual-level and contextual factors and sources of bias may be responsible for differences in the effect estimates observed across ECHO cohorts. https://doi.org/10.1289/EHP13961.