scholarly journals Adversarial balancing-based representation learning for causal effect inference with observational data

2021 ◽  
Author(s):  
Xin Du ◽  
Lei Sun ◽  
Wouter Duivesteijn ◽  
Alexander Nikolaev ◽  
Mykola Pechenizkiy
Keyword(s):  
Health Care ◽  
Observational Data ◽  
Causal Effect ◽  
Representation Learning ◽  
Average Treatment Effect ◽  
Treatment Selection ◽  
Causal Effects ◽  
Control Group ◽  
Confounding Bias ◽  
The Impact

AbstractLearning causal effects from observational data greatly benefits a variety of domains such as health care, education, and sociology. For instance, one could estimate the impact of a new drug on specific individuals to assist clinical planning and improve the survival rate. In this paper, we focus on studying the problem of estimating the Conditional Average Treatment Effect (CATE) from observational data. The challenges for this problem are two-fold: on the one hand, we have to derive a causal estimator to estimate the causal quantity from observational data, in the presence of confounding bias; on the other hand, we have to deal with the identification of the CATE when the distributions of covariates over the treatment group units and the control units are imbalanced. To overcome these challenges, we propose a neural network framework called Adversarial Balancing-based representation learning for Causal Effect Inference (ABCEI), based on recent advances in representation learning. To ensure the identification of the CATE, ABCEI uses adversarial learning to balance the distributions of covariates in the treatment and the control group in the latent representation space, without any assumptions on the form of the treatment selection/assignment function. In addition, during the representation learning and balancing process, highly predictive information from the original covariate space might be lost. ABCEI can tackle this information loss problem by preserving useful information for predicting causal effects under the regularization of a mutual information estimator. The experimental results show that ABCEI is robust against treatment selection bias, and matches/outperforms the state-of-the-art approaches. Our experiments show promising results on several datasets, encompassing several health care (and other) domains.

You Can’t Drive a Car With Only Three Wheels

2019 ◽  
Vol 188 (9) ◽  
pp. 1682-1685 ◽  
Author(s):  
Hailey R Banack
Keyword(s):  
Causal Inference ◽  
Observational Data ◽  
Causal Effect ◽  
Causal Effects ◽  
Measurement Bias ◽  
Exposure Misclassification ◽  
Unmeasured Confounding ◽  
Obstetrical Care ◽  
The Impact ◽  
Fundamental Requirement

Abstract Authors aiming to estimate causal effects from observational data frequently discuss 3 fundamental identifiability assumptions for causal inference: exchangeability, consistency, and positivity. However, too often, studies fail to acknowledge the importance of measurement bias in causal inference. In the presence of measurement bias, the aforementioned identifiability conditions are not sufficient to estimate a causal effect. The most fundamental requirement for estimating a causal effect is knowing who is truly exposed and unexposed. In this issue of the Journal, Caniglia et al. (Am J Epidemiol. 2019;000(00):000–000) present a thorough discussion of methodological challenges when estimating causal effects in the context of research on distance to obstetrical care. Their article highlights empirical strategies for examining nonexchangeability due to unmeasured confounding and selection bias and potential violations of the consistency assumption. In addition to the important considerations outlined by Caniglia et al., authors interested in estimating causal effects from observational data should also consider implementing quantitative strategies to examine the impact of misclassification. The objective of this commentary is to emphasize that you can’t drive a car with only three wheels, and you also cannot estimate a causal effect in the presence of exposure misclassification bias.

Propensity score specification for optimal estimation of average treatment effect with binary response

2020 ◽  
Vol 29 (12) ◽  
pp. 3623-3640
Author(s):  
John A Craycroft ◽  
Jiapeng Huang ◽  
Maiying Kong
Keyword(s):  
Propensity Score ◽  
Treatment Effect ◽  
Propensity Scores ◽  
Optimal Estimation ◽  
Theoretical Explanation ◽  
Average Treatment Effect ◽  
Adaptive Lasso ◽  
Confounding Bias ◽  
Average Treatment ◽  
The Impact

Propensity score methods are commonly used in statistical analyses of observational data to reduce the impact of confounding bias in estimations of average treatment effect. While the propensity score is defined as the conditional probability of a subject being in the treatment group given that subject’s covariates, the most precise estimation of average treatment effect results from specifying the propensity score as a function of true confounders and predictors only. This property has been demonstrated via simulation in multiple prior research articles. However, we have seen no theoretical explanation as to why this should be so. This paper provides that theoretical proof. Furthermore, this paper presents a method for performing the necessary variable selection by means of elastic net regression, and then estimating the propensity scores so as to obtain optimal estimates of average treatment effect. The proposed method is compared against two other recently introduced methods, outcome-adaptive lasso and covariate balancing propensity score. Extensive simulation analyses are employed to determine the circumstances under which each method appears most effective. We applied the proposed methods to examine the effect of pre-cardiac surgery coagulation indicator on mortality based on a linked dataset from a retrospective review of 1390 patient medical records at Jewish Hospital (Louisville, KY) with the Society of Thoracic Surgeons database.

scholarly journals The Impact of Online Computer Assisted Learning at Home for Disadvantaged Children in Taiwan: Evidence from a Randomized Experiment

2020 ◽  
Vol 12 (23) ◽  
pp. 10092
Author(s):  
Bin Tang ◽  
Te-Tien Ting ◽  
Chyi-In Wu ◽  
Yue Ma ◽  
Di Mo ◽  
...  
Keyword(s):  
Academic Performance ◽  
Preliminary Evidence ◽  
Average Treatment Effect ◽  
Control Group ◽  
Computer Assisted ◽  
Computer Assisted Learning ◽  
Intention To Treat ◽  
Catch Up ◽  
Standard Deviations ◽  
The Impact

In Taiwan, thousands of students from Yuanzhumin (aboriginal) families lag far behind their Han counterparts in academic achievement. When they fall behind, they often have no way to catch up. There is increased interest among both educators and policymakers in helping underperforming students catch up using computer-assisted learning (CAL). The objective of this paper is to examine the impact of an intervention aimed at raising the academic performance of students using an in-home CAL program. According to intention-to-treat estimates, in-home CAL improved the overall math scores of students in the treatment group relative to the control group by 0.08 to 0.20 standard deviations (depending on whether the treatment was for one or two semesters). Furthermore, Average Treatment Effect on the Treated analysis was used for solving the compliance problem in our experiment, showing that in-home CAL raised academic performance by 0.36 standard deviations among compliers. This study thus presents preliminary evidence that an in-home CAL program has the potential to boost the learning outcomes of disadvantaged students.

scholarly journals Retrospective observational study of the impact on emergency admission of telehealth at scale delivered in community care in Liverpool, UK

BMJ Open ◽  
2019 ◽  
Vol 9 (7) ◽  
pp. e028981
Author(s):  
Cees van Berkel ◽  
Peter Almond ◽  
Carol Hughes ◽  
Maurice Smith ◽  
Dave Horsfield ◽  
...  
Keyword(s):  
Observational Study ◽  
Subgroup Analysis ◽  
Intervention Group ◽  
Average Treatment Effect ◽  
Retrospective Observational Study ◽  
Control Group ◽  
Chronic Obstructive ◽  
Emergency Admissions ◽  
Percentage Decrease ◽  
The Impact

ObjectiveTo assess the effect of a real world, ongoing telehealth service on the use of secondary healthcare.DesignA retrospective observational study with anonymous matched controls.SettingPrimary and community healthcare. Patients were recruited over 4 years in 89 general practices in Liverpool, UK and remotely managed by a dedicated clinical team in Liverpool Community Health.Participants5154 patients with chronic obstructive pulmonary disease, heart failure or diabetes were enrolled in the programme, of whom 3562 satisfied the inclusion criteria of this study.InterventionAt least 9 weeks of telehealth including vital sign collection, questionnaires, education, support and informal coaching by clinical staff.Primary outcomeReduction in the number of emergency admissions in the 12 months after start, compared with the year before start. Secondary subgroup analysis to improve future targeting and personalisation of the service.ResultThe average number of emergency admissions for the intervention group at baseline is 0.35, 95% CI 0.32 to 0.38. The differential decrease in emergency admissions in the intervention group in comparison with the control group, the average treatment effect, is 0.08, 95 CI 0.05 to 0.11, corresponding to an average percentage decrease of 22.7%. In subgroup analysis, a score is calculated that can be used prospectively to predict individual benefit from the intervention. Patients with an above median score (37%) are predicted average reduction in emergency admissions of 0.15, 95% CI 0.09 to 0.2, corresponding to a percentage decrease in admissions of 25.3%.ConclusionThe telehealth intervention has a positive impact across a wide cohort of patients with different diseases. Prospective scoring of patients and allocation to targeted telehealth interventions is likely to improve the effectiveness and efficiency of the service.

scholarly journals Using register data to estimate causal effects of interventions: An ex post synthetic control-group approach

2017 ◽  
Vol 45 (17_suppl) ◽  
pp. 50-55 ◽  
Author(s):  
Magnus Bygren ◽  
Ryszard Szulkin
Keyword(s):  
Observational Data ◽  
Treatment Effect ◽  
Job Training ◽  
Panel Data Analysis ◽  
Register Data ◽  
Causal Effects ◽  
Control Group ◽  
Synthetic Control ◽  
Regression To The Mean ◽  
Ex Post

Aims: It is common in the context of evaluations that participants have not been selected on the basis of transparent participation criteria, and researchers and evaluators many times have to make do with observational data to estimate effects of job training programs and similar interventions. The techniques developed by researchers in such endeavours are useful not only to researchers narrowly focused on evaluations, but also to social and population science more generally, as observational data overwhelmingly are the norm, and the endogeneity challenges encountered in the estimation of causal effects with such data are not trivial. The aim of this article is to illustrate how register data can be used strategically to evaluate programs and interventions and to estimate causal effects of participation in these. Methods: We use propensity score matching on pretreatment-period variables to derive a synthetic control group, and we use this group as a comparison to estimate the employment-treatment effect of participation in a large job-training program. Results: We find the effect of treatment to be small and positive but transient. Conclusions: Our method reveals a strong regression to the mean effect, extremely easy to interpret as a treatment effect had a less advanced design been used (e.g. a within-subjects panel data analysis), and illustrates one of the unique advantages of using population register data for research purposes.

scholarly journals Interpretation of mendelian randomization using one measure of an exposure that varies over time.

2021 ◽  
Author(s):  
Tim T Morris ◽  
Jon Heron ◽  
Eleanor Sanderson ◽  
George Davey Smith ◽  
Kate Tilling
Keyword(s):  
Blood Pressure ◽  
Systolic Blood Pressure ◽  
Observational Data ◽  
Genetic Variant ◽  
Mendelian Randomization ◽  
Causal Effect ◽  
Causal Effects ◽  
Time Points ◽  
The Life Course ◽  
Over Time

Background Mendelian randomization (MR) is a powerful tool through which the causal effects of modifiable exposures on outcomes can be estimated from observational data. Most exposures vary throughout the life course, but MR is commonly applied to one measurement of an exposure (e.g., weight measured once between ages 40 and 60). It has been argued that MR provides biased causal effect estimates when applied to one measure of an exposure that varies over time. Methods We propose an approach that emphasises the liability that causes the entire exposure trajectory. We demonstrate this approach using simulations and an applied example. Results We show that rather than estimating the direct or total causal effect of changing the exposure value at a given time, MR estimates the causal effect of changing the liability as induced by a specific genotype that gives rise to the exposure at that time. As such, results from MR conducted at different time points are expected to differ (unless the liability of exposure is constant over time), as we demonstrate by estimating the effect of BMI measured at different ages on systolic blood pressure. Conclusions Practitioners should not interpret MR results as timepoint-specific direct or total causal effects, but as the effect of changing the liability that causes the entire exposure trajectory. Estimates of how the effects of a genetic variant on an exposure vary over time are needed to interpret timepoint-specific causal effects.

scholarly journals An Assessment of Impact of Leadership Training on Health System Performance in Selected Counties in Kenya

2021 ◽  
Vol 8 ◽  
Author(s):  
Tecla Chelagat ◽  
James Rice ◽  
Joseph Onyango ◽  
Gilbert Kokwaro
Keyword(s):  
Health Care ◽  
Health System ◽  
Leadership Training ◽  
Performance Indicators ◽  
System Performance ◽  
Individual Development ◽  
Control Group ◽  
Health System Performance ◽  
For Profit ◽  
The Impact

Introduction: The provision of health care services in Kenya was devolved from the national government to the counties in 2013. Evidence suggests that health system performance in Kenya remains poor. The main issue is poor leadership resulting in poor health system performance. However, most training in Kenya focuses on “leaders” (individual) development as opposed to “leadership” training (development of groups from an organization). The purpose of that study was to explore the impact of leadership training on health system performance in selected counties in Kenya.Methods: A quasi-experimental time-series design was employed. Pretest, posttest control-group design was utilized to find out whether the leadership development program positively contributed to the improvement of health system performance indicators compared with the non-trained managers. Questionnaires were administered to 31 trained health managers from the public, private for-profit, and private not-for-profit health institutions within the same counties.Results: The pretest and posttest means for all the six health system (HS) pillar indicators of the treatment group were higher than those of the control group. The regression method to estimate the DID structural model used to calculate the “fact” and “counterfactual” revealed that training had a positive impact on the intended outcome on the service delivery, information, leadership and governance, human resources, finance, and medical products with impact value ≥1 (57.2).Conclusion: The study findings support both hypotheses that trained health care management teams had a significant difference in the implementation status of priority projects and, hence, had a significant impact on health system performance indicators compared with non-trained managers.

scholarly journals 1218Estimating the causal effect of mobility on Dengue during the COVID-19 pandemic

2021 ◽  
Vol 50 (Supplement_1) ◽  
Author(s):  
Kirstin Roster ◽  
Colm Connaughton ◽  
Francisco A Rodrigues
Keyword(s):  
Causal Effect ◽  
Experimental Situation ◽  
Human Mobility ◽  
Sao Paulo ◽  
Average Treatment Effect ◽  
São Paulo ◽  
Paulo State ◽  
Quasi Experimental ◽  
São Paulo State ◽  
The Impact

Abstract Background The COVID-19 pandemic led to a reduction in human mobility which occurred randomly (in time) and is not linked to any other Dengue risk factors. This gives rise to a quasi-experimental situation to assess the impact of mobility reduction on Dengue Fever in Brazilian cities using propensity score matching. Methods We match weeks during the peak pandemic period for 37 cities in São Paulo state with comparable prior periods based on instruments for mosquito population size and human susceptibility. By matching within cities, we also control for city-level characteristics, such as landscape or population density. We compute propensity scores using logistic regression and Random Forests and implement both one-to-one and one-to-many matching with calipers. Results We compare the Sample Average Treatment Effect on the Treated (SATT) across models and find variation in the direction of the causal effect. In 12 cities, mobility reductions are linked to more Dengue cases, while fewer cases are reported in 9 cities. The remaining cities are sensitive to the model chosen. Conclusions The SATT of mobility on Dengue varies across the cities in our sample, with more cities experiencing an increase in cases during the COVID-19 pandemic. Key messages A quasi-experimental analysis suggests that there is a a causal effect of mobility on Dengue that varies across cities in São Paulo state.

The Journey of Recovery—Caregivers’ Perspectives From a Hip Fracture Telerehabilitation Clinical Trial

2020 ◽  
Author(s):  
Patrocinio Ariza-Vega ◽  
Herminia Castillo-Pérez ◽  
Mariana Ortiz-Piña ◽  
Lena Ziden ◽  
Jerónimo Palomino-Vidal ◽  
...  
Keyword(s):  
Older Adults ◽  
Health Care ◽  
Clinical Trial ◽  
Hip Fracture ◽  
Family Member ◽  
Family Caregivers ◽  
Public Health Care ◽  
Control Group ◽  
The Impact ◽  
At Home

Abstract Objective To explore family caregivers’ perspectives of the recovery process of older adults with hip fracture, and describe experiences from caregivers who (i) used the online intervention or (ii) received home-based care provided by the Andalusian Public Health Care System. Methods This was an exploratory secondary study with informal family caregivers who had an older adult family member with hip fracture enrolled in a novel telerehabilitation (telerehab) clinical trial. Forty-four caregivers of older adults with hip fracture were interviewed at 6 to 9 months after their family member’s hip fracture. Results Caregivers shared concerns of family members’ survival and recovery; they recounted increased stress and anxiety due to the uncertainty of new tasks associated with providing care and the impact on their lifestyle. Although most caregivers were satisfied with the health care received, they made suggestions for better organization of hospital discharge, and requests for home support. The main reasons why caregivers and their family member chose the telerehab program were to, enhance recovery after fracture, gain knowledge for managing at home, and the convenience of completing the exercises at home. There were more family caregivers in the control group who expressed a high level of stress and anxiety, and they also requested more social and health services compared with caregivers whose family member received telerehab. Conclusions Family caregivers are an essential component of recovery after hip fracture by providing emotional and physical support. However, future clinical interventions should evaluate person-centered interventions to mitigate possible stress and anxiety experienced by family caregivers. Impact Family caregivers’ perspectives are necessary in the co-design of management strategies for older adults after hip fracture.

Analyzing Individual and Average Causal Effects via Structural Equation Models

Methodology ◽  
2005 ◽  
Vol 1 (1) ◽  
pp. 39-54 ◽  
Author(s):  
Rolf Steyer
Keyword(s):  
Measurement Error ◽  
Untreated Control ◽  
Structural Equation ◽  
Structural Equation Models ◽  
Causal Effect ◽  
Latent Trait ◽  
Causal Effects ◽  
Control Group ◽  
Expected Outcome ◽  
The Individual

Abstract. Although both individual and average causal effects are defined in Rubin's approach to causality, in this tradition almost all papers center around learning about the average causal effects. Almost no efforts deal with developing designs and models to learn about individual effects. This paper takes a first step in this direction. In the first and general part, Rubin's concepts of individual and average causal effects are extended replacing Rubin's deterministic potential-outcome variables by the stochastic expected-outcome variables. Based on this extension, in the second and main part specific designs, assumptions and models are introduced which allow identification of (1) the variance of the individual causal effects, (2) the regression of the individual causal effects on the true scores of the pretests, (3) the regression of the individual causal effects on other explanatory variables, and (4) the individual causal effects themselves. Although random assignment of the observational unit to one of the treatment conditions is useful and yields stronger results, much can be achieved with a nonequivalent control group. The simplest design requires two pretests measuring a pretest latent trait that can be interpreted as the expected outcome under control, and two posttests measuring a posttest latent trait: The expected outcome under treatment. The difference between these two latent trait variables is the individual-causal-effect variable, provided some assumptions can be made. These assumptions - which rule out alternative explanations in the Campbellian tradition - imply a single-trait model (a one-factor model) for the untreated control condition in which no treatment takes place, except for change due to measurement error. These assumptions define a testable model. More complex designs and models require four occasions of measurement, two pretest occasions and two posttest occasions. The no-change model for the untreated control condition is then a single-trait-multistate model allowing for measurement error and occasion-specific effects.

Sign in / Sign up

Export Citation Format

Share Document