Doubly robust adaptive LASSO for effect modifier discovery

Effect modification occurs when the effect of a treatment on an outcome differsaccording to the level of some pre-treatment variable (the effect modifier). Assessing an effect modifier is not a straight-forward task even for a subject matter expert. In this paper, we propose a two-stageprocedure to automatically selecteffect modifying variables in a Marginal Structural Model (MSM) with a single time point exposure based on the two nuisance quantities (the conditionaloutcome expectation and propensity score). We highlight the performance of our proposal in a simulation study. Finally, to illustrate tractability of our proposed methods, we apply them to analyze a set of pregnancy data. We estimate the conditional expected difference in the counterfactual birth weight if all women were exposed to inhaled corticosteroids during pregnancy versus the counterfactual birthweight if all women were not, using data from asthma medications during pregnancy.

An Exponential Endogenous Switching Regression with Correlated Random Coefficients

This paper presents a method for estimating the average treatment effects (ATE) of an exponential endogenous switching model where the coefficients of covariates in the structural equation are random and correlated with the binary treatment variable. The estimating equations are derived under some mild identifying assumptions. We find that the ATE is identified, although each coefficient in the structural model may not be. Tests assessing the endogeneity of treatment and for model selection are provided. Monte Carlo simulations show that, in large samples, the proposed estimator has a smaller bias and a larger variance than the methods that do not take the random coefficients into account. This is applied to health insurance data of Oregon.

Global vs local: analysis of the consumer-brand relationships in India

PurposeThe purpose of this study is to analyze two global brands (i.e. Benetton and Tommy Hilfiger) and one Indian brand (i.e. Wills Lifestyle) in terms of general brand impression, brand specific associations and brand commitment. In addition, the study investigates how the regional differences in India and Indian consumers' affinity towards global brands influence the consumer-brand relationships.Design/methodology/approachThe research framework has been developed based on consumer-brand relationship theory. The consumer–brand relationship is an important indicator of the success of brands, especially when brands attempt to expand to other markets (Roper and Parker, 2006; Bastos and Levy, 2012). Three brand types were chosen for this study. The choice of the US global brand is Tommy Hilfiger, the European global brand is United Colors of Benetton, and the Indian domestic brand is Wills Lifestyle. The study utilized a repeated measure (split-plot) design involving more than two independent groups. A split-plot analysis of variance analyses a design in which a repeated measure (i.e. within subjects) factor is crossed with a between-subjects (i.e. treatment variable) factor.FindingsThe results confirm the importance of global brands over local brands in the Indian apparel consumer market. This study also examined how Indian consumers' affinity for global brands influences their evaluation of the global brands and the local Indian brands.Originality/valueThe study expands the literature on Indian consumer brand preferences through the investigation of three brands. The theoretical background of the study is the consumer-brand relationship theory that explains the importance of consumer–brand relationship when brands attempt to expand to other markets.

Entropy Balancing for Continuous Treatments

Interest in evaluating the effects of continuous treatments has been on the rise recently. To facilitate the estimation of causal effects in this setting, the present paper introduces entropy balancing for continuous treatments (EBCT) – an intuitive and user-friendly automated covariate balancing scheme – by extending the original entropy balancing methodology of Hainmueller, J. 2012. “Entropy Balancing for Causal Effects: A Multivariate Reweighting Method to Produce Balanced Samples in Observational Studies.” Political Analysis 20 (1): 25–46. In order to estimate balancing weights, the proposed approach solves a globally convex constrained optimization problem, allowing for computationally efficient software implementation. EBCT weights reliably eradicate Pearson correlations between covariates (and their transformations) and the continuous treatment variable. As uncorrelatedness may not be sufficient to guarantee consistent estimates of dose–response functions, EBCT also allows to render higher moments of the treatment variable uncorrelated with covariates to mitigate this issue. Empirical Monte-Carlo simulations suggest that treatment effect estimates using EBCT display favorable properties in terms of bias and root mean squared error, especially when balance on higher moments of the treatment variable is sought. These properties make EBCT an attractive method for the evaluation of continuous treatments. Software implementation is available for Stata and R.

Risk and Health Policy Preferences: Evidence from the UK COVID-19 Crisis

The onset of the COVID-19 pandemic constituted a large shock to the risk of acquiring a disease that represents a meaningful threat to health. We investigate whether individuals subject to larger increases in objective health risk -- operationalised by occupation-based measures of proximity to other people -- became more supportive of increased government healthcare spending during the crisis. Using panel data which tracks UK individuals before and after the outbreak of the pandemic, we implement a fixed-effect design which was pre-registered before the key treatment variable was available to us. While individuals in high-risk occupations were more worried about their personal risk of infection, and had higher COVID death rates, there is no evidence that increased health risks during COVID-19 shifted attitudes on government spending on healthcare, nor broader attitudes relating to redistribution. Our findings are consistent with recent research demonstrating the limited effects of the pandemic on political attitudes.

Using predicted length of stay to define treatment and model costs in hospitalized adults with serious illness: an evaluation of palliative care

Background Economic research on hospital palliative care faces major challenges. Observational studies using routine data encounter difficulties because treatment timing is not under investigator control and unobserved patient complexity is endemic. An individual’s predicted LOS at admission offers potential advantages in this context. Methods We conducted a retrospective cohort study on adults admitted to a large cancer center in the United States between 2009 and 2015. We defined a derivation sample to estimate predicted LOS using baseline factors (N = 16,425) and an analytic sample for our primary analyses (N = 2674) based on diagnosis of a terminal illness and high risk of hospital mortality. We modelled our treatment variable according to the timing of first palliative care interaction as a function of predicted LOS, and we employed predicted LOS as an additional covariate in regression as a proxy for complexity alongside diagnosis and comorbidity index. We evaluated models based on predictive accuracy in and out of sample, on Akaike and Bayesian Information Criteria, and precision of treatment effect estimate. Results Our approach using an additional covariate yielded major improvement in model accuracy: R2 increased from 0.14 to 0.23, and model performance also improved on predictive accuracy and information criteria. Treatment effect estimates and conclusions were unaffected. Our approach with respect to treatment variable yielded no substantial improvements in model performance, but post hoc analyses show an association between treatment effect estimate and estimated LOS at baseline. Conclusion Allocation of scarce palliative care capacity and value-based reimbursement models should take into consideration when and for whom the intervention has the largest impact on treatment choices. An individual’s predicted LOS at baseline is useful in this context for accurately predicting costs, and potentially has further benefits in modelling treatment effects.

“Thanks in advance” – The negative effect of a polite phrase on compliance with a request

This paper studies the effect of the commonly used phrase “thanks in advance” on compliance with a small request. In a controlled laboratory experiment we ask participants to give a detailed answer to an open question. The treatment variable is whether or not they see the phrase “thanks in advance.” Our participants react to the treatment by exerting less effort in answering the request even though they perceive the phrase as polite.

Addressing the issue of bias in observational studies: Using instrumental variables and a quasi-randomization trial in an ESME research project

Purpose Observational studies using routinely collected data are faced with a number of potential shortcomings that can bias their results. Many methods rely on controlling for measured and unmeasured confounders. In this work, we investigate the use of instrumental variables (IV) and quasi-trial analysis to control for unmeasured confounders in the context of a study based on the retrospective Epidemiological Strategy and Medical Economics (ESME) database, which compared overall survival (OS) with paclitaxel plus bevacizumab or paclitaxel alone as first-line treatment in patients with HER2-negative metastatic breast cancer (MBC). Patients and methods Causal interpretations and estimates can be made from observation data using IV and quasi-trial analysis. Quasi-trial analysis has the same conceptual basis as IV, however, instead of using IV in the analysis, a “superficial” or “pseudo” randomized trial is used in a Cox model. For instance, in a multicenter trial, instead of using the treatment variable, quasi-trial analysis can consider the treatment preference in each center, which can be informative, and then comparisons of results between centers or clinicians can be informative. Results In the original analysis, the OS adjusted for major factors was significantly longer with paclitaxel and bevacizumab than with paclitaxel alone. Using the center-treatment preference as an instrument yielded to concordant results. For the quasi-trial analysis, a Cox model was used, adjusted on all factors initially used. The results consolidate those obtained with a conventional multivariate Cox model. Conclusion Unmeasured confounding is a major concern in observational studies, and IV or quasi-trial analysis can be helpful to complement analysis of studies of this nature.

Exploring Treatment for Depression in Parkinson’s Patients: A Cross-Sectional Analysis

Depression is a highly prevalent, often underrecognized and undertreated comorbidity of Parkinson’s disease closely correlated to health-related quality of life. National trends in depression care for patients with Parkinson’s disease are not well documented. This paper identifies a cohort of patients with Parkinson’s disease from nationally representative survey data and analyzes trends in depression care. Using data from the 2005–2006 through 2015–2016 waves of the National Health and Nutrition Examination Survey (NHANES), individuals were classified as Parkinson’s patients by reported medication use. PHQ-9 scores were used to identify individuals screening positive for depression. A composite treatment variable examined the reported use of mental health services and antidepressant medication. Survey participants with probable PD screened positive for depression, reported the use of antidepressant medication, and reported visits to mental health services more frequently than the control group. Survey participants with PD who screened positive for depression were more likely to report limitations in physical functioning due to an emotional problem than controls. While depression is highly prevalent among individuals with Parkinson’s disease, they are more likely to receive any treatment. Further research is required to investigate differences in patterns of treatment, contributing factors of emotions to limitations in physical functioning, and appropriate interventions.

The association between a history of anxiety or depression and utilization of diagnostic imaging

Objective While prior research shows that mental illness is associated with lower utilization of screening imaging, little is known about how mental illness impacts use of diagnostic imaging, other than for screening. This study explores the association between a history of anxiety or depression in the prior year and utilization of diagnostic imaging. Methods Commercial and Medicare Advantage health plan claims from 2017 and 2018 from patients with plans from one national organization were extracted. Exclusions were made for patients without continuous plan enrollment. History of anxiety or depression was determined using 2017 claims, and downstream diagnostic imaging was determined using 2018 claims. Univariate associations were assessed with Chi-square tests. A matched sample was created using Coarsened Exact Matching, with history of mental illness serving as the treatment variable. Logistic regressions were used to calculate adjusted odds ratios, before and after matching, controlling for age, sex, urbanicity, local income, comorbidities, claims history, region, and health plan characteristics. Associations between mental illness and chest imaging, neuroimaging, and emergency department imaging were also evaluated. Results The sample included 2,381,851 patients before matching. Imaging was significantly more likely for patients with a history of anxiety (71.1% vs. 55.7%, P < .001) and depression (73.2% vs. 55.3%, P < .001). The adjusted odds of any imaging were 1.24 (95% confidence interval [CI]: 1.22–1.26) for patients with a history of anxiety, and 1.43 (CI: 1.41–1.45) for patients with a history of depression before matching, and 1.18 (CI: 1.16–1.20) for a history of anxiety and 1.33 (CI: 1.32–1.35) for a history of depression after matching. Adjusted analyses found significant, positive associations between mental illness and chest imaging, neuroimaging, and emergency department imaging both before and after matching. Discussion In contrast to prior findings on screening, anxiety and depression were associated with greater likelihood of diagnostic imaging within the population studied.