Weighted M-statistics With Superior Design Sensitivity in Matched Observational Studies With Multiple Controls

Causal inference analysis of the radiologic progression in the chronic obstructive pulmonary disease

There is limited evidence regarding the causal inference of emphysema and functional small airway disease in the subsequent progression of chronic obstructive pulmonary disease (COPD). Patients consisting of two independent cohorts diagnosed with COPD and underwent two serial chest CT scans were included. Total percent emphysema (PRMEmph) and fSAD (PRMfSAD) was quantified via PRM. To investigate the progression of emphysema, we divided COPD patients with PRMEmph < 10% into low and high PRMfSADgroup, matched with similar baseline characteristics, and conducted nonparametric hypothesis tests based on randomization inference using Wilcoxon signed rank test and Huber's M statistics. In patients with baseline PRMEmph < 10%, there were 26 and 16 patients in the low PRMfSA group and 52 and 64 patients in the high PRMfSA in the derivation and validation cohorts, respectively. In the both low and high PRMfSAD groups, there were 0.11 and 1.43 percentage point increases (Huber's M statistic p = 0.016) and 0.58 and 2.09 percentage point increases (p = 0.038) in the proportion of emphysema in the derivation and validation cohorts, respectively. On the contrary, among patients with baseline PRMfSAD < 20%, there was no significant differences in the interval changes of PRMfSAD between the low and high PRMEmph groups in both cohorts. In COPD patients with low emphysema, group with baseline high PRMfSAD showed greater change of PRMEmph than those with low PRMfSAD in both the derivation and validation cohorts. Imaging-based longitudinal quantitative analysis may provide important evidence that small airway disease precedes emphysema in CT-based early COPD patients.

BOUNDS ON THE CONDITIONAL AND AVERAGE TREATMENT EFFECT WITH UNOBSERVED CONFOUNDING FACTORS

For observational studies, we study the sensitivity of causal inference when treatment assignments may depend on unobserved confounders. We develop a loss minimization approach for estimating bounds on the conditional average treatment effect (CATE) when unobserved confounders have a bounded effect on the odds ratio of treatment selection. Our approach is scalable and allows flexible use of model classes in estimation, including nonparametric and black-box machine learning methods. Based on these bounds for the CATE, we propose a sensitivity analysis for the average treatment effect (ATE). Our semiparametric estimator extends/bounds the augmented inverse propensity weighted (AIPW) estimator for the ATE under bounded unobserved confounding. By constructing a Neyman orthogonal score, our estimator of the bound for the ATE is a regular root-n estimator so long as the nuisance parameters are estimated at the o p n - 1 / 4 rate. We complement our methodology with optimality results showing that our proposed bounds are tight in certain cases. We demonstrate our method on simulated and real data examples, and show accurate coverage of our confidence intervals in practical finite sample regimes with rich covariate information.

A Propensity Score Method for Investigating Differential Item Functioning in Performance Assessment

This study introduces a novel differential item functioning (DIF) method based on propensity score matching that tackles two challenges in analyzing performance assessment data, that is, continuous task scores and lack of a reliable internal variable as a proxy for ability or aptitude. The proposed DIF method consists of two main stages. First, propensity score matching is used to eliminate preexisting group differences before the test, ideally creating equivalent groups as in a randomized experimental study. Then, linear mixed effects models are adopted to perform DIF analysis based on the matched data set. We demonstrate this propensity DIF method using a high-stakes functional English language proficiency test. DIF due to education was investigated in the writing component, which consists of two continuously scored performance-based tasks. Although the proposed method is demonstrated in the context of language testing, it can be applied to other types of performance assessments.

Educational and criminal justice outcomes 12 years after school suspension

A third of US students are suspended over a K-12 school career. Suspended youth have worse adult outcomes than non-suspended students, but these outcomes could be due to selection bias: that is, suspended youth may have had worse outcomes even without suspension. This study compares the educational and criminal justice outcomes of 480 youth suspended for the first time with those of 1193 matched non-suspended youth from a nationally representative sample. Prior to suspension, the suspended and non-suspended youth did not differ on 60 pre-suspension variables including students' self-reported delinquency and risk behaviors, parents' reports of socioeconomic status, and administrators' reports of school disciplinary policies. Twelve years after suspension (ages 25-32), suspended youth were less likely than matched non-suspended youth to have earned bachelors degrees or high school diplomas, and were more likely to have been arrested and on probation, suggesting that suspension rather than selection bias explains negative outcomes.

School suspension predicts trichomoniasis five years later in a matched sample

BACKGROUND

Young adults who were suspended from school during adolescence are more likely than matched non-suspended youth to be arrested, on probation, or not graduate from high school, which are STI risk factors. This study evaluates whether suspension is a marker for STI risk among young adults who avoid subsequent negative effects.

METHODS

This study evaluated whether suspension predicts a positive test for chlamydia, gonorrhea, or trichomoniasis in a urine sample using matched sampling in the National Longitudinal Study of Adolescent and Adult Health (Add Health), and evaluated potential mediators between suspension and STI status using causal mediation analysis. We used Mahalanobis and exact matched sampling within propensity score calipers to compare 381 youth suspended for the first time in a 1-year period with 980 non-suspended youth. The suspended and non-suspended youth were similar on 67 pre-suspension variables. We evaluated STI outcomes 5 years after suspension.

RESULTS

Before matching, suspended youth were more likely to test positive for trichomoniasis and gonorrhea, but not chlamydia, than non-suspended youth. Suspended youth were more likely to test positive for trichomoniasis 5 years after suspension than matched non-suspended youth (OR = 2.87 (1.40, 5.99)). Below-median household income before suspension explained 9% of the suspension-trichomoniasis association (p = 0.02), but criminal justice involvement and educational attainment were not statistically significantly mediators.

CONCLUSIONS

School suspension is a marker for STI risk. Punishing adolescents for initial deviance may cause them to associate with riskier sexual networks even if they graduate high school and avoid criminal justice system involvement. Suspension may compound disadvantages for youth from below-median-income families, who have fewer resources for recovering from setbacks.