Stronger Instruments and Refined Covariate Balance in an Observational Study of the Effectiveness of Prompt Admission to Intensive Care Units

Abstract

Instrumental Variable (IV) methods, subject to appropriate identification assumptions, allow for consistent estimation of causal effects in the presence of unobserved confounding. Near-far matching has been proposed as one analytic method to improve inference by strengthening the effect of the instrument on the exposure and balancing observable characteristics between groups of subjects with low and high values of the instrument. However, in settings with hierarchical data (e.g. patients nested within hospitals), or where several covariate interactions must be balanced, conventional near-far matching algorithms may fail to achieve the requisite covariate balance. We develop a new matching algorithm, that combines near-far matching with refined covariate balance, to balance large numbers of nominal covariates while also strengthening the IV. This extension of near-far matching is motivated by a case study that aims to identify the causal effect of prompt admission to the Intensive Care Unit on 7-day and 28-day mortality. * We thank Anirban Basu for comments and discussion.

Survivor-Complier Effects in the Presence of Selection on Treatment, With Application to a Study of Prompt ICU Admission

“…However, the bounds clearly favor the possibility that prompt ICU care reduces mortality. Moreover, analyses of these same data using an IV analysis found that ICU care generally lowers mortality rates (Keele et al 2016). While that estimate is based on a different sub-population, finding that ICU care is generally beneficial in conjunction with our analysis provides further evidence that prompt ICU care is beneficial for patients.…”

Analysis of Benefit of Intensive Care Unit Transfer for Deteriorating Ward Patients

“…We paired patients with similar baseline characteristics and matched patients assessed when there were “many” vs “few” ICU beds available, excluding matched pairs with a difference of less than 3 ICU beds available (eAppendix 1 in the Supplement ). 31 This near-far matching algorithm attempted to ensure that measured confounders were balanced and the instrument (ICU bed availability) was strong (ie, predicted ICU transfer). 32 , 33 , 34 , 35 To assess the quality of the matches, we reported covariate balance according to standardized mean differences.…”

Profiling Compliers in Instrumental Variables Designs

“…To conduct complier profiling conditional on covariates, researchers can use, for example, weighting-based semiparametric instrumental variable estimation. 15 An alternative approach is to use matching to adjust for IV-outcome confounders; see Baiocchi et al 3 and Keele et al 16 for examples. With matched data, complier profiling can be done using estimators unconditional of W .…”

“…In their analysis, they adjusted for possible IVoutcome confounders using a variant of matching that combines near-far matching with refined covariate balance. 21,22 More specifically, patients were exactly matched on hospital such that across-hospital differences could not bias the analysis. They were also exactly matched on an indicator for sepsis.…”