Treatment Effects in the Presence of Unmeasured Confounding: Dealing With Observations in the Tails of the Propensity Score Distribution--A Simulation Study

Abstract

Frailty, a poorly measured confounder in older patients, can promote treatment in some situations and discourage it in others. This can create unmeasured confounding and lead to nonuniform treatment effects over the propensity score (PS). The authors compared bias and mean squared error for various PS implementations under PS trimming, thereby excluding persons treated contrary to prediction. Cohort studies were simulated with a binary treatment T as a function of 8 covariates X. Two of the covariates were assumed to be unmeasured strong risk factors for the outcome and present in persons treated contrary to prediction. The outcome Y was simulated as a Poisson function of T and all X's. In analyses based on measured covariates only, the range of PS's was trimmed asymmetrically according to the percentile of PS in treated patients at the lower end and in untreated patients at the upper end. PS trimming reduced bias due to unmeasured confounders and mean squared error in most scenarios assessed. Treatment effect estimates based on PS range restrictions do not correspond to a causal parameter but may be less biased by such unmeasured confounding. Increasing validity based on PS trimming may be a unique advantage of PS's over conventional outcome models.

Survival outcomes with and without splenectomy in splenic marginal zone lymphoma

“…Over 90% of the present cohort was diagnosed in the era of availability of efficacious chemoimmunotherapy, the principal therapeutic alternative beyond supportive care, palliative radiation, or interferon (for hepatitis C-associated cases). In an attempt to reduce the indication bias, patients at the extremes of PS were removed from the analysis, a method partly compensating for unmeasured confounding related to medical frailty [20]. The sensitivity analysis conducted in the highest-risk group of patients who died of SMZL demonstrated a consistently negative result.…”

Interpreting the quality of health care database studies on the comparative effectiveness of oral anticoagulants in routine care

“…33 It is essential to exclude patients in the extreme PS ranges where there is little clinical ambivalence in treatment choice. 22 These tails of the PS distribution often harbor extreme patient scenarios that are not representative of the majority of patients in clinical practice, and keeping them in the analyses may lead to less clinically relevant findings. 34,35 Exclusion of patients in areas of nonoverlap of the PS can be accomplished through matching with an appropriately tight caliper or PS adjustment after symmetric or asymmetric trimming.…”

“…Second, an important advantage of PS analyses is that they can easily enable investigators to avoid comparing patients who are dissimilar. This can be achieved by trimming on the PS, ie, by excluding patients in one group who have a PS value that is larger (or smaller) than that of any patients in the comparison group, 22 or by matching within a defined caliper. 21 If a patient cannot find a subject from the comparison group with a PS value within that caliper, then that patient will be excluded.…”

Observational Research Using Propensity Scores

“…9 If there are differences in treatment effects across strata, this could imply either effect modification 10 or the presence of unmeasured confounding (particularly in the tails of the PS distribution). 11 Effect modification means that the effects of treatment depend on the likelihood of treatment, a particular stratum of patients that is unlikely to be treated may only show small or no benefits of actual treatment while another stratum of patients that is highly likely to be treated may have significantly different benefits. If treatment effects are consistent across strata, researchers can collapse stratum-specific estimates into a single overall estimate.…”