Prognostic Models and the Propensity Score
Subjects in observational studies of exposure effects have not been randomized to exposure groups and may therefore differ systematically with regard to variables related to exposure and/or outcome. To obtain unbiased estimates and tests of exposure effects one needs to adjust for these variables. A common method is adjustment via a parametric model incorporating all known prognostic variables. Rosenbaum and Rubin propose adjustment by the conditional exposure probability given a set of covariates which they call the propensity score. They show that, at any value of the propensity score, covariates are on average balanced between exposure groups. Thus matching on the propensity score leads to unbiased estimators and tests of exposure effect. However, the validity of the method depends on knowing the exposure probability. This quantity is usually not known in observational studies and needs to be estimated.
Available from: Martin Stevens
- "However, it is not always feasible to use random assignment. For such situations, propensity scores are considered an useful alternative [19,20]. With propensity scores, adjustments can be made for observed variables to minimize or remove the bias, which normally disturbs the results in observational studies. "
[Show abstract] [Hide abstract]
ABSTRACT: Hip fractures constitute an economic burden on healthcare resources. Most persons with a hip fracture undergo surgery. As morbidity and mortality rates are high, perioperative care leaves room for improvement. Improvement can be achieved if it is organized in comprehensive care pathways, but the effectiveness of these pathways is not yet clear. Hence the objective of this study is to compare the clinical effectiveness of a comprehensive care pathway with care as usual on self-reported limitations in Activities of Daily Living.
A controlled trial will be conducted in which the comprehensive care pathway of University Medical Center Groningen will be compared with care as usual in two other, nonacademic, hospitals. In this trial, propensity scores will be used to adjust for differences at baseline between the intervention and control group. Propensity scores can be used in intervention studies where a classical randomized controlled trial is not feasible. Patients aged 60 years and older will be included. The hypothesis is that 15% more patients at University Medical Center Groningen compared with patients in the care-as-usual condition will have recovered at least as well at 6 months follow-up to pre-fracture levels for Activities of Daily Living.
This study will yield new knowledge with respect to the clinical effectiveness of a comprehensive care pathway for the treatment of hip fractures. This is relevant because of the growing incidence of hip fractures and the consequent massive burden on the healthcare system. Additionally, this study will contribute to the growing knowledge of the application of propensity scores, a relatively novel statistical technique to simulate a randomized controlled trial in studies where it is not possible or difficult to execute this kind of design.Trial registration: Nederlands Trial Register NTR3171.
Available from: nih.gov
- "It has been suggested that rehabilitative treatments of the type developed specifically for patients with severe and persistent mental illnesses (Mueser, Drake, and Bond 1997; Lehman 1995) may be effective for veterans with chronic PTSD (Rosenheck and Fontana 1996; Friedman and Rosenheck 1996). Conventional psychotherapies for PTSD focus on cathartic recollection , progressive desensitization, deconditioning, or cognitive reframing of traumatic memories. "
[Show abstract] [Hide abstract]
ABSTRACT: To evaluate the effectiveness of a work therapy intervention, the Department of Veterans Affairs (VA) Compensated Work Therapy program (CWT), in the treatment of patients suffering from chronic war-related post-traumatic stress disorder (PTSD); and to demonstrate methods for using outcomes monitoring data to screen previously untested treatments.
Baseline and four-month follow-up questionnaires administered to 3,076 veterans treated in 52 specialized VA inpatient programs for treatment of PTSD at facilities that also had CWT programs. Altogether 78 (2.5 percent) of these patients participated in CWT during the four months after discharge.
The study used a pre-post nonequivalent control group design.
Questionnaires documented PTSD symptoms, violent behavior, alcohol and drug use, employment status, and medical status at the time of program entry and four months after discharge from the hospital to the community. Administrative databases were used to identify participants in the CWT program. Propensity scores were used to match CWT participants and other patients, and hierarchical linear modeling was used to evaluate differences in outcomes between treatment groups on seven outcomes.
The propensity scaling method created groups that were not significantly different on any measure. No greater improvement was observed among CWT participants than among other patients on any of seven outcome measures.
Substantively this study suggests that work therapy, as currently practiced in VA, is not an effective intervention, at least in the short term, for chronic, war-related PTSD. Methodologically it illustrates the use of outcomes monitoring data to screen previously untested treatments and the use of propensity scoring and hierarchical linear modeling to adjust for selection biases in observational studies.
Available from: chrp.org
[Show abstract] [Hide abstract]
ABSTRACT: There has been rising interest in approaches designed to draw causal inferences from observational or quasi-experimental data in studies of health outcomes and policy. Causal interpretations of observed associations in such settings can be tricky, largely due to the problems of selection bias. Methodologies such as propensity scores and instrumental variables are increasingly part of the statistician's toolbox. Statisticians involved with health policy research need to communicate with an audience of physicians and other consumers, who are often skeptical of these techniques. At this roundtable, we will discuss the role of statisticians in (1) arguing the need for appropriate methodologies for dealing with self-selection in observational studies; (2) producing effective displays for validating assumptions and documenting findings, and (3) describing an observational study's results, assumptions, caveats and conclusions accurately and usefully for a clinical or policy-oriented audience. Arguing the Need for Dealing with Self-Selection in Observational Studies Randomized experiments are the "gold standard" - they ensure that subjects receiving different exposures are comparable. Yet, we cannot always do experiments - exposures may be harmful, controlled by a systemic process that will not yield control, beyond reach legally or financially. We're frequently interested in phenomena that do not lend themselves to randomized trials. Such trials often have limited external validity as well - due in many cases to exclusion criteria, and other phenomena that limit our ability to study "entrenched practices". In an observational study concerning exposures and their effects, the researcher does not control the assignment of exposures. Despite this, we want to be able to compare groups who "looked similar" prior to exposure assignment - thus, analytical adjustments are need to account for baseline differences in covariates. A study is biased if the exposed and unexposed groups differ in ways that matter for the outcomes of interest. We need to think hard about how exposure was determined. Patients who receive an exposure are usually different from patients who don't receive it in important ways. We capture reasons behind exposure assignment in covariates, then adjust for covariate differences in estimating effects on outcomes. "… the elucidation of causal relationships from observational studies must be shaped by knowledge (or assumptions) about how the data were generated; such assumptions are crucial to causal inference." -
Data provided are for informational purposes only. Although carefully collected, accuracy cannot be guaranteed. The impact factor represents a rough estimation of the journal's impact factor and does not reflect the actual current impact factor. Publisher conditions are provided by RoMEO. Differing provisions from the publisher's actual policy or licence agreement may be applicable.