Methodological progress note: A clinician's guide to propensity scores

Antibiotic use and outcomes among children hospitalized with suspected pneumonia

“…The IPTW method was selected to address potential bias due to confounding by indication (e.g., children who are more ill appearing are more likely to get antibiotics) and adjust for systematic differences in patient characteristics between those who did and did not receive ED antibiotics. 24,25 Propensity scores were estimated as the likelihood of receiving antibiotics using multivariable logistic regression. Covariates were selected a priori to address confounding by severity of illness and likelihood of bacterial etiology, including age <5 years, sex, race, influenza season (winter/fall), history of CAP, fever at home, days of illness, radiographic CAP, the maximum temperature in ED, hypoxemia (≤90%) in ED, oxygen administered in ED, and decreased breath sounds, wheeze, crackles, and ill appearance on ED examination.…”

“…We performed propensity score analysis using inverse probability of treatment weighting (IPTW) to evaluate the relationship of the antibiotic use and outcomes. The IPTW method was selected to address potential bias due to confounding by indication (e.g., children who are more ill appearing are more likely to get antibiotics) and adjust for systematic differences in patient characteristics between those who did and did not receive ED antibiotics 24,25 . Propensity scores were estimated as the likelihood of receiving antibiotics using multivariable logistic regression.…”

Strategies to evaluate the quality of hospital design with clinical data

“…Rather than "adjust the outcome," propensity score matching attempts to "adjust the cohort" upfront before comparisons are made. 16 This approach is particularly useful when outcomes are sparse relative to the number of covariates.…”

Propensity scores to estimate treatment effects of thiamine in alcohol use disorder

“…A propensity score is each patient's probability of receiving a given treatment based on an underlying group of factors that may influence the intervention being given (e.g., age, comorbidities, illness severity). 6 This probability is distinct from whether the treatment was actually received or not (i.e., some patients with a high propensity score may not receive the treatment). Propensity scores can then control for confounding as individuals with comparable propensity scores from the treated and nontreated groups will have a similar Residual confounding (the outcome not being completely explained by the primary exposure and the covariates) may exist if variables are either not included in the model or the variable is not well defined.…”

“…A propensity score is each patient's probability of receiving a given treatment based on an underlying group of factors that may influence the intervention being given (e.g., age, comorbidities, illness severity) 6 . This probability is distinct from whether the treatment was actually received or not (i.e., some patients with a high propensity score may not receive the treatment).…”