Which Propensity Score Method Best Reduces Confounder Imbalance? An Example From a Retrospective Evaluation of a Childhood Obesity Intervention.
Krista Schroeder, Haomiao Jia, Arlene Smaldone
PMID 27801717WHAT IT FOUND
Matching children by how likely they were to have joined a nurse-led school obesity program balanced measured background differences best.
Other balancing approaches left more differences, and the estimated BMI change shifted depending on the approach.
Key findings
01Propensity score matching created two equal groups of 1,049 children and removed all significant differences between intervention and control groups.
02Stratification and weighting each left significant differences for eight of 11 measured background factors, and both created new differences.
03The estimated BMI percentile change differed by method: without adjustment, the intervention group had a 0.12 smaller decrease; with matching, 0.05; with stratification, 0.14; with weighting, 0.02.
STILL TO COME
How it was doneWhat they foundWhat it means for RNs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The comparison was made in one dataset, so it may not apply to other studies or populations. Data for at least one variable were missing for 75 children (3.7%). Only measured factors were balanced; unmeasured factors such as parental attitudes were not available. The analysis did not limit participants to the range of propensity scores where intervention and control children overlapped well. The propensity score model was fixed in advance and only one model was used. Very few children received the program (5%), which made weights extreme and may have harmed stratification and weighting. The matching analysis used 2,058 children, while stratification and weighting used 20,443, which may have made it easier for matching to show no significant differences. Balance was judged only by whether differences were statistically significant, not by how large they were. The strict rule for stratification counted a confounder as unbalanced if it differed in any one of the five strata, which may have made stratification look worse. Because participation was voluntary, the study cannot show that the program caused any BMI change.
Declared interests
The authors reported no conflicts of interest.
The easy way to misread this
Do not read this as evidence that the nurse-led obesity program improved BMI. The study compared statistical balancing methods in one observational dataset, and the estimated BMI percentile difference changed with the method; it cannot prove causation.