A matched vaping study: who remained in the comparison?

Read how the comparison sample was built

A propensity-score method uses recorded characteristics to help construct a comparison. NCI’s matching guidance emphasizes observed covariate balance, overlap, and the participants excluded by the procedure. These are methodological principles; its cancer examples and variable recommendations are not applied to vaping here.

Find the paper’s selection steps before looking only at the final effect estimate. A study may begin with a large database but report a matched analysis of fewer people. That is a defined comparison population, not automatically the entire original database. Record how the paper describes that population instead of repeating the largest sample number from its introduction.

US National Cancer Institute: SEER-MHOS Propensity Score Matching Guidance

Use an original before-and-after sample card

Imagine a fictional comparison of adults who chose two support arrangements. Before matching, their recorded work schedules differ. The analysis forms matches using schedule and other documented baseline variables. On one card, write the original sample; on another, write the matched sample and the variables actually checked for balance.

Add a third line for adults who had no suitable match. This exercise contains no estimates or real participants. It shows why a claim about matched adults need not describe everyone seeking support. Do not assign missing characteristics to the unmatched people or conclude that a program is unsuitable for them simply because the analysis did not retain them.

Ask what similar means in this report

Look for the reported balance information rather than assuming that matched means identical. Also check when the variables were measured and what information was unavailable. If motivation, access, or another relevant factor was not recorded, the matching label cannot establish that it was balanced. Nor does that absence prove it explains a particular result.

The same study can use matching and later statistical adjustment; read both procedures without conflating them. An introductory reader should not redesign the model or apply a universal balance cutoff. Methodological appraisal requires the actual question, measurements, and assumptions. Your useful task is to locate what was demonstrably compared and what remains uncertain.

Carry the retained population into the claim

A bounded summary could say that the estimate concerns the matched comparison under the stated variables and exclusions. If the publication does not explain unmatched cases or balance, leave that question open. Do not silently extend the result to people unlike those retained or relabel the study randomized because its final groups look similar.

For a care decision, ask a qualified professional about the full evidence and its applicability. Matching your personal log with a friend’s log cannot produce the counterfactual outcome you would have had under another support choice. Both records can still be useful for practical conversations without becoming a controlled intervention comparison.