Selecting primary studies

Once the search is complete, the retrieved references — many of them duplicates or clearly out of scope — go through several subphases to arrive at the final list of primary studies:

1. Deduplication — the same reference retrieved through multiple search strategies is identified and removed.
2. Title/abstract screening — remaining references are checked against the eligibility criteria; anything that could plausibly match is retained for the next phase.
3. Full-text assessment — retained references are read in full; each is either included or excluded, with the exclusion reason recorded.

A study can match every eligibility criterion and still not report the data needed for an effect size computation. There are three ways to handle it: exclude it outright and note why; keep it in the systematic review's narrative synthesis but leave it out of the statistical pooling; or contact the study's authors for the missing numbers, and include it if they can provide them.

It's good practice to compute interrater reliability at each subphase: two or more researchers screen the same references independently, and their agreement is reported as a percentage or Cohen's κ (values above .60 are considered acceptable, above .80 good). All these numbers — records identified, screened, excluded, and included — are exactly what a PRISMA flow diagram documents.

The result of this step is the final list of primary studies, ready to be examined in detail in Coding primary studies.