Meta Analysis Part 2

Meta-analysis work is getting harder to start because the field is crowded with reviews and network meta-analyses, leaving fewer genuinely “original” trials and cohort studies to synthesize. The practical response is to treat topic selection as the first bottleneck, then build a tight PICO (Population, Intervention, Comparator, Outcomes) and a search string that reliably captures all relevant studies—without drowning the screening team in irrelevant hits.

Once a viable topic is chosen, the workflow moves into screening. The session stresses that PICO must be precise and consistent from the search string through inclusion decisions. A sample framing used throughout is cyclophosphamide versus corticosteroids in lupus nephritis (nephritic patients as the population; steroids as the comparator; outcomes such as kidney failure or mortality). After PICO is set, the next step is translating it into a PubMed search strategy using synonyms and controlled combinations: “OR” connects alternative terms for the same concept (e.g., cyclophosphamide and its alternative names), while “AND” links different concepts (drug + steroids + disease + population). Bracketed groups are recommended to keep the logic correct. The guidance also warns against over-trusting automated counts from tools like “Chargy/ChatGPT,” and against applying heavy filters early—filters can be unreliable when the goal is to avoid missing eligible studies.

The session then shifts to operationalizing screening using Rayyan, described as a free online tool for systematic reviews and meta-analyses. The process begins by running the search in PubMed, exporting results via “Send to” and “Citation manager,” and importing them into Rayyan using the “All results” option (not just the first page). Rayyan supports team screening: the owner can invite reviewers by email, and the platform can run in “blind on” mode so individual screeners cannot see each other’s decisions. Duplicates—common when importing from multiple databases—must be resolved by keeping one record.

Screening is split into primary and secondary stages. Primary screening is title/abstract-based: reviewers choose Include, Exclude, or Maybe (for uncertainty), and they can add comments explaining decisions. Keyboard shortcuts (like “E” for exclude) speed up work. The session emphasizes that primary screening should focus on matching PICO and excluding non-original study types such as case reports. When two reviewers disagree, the article lands in a conflict queue.

Secondary screening (full text screening) is where conflicts are resolved and where inclusion becomes final. Reviewers check the full manuscript—especially methods and results—and verify that the study truly matches the PICO and includes relevant outcomes. Tables are treated as the backbone of this step: baseline characteristics (often “Table 1”) and outcome reporting (often “Table 2” and beyond) must align with the meta-analysis outcomes. Supplementary files are also flagged as necessary when key outcomes or data are missing from the main text. If a study lacks any of the outcomes of interest, it may be excluded because extraction and analysis depend on outcome availability.

The session closes by reinforcing the end-to-end logic: build a comprehensive search string, import all results into Rayyan, run independent primary screening to reduce error, resolve disagreements through secondary screening using full-text and outcome checks, and then proceed to data extraction in the next session.