Funnel plot asymmetry signals small-study effects, and publication bias is only one cause. Unblinded outcome assessment, true heterogeneity or chance can also cause it. Cochrane reserves formal tests for meta-analyses of 10 or more studies. Of 300 sampled reviews, 28% had one.

Four checks on a forest plot help you judge someone else’s meta-analysis. - The axis labels. The authors choose which side favours which arm, and some plots put benefit on the right. - The diamond. Its width is the pooled interval. Check where it sits against the line of no effect. - The number of studies. With few studies, tau² is poorly estimated and I² carries a wide interval of its own. - The heterogeneity statistics. I² gives inconsistency as a proportion. Tau² shows how far apart the true effects lie. The diamond shows what pooling adds. In the guide’s example, five of the seven study intervals cross the line at 1.0. The diamond sits clear of it, at 0.83 (0.74 to 0.92). Each of those five is compatible with no effect, which is weaker than showing there is none. One of them runs from 0.44 to 1.05, too imprecise on its own to separate a large benefit from none. Pooling exists to deal with that imprecision. Pooling also carries every flaw of the pooled studies into the result. That is why risk of bias is assessed before the diamond is drawn. A careful reader checks that it was.

“…one human could do the work of two with the same quality and in less time…” Ley Muller (Values-Driven AI) on a RobotReviewer study at NIPH. Some colleagues still wanted two humans. #SystematicReview #RiskOfBias #AI

The calculation is the easy part of a meta-analysis. We have published a guide on the PICO Portal blog. It covers how to read a forest plot, what I² measures, fixed and random effects, and the data a team needs before the analysis.

“We’re expected to use AI, but they want us to follow completely outdated guidance.” Karen Robinson, PICO Portal. “They” are journal editors. #SystematicReview #EvidenceSynthesis #AI

Four subtractions check a PRISMA diagram. - Identified minus removed equals screened. - Screened minus excluded equals sought. - Sought minus not retrieved equals assessed. - Exclusion reasons sum to assessed minus reports going forward. Reviewers run the second one first.

The PRISMA 2020 flow diagram has three phases, with Eligibility folded into Screening. A 2009 file looks close enough to reuse. It has nowhere to put several counts that 2020 expects, and the 2020 statement says the 2009 version should no longer be used. - Registers get a line of their own, apart from databases. - Websites, organisations and citation searching get a second column. - Removals are counted on three lines: duplicates, automation, other reasons. - Reports not retrieved get a box of their own. The full-text work still happens, now under Screening, as reports sought, not retrieved, assessed and excluded. The word Eligibility survives in a single box label: “Reports assessed for eligibility.” The lower half of the diagram changed too. The 2009 version counted full-text articles and split included studies into qualitative and quantitative synthesis. The 2020 version ends with studies and the reports of those studies. One template became four: new or updated review, each with or without sources beyond databases and registers.

Each box in a PRISMA flow diagram holds a count, and each arrow a subtraction. We have published the second guide in our review series on the PICO Portal blog. It walks the 2020 template box by box and names three errors peer reviewers flag.

“…the conducting guidelines that we have for systematic reviews should change.” Farhad Shokraneh, University of Oxford. PRISMA 2020 already has a line for automation tools. #SystematicReview #PRISMA #AI

“…if you do a PubMed search, you’re not gonna find the same thing.” Jennifer McKinnell, McMaster University: AI search and PubMed differ on intermittent fasting. picoportal.org/2026/03/12/fr…

Eight stages of a systematic review, with honest durations. The surprise for most teams: screening is not the longest stage. Full-text retrieval, extraction, and appraisal reliably are.

We tested GPT-5 on 541 full-text articles. Even at its most inclusive setting, it missed 16% of eligible studies, or one in six. Reliable full-text screening remains a hard problem to solve. What did we learn, and where might AI still help? More in chapters 2 through 5.

Four commitments separate a systematic review from a literature review. - A protocol before the first record is screened - A search someone else can re-run - Criteria applied the same way to every record, borderline ones included - Two independent assessors. The full guide, with the eight stages, the five roles, and honest durations: picoportal.org/2026/09/15/sy…

The mistakes that recur in systematic reviews are failures of record-keeping. We have published a guide on the PICO Portal blog: the eight stages, who does what, PRISMA 2020, and where reviews go wrong. Read it here: picoportal.org/2026/09/15/sy…

At #ISPORAnnual in Montréal this May, we’re sharing new findings on LLMs for accelerating #evidencesynthesis without compromising rigor. 1️⃣ Contextual screening 2️⃣ Full-text data extraction Learn more and book a 1:1 at the event: picoportal.org/2025/03/31/pi… #systematicreview #EBM

Attending the Global Evidence Summit in Prague next week? Let's connect! There's never been a more exciting time for technology in evidence-based practice. Meet with us to dive deep into how AI is transforming the field! 📆 Schedule your time: bit.ly/4eaHDBY #GES2024

🚀 #GenerativeAI is transforming Systematic Review! Join our webinar on Sept 18: "Is the Future of AI in Systematic Review Here?" 🧠💡 Speakers from JHU, CU & PICO Portal will discuss AI's role in #SLR, #RiskofBias, & #DataExtraction. Register now! 🔗 bit.ly/47gRCDK

📣 Webinar: “Coupling Human & AI to Overcome Resource Constraints for Evidence Syntheses” 🗓️ Aug 22, 11am ET. Join Lisa Brown from @theNASEM to hear about a #scopingreview on defining Long COVID using PICO Portal PLUS+. bit.ly/4cd7Sq0 #evidencebasedmedicine

⏱ T minus 1 week for our next webinar on using AI for RoB and Data Extraction at scale. 💡 We also welcome our additional speaker to the presentation 📈 Hundreds have registered to stay up to date on this growing field. Have you? 🔗 Register here: picoportal.org/2024/04/27/pi…

🔍 Recent article in #PharmacoEcomonics leverages PICO Portal to screen 3000+ citations & conduct a #ScopingReview on costing of DHIs. It also offers a specialized DHI program costs checklist for #HEOR and #DigitalHealth professionals. Link to article: lnkd.in/ee4NpcC8