←
CoreSR1,152 posts
CoreSR avatar

CoreSR

@CoreSRR

An AI assisted platform that will take your systematic review idea from protocol to publication. Visit us: coresr.ai

◷ Joined July 2026▧ 41 Photos and videos
CoreSR avatar
CoreSR@CoreSRR·10m

A random-effects 95% CI tells you where the average effect probably sits, not what the next setting will see. That's the prediction interval. With real heterogeneity it can cross the null while the CI doesn't. Report both: readers apply effects to settings, not to averages.

CoreSR avatar
CoreSR@CoreSRR·3h

Where additivity fails, an interaction term for the pair that synergizes (or cancels) beats forcing the sum. CoreSR runs component NMA in real R inside the review project. 7-day free trial: coresr.ai #NetworkMetaAnalysis #ComponentNMA #EvidenceSynthesis

CoreSR avatar
CoreSR@CoreSRR·3h

Multicomponent interventions (exercise + diet + education) usually get lumped into one NMA node. Component NMA estimates each component, but leans on additivity: a combination works as the sum of its parts. Check it against the standard NMA fit before trusting the split.

CoreSR avatar
CoreSR@CoreSRR·6h

ROBINS-I, applied to each component of the target trial, flagged why: eligibility and outcome measurement couldn't be fully copied, plus residual confounding. Open access: doi.org/10.1186/s12874-026-0… #TargetTrialEmulation #RiskOfBias #CausalInference

CoreSR avatar
CoreSR@CoreSRR·6h

Following target trial emulation guidance doesn't guarantee the trial's answer. Giallongo et al. (BMC Med Res Methodol) emulated the FIRST-ABC step-up RCT from PICANet, 5,405 children, doubly robust. Liberation RR 1.33 (1.27-1.40); the RCT found 1.03 (0.90-1.18).

CoreSR avatar
CoreSR@CoreSRR·9h

A living systematic review is a set of commitments, not a label: a stated search schedule, a rule for when new evidence triggers re-analysis, and a plan for repeated looks at the pooled estimate. Without those, it's a static review with a recent date.

CoreSR avatar
CoreSR@CoreSRR·Nov 17

A "high vs low dose" meta-analysis asks one narrow question of a few direct trials. A dose-response meta-analysis uses every dose arm, placebo-controlled and head-to-head, and models the curve (e.g. restricted cubic splines), so you can see where benefit flattens, if it does.

CoreSR avatar
CoreSR@CoreSRR·Nov 17

At CoreSR, we build task-specific models and fine-tuned adapters in-house, trained and validated for screening, data extraction, and analysis rather than relying on generic wrappers.

CoreSR avatar
CoreSR@CoreSRR·Nov 16

An NMA doesn't get one certainty rating. Each comparison does: rate the direct and indirect evidence, then the network estimate, with intransitivity and incoherence as added reasons to rate down. CoreSR does NMA GRADE natively, per comparison, beside the R output.

CoreSR avatar
CoreSR@CoreSRR·Nov 15

RoB 2 domain 2 isn't one question. For the effect of assignment (ITT): deviations that arose from the trial context, and whether people were analysed as randomized. For the effect of adhering: non-adherence, failed implementation, unbalanced co-interventions. Pick one first.