Navigating the 2026 NIHR Evidence Synthesis Programme: Strategic Alignment and Methodological Expectations for Meta-Analysis Teams

The NIHR Evidence Synthesis Programme (ESP) funds research that systematically identifies, appraises and combines existing evidence on health and social care questions. Its researcher-led opportunity 2026/361 opened on 26 January 2026 and closed on 18 May 2026 [1]. It excluded very narrow questions and proposals limited to rapid reviews, scoping reviews or evidence maps, and allowed an economic analysis or modelling component without requiring one [1]. Future calls may differ, so check the current specification before planning.

Start from the decision, not the method: the population, intervention, comparator, outcomes and who will act on the answer. National relevance has to be shown through implications for policy, practice or services. A NICE evidence gap or a published priority helps, but alignment alone does not secure NIHR funding; the proposal must say what new knowledge the evidence synthesis adds.

Specify eligibility, databases, supplementary searching, screening, extraction, risk-of-bias assessment and synthesis. Structure the protocol with PRISMA-P and plan reporting to PRISMA 2020 [2, 3]; both are reporting guidelines, not methodological justification. The statistical analysis plan should define the estimand, effect measure, model, missing-data strategy and sensitivity analyses, and separate pre-specified from exploratory analyses.

A random-effects meta-analysis assumes true effects vary around a mean, with between-study variance τ². Estimating τ² (for example by REML) and making valid inference about the mean are separate tasks. With few studies, τ² is highly uncertain, and a small estimate does not prove homogeneity.

Conventional intervals treat the estimated variance as known and can under-cover with few studies. The Hartung–Knapp adjustment uses a t-distribution and often performs better, particularly with appreciable heterogeneity [4], but it is not superior in every configuration. Justify the choice rather than promise that one method solves small-sample problems. Treat I² and prediction intervals as descriptions, read alongside study characteristics, not as automatic rules for pooling.

Rare adverse events produce trials with zero events in one or both arms. Under conventional relative-effect methods, double-zero-event studies give no within-study estimate, but likelihood-based models using arm-level data can draw information from them, so neither automatic exclusion nor automatic inclusion is a rule. A continuity correction such as adding 0.5 makes the arithmetic work but adds artificial information; an analysis across the Cochrane Database found that the method chosen can change the results of real meta-analyses [5]. Pre-specify a primary model suited to the data, test the alternatives in sensitivity analyses, and say so when the data are too sparse to support firm estimates.

A relative effect means little without baseline risk. State which baseline-risk source applies, how its uncertainty will be handled, and the timeframe for absolute effects. Where a minimum important difference exists, interpret estimates against it, and carry treatment effects and their uncertainty into any economic model. This is what turns a pooled estimate into decision-relevant evidence.

A network meta-analysis is only as credible as its transitivity assumption: if A-versus-B trials recruit mild disease and B-versus-C trials severe disease, an indirect A-versus-C comparison may be biased when severity modifies the effect. Describe effect modifiers, consistency checks and how sparse networks limit rankings. Individual participant data meta-analysis allows harmonised analyses and covariate interactions, but depends on data access, governance and consistent definitions. Neither is automatically preferred to a well-designed pairwise synthesis.

Use risk-of-bias tools suited to each design and assess certainty across risk of bias, inconsistency, indirectness, imprecision and publication bias [6]. The 2026 call required applicants to explain how inclusive research design and sex and gender were considered across the research lifecycle, with the associated costs, and encouraged co-production with practitioners, patients, carers and the public [1].

Research proposal feasibility means deliverable within the time, budget and team, not just statistically sound. The call asked for timelines and milestones, value for money and pathways to impact, and encouraged applicants to embed a methodological sub-study, such as a Study Within A Review (SWAR), subject to programme rules [1]. A SWAR needs its own question and measurable outcome, and must not compromise the main synthesis.