Meta-synthesis: integrating qualitative findings across studies

The method’s working vocabulary: first-order constructs are participants’ own accounts (the quotes in primary papers); second-order constructs are the primary authors’ interpretations (their themes); third-order constructs are the synthesist’s interpretation of those interpretations.

The synthesist mostly works with second-order material — which is why the quality and reporting depth of the primary studies bounds everything. A primary paper that presents three thin themes and two quotes contributes almost nothing synthesisable, however large its sample.

Keeping the three levels distinct in the analysis is what separates synthesis from paraphrase: a third-order construct must be traceable down through named second-order themes to first-order evidence.

Thematic synthesis — the most used approach — was formalised by Thomas and Harden in Methods for the thematic synthesis of qualitative research in systematic reviews: line-by-line coding of findings, descriptive themes, then analytic themes that go beyond the primary studies.

The broader craft — sampling, appraisal, and the varieties of synthesis — is laid out in Sandelowski and Barroso’s Handbook for Synthesizing Qualitative Research (Springer Publishing), still the fullest treatment of the judgement calls involved.

Confidence in synthesised findings now has a formal system: GRADE-CERQual, introduced by Lewin and colleagues in Using qualitative evidence in decision making for health and social interventions, which grades each synthesised finding on methodological limitations, coherence, adequacy, and relevance.

A synthesis of qualitative studies on TB treatment adherence included 28 papers from 14 countries. Line-by-line coding of the findings sections produced 214 codes, consolidated into eleven descriptive themes — stigma, cost, side effects, provider relationships, and so on — all recognisable from the primary studies.

The analytic step produced the third-order construct that made the synthesis cited: adherence behaviour tracked the visibility of treatment, not its burden. Clinic queues visible to neighbours, workplace absences, packaging identifiable at home — across settings, patients managed exposure first and regimen second. No single primary study had stated this; seven had second-order themes consistent with it, and the deviant cases (two studies where treatment was home-delivered and discreet, with high adherence despite severe side effects) strengthened rather than broke the construct.

Each synthesised finding was CERQual-graded; the visibility construct carried high confidence (coherent across 19 studies of adequate thickness), while a finding about gendered adherence differences was graded low — four studies, two with serious limitations — and flagged as a research gap rather than a conclusion.

• Stopping at descriptive themes. A list of themes that any included study already contained is aggregation, not synthesis.
• Untraceable third-order claims. Every synthesised construct needs its chain of custody down to primary evidence.
• Treating study count as confidence. Twelve thin studies can ground less than four thick ones; CERQual exists precisely to replace counting.
• Synthesising across incompatible questions. Studies about experiencing a disease and studies about experiencing its treatment are adjacent, not identical; forcing them together produces themes true of neither.
• Ignoring reflexivity. The synthesist’s framing shapes the third-order constructs; the write-up should say who read what and how disagreements were handled.

Meta-synthesis inherits and compounds the flattening of primary analysis: participants’ words arrive twice interpreted, and context — the soul of qualitative work — survives only as much as primary reporting allowed. The method reads best as building transferable mid-range theory, not as recovering experience.

It is also hostage to reporting conventions: word limits in primary journals systematically thin the material available to synthesise, and the synthesist cannot repair what was never printed.

And the interpretive step resists standardisation by design. Two teams can code identically and build different third-order constructs; transparency about the chain of interpretation is the method’s honesty mechanism, not a guarantee of convergence.

Meta-ethnography is the interpretive ancestor — translation of studies into one another — where thematic synthesis is the more procedural descendant.

A systematic literature review with narrative synthesis handles mixed or quantitative bodies of evidence; meta-synthesis is its qualitative counterpart.

An integrative review admits qualitative and quantitative sources together when the question needs both.

Line-by-line coding of 28 findings sections is exactly the corpus-scale coding work AI does well: Evidano codes the findings passages inductively or against your developing frame and links every code to its source passage, so descriptive themes are built on retrievable evidence and the traceability chain CERQual asks for exists by construction.

The analytic themes — the third-order move — are the synthesist’s interpretation and should stay that way; the tool’s role is to make the second-order layer completely searchable while you make the leap.