shield_lock
JinDaGe
← Back to skills
extension
Category: Productivity & OfficeNo API key required

sr-complex-interventions

Systematic reviews of complex interventions per Cochrane Handbook Ch 17. Use when the intervention has multiple interacting components, is context-dependent, or involves behaviour change. Logic models, component analysis, process evaluation integration, and qualitative comparative analysis (QCA). Trigger on "complex intervention", "logic model", "component analysis", "process evaluation", "behaviour change", "BCT", "qualitative comparative analysis", "QCA", "programme theory", "implementation". Use ONLY when the intervention being reviewed is complex per Ch 17 criteria.

personAuthor: TashanworldhubOpenAPI
open_in_newRepositorydownloadDownload package

Complex Interventions (Cochrane Handbook Ch 17)

Your Role

Guide systematic reviews of complex interventions: define complexity, build logic models, deconstruct interventions into components, integrate process evaluations, and apply QCA when appropriate.

Prerequisites

  • Extraction data from included studies
  • Intervention descriptions (from extraction Block C)
  • Process evaluation data if available

Workflow

Step 1: Define Complexity

CHECK against the Ch 17 complexity criteria:

  • Number of interacting components within the experimental and comparator interventions
  • Number and difficulty of behaviours required by those delivering or receiving the intervention
  • Number of groups or organisational levels targeted by the intervention
  • Degree of flexibility or tailoring of the intervention permitted
  • Degree of variation in how the intervention is delivered in practice

DOCUMENT: "The intervention is complex because [criteria met]. We will use [logic model / component analysis / process evaluation] because [rationale]."

Step 2: Build a Logic Model

ASK the user to define:

  • Inputs: resources needed (staff, equipment, training)
  • Activities: what is delivered (sessions, drugs, procedures)
  • Mechanisms: how change is expected to happen (knowledge, motivation, physiology)
  • Outcomes: short-term, medium-term, long-term (per PICO)
  • Contextual moderators: factors that influence effectiveness (setting, provider type, patient characteristics)

GENERATE text logic model:

INPUTS → ACTIVITIES → MECHANISMS → OUTCOMES
                                         ↑
                                   CONTEXT

SAVE as sr-logic-model.md.

Step 3: Deconstruct Into Components

IDENTIFY distinct components of the intervention (e.g., education + feedback + protocol change).

For each component, code as present/absent per study.

RUN component meta-regression:

Rscript scripts/component_meta_reg.R --data extraction.csv --components education,feedback,protocol

INTERPRET:

  • p < 0.10 for a component → component modifies effect
  • p ≥ 0.10 → component not associated with effect

DOCUMENT LIMITATION: "Component meta-regression is observational — components are correlated and confounded by study-level factors."

Step 4: Integrate Process Evaluations

If process evaluations are available alongside outcome studies:

EXTRACT:

  • Fidelity: was the intervention delivered as intended?
  • Dose: how much was delivered/received?
  • Reach: who participated?
  • Adaptation: was it modified for local context?
  • Context: what contextual factors influenced outcomes?

SYNTHESIZE: "Process evaluations indicated that fidelity was [high/moderate/low]. Interventions with higher fidelity showed [pattern]."

Step 5: Apply QCA if Appropriate

QCA is appropriate when:

  • Multiple configurations of components exist across studies
  • Effect patterns are not linear/additive
  • You have ≥10 studies with clear component descriptions

GENERATE truth table: all possible combinations of components → outcome present/absent.

Scripts

scripts/logic_model_builder.py

Interactive text logic model builder: prompts for inputs, activities, mechanisms, outcomes, context. Outputs structured markdown. Usage: python3 logic_model_builder.py --output sr-logic-model.md

scripts/component_meta_reg.R

Meta-regression by presence/absence of each intervention component. Dummy-coded variables, random-effects meta-regression with knha adjustment. Usage: Rscript component_meta_reg.R --data extraction.csv --components education,feedback --outcome mortality

scripts/process_eval_extract.py

Structured extraction for process evaluations: fidelity, dose, reach, adaptation, context. Usage: python3 process_eval_extract.py --study Smith2021 --fidelity high --dose 80

Assets

assets/logic-model-template.md

Structured template with input/activity/mechanism/outcome/context columns and EM examples.

assets/bct-taxonomy-em.md

Behaviour change technique taxonomy (Michie 2013) with EM-specific examples (e.g., feedback on resuscitation performance, goal-setting for sepsis care).

assets/rameses-checklist.md

RAMESES II reporting standards for complex intervention systematic reviews.

Guardrails

  1. "Do NOT assume components are additive — interaction and emergence are central to complexity."
  2. "Component meta-regression is observational — interpret cautiously, even across RCTs."
  3. "Logic model must be developed BEFORE seeing results to avoid confirmation bias."
  4. "If component analysis yields no clear pattern, report this — null findings inform future research."
  5. "Process evaluations can explain heterogeneity — do not ignore them in synthesis."
  6. "QCA requires ≥10 studies with clear component descriptions — not suitable for small reviews."

Handoff

→ sr-synthesis (component MA results), → sr-interpretation (contextualized conclusions), → sr-writing