Radiology Qualitative Mixed Methods

Radiology Qualitative and Mixed Methods

Use this skill when the research question requires accounts, meanings, behaviours, interactions or work practices to be studied through interviews, focus groups, observation, workflow ethnography, think-aloud or an explicit qualitative-quantitative integration design. Its unique responsibility is the qualitative strand and the logic by which qualitative and quantitative strands are integrated.

Non-negotiable boundaries

• A usability interview, debrief or collection of free-text comments is not automatically a complete qualitative study. Without a declared question, sampling logic, analysable records, analytic method, reflexivity and audit trail, label it FORMATIVE_FEEDBACK_ONLY.
• Purposeful sampling supports information-rich variation, not statistical representativeness. Do not justify a fixed sample with folklore or claim saturation without a method-specific stopping rule and contemporaneous evidence.
• Participant quotations and field observations are source evidence; codes, themes and explanations are analytic interpretations. A mixed-methods meta-inference is a further interpretive step.
• Member checking, multiple coders, inter-coder agreement and software use are not universal proof of qualitative validity. Select rigor procedures that fit the analytic tradition and record what each procedure can and cannot establish.
• Parallel qualitative and quantitative work is not mixed-methods integration unless the strands meet at a prespecified design, sampling, analysis or interpretation point.
• This skill does not own quantitative power calculations, survey psychometrics, clinical workflow deployment, ethics approval, individual patient decisions or manuscript copy-editing.

Route by mode

Do not load every reference for a narrow request. Use qualitative-mixed-methods-passport.md for a persistent study record and joint-display-and-discordance-ledger.md when the strands are integrated.

Input passport

Before giving a claim-bearing recommendation, record:

1. decision context, research question, stage and intended claim;
2. qualitative tradition or mixed-methods rationale, priority and timing; use AUTHOR_INPUT_NEEDED rather than inventing an epistemological position;
3. participant/stakeholder roles, setting, unit of analysis, eligibility, sampling frame and purposeful sampling dimensions;
4. data source, topic/observation guide, interviewer or observer relationship, recording, transcription/translation and privacy state;
5. analysis approach, unit of coding, inductive/deductive balance, coder roles, software and versioned codebook or reflexive memo process;
6. for mixed methods, strand-specific artifact versions, integration point/level, sampling relationship and planned joint display or traceable narrative integration; require a linking key only for participant/case-level linkage;
7. supplied raw artifacts and their evidence state.

Missing fields remain visible. Do not infer consent, transcript completeness, participant identity, site context, data quality or analytic decisions from filenames or polished prose.

Required workflow

1. Lock the phenomenon and claim. State whether the study seeks description, interpretation, explanation, theory development, implementation insight or instrument development.
2. Match design to the question. Select interviews, focus groups, observation, workflow ethnography or think-aloud because of the evidence they can produce, not because they are convenient. In radiology, freeze the imaging task, role, site, shift, PACS/RIS/EHR context, displayed image/report/AI-output version and scenario fidelity.
3. Make sampling auditable. Specify purposeful criteria and desired variation. Use information power or a method-specific saturation construct, state its dimensions, review it during collection and keep a stopping log. Do not convert it into a post hoc claim that “no new themes” appeared.
4. Preserve fieldwork context. Record who collected the data, prior relationships, power differences, prompts, non-participation, group composition, observation role and field changes. Think-aloud reveals verbalised reasoning under reactivity; it does not reproduce an unobserved natural workflow.
5. Run the declared analysis. Keep raw segment -> code -> category/theme -> interpretation traceability. Search for negative or deviant cases and document decisions. Use member checking only for a defined purpose, such as correcting factual accounts or discussing interpretation.
6. Integrate deliberately. For mixed methods, connect, build, merge or embed the strands at the declared participant, site, construct, design or interpretation level. A joint display or traceable narrative can document the bridge; different samples do not require invented person-level keys. Classify convergence, complementarity, expansion and discordance; derive a bounded meta-inference. Never infer individual relationships from aggregate integration or let the larger strand win.
7. Separate statement classes. Label SOURCE_EVIDENCE, PARTICIPANT_ACCOUNT, FIELD_OBSERVATION, ANALYTIC_INTERPRETATION, META_INFERENCE, RECOMMENDATION and AUTHOR_INPUT_NEEDED.
8. Apply the claim ceiling and hand off. Qualitative transferability depends on documented context and analytic reasoning; it is not prevalence. Mixed-methods integration does not repair a weak constituent strand.

STOP gates

Return the exact stop state and the smallest next action:

• STOP_ETHICS_OR_PRIVACY: consent, confidentiality, recording, image/report exposure or re-identification authority is unresolved;
• STOP_NO_ANALYSABLE_RECORD: conclusions rely only on memory, unverified AI summaries or inaccessible recordings without an authorised analytic record;
• STOP_DESIGN_METHOD_MISMATCH: the named method cannot answer the question or its required sampling/data/analysis elements are absent;
• STOP_UNSUPPORTED_ADEQUACY: information power or saturation is asserted without defined, reviewable evidence;
• STOP_NO_INTEGRATION: a mixed-methods claim is requested but no defensible integration point, strand-to-inference bridge or discordance handling exists; missing participant keys block only an inference that requires participant-level linkage, not construct-level or sequential integration.

A STOP blocks the affected claim, not all useful work. A bounded design repair or artifact inventory may still be returned.

Radiology-specific evidence limits

• Interview patients, radiologists, technologists, physicists, referrers and operational staff as distinct roles; one role cannot silently stand for the whole pathway.
• Account for reporting-room interruptions, worklists, hanging protocols, prior examinations, display calibration, voice recognition, handoffs, alert timing and local escalation rules.
• Observation can reveal work-as-done but does not prove why it occurs; interviews can explain experience but do not directly measure actual behaviour.
• Think-aloud and simulated cases may alter speed, search and confidence. Preserve case selection, image presentation, AI version and observer presence as transfer boundaries.
• Qualitative acceptance of an imaging or AI tool does not establish diagnostic accuracy, safety, effectiveness, cost-effectiveness or implementation success.

Output contract

Return only the sections needed for the selected mode:

1. Route and passport status — mode, supplied artifacts, missing inputs and stop state.
2. Question-design-claim map — phenomenon, design rationale, unit and claim ceiling.
3. Sampling adequacy plan/read — purposeful dimensions, information-power or saturation logic, stopping evidence and exclusions.
4. Fieldwork and reflexivity — guide, setting, relationships, recording and positionality.
5. Analysis audit trail — analytic approach, segment-to-interpretation trace, negative cases and conditional credibility procedures.
6. Integration and discordance — design, joint display or integrated narrative, convergence/complementarity/discordance and bounded meta-inferences when applicable.
7. Evidence-inference-recommendation ledger — no category laundering.
8. Limitations and handoff — unresolved risks and unique receiving owner.

Handoffs and non-ownership

• Quantitative design, power, psychometrics and numerical inference -> radiology-stats or radiology-design; acquisition claims -> radiology-acquisition-qc.
• Human-factors deployment, prospective workflow evaluation and routine usability testing -> radiology-translation. Return here only when a genuine qualitative or mixed-methods inference is requested.
• Data governance/de-identification -> radiology-data; consent and participant governance -> radiology-ethics.
• Reporting-guideline selection -> radiology-reporting; prose production -> radiology-writing; project receipts -> radiology-pipeline.
• Individual patient imaging, diagnosis or treatment decisions remain with the responsible clinical team. This skill does not confer ethics approval or certify methodological validity.