Realist Evaluation and QCA of Complex Interventions in Oncology Management: MDT, Telehealth/ePRO and QA/QI

For a new chemotherapy drug, the causal question is fairly simple: does it work? For a complex organisational intervention such as a multidisciplinary tumour board, a telehealth or electronic patient-reported outcome (ePRO) service, or a quality assurance and improvement programme, that question is the wrong one. Such interventions work in some places and not others, for some patients and not others, through mechanisms that depend on how they are implemented. The better question is: in what context, through which mechanism, and for whom does it work?

This chapter, the third report in a causal chain analysis series, addresses that question. The earlier reports showed that the oncology management literature leaves the organisational links of the causal chain largely empty, and that chains such as “MDT to outcomes” and “AI and QA to safety” cannot be identified by statistical adjustment alone because of unmeasured confounding. Here, Pawson and Tilley’s realist evaluation is used to extract context-mechanism-outcome (CMO) configurations for each intervention family, to define cross-family middle-range mechanisms, and to test those theoretical patterns with Ragin’s crisp-set qualitative comparative analysis (csQCA).

Method

From the 856-article Web of Science corpus, 147 records in the MDT (66), telehealth and ePRO (29) and QA/QI (67) families were screened, and 59 studies reporting an implementation or evaluation were retained as the evidence base (MDT 23, telehealth 12, QA 24). Context, mechanism and outcome propositions were extracted from each abstract and merged within families into 16 CMO configurations: 12 positive and 4 negative or rival.

For the QCA, 46 studies with comparative or before-after designs were coded on five binary conditions:

• HIC: high-income country.
• ACAD: academic or tertiary centre, or network.
• DIG: a digital tool component.
• DEDIC: a dedicated team, coordinator or training.
• FEED: a measurement-feedback loop, such as a dashboard, audit, ePRO alert, FMEA or peer review.

The outcome was coded 1 when a comparative improvement in a patient-level or operational outcome was reported. A truth table was built with a consistency threshold of 0.80, conservative and parsimonious solutions were derived with the Quine-McCluskey algorithm, and necessity was analysed separately.

Programme theories for the three families

Multidisciplinary team. The MDT meeting lets different specialties assess the same case at once (collective sense-making), moves the decision closer to evidence and guidelines, makes coordination visible, and reaches outcomes through implementation of the recommendation and timeliness. Its effect depends on who coordinates the meeting, whether a mechanism follows up implementation, and case complexity. Of 18 MDT studies in the QCA, 10 (56%) reported a positive outcome.

Telehealth and ePRO. These services make information flow between patient and team more frequent and asynchronous, surface symptoms early and in structured form (early detection), engage patients in self-monitoring and decisions (activation), and reduce travel and clinic time (access and cost). Results depend on whether data are tied to a response loop (alert, then nurse or physician action) and on patients’ digital and health literacy. Of 8 studies in the QCA, 7 (88%) were positive.

Quality assurance and improvement. QA/QI programmes make the process measurable (visibility), compare deviations with a standard (feedback) and attach corrective action to an institutional owner (accountability), reaching patient safety and efficiency through errors, waiting and plan quality. Effects depend on closing the measurement-feedback loop, real-time data and peer comparison at network or consortium level. Of 20 studies in the QCA, 15 (75%) were positive.

Five cross-family mechanisms

• M1. The visibility-feedback loop: measuring a process or symptom, comparing it with a standard and linking it to action. The strongest mechanism: 25 of 28 studies with FEED (0.89) were positive, against 9 of 17 without it (0.53).
• M2. Ownership and defined roles: attaching responsibility to a person through a coordinator, nurse, dedicated team or training. Not sufficient alone (digital tool plus dedicated team without feedback: 0.33); combined with feedback it reached 1.00.
• M3. Collective sense-making and peer pressure: decisions and plans being seen by others in a team or network. It consistently changes decisions; whether that becomes a patient outcome depends on M1 and M2.
• M4. Patient activation: patients joining their own care through information, self-efficacy and communication channels. Supported in the digital-plus-feedback configuration (0.91), with literacy and device access as preconditions.
• M5. Standardisation and forcing functions: reducing error by design through software or protocol. Requires local customisation and FMEA-based prioritisation.

QCA results

No single condition was strictly necessary, although FEED had high coverage (0.87) and ACAD high necessity consistency (0.88). In the truth table reduced to the three core conditions (DIG, DEDIC, FEED), all three configurations containing FEED exceeded the 0.80 threshold (0.83 to 1.00), and none without FEED did. The striking finding is that a digital tool plus a dedicated team without a feedback loop showed only 0.33 consistency: the studies in that cell built a platform or training programme but did not measure outcomes. Investment in structure and resources without measurement does not produce results.

The five-condition parsimonious solution (consistency 0.96, coverage 0.72) suggests complementary pathways:

• Digital tool with feedback, without a dedicated coordinator (11 studies, 0.91): when digital monitoring or a data dashboard feeds directly into feedback, no extra coordinator is needed.
• Dedicated team with feedback, without a digital component (3 studies, 1.00): audit and risk matrices driven by a dedicated team.
• Low- and middle-income settings: both feedback-loop and purely team-based interventions were consistently positive, although publication bias may explain part of this.

The conservative solution for a negative outcome was an academic setting without digital tools, dedicated team or feedback loop: studies that reported decision change but did not measure patient outcomes.

A refined programme theory: Structure x Loop x Owner

The synthesis proposes one middle-range theory for all three families. A complex oncology management intervention produces patient-level or operational results when it (a) establishes a structure (an MDT meeting, an ePRO application, QA software or a protocol), (b) defines a loop that compares the information the structure produces with a standard and links it to action, and (c) has a defined owner who runs the loop (a coordinator, nurse, quality lead or peer network). Context changes which component is critical: in resource-constrained systems, structure and standardisation alone bring large gains because the baseline is low; in mature academic systems, the structure already exists and gains come from the loop and the owner.

Managerial implications

• When setting up an MDT, define alongside the meeting calendar an indicator of whether recommendations were implemented, and its owner; add human factors training for chairing meetings and managing fatigue.
• Buy ePRO and telehealth not as data collection tools but as a threshold, alert and nurse-action chain; avoid telehealth that relies on telephone calls alone.
• Plan QA/QI with a real-time dashboard, FMEA-prioritised control points and peer comparison through a network or consortium together; an initial rise in near-miss reporting signals visibility, not failure.
• In resource-limited settings prioritise structure and standardisation (molecular tumour boards, plan QA plus training); in mature centres prioritise the loop and ownership.

Evidence assessment and limitations

Published evidence for the organisational links is mainly before-after and single-centre; the QCA shows which components co-occur when outcomes are reported, not effect sizes. Next steps include full-text recoding of the 59 studies with a second coder and agreement statistics, fuzzy-set QCA to grade conditions such as how fully the feedback loop is closed, and adding CMO-derived mechanism indicators (recommendation adherence, alert-response time, near-miss rate) to the interrupted time series and stepped-wedge designs proposed for these chains. CMO extraction and coding were done by a single analyst at abstract level; the outcome was binary, ignoring effect size and durability; and with 44 cases, 15 unobserved configurations were treated as logical remainders. The Structure x Loop x Owner theory can be used directly as a checklist in the oncology hospital management training series.

Source chapter: Kurutkan, M. N. (2026). Realist evaluation (CMO) and qualitative comparative analysis of complex interventions in oncology management: MDT, telehealth/ePRO and QA/QI. Oncology Hospital Management Training Series, supplementary chapter (original in Turkish). Accompanied by QCA workbook and csQCA engine.

References