Causal Chain Analysis in Oncology Management Research: Dependent, Independent, Mediating, Moderating and Control Variables

If an oncology manager invests in a tumour board, a telehealth service or a quality improvement programme, how does that decision eventually reach the patient’s survival, quality of life or cost of care? Which intermediate links carry the effect, for whom does it work, and what must be controlled before anyone can claim it worked? This chapter reads an entire literature to answer those questions. It maps the implicit causal chain in a Web of Science corpus of 856 articles (2020 to 2026) on radiotherapy and systemic treatment management, classifying the variables the field uses as dependent, independent, mediating, moderating or control variables, and justifying each choice.

Three theoretical anchors

• Donabedian’s structure-process-outcome model organises how organisational and technological inputs reach patient outcomes through intermediate links.
• Value-based oncology brings clinical, patient-centred and economic outcomes into a single outcome-over-cost equation.
• Causal mediation and moderation theory, from the Baron-Kenny tradition to VanderWeele’s counterfactual definitions, separates the mechanism (how) from the context (for whom and where).

The corpus at a glance

Publications grew steadily, from 101 in 2020 to 156 in 2025. Fifty-seven per cent of records sit in Oncology and 22% in Radiology and Nuclear Medicine, while the categories where a management perspective is directly visible are much smaller: Health Care Sciences and Services (9.7%), Health Policy and Services (3.7%) and Economics. Candidate variables were derived from theory and then counted in titles, abstracts and keywords using a regular-expression dictionary, producing an inventory of 69 variables: 19 dependent, 20 independent, 8 mediating, 14 moderating and 8 control. An important caveat: frequencies measure the visibility of a concept, not proof that it was measured. They indicate density of interest, not density of evidence.

The variable inventory

Dependent (outcome) variables fall into four value-based families. Clinical: overall survival (15.2% of articles), progression- or disease-free survival (12.7%), toxicity (28.9%), mortality. Patient-centred: health-related quality of life (24.3%), symptom burden (23.8%), patient-reported outcomes, distress, satisfaction (3.7%). Economic: cost (24.1%), cost-effectiveness via ICER and QALY (16.7%), resource use. Operational and quality: efficiency and workflow (14.7%), plan quality, timeliness (6.1%), guideline adherence (2.8%) and safety events (3.7%). Several of these, such as toxicity, treatment completion and delay, act both as outcomes and as intermediate links toward more distal ones.

Independent (intervention) variables are the levers a manager or clinician can actually pull. Clinical levers dominate: chemotherapy regimen (56.7%), surgery and organ preservation (22.7%), hypofractionation (12.9%) and immunotherapy (10.5%). Organisational levers are far less visible: the multidisciplinary team (7.7%), QA/QI programmes including Lean (7.5%), AI and automation (7.0%), palliative care integration (7.0%) and telehealth (1.4%). From a business-school viewpoint, this gap is where the largest contribution can be made.

Mediators explain the mechanism. Only ten studies in the corpus tested mediation explicitly, almost all structural equation models in the quality-of-life and self-management cluster (self-efficacy, illness perception, emotion regulation). Organisational mediators such as care coordination and navigation (0.1%), health literacy (0.1%) and staff workload and burnout (2.3%) are theoretically strong but almost never tested.

Moderators answer “for whom and in what context”. Patient-level: stage and metastatic status (38.9%), sex, age and frailty, biomarkers. System-level: country and health system (29.6%), the willingness-to-pay threshold (12.3%) and facility type. Only two studies tested moderation formally with an interaction term, while 56 reported subgroup analyses: moderating effects are discussed but rarely tested statistically.

Control variables close confounding paths. Although 27.1% of the corpus reports multivariable adjustment, the rationale for the adjustment set is usually not reported. Histology and tumour site (67.8%) and calendar year are the most widely applicable controls; time horizon, discount rate and perspective are the structural controls in economic models; centre effects matter in multicentre data.

Eight candidate causal chains

The inventory was turned into eight chains (X intervention, M mechanism, Y outcome, W moderator, C control), each anchored in at least one empirical cluster:

• Z1. Treatment technology (hypofractionation, IMRT/VMAT, SBRT) to toxicity and plan quality to quality of life.
• Z2. Hypofractionation to capacity and resource use to cost and ICER.
• Z3. Immunotherapy or new drugs to PFS and OS gains to cost per QALY.
• Z4. Self-management education to self-efficacy to adherence to quality of life.
• Z5. Social determinants and access to treatment delay to survival.
• Z6. MDT and coordination to guideline adherence and timeliness to survival and satisfaction.
• Z7. AI, automation and QA to plan quality and efficiency to safety and cost.
• Z8. Supportive care (exercise, nutrition, palliative care) to symptom burden to hospital admissions and cost.

Each chain comes with a recommended analysis, from serial mediation and PLS-SEM (Z4) to Markov models with threshold analysis (Z3), multilevel patient-centre models (Z6) and interrupted time series or difference-in-differences designs (Z5, Z7).

Where the evidence is, and is not

A co-occurrence heat map shows evidence concentrated in the cells for chemotherapy and toxicity, chemotherapy and quality of life, and chemotherapy or immunotherapy and cost-effectiveness. The columns for satisfaction, patient safety, guideline adherence and hospital admission are nearly empty for almost every intervention. Methodologically the corpus is rich in randomised trials (162), prospective cohorts (137) and Markov or partitioned survival models (99), but poor in mediation (10), moderation (2), SEM (5) and quasi-experimental designs (7). The ends of the chain (X and Y) are well documented; the middle links (M) and the conditions (W) remain largely theoretical.

Priority research gaps

• Linking organisational interventions (MDT, telehealth, QA/QI, AI) to outcomes, especially chains Z6 and Z7, with ITS or DiD designs and multilevel models.
• Systematic measurement of satisfaction, guideline adherence and patient safety outcomes, currently visible in only 1% to 4% of articles.
• Testing health literacy, care coordination and staff workload as mediators.
• Testing moderation hypotheses with interaction terms rather than subgroup description, and adapting equity moderators such as socioeconomic status and facility type to the Turkish context.

Eight frameworks for stronger evaluation

The chapter proposes a sequence of advanced analyses on the same corpus: an evidence gap map; article-level role tagging with a large language model and human verification; DAG-based causal identification with target trial emulation; causal mediation (Imai, VanderWeele); meta-analytic SEM; Donabedian-based multilevel models; value-based outcome-over-cost analysis using ICHOM standard sets; and realist evaluation of context, mechanism and outcome. The DAG and realist evaluation strands are developed in their own companion chapters.

The managerial conclusion

The literature documents the start of the oncology causal chain (treatment and technology choices) and its end (survival, quality of life, cost-effectiveness) in depth. It largely leaves empty the area where managers actually intervene: organisational levers and the mechanisms through which they reach outcomes. That empty middle is both a research agenda and a warning against assuming that an organisational investment “obviously” works.

Source chapter: Kurutkan, M. N. (2026). Causal chain analysis in the oncology management literature: Identifying and justifying dependent, independent, mediating, moderating and control variables. Oncology Hospital Management Training Series, supplementary chapter (original in Turkish).

References