Oncology Research Workflow

Name the outcome and its boundary

Distinguish prediction, atlas construction, literature work and data preparation. Start with the research question, dataset and cohort boundary rather than a model name.

Keep data rights, consent, provenance and qualified scientific review attached to the work. A research example establishes no diagnostic accuracy, clinical benefit or medical authorisation.

Compare different causal paths

Compare process redesign, changes to an existing system, bounded rules, AI assistance and automation. Each option needs a reason it could change this piece of work, and assumptions that can be tested.

Keep the workload, acceptable quality, human review, integration and ongoing costs visible. A modelled capacity gain becomes financial value only under explicit conversion conditions.

Design a test that can change the decision

• Agree the scientific owner, evaluation method and evidence that would invalidate the hypothesis.
• Check cohort relevance, leakage and reproducibility before extending a finding.

Agree the metric, baseline, threshold, observation source and person responsible for the next decision. A result may support a bounded pilot, more evidence, redesign or stopping.

Keep a testable decision record

Metric and baseline

Agree the scientific owner, evaluation method and evidence that would invalidate the hypothesis.

Threshold and failure signal

Agree what would change the decision before observing results. Include unacceptable quality, rights or operating cost, and the evidence that would disprove the hypothesis.

Measurement and responsibility

Name the observation source, period, reviewer and person who accepts the outcome. Preserve assumptions separately from observed results.

Next action and review

Decide whether to test, redesign, expand or stop. A further operating step requires its own permissions and evidence; revisit it when the context changes.

Continue with this question

Explore the calculator · Research method · All research guides