Implementing Knowledge-Based Planning In Clinical Workflow

Knowledge-based planning (KBP) is changing how radiation oncology teams use historical treatment data. Rather than treating every plan as an isolated expert exercise, a KBP system learns relationships between patient anatomy, dose objectives and achievable plan quality. It can estimate dose-volume outcomes, identify unrealistic targets and help planners produce consistent IMRT, VMAT or stereotactic plans.

The clinical value is greatest when the technology is fitted to everyday work. A department needs carefully curated data, clear clinical ownership, independent validation and a workflow that supports rather than interrupts professional judgement. Lessons presented through the Canadian Organization of Medical Physicists’ 2014 meeting remain relevant to Australian services managing varied caseloads, distributed teams and pressure for safe, efficient treatment.

Why Knowledge-Based Planning Matters In Practice

Traditional planning relies heavily on individual experience. An experienced planner may recognise that a particular pelvic geometry will make bowel sparing difficult, while a newer team member may spend hours pursuing an objective that is unlikely to be achieved. KBP provides a data-informed starting point by comparing the current patient with previously treated cases and estimating realistic dose distributions.

This does not make the planner or radiation oncologist redundant. Instead, it gives the team a reference range for achievable quality. A model can flag unusual anatomy, reveal when a plan is an outlier and support peer review with objective evidence. In a busy metropolitan service in Sydney or Melbourne, that consistency can help departments manage high treatment volumes without allowing efficiency to replace clinical scrutiny.

KBP also creates a common language across professions. Radiation oncologists, dosimetrists, radiation therapists and medical physicists can discuss whether an objective is clinically necessary, technically feasible or unsupported by the department’s own evidence. That shared understanding is especially useful when a service introduces new treatment techniques across public and private facilities.

Build A Reliable Planning Dataset

The quality of a knowledge-based model depends on the quality of the plans used to train it. A dataset should contain approved plans that reflect current contouring protocols, prescription practices, immobilisation methods and treatment technologies. Mixing outdated plans with modern VMAT cases can teach the system historical compromises rather than present-day standards.

Data preparation should include checks for missing structures, inconsistent naming, unusual prescriptions and plans that were accepted only because no better option was available. Outliers require investigation, not automatic deletion. Some represent data errors; others may identify valuable clinical exceptions, such as previous irradiation, very large targets or unusual postoperative anatomy.

Australian departments should also consider how local practice affects generalisability. A model developed from a large tertiary centre may not transfer directly to a regional service in Queensland, Western Australia or Tasmania. Patient mix, scanner capability, staffing, referral patterns and access to specialist review can all influence the planning population. Local validation is essential before a vendor model is treated as authoritative.

Data governance must be agreed early. Patient information should be de-identified where possible, stored securely and handled under the organisation’s privacy and cybersecurity policies. Procurement teams should clarify where data are processed, who can access model outputs and how updates are controlled. The Australian market includes systems with different levels of interoperability, so a technically impressive platform may still fail if it cannot exchange data reliably with the oncology information system and treatment planning system.

Fit The Model To Clinical Decisions

A successful implementation begins with a defined clinical problem. “Improve plan quality” is too broad to guide configuration or evaluation. A department might instead target reduced dose to the rectum in prostate radiotherapy, improved heart and lung sparing in breast treatment, or shorter planning time for head and neck cases.

The clinical team should decide which objectives are hard constraints, which are desirable preferences and which should be reviewed individually. KBP predictions can show what is commonly achievable, but they do not establish that a lower dose is clinically required for every patient. A radiation oncologist must still balance target coverage, organ-at-risk protection, treatment intent and patient-specific priorities.

Workflow mapping should cover the entire pathway: simulation, contour review, planning, plan evaluation, peer review, approval and treatment delivery. If the model is accessed in a separate application and requires repeated manual exports, adoption will suffer. Integration with existing systems, role-based access and clear escalation routes are more important than adding a long list of optional features.

Education should be practical and role-specific. Planners need to understand model predictions and failure modes; clinicians need to interpret achievable-dose information; physicists need to monitor performance and manage technical change. The broader discussion about medical physics education is useful here because implementation depends on continuing professional development, not a single software demonstration.

Validate Before Routine Use

Validation should compare KBP-assisted plans with an agreed baseline. Useful measures include target coverage, organ-at-risk dose, conformity, homogeneity, planning time, frequency of rework and the number of plans requiring significant clinical override. A department should establish acceptance criteria before reviewing results, rather than moving the goalposts after seeing the data.

The test set must be separate from the cases used to build the model. It should include routine cases and challenging anatomies, with review by experienced clinicians. Dose predictions should be examined alongside actual deliverable plans because an attractive predicted distribution may not be achievable after optimisation, machine constraints or quality assurance.

Safety evaluation should include independent dose calculation, patient-specific quality assurance where required and checks for incorrect structure mapping. An automated system may produce a plausible result while applying the wrong objective to a renamed or incomplete contour. Alert thresholds should be visible, documented and linked to an action, such as physicist review or a repeat contour check.

Performance monitoring continues after deployment. Departments can track whether plan quality drifts as scanners, treatment machines, contouring guidelines or patient populations change. A model that performs well in Melbourne may require recalibration for a smaller service in Adelaide. Governance meetings should review overrides, near misses, outliers and user feedback without turning the process into a blame exercise.

Make Adoption Sustainable

KBP becomes part of clinical practice when accountability is explicit. A named medical physicist may oversee model performance, while a multidisciplinary group approves clinical objectives and change requests. Version control should record the training dataset, software release, configuration, validation results and date of approval.

Australian services also need a realistic resourcing plan. Vendor support may be strong in capital cities but slower for regional sites, particularly when issues require time-zone coordination or specialist configuration. Contracts should address implementation assistance, data portability, cybersecurity, response times and the cost of future upgrades. Local market comparisons should include the full operating burden rather than the licence price alone.

Professional education can reinforce safe use. Staff preparing for or maintaining specialist credentials should understand how automated planning relates to optimisation, dosimetry, quality assurance and clinical governance. The archived CCPM examination information illustrates the continuing importance of structured professional standards, even as software changes the way technical work is performed.

Adoption should be measured in patient and service outcomes, not novelty. A model that saves planning time but increases review complexity may need redesign. A model that produces modest efficiency gains while reducing unwarranted variation may be highly valuable. Regular feedback sessions, concise local protocols and protected training time help make the technology dependable during busy periods, including end-of-financial-year procurement cycles and holiday staffing changes.

Practical Implementation Priorities

A staged programme gives teams time to test assumptions and build trust. Begin with one treatment site or disease group, document the baseline process, and involve the people who will use and review the plans every day. Expansion should follow evidence of safe performance rather than a fixed calendar.

The following priorities provide a practical foundation:

• Select a clearly defined clinical use case with measurable quality and efficiency outcomes.
• Audit and clean the training dataset before configuring model objectives.
• Establish multidisciplinary ownership for approval, overrides, monitoring and updates.
• Validate against independent routine and difficult cases using agreed acceptance criteria.
• Integrate the tool with existing planning and oncology information workflows wherever possible.
• Record local performance, user feedback and safety events in a regular governance review.

The implementation team should communicate what KBP can and cannot do. It can estimate achievable outcomes, highlight variation and support optimisation. It cannot replace contour approval, treatment intent decisions, patient-specific judgement or independent quality assurance. Clear boundaries reduce both overconfidence and unnecessary resistance.

Begin with a documented baseline in one Australian service, select a representative dataset and run a prospective evaluation with clinical oversight. Use the results to refine the model, train the team and decide whether wider deployment is justified. That disciplined approach turns knowledge-based planning from an attractive software feature into a measurable improvement in clinical workflow.