Добавил:
kiopkiopkiop18@yandex.ru t.me/Prokururor I Вовсе не секретарь, но почту проверяю Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:
Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_5525_Библиотеки_им_академика_М_И_Перельмана.pdf
Скачиваний:
0
Добавлен:
31.08.2026
Размер:
29 Мб
Скачать
Artificial Intelligence in Adaptive Radiation Therapy
Addressing these issues requires robust validation studies, adherence to ethical guidelines, and the development of comprehensive regulatory frameworks. Emerging innovations, such as AI-powered adaptive radiotherapy and quantum computing, further expand the potential of clinical trials, offering transformative opportunities for precision medicine. By combining computational intelligence with rigorous scientic standards, AI is set to revolutionize evidence-based healthcare, enabling more effective and patient-centered clinical research.

References

[1] Kandi V, Vadakedath S, Addanki P S, Godishala V and Pinnelli V B 2023 Clinical trials: the
role of regulatory agencies, pharmacovigilance laws, guidelines, risk management, patenting, and publicizing results Borneo. J. Pharm.
[2] Trigueiros B A F D S, Ávila A R and Pimenta F P 2022 An examination of the use of clinical
trials as a source of information in scientic research Transinformação
[3] Bhattamisra S K, Banerjee P, Gupta P, Mayuren J, Patra S and Mayuren C 2023 Articial
intelligence in pharmaceutical and healthcare research Big. Data. Cogn. Comput.
[4] Robbins W 1994 Rapporteur summary: role of technology in clinical trialsvalidation of
information Health Care Technology Policy I: The Role of Technology in the Cost of Health Care (Piscataway, NJ: IEEE)
[5] Max M 1994 Rapporteur summary: role of technology in the cost of health care Health Care
Technology Policy I: The Role of Technology in the Cost of Health Care
SPIE) pp 134–48 [6] Bleicher P 2003 Clinical trial technology: at the inection point BIOSILICO 1 163–8 [7] Royle J K, Hughes A, Stephenson L and Landers D 2021 Technology clinical trials: turning
innovation into patient benet Digit. Health. [8] Marquis-Gravel G et al 2019 Technology-enabled clinical trials Circulation 140 1426–36 [9] Turner J R and Hoofwijk T J 2013 Clinical trials in new drug d evelopment
J. Clin. Hypertens.
[10] Ying A J 2023 The importance of the clinical trialing process in the development of modern-
day drugs Proc. SPIE
[11] Raber-Johnson M L, Gallwitz W E, Sullivan E J and Storer P 2019 Innovation in clinical
trial design and product promotion: evolving the patient perspective with regulatory and
technological advances Ther. Innov. Amp. Regul. Sci.
[12] Chow S and Pei Liu J 1998 Design and Analysis of Clinical Trials: Concept and
Methodologies (Hoboken, NJ: Wiley)
[13] Onken J E and Brazer S R 1994 Clinical trials: how should they be designed? Gastrointest.
Endosc. Clin. N. Am.
[14] Sverdlov O, Ryeznik Y and Wong W K 2019 On optimal designs for clinical trials: an
updated review J. Stat. Theory Pract.
[15] Hee S W et al 2015 Decision-theoretic designs for small trials and pilot studies: a review Stat.
Methods Med. Res.
[16] Sverdlov O, Ryeznik Y and Wong W K 2020 On optimal designs for clinical trials: an
updated review J. Stat. Theory Pract.
[17] Hung H M J and Wang S-J 2014 Emerging challenges of clinical trial methodologies in
regulatory applications Clinical Trial Biostatistics and Biopharmaceutical Applications (Boca
Raton, FL: CRC Press) pp 39–76
15 306–9
12611 1261166–16
4 423–34
25 1022–38
6 93–109
34 e210065
7 10–0
(Bellingham, WA:
7 205520762110121
54 519–27
14 10
14 10
19-16
Artificial Intelligence in Adaptive Radiation Therapy
[18] Xu J, Psioda M A and Ibrahim J G 2022 Bayesian design of clinical trials using joint models
for recurrent and terminating events Biostatistics
[19] David L E, Albert A T, Danielle B and Keymer Michael A 2020 Systems and methods for
designing clinical trials Justia Patents Patent # 11,139,051;
11139051
[20] Phadnis M A, Wetmore J B and Mayo M S 2017 A clinical trial design using the concept of
proportional time using the generalized gamma ratio distribution Stat. Med.
[21] Silverman R and Kwiatkowski T 1998 Research fundamentals: III. Elements of a research
protocol for clinical trials Acad. Emerg. Med.
[22] Holloway P J and Mooney J A 2004 Whats a research protocol? Health Educ. J. 63 374–84 [23] Karsh L I 2012 A clinical trial primer: historical perspective and modern implementation
Urol. Oncol. Semin. Orig. Investig.
[24] Hess A S and Abd-Elsayed A 2019 Components of clinical trials Pain ed A Abd-Elsayed
(Cham: Springer) pp 83–5
[25] Reddy P and Bhadauria U S 2019 Integral elements of a research protocol J. Indian Acad.
Oral Med. Radiol.
[26] Xun Y et al 2022 Protocols for clinical practice guidelines J. Evid.-Based Med. 16 3–9 [27] An M-W, Duong Q, Le-Rademacher J and Mandrekar S J 2020 Principles of good clinical
trial design J. Thorac. Oncol.
[28] Park J-S et al 2020 An interactive retrieval system for clinical trial studies with context-
dependent protocol elements PLoS One
[29] Harrer S 2020 Articial intelligence for clinical trial design 2020 IEEE Signal Processing in
Medicine and Biology Symposium (SPMB)
[30] Woo M 2019 An AI boost for clinical trials Nature 573 S100–2 [31] Miyasato G, Kasivajjala V, Kumar K, Kadam A S and Friedman H S 2023 AI-driven real-
time patient identication for randomized controlled trials J. Clin. Oncol.
[32] Askin S, Burkhalter D, Calado G and El Dakrouni S 2023 Articial intelligence applied to
clinical trials: opportunities and challenges Health Technol.
[33] Tsuchiwata S and Tsuji Y 2023 Computational design of clinical trials using a combination
of simulation and the genetic algorithm CPT Pharmacomet. Syst. Pharmacol.
[34] Yin J, Ngiam K Y and Teo H H 2021 Role of articial intelligence applications in real-life
clinical practice: systematic review J. Med. Internet Res.
[35] Anran Wang X X, Shengyu Liu Q Q and Zhu Wu S i 2022 Characteristics of articial
intelligence clinical trials in the eld of healthcare: a cross-sectional study on ClinicalTrials.
gov Int. J. Environ. Res. Public. Health
[36] Herson J 2023 Digital twins: a futuristic articial intelligence methodology for design and
analysis of clinical trials Ann. Biostat. Biom. Appl.
[37] Vatankhah Barenji R and Ebrahimi Hariry R 2023 Blockchain-enabled quality improve-
ment digital twin for clinical trials Preprint
[38] Susilo M E et al 2023 Systems-based digital twins to help characterize clinical dose–response
and propose predictive biomarkers in a phase I study of bispecic antibody, mosunetuzu-
mab, in NHL Clin. Transl. Sci.
[39] Camps J et al 2023 Digital twinning of the human ventricular activation sequence to clinical
12-lead ECGs and magnetic resonance imaging using realistic Purkinje networks for in silico
clinical trials Med. Image Anal.
31 167
30 S28–32
15 1277–80
15 e0238290
19 13691
16 1134–48
94 103108
24 866–84
https://patents.justia.com/patent/
36 4121–40
5 1218–23
(Piscataway, NJ: IEEE)
41 e13565–5
13 203–13
12 522–31
23 e25759
5
19-17
Artificial Intelligence in Adaptive Radiation Therapy
[40] Wang Z, Gao C, Glass L M and Sun J 2022 Articial intelligence for in silico clinical trials: a
review arXiv:
[41] Baer A R, Bridges K D, ODwyer M, Ostroff J and Yasko J 2010 Clinical research site
infrastructure and efciency J. Oncol. Pract.
[42] NCIs National Clinical Trials Network (NCTN) NCI https://www.cancer.gov/research/
infrastructure/clinical-trials/nctn (Accessed: 15 January 2024)
[43] Zou W, Geng H, Teo B K, Finlay J and Xiao Y 2018 NCTN clinical trial standardization
for radiotherapy through IROC and CIRO Med. Phys.
[44] Lee S H, Geng H and Xiao Y 2022 Radiotherapy standardisation and articial intelligence
within the National Cancer Institutes Clinical Trials Network Clin. Oncol.
[45] Viceconti M, De Vos M, Mellone S and Geris L 2023 From the digital twins in healthcare to
the virtual human twin: a moon-shot project for digital health research arXiv:
[46] Ohri N, Shen X, Dicker A P, Doyle L A, Harrison A S and Showalter T N 2013
Radiotherapy protocol deviations and clinical outcomes: a meta-analysis of cooperative
group clinical trials J. Natl. Cancer Inst.
[47] Doot R K et al 2012 Design considerations for using PET as a response measure in single site
and multicenter clinical trials Acad. Radiol.
[48] Xiao Y, Rosen M, Xiao Y and Rosen M A 2017 The role of imaging and radiation oncology
core for precision medicine era of clinical trial Transl. Lung Cancer Res.
[49] Glide-Hurst C K et al 2021 Adaptive radiation therapy (ART) strategies and technical
considerations: a state of the ART review from NRG oncology Int. J. Radiat. Oncol. Biol.
Phys.
109 1054–75
[50] Nie K et al 2019 NCTN assessment on current applications of radiomics in oncology
Int. J. Radiat. Oncol. Biol. Phys.
[51] Zou W et al 2023 Framework for quality assurance of ultrahigh dose rate clinical trials
investigating FLASH effects and current technology gaps Int. J. Radiat. Oncol. Biol. Phys.
116 1202–17
[52] Ge Y and Wu Q J 2019 Knowledge-based planning for intensity-modulated radiation
therapy: a review of data-driven approaches Med. Phys.
[53] Li N, Carmona R and Sirak I 2017 Highly efcient training, renement, and validation
of a knowledge-based planning quality-control system for radiation therapy clinical trials
Int. J. Radiat. Oncol. Biol. Phys.
[54] Giaddui T 2016 A feasibility study of the dosimetric compliance criteria of the NRG-HN002
head and neck clinical trial across different radiotherapy treatment planning systems and
delivery techniques: a model for optimizing initial trial launch J Cancer Prev. Curr. Res.
341–5
[55] Giaddui T, Chen W and Yu J 2016 Establishing the feasibility of the dosimetric compliance
criteria of RTOG 1308: phase III randomized trial comparing overall survival after photon
versus proton radiochemotherapy for inoperable stage II-IIIB NSCLC Radiat. Oncol.
[56] Yusufaly T, Miller A and Medina-Palomo A 2020 A multi-atlas approach for active bone
marrow sparing radiation therapy: implementation in the NRG-GY006 trial Int. J. Radiat.
Oncol. Biol. Phys.
[57] Younge K C, Marsh R B and Owen D 2018 Improving quality and consistency in NRG
oncology radiation therapy oncology group 0631 for spine radiosurgery via knowledge-based
planning Int. J. Radiat. Oncol.
2209.09023
6 249–52
45 e850–3
34 128–34
2304.06678
105 387–93
19 184–90
6 621–4
104 302–15
46 2760–75
97 164–72
11 66
108 1240–7
100 1067–74
5
19-18
Artificial Intelligence in Adaptive Radiation Therapy
[58] Geng H, Liao Z and Nguyen Q N 2023 Implementation of machine learning models to
ensure radiotherapy quality for multicenter clinical trials: report from a phase III lung cancer
study Cancers
[59] Giaddui T, Geng H and Chen Q 2020 Ofine quality assurance for intensity modulated
radiation therapy treatment plans for NRG-HN001 head and neck clinical trial using
knowledge-based planning Adv. Radiat. Oncol.
[60] Geng H et al 2021 A comparison of two methodologies for radiotherapy treatment plan
optimization and QA for clinical trials J. Appl. Clin. Med. Phys.
[61] Men K et al 2020 Automated quality assurance of OAR contouring for lung cancer based on
segmentation with deep active learning Front. Oncol.
[62] Lee S H et al 2023 Interpretable machine learning for choosing radiation dose-volume
constraints on cardio-pulmonary substructures associated with overall survival in NRG
oncology RTOG 0617 Int. J. Radiat. Oncol. Biol. Phys.
[63] Russell A M et al 2023 Complex and alternate consent pathways in clinical trials:
methodological and ethical challenges encountered by underserved groups and a call to
action Trials
[64] Doan X, Florea M and Carter S E 2023 Legal-ethical challenges and technological solutions
to e-health data consent in the EU Front. Artif. Intell. Appl.
[65] Dietz H P 2006 Bias in research and conict of interest: why should we care? Int. Urogynecol.
J.
18 241–3
[66] Meerpohl J J et al 2015 Evidence-informed recommendations to reduce dissemination bias in
clinical research: conclusions from the OPEN (Overcome failure to Publish nEgative
Ndings) project based on an international consensus meeting BMJ Open
[67] Shabani M and Obasa M 2019 Transparency and objectivity in governance of clinical trials
data sharing: current practices and approaches Clin. Trials
[68] Strech D 2022 Transparenz in der klinischen Forschung: welchen Beitrag leistet die neue
EU-Verordnung 536/2014? Bundesgesundheitsbl
[69] DeVito N J and Goldacre B 2023 New UK clinical trials legislation will prioritise
transparency Brit. Med. J.
[70] ICH 2016 ICH harmonised guideline: integrated addendum to ICH E6(R1): guideline for
good clinical practice E6(R2)
[71] World Medical Association 2025 WMA Declaration of Helsinki: Ethical Principles for
Medical Research Involving Human Subjects Declaration World Medical Association
https://wma.net/policies-post/wma-declaration-of-helsinki-ethical-principles-for-medical-
research-involving-human-subjects/
[72] National Archives 2025 Title 21Food and Drugs Code of Federal Regulations www.ecfr.
gov/current/title-21
[73] Clinical Trials Regulation European Medicines Agency https://ema.europa.eu/en/human-
regulatory/research-development/clinical-trials/clinical-trials-regulation
[74] Xiao Y et al 2021 Toward individualized voxel-level dosimetry for radiopharmaceutical
therapy Int. J. Radiat. Oncol. Biol. Phys.
[75] Li H et al 2023 Overview and recommendations for prospective multi-institutional clinical
trials of spatially fractionated radiation therapy (SFRT) Int. J. Radiat. Oncol.
[76] Paganetti H 2014 Relative biological effectiveness (RBE) values for proton beam therapy.
Variations as a function of biological endpoint, dose, and linear energy transfer Phys. Med.
Biol.
59 R419–72
15 1–11
5 1342–9
22 329–37
10 986
117 1270–86
24 151
368 243–53
5 e006666–e6
16 547–51
66 52–9
382 1547–7
https://ich.org/page/efcacy-guidelines
109 902–4
737–49
19-19
Artificial Intelligence in Adaptive Radiation Therapy
[77] Subcommittee on Quantum Information Science 2018 National strategic overview for quantum
information science Strategic Overview National Science and Research Council
gov/wp-content/uploads/2020/10/2018_NSTC_National_Strategic_Overview_QIS.pdf
[78] Vandewinckele L et al 2020 Overview of articial intelligence-based applications in radio-
therapy: recommendations for implementation and quality assurance Radiother. Oncol.
55–66
[79] Maspero M et al 2018 Dose evaluation of fast synthetic-CT generation using a generative
adversarial network for general pelvis MR-only radiotherapy Phys. Med. Biol.
[80] Chapman J W, Lam D, Cai B and Hugo G D 2022 Robustness and reproducibility of an
articial intelligence-assisted online segmentation and adaptive planning process for online
adaptive radiation therapy J. Appl. Clin. Med. Phys.
[81] Casola L (ed) 2023 Opportunities and Challenges for Digital Twins in Biomedical Research:
Proceedings of a Workshop-in Brief
(Washington, DC: National Academies Press)
23 e13702
https://quantum.
153
63 235007
19-20
IOP Publishing
Artificial Intelligence in Adaptive Radiation Therapy
Yi Wang and X. Sharon Qi
Chapter 20
Safety and training considerations in the clinical
implementation of artificial intelligence
adaptive radiation therapy
Kelly Nealon and Jennifer Pursley
In the rapidly evolving landscape of radiation therapy, the integration of articial intelligence (AI) in adaptive radiation therapy (ART) brings unprecedented advancements but also necessitates a comprehensive understanding of the associated risks and safety considerations that must be made. When implementing new technology into clinical practice, the departmental risk management and staff training program must be re-evaluated to accommodate the changing workow [1]. The process of utilizing AI ART methods for patient treatment differs substantially from standard linac-based external beam workows [2, 3]. Specically, real-time plan adaptation requires an in-depth understanding of patient anatomy, contour quality, and treatment planning. This dynamic adaptation process introduces distinct roles for therapists, physicists, and physicians that may differ from their prior practice. The intricacies of real-time plan adaptation necessitate specialized training to ensure that healthcare professionals have the expertise required to navigate the nuances of patient-specic anatomy and optimize treatment plans effectively. Therefore, to safely and effectively implement AI ART, updates should be made to the risk management and staff training programs, and appropriate end-to-end testing should be performed before clinical deployment.

20.1 Risk management

Risk management programs are used to identify and correct points of weakness in a workow that could introduce risk to patients and staff. While there are many possible components of an effective risk management program, both prospective and reactionary techniques should be included to ensure that the safety and efciency of the workow is optimized.
doi:10.1088/978-0-7503-6119-4ch20 20-1 ª IOP Publishing Ltd 2025. All rights,
including for text and data mining (TDM), artificial intelligence (AI) training, and similar technologies, are reserved.
Artificial Intelligence in Adaptive Radiation Therapy
20.1.1 Prospective risk assessments
Prior to introducing ART into clinical practice, a series of prospective risk assess­ments should be performed in order to proactively anticipate and mitigate issues before they occur. These risk assessments should be performed by a multidiscipli­nary team. To determine the appropriate specialties that should be represented on the team, the process should rst be examined from start to end to identify all role groups that lend a hand to the process. For example, successful implementation of AI ART requires contributions from radiation therapists, both during CT simu­lation and at the treatment machine, dosimetrists, medical physicists, and radiation oncologists [3]. Therefore, a representative should be nominated from each role group to participate in the risk assessment process. At a minimum, we recommend that a failure mode and effects analysis and hazard testing of the major components of the ART workow be performed prior to go-live. Each of these methods will be discussed in the following sections.
20.1.1.1 Failure mode and effects analysis
One type of prospective risk assessment that is used in many industries to anticipate and limit risk is called failure mode and effects analysis (FMEA). FMEA is a systematic approach that can be used to ensure that potential weaknesses have been identied and mitigated before their occurrence in a workow [4]. During FMEA, a multidisciplinary team of representatives from all participating role groups is assembled. This team then creates a process map, during which each step of the workow is identi ed and visually mapped out. An example of the steps that would be included in a process map detailing a persons daily drive to work is shown in gure 20.1. For a workow to be completed correctly, each step of the process map must be completed without error.
Within a radiation oncology department, process maps should be created that encompass the entirety of the workow, from the patients rst appointment at CT simulation to their nal treatment fraction. When implementing AI ART, it is important to create additional process maps that focus on how the updated workow differs from the standard clinical procedure.
For each step identied in the process map, the team then attempts to predict any potential error, or failure mode, that could occur while completing each given task. These failure modes should include any action that leads to an undesired outcome, including both large-scale issues that could potentially impact nal patient treatment and small issues that may only temporarily inconvenience staff. For each failure mode, a fault tree is then created. Fault trees are visual diagrams used to identify potential causes for each given error. For example, if the failure mode identied is someone running a red light while driving, potential causes could include distracted
Figure 20.1. An example process map, created to detail the steps involved in commuting to work by car.
20-2
Artificial Intelligence in Adaptive Radiation Therapy
Figure 20.2. An example of a fault tree, generated to identify causes associated with the failure mode of running a red light while driving (red). Identied causes are shown in blue, with possible contributing factors to each cause are shown in yellow.
driving, mechanical issues, and lack of visibility (gure 20.2). Failure modes should be evaluated independently for each identied cause of the error, as some causes are more likely to lead to an event occurring than others.
To maximize the safety of this process, each of these causes should be appropriately addressed prior to the driver returning to the roads.
To quantify the risk, each failure mode is then assigned three numerical scores. First, occurrence (O), which describes the likelihood of that error occurring. Next, severity ( S) describes how dangerous or detrimental the effect would be if that error were to occur without being detected. Finally, detectability (D), describes the likelihood that the failure mode will not be detected in time to prevent an event from occurring. It is important to note that scoring for FMEA is a subjective process and can be based on a combination of participants clinical experience, as well as examples of error occurrence in the literature. By instructing all members of the FMEA team to follow a clear set of scoring guidelines throughout the evaluation, some subjectivity can be eliminated and consistency can be expected. An example of scoring guidelines, similar to those recommended by TG-100, that could be used is shown in table 20.1 [4].
All three scores are then multiplied together to obtain a metric called the risk priority number (RPN), which is a surrogate for the amount of risk that the given error poses to patients. A higher RPN indicates higher risk, and therefore mitigating failure modes with higher scores should be prioritized.
By performing an FMEA prior to the clinical implementation of a new tool or process, the workow can be designed with proper quality control process steps to eliminate the risk that could be passed down to patients undergoing treatment. Staff training can also be tailored to highlight points of potential risk and limit the
20-3
Artificial Intelligence in Adaptive Radiation Therapy
Table 20.1. Example of FMEA scoring guidelines that could be used to create consistency in scoring across all participants.
Score
Qualitative Frequency Impact
1 Failure unlikely 0.1% No effect 0.01% 2 0.02% Inconvenience 0.2% 3 Failure infrequent 0.04% 0.5% 4 0.1% Minor dosimetric error 1.0% 5 < 0.2% Limited toxicity or tumor 6 Occasional failure < 0.5% 5.0% 7 < 1% Potentially serious toxicity 8 Frequent failure < 2% 15% 9 < 5% Potentially very serious
10 Failure
Occurrence (O)
inevitable
Severity (S) Detectability (D)
Probability of failure mode going undetected
2.0%
underdose
10%
or tumor underdose
20% toxicity or tumor underdose
> 5% Catastrophic > 20%
likelihood of these errors occurring. By prospectively identifying and mitigating points of risk, a culture of safety can be created among the clinical team.
20.1.1.2 Example of improvements made to workflow based on FMEA results
In radiation oncology, FMEA has been proven effective at eliminating points of risk prior to impacting patient care when introducing technologies such as auto­contouring, Gamma Knife and the Halcyon into the clinical workow [58]. Similarly, several groups have also identied the benets of applying FMEA to evaluate the deployment of AI ART tools [2, 9, 10].
One group that has shown the benets of using FMEA to limit risk in AI ART processes is Liang et al who performed an FMEA to evaluate the proposed workow for an MR-linac being deployed into a clinic that had previously only made use of standard, non-adaptive, treatment machines [2]. The authors noted that the workow for the adaptive system was radically different than the conventional treatment workow, and therefore additional quality management resources were needed. An online reference form was created to guide clinical team members through the treatment planning, quality assurance, and delivery processes. The FMEA revealed that while this reference guide was useful for new users, a more concise checklist should be developed to be completed by physics staff for each fraction of patient treatment. The checklist was made to comply with recommen­dations from AAPM and contains essential information to be checked before, during, and after treatment to ensure the treatment was delivered as intended and to mitigate several identied high-risk failure modes.
20-4
Artificial Intelligence in Adaptive Radiation Therapy
20.1.1.3 Hazard analysis
Another type of prospective risk assessment that can be used to optimize safety when implementing an AI ART workow into the clinic is hazard analysis, as recom­mended by IEC 62 366: Application of usability engineering to medical devices [11]. A hazard scenario refers to a potentially dangerous or risky situation that may occur when an error is introduced into a workow. If this error goes unnoticed by members of a radiation therapy team, it could heighten the risk and jeopardize patient safety. Conducting a hazard analysis allows the identication of the root causes of these scenarios, enabling the implementation of additional safeguards to mitigate and address potential risks. An illustration of hazard analysis is evident in the research conducted by Pawlicki et al where they employed a tool known as system theoretic process analysis (STPA) to pinpoint and mitigate potential hazards within clinical radiation therapy workows [12]. Other studies demonstrated the advantages of applying hazard analysis to evaluate the clinical safety associated with the use of the Halcyon and an automated contouring and treatment planning tool [13, 14].
In order to perform a hazard analysis, errors that are likely to occur in a given workow should be simulated to evaluate their detectability. When performed in conjunction with an FMEA, the list of high RPN failure modes can be used to inform which errors, or hazard scenarios, should be inserted into the workow. An end-to-end test of the workow should then occur during which all members of the clinical team complete their corresponding task, with the hazard present in the process to determine if or when the error hazard is detected. If the hazard is detected and corrected prior to impacting the nal output, such as high-quality patient treatment, then the process is working as designed and intended.
If undetected, feedback should be requested from participants to inform what changes in the process or quality management steps should be made to increase detectability. Following changes to the workow, the testing should be repeated with a new, blinded, set of participants to conrm that the changes made were effective in reducing risk.
20.1.2 Root cause analysis
To maximize the effectiveness of a quality management program, retrospective or reactionary evaluation techniques should also be utilized. Root cause analysis (RCA) is one retrospective risk assessment technique that should be incorporated into radiation therapy programs, including those that utilize AI ART systems [15]. RCA is a process used to identify the cause of a safety event, or failure mode, that has occurred in clinical practice. Performing an RCA requires users to step backward through a process, originating from the safety event until the decision point is identied which causes the workow to divert from the intended outcome, i.e. successful patient treatment. To successfully perform an RCA, members of all subgroups of the clinical team must participate in order to represent all perspectives of the event that occurred. The team must then work to identify both what happened at each step in the process, and also why each decision was made that allowed the
20-5