Volume 57, Number 1, January-March 2022
Page(s) 33 - 40
Published online 17 January 2022

© SFRP, 2022

1 Introduction

Radiotherapy is one of the most common cancer treatment methods worldwide. This method is applied either as a single treatment or in different combinations with surgery and chemotherapy, taking into account the type and stage of the disease, the age and gender of the patient, and the patient’s condition (Beyzadeoglu et al., 2010). More than a century of scientific studies and developing technology have advanced RT in terms of application, treatment success and safety, thus, strengthening its contribution to the fight against cancer.

With the introduction of image-guided radiotherapy (IGRT) methods and systems that can synchronize patient movements, the reliability of RT has undoubtedly increased. However, despite the care taken to ensure accurate and correct management in RT, many unsuccessful treatments and resulting deaths have been reported over the years (International Atomic Energy Agency [IAEA], 1998; Mettler and Ortiz López, 2001; Johnston, 2006; ASN Report, 2007; Bogdanich, 2010; Johnston, 2019).

As a result of the rapid development of RT technologies, traditional quality control methods, published by international organizations such as the AAPM, ASTRO, ACR, ESTRO, and IAEA are no longer adequate. These methods have focused on functional performance of RT equipment, but the majority of accidents and incidents in RT are observed during the workflow and application processes, independent of devices and equipment (Derreumaux et al., 2008; ICRP, 2009; Huq et al., 2016; Thellier et al., 2021).

Risk assessment methods are used in many work areas considered high-risk areas, such as in the nuclear, space exploration, aviation and chemical industries. Fault tree method is one of the probabilistic risk assessment (PRA) and security analysis methods, and uses logical diagrams to highlight the interconnections between system errors, system defects and system components errors. Approaches to fault trees can be described as either a model, or a graphic. This model aims to ensure that events and errors responsible for undesired events (top events or accidents) in the system are predictable (Signoret and Leroy, 2021).

Fault trees can be constructed using either or both retrospective and prospective analyses, and this flexibility makes them useful tools in safety analysis (Ortiz López, 2012). FTA is used to perform risk assessment, calculate risk level, identify critical security components, define the functions of security components and measure the impact of designed changes. The fault tree covers the following aspects: failures, faulty events, normal events, environmental effects, systems, subsystems and components, and system elements (software, hardware, human, instructions etc.) (Vesely, 2002).

As briefly mentioned above, it is very important to evaluate RT applications, considered as high-risk working areas, in terms of risk and security, and to make the necessary improvements (Ekaette et al., 2007; Gilmore and Rowbottom, 2021). In this study, it is aimed to predict possible accidents in RT applications at the radiation oncology department of Medicalpark İzmir Hospital, and to take preventive actions. This will be achieved by conducting a quantitative and qualitative evaluation of the system using the fault tree method, a well-known probabilistic risk analysis method.

2 Materials and methods

In this study, probabilistic risk analysis was performed for RT applications in İzmir Medicalpark Hospital between 2015 and 2017. The department of analyzed radiation oncology consists of 1 radiation oncologist, 3 medical physicists, 5 radiotherapy technicians, 1 oncology nurse and 3 department secretaries. In this department, 1 PET-CT and 1 linear accelerator device capable of performing volumetric arc treatment are used to treat approximately 1000 patients in approximately 20 000 treatment sessions annually.

2.1 Determination of workflow

The first step of FTA is to determine the workflow. Workflow is used for systems in which certain procedures are performed in specific order, in specific ways and at specific times from the beginning to the completion. In such systems, there is no variation in the location, duration or form of the application steps, except for special cases. The work flow charts created by schematizing the application processes serve to provide an overall view of the process and allow evaluation of the system.

As in the standard, RT application stages are divided into three main parts in the study (1. simulation [fixation and imaging]; 2. treatment planning (dose planning); 3. treatment delivery). This is followed by the determination of the work flow charts of the RT application steps and the disciplines involved (Fig. 1). At this stage, a detailed work flow chart was created by dividing each main section into sub-application steps. This allowed each link in the radiotherapy chain to become more visible in terms of predictability of malpractices.

thumbnail Fig. 1

Simulation (a), treatment planning (b) and treatment delivery (c) workflow charts and which disciplines involved.

2.1.1 Simulation

The simulation process is performed to determine the target area on the patient and to prepare the physical parameters required for treatment. Currently, it is performed on computed tomography (CT) devices providing DICOM (Digital Imaging and Communication in Medicine) data required for three-dimensional (3D) dose calculation. At the start of the procedures, equipment is selected that will allow the patient to remain comfortably in the appropriate position for RT. Then, the patient is placed on the CT device with this equipment, in the most suitable position for RT. The projections of the wall lasers on the patient are marked with radiopaque markers in order to ensure the spatial coordinate alignment between the RT device and the CT-simulation device. This is followed by tomography scan that is most appropriate for the RT treatment method. After the necessary controls, the CT data are transferred to the treatment planning system. Although the application steps are the same, the simulation process is patient-specific. The suitability and reproducibility of the simulation for the RT technique is very important in terms of the accuracy and quality of the treatment. An effective simulation process is the first step towards an effective RT treatment application.

2.1.2 Treatment planning

In this study, the dose planning section is divided into 11 steps (Fig. 1). DICOM data of the patient are sent from the CT simulator and import to TPS. In this process, it is important that all data is accurate and it may be necessary to fuse a patient’s CT images and magnetic resonance (MR) images. Target volume (PTV) and organ at risk (OAR) volumes are determined on CT images by the radiation oncologist, and then the patient’s RT prescription is determined. After these processes, necessary contour separation, joints and drawings are made in the TPS system to allow the optimization algorithms to function properly. In the dose calculation step, the medical physicist determines the most appropriate treatment technique for the patient. In the plan acceptance stage, the radiation oncologist and medical physicist examine the treatment plan. In this section, re-planning or re-contouring can be performed when an unwanted situation is encountered. The accepted plan is evaluated in the dosimetric quality control process before being applied to the patient. When it passes dosimetric quality control, the plan is electronically signed in an interface program where the treatment is regularly recorded and all treatment parameters are continuously monitored. In this way, it is aimed to prevent under or overdose delivery.

2.1.3 Treatment delivery

In this stage, the treatment plan obtained following preparations is applied to the patient. This part is also called set-up, which means first day treatment, configuration or adjustment. In this section, radiation oncologists and medical physicists accompanying the technicians review all treatment parameters. Firstly, the patient is positioned on the treatment device under simulation conditions and the necessary table shifts are made to match the target volume (PTV) center with the beam center. The imaging system in the RT device displays the target area of the patient. The image obtained is matched with the TPS images on which the entire calculation is performed. Precise matching of the target area and surrounding tissues is enabled by the image guide system. After spatial verification of the tumor location, all parameters of the treatment plan are checked. During the treatment, the patient and treatment parameters are monitored from the control room.

2.2 Creating the fault tree

Fault tree analysis (FTA) starts with graphical evaluation. The top event is the main subject of analysis and refers to the most significant impact, performance, injury, damage or loss. FTA includes application-related factors. Other events or faults that occur under the direct or indirect effect of these factors constitute a top event. The factors thought to affect the top event are placed in the diagram. The fault tree is completed by building the diagram until it covers all sub-factors (Jakóbczak, 2016). In this study, the top event was determined as the case of incorrect RT dose, or RT dose distribution region that caused unsuccessful administration as stated in Task Group 100 of the AAPM (Huq et al., 2016) Then, the three main triggers that may cause this top event are categorized as simulation errors, planning errors or treatment delivery errors. In fault tree terminology, these steps are called intermediate events. Intermediate events may consist of sub-intermediate events or basic events that may give rise to these. The top event, intermediate events, sub-intermediate events and basic events in the study are shown in Table 1. Independent basic events that can cause intermediate events are combined using Boolean mathematical logic circuits, one of the most common elements in these are AND and OR gates. In a circuit connected with the AND gate, for an event at the top to occur, the underlying cause events must occur simultaneously. Therefore, probability of the upper event occurrence is equal to product of the lower events probabilities. At the OR gate, for the upper event to occur, it is sufficient to have any one of the cause events under it. Therefore, probability of the upper event occurrence is equal to the sum of the lower events probabilities. In this study, each intermediate event was analyzed down to sub-intermediate events and basic events. To determine the analysis and error sources, the following were utilized: The Nuclear Regulatory Commission (NRC), Radiation Oncology Safety Information System (ROSIS), IAEA safety reports, and clinical experience and knowledge. In addition, the definitions provided in Ford et al. (2012) were used to determine the detailed workflow and error sources.

For quantitative analysis, the fault tree model created using SAPHIRE 8 software was obtained from Oak Ridge National Laboratory (Fig. 2). SAPHIRE has features suitable for making risk analysis studies and evaluations. Probabilistic risk analysis is performed to assess the risk in a complex system.

In the quantitative calculations, the top event needs to be analyzed in detail to reveal the most fundamental factors, depending on the nature of the study. It is necessary to determine the occurrence probability of the basic events obtained. However, there are some difficulties for RT at this point. Developing PRA models in RT applications is challenging due to the lack of data on the occurrence of faults. Therefore, most of the risk analysis studies in such systems rely on expert consensus to estimate the required probability data (Olch, 2014). Similarly, in the study, the quantitative analysis of the system was carried out by expert judgment method.

Table 1

Top event, intermediate, sub-intermediate and basic events.

thumbnail Fig. 2

Top (green), intermediate (blue), sub-intermediate (gray) and basic (yellow) events of the fault tree created for RT applications in SAPHIRE software.

2.3 Expert judgment method

Experts are indispensable for modern organizations, because of their ability to fill gaps in knowledge and establish a bridge between pieces of information when there is a lack of data in an important event, method or system. This knowledge can reduce uncertainty (Benini et al., 2017). In fault tree analyzes, statistical data can be found in terms of frequency estimates for event initiators, but such data is very rare when estimating conditional probabilities in a model. Therefore, expert judgment is essential when the risk model involves quantitative evaluation (Rosqvist and Tuominen, 1999). In this study, expert judgment was obtained via face-to-face interviews in order to reduce the uncertainty of the data to be collected, and to eliminate the worthless data. These interviews were conducted using a specific methodology, with uniform steps applied to each expert.

In this study, the expert judgment method was carried out below in the following order.

2.3.1 Selection of experts

In the setting of this study, 1 radiation oncologist, 2 medical physicists, 1 intern medical physicist, and 5 radiotherapy technicians work in the radiation oncology clinic. The study involved evaluating the data of patients treated between 2015 and 2017. For this reason, all employees were included in the study, except 1 intern medical physicist and 1 technician each with less than one year of work experience. The role of the nurse is patient care and psychological support in RT clinic, so no nurses were included in the study, as their role did not affect the “mistreatment” event.

2.3.2 Informing the experts

Face-to-face interviews were conducted individually, in the radiation oncology clinic. Before the questions, each expert was informed about the purpose of the study, its rationale, expectations and method.

2.3.3 Data collection and assumptions

In the analyzed radiotherapy system, preparation of radiotherapy and first day of treatment (set-up) were evaluated. After set-up, other fractions were not included in the fault tree. 2707 RT application performed between 2015 and 2017 were evaluated. The data required for the fault tree were obtained from expert interviews. After performing the fault tree, intermediate-events and basic events that may cause the undesired top event were determined, and questions about each basic event were prepared. Then, experts were asked each question in a clear and understandable manner associated with that basic event (fault). In addition, during the interview, the questions were individually explained and information was provided to increase the quality of the data. In order to increase the significance of the numerical values given to the error models, the information was given on how many times the application was carried out by each expert. In addition, in order to increase the reliability of the expert’s answers, three values were requested for each question as maximum, minimum and most probable.

2.3.4 Data aggregation and evaluation

Weighting method was performed when combining the data provided by the different experts from the group. This method includes the following information; the number of years the expert worked, the number of RT applications after 2015, and the reliability coefficient given by the expert physicist. This merging procedure was performed with arithmetic mean for medical physicists. Since there is only one radiation oncologist, his data were used without weighting.

2.4 Solving of fault tree

All basic event probabilities provided from expert assessment were entered in the SAPHIRE program. After the calculations were performed by the software, the probabilities of top events determined as incorrect dose or dose distribution in RT applications and sub events were calculated.

3 Results

When the fault tree was resolved, 87 CUT-SET were calculated (A CUT-SET in a fault tree is a set of basic events whose simultaneous occurrence ensures that the top event occurs). Table 2 shows the first 5 CUT-SET that made the most contribution to the analysis result for incorrect dose or dose distribution (top event).

As a result of the analysis, the occurrence probability of the top event was calculated as 5.371 × 10−3. The analysis results of the three subsections under the top event are given in Table 3.

It was observed that the greatest contribution to the top event was the image guidance matching error, with a rate of 7.88%. The probability value is calculated as 4.23 × 10−4. The 2nd highest contribution was the examination information error of the patient, with a rate of 7.21% and its probability was calculated as 3.87 × 10−4 (Tab. 2).

In the analysis, it was calculated that the step contributing the 3rd highest amount to the top event may occur in the patient follow-up during set-up. The contribution to the top event was found to be 5.57% and the error probability was calculated as 2.99 × 10−4 (Tab. 2).

It was calculated that the step contributing the 4th highest amount to the top event could be sending the wrong images taken for simulation. The contribution of this error to the top event was found to be 5.43% and its probability was calculated as 2.92 × 10−4.

As a result of the analysis, it was seen that the 5th highest contribution to the top event may develop due to the defects in the quality of the images. The contribution of such events to the top event was calculated as 4.33% and the probability as 2.32 × 10−4.

Table 2

CUT-SET report of incorrect dose or dose distribution region.

Table 3

Analysis results of simulation, treatment planning and treatment delivery.

4 Discussion

In this study, the RT clinic was evaluated using the FTA method. This method enables qualitative and quantitative risk analysis of systems. In quantitative part, the expert judgment method was used for error probabilities. The technicians, medical physicists and radiation oncologist provided data on their areas of expertise.

The analysis results show that the greatest contribution to the top event was the image guidance matching error. There are two control mechanisms in this section, the oncologist and the practitioner technician, but the practitioner’s self-control is the weak point of the control mechanism. In addition, increases in the clinical workload weaken the control mechanism, as the technician’s image matching control time decreases. Especially during the busy hours of the clinic, inappropriate matching may result from acting hastily (human failure). The literature confirms that one of the most important problems of new systems is caused by human error in image guidance (Baeza, 2012).

Gross Tumor Volume (GTV) is defined based on patient diagnostic information, such as PET, MRI, surgical margin, and the number of lymph nodes involved. An error in this information can lead to the wrong treatment from the very beginning (for example, if diagnostic patient information is outdated or missing). In clinical practice, the only control point in GTV determination is the radiation oncologist. The radiation oncologist stated that it is difficult to detect this error, especially when there is inaccuracy in patients’ MRI, PET and surgical information. In the study, this was confirmed by the patient’s examination information error being the second highest contributor to the top event.

Movement of the patient during the treatment means not being able to deliver the desired dose to the target area, and may cause a critical organ to be exposed to an extra dose. This eventuality directly affects the values of TCP (tumor control possibility) and NTCP (normal tissue complication possibility), which are extremely important in radiotherapy. Therefore, the patient should be followed during the treatment. This was supported by the results showing that the third highest contribution to the top event came from patient follow-up error during treatment.

In the clinic, information about the simulation process is verbally communicated to the technician by the radiation oncologist. Due to communication problems, more than one CT scan may be performed due to the wrong CT scanning parameters. In addition, more than one CT scan may be performed for patient-specific reasons (rectal fullness, insufficient filling of the bladder, intestinal gas, etc.). Human error resulting in the transfer of the wrong CT image to the planning contributed to the top event in the 4th order.

With the development of IGRT methods, more precise detection of target regions and critical organs can be made. In this way, because the treated areas can be seen more clearly, the prescription doses in radiotherapy have been increased. However, errors in target detection may occur due to problems in imaging systems (such as deterioration in image quality, artifacts caused by the presence of implant materials, blurring due to patient movement, incomplete or incorrect calibration, etc.). Therefore, dose uncertainties and erroneous treatments may occur. The statistical analysis also supported this situation and it was observed that the fifth highest contribution to the top event was the defects in image quality.

In the literature, there are a small number of quantitative PRA studies using fault tree method for RT clinics. The comparison of the current analysis results with those of Ekaette et al. (2007) is given in Table 4. The evaluation of the main fault tree results shows that the error probabilities are very close to each other, and it can be concluded that the top event FTA results are compatible with the literature. Ekaette et al. (2007) used 4-stages procedure: prescription, simulation, treatment planning and data entry. In the current study, there were 3-stages: simulation, treatment planning and treatment delivery. In the comparison, the simulation-based fault probabilities are generally consistent across studies, and discrepancies are likely to be due to differences in clinical conditions, application, and immobilization equipment. The main reason for the difference between the analysis results of the treatment planning part is likely caused by expert estimates. In addition, block use was not applied in this study, unlike in Ekaette et al. (2007). It appears that differences between clinics affect the results. The small discrepancies may be because the treatment delivery step was additionally evaluated in this study. In addition, in the study of Ekaette et al. (2007), there are more personnel in the expert judgment team and more patients, which is a drawback of the current study in terms of modeling the system and the reliability of the data.

Table 4

Comparison of fault tree analysis results.

5 Conclusion

In this study, PRA of the RT clinic was carried out using the fault tree method. This proactive method operates differently from deterministic methods, and provides quantitative evaluation of the aspects closest to the fault. Each radiotherapy clinic has its own characteristics in terms of the workflow, the number and variety of staff, the distribution of tasks, the devices and equipment used, and the application techniques. These differences show a need for every clinic to perform parallel studies. In addition, AAPM TG 100 recommended that each clinic should create and evaluate studies in proactive quality management in radiotherapy (Huq et al., 2016).

One of the greatest challenges for this type of study is obtaining the probability data for the fault tree. In the analysis of such systems, the best data source is the expert judgment method, and the insufficient number of experts engaged in expert evaluation method may have caused inaccurate estimates.

The transfer of deficiencies and mistakes by new employees due to workload, patient density and insufficient number of staff may increase the risk of accidents in the long-term. Thus, PRA is important for safe application and management, especially when a new clinic is established, or a new device, equipment or application method is commissioned. After the risk analysis, the appropriate quality control method for weak points should be determined and implemented. This study, which is the first in our country, is a preliminary study aimed at improving quality management with probabilistic safety analysis in radiotherapy.

Conflict of interest

The authors declare that they have no conflict of interest.


This research did not receive any specific funding.

Ethical approval

Ethical approval was not required.

Informed consent

This article does not contain any studies involving human subjects.

Authors contributions

Ç. Özbay: conceptualization, methodology, investigation, writing original draft, writing-reviewing and editing; T. Özbay: investigation, writing-reviewing and editing; A. Güler Yiğitoğlu: visualization, investigation; M. Bayburt: supervision.


The authors thank technicians and experts, who provided information regarding the RT system and provided subjective probabilities for the analysis, in department of Radiotherapy at the İzmir Medicalpark Hospital. The authors also thank Oak Ridge National Laboratory for providing SAPHIRE software. The authors also thank Simon Mumford for proofreading the article.


Cite this article as: Özbay Ç, Özbay T, Güler Yiğitoğlu A, Bayburt M. 2022. Probabilistic risk assessment of radiotherapy application. Radioprotection 57(1): 33–40

All Tables

Table 1

Top event, intermediate, sub-intermediate and basic events.

Table 2

CUT-SET report of incorrect dose or dose distribution region.

Table 3

Analysis results of simulation, treatment planning and treatment delivery.

Table 4

Comparison of fault tree analysis results.

All Figures

thumbnail Fig. 1

Simulation (a), treatment planning (b) and treatment delivery (c) workflow charts and which disciplines involved.

In the text
thumbnail Fig. 2

Top (green), intermediate (blue), sub-intermediate (gray) and basic (yellow) events of the fault tree created for RT applications in SAPHIRE software.

In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.