Fuzzy Analytical Hierarchy Process-based Risk Priority Number Approach in Failure Modes and Effects Analysis for Magnetic Resonance Imaging-guided High-dose-rate Brachytherapy for Gynecologic Cancer

Abstract

Purpose

Failure modes and effects analysis (FMEA) is commonly used to identify, prioritize, and mitigate potential failure modes (FMs) by assigning a risk priority number (RPN). However, traditional RPN-based FMEA has limitations, particularly when handling the degree of interdependency within processes. To address this, we propose a fuzzy analytical hierarchy process (AHP)-based RPN method, designed to prioritize FMs by accurately weighing risk factors in magnetic resonance imaging (MRI)-guided high-dose-rate brachytherapy (BT) for gynecologic (GYN) cancer.

Methods and Materials

A process map covering all steps was developed for MRI-based GYN BT, and potential FMs were identified. Evaluators were given 2 questionnaires, one for conventional FMEA and another for AHP evaluation. For the AHP method, substeps were grouped by job specialty, with identical weights applied to FMs within each specialty group. Fuzzy linguistic terms helped evaluators handle uncertainties, and final fuzzy AHP-based RPN values were calculated by applying weighted risk factor scores.

Results

The process map included 6 primary steps, 33 substeps, and 82 FMs. In the fuzzy AHP RPN analysis, the top 5 FMs were identified as incorrect/suboptimal applicator insertion, incorrect applicator reconstruction, dose-volume histogram not meeting the physician's intent, incorrect/missing contours, and applicator/patient movement. By comparison, the conventional FMEA ranked the top 5 as incorrect/missing contours, incorrect/suboptimal applicator insertion, dose-volume histogram not meeting the physician's intent, applicator/patient movement, and incorrect applicator reconstruction. FMs with rank differences of 10 or more between methods were mostly related to applicator insertion and MRI.

Conclusions

This study demonstrates the feasibility and effectiveness of a fuzzy AHP-based RPN method for comprehensive FM prioritization, tailored to the clinical workflow of MRI-based GYN BT. Our findings provide a valuable reference for implementing fuzzy AHP-based risk assessment in MRI-guided BT.

Introduction

Magnetic resonance imaging (MRI) is considered the gold standard imaging modality for cervical cancer because of its superior soft tissue imaging characteristics, allowing for the brachytherapy (BT) to be tailored individually with accurate delineation of the gross tumor volume and the high-risk clinical target volume. The Gynecologic Groupe European de Curiethérapie—European Society for Radiation Therapy and Oncology working group1, 2 has emphasized the important role of MRI in the successful implementation of 3-dimensional image-based adaptive treatment planning in cervical cancer BT to improve local control and reduce high-grade toxicities.

Despite the advantages of MRI-based BT, its implementation is limited by the accessibility of MRI in the radiation oncology department, the increased time and expertise required for MR-based BT, and the elevated costs associated with additional imaging.3, 4 Moreover, high-dose-rate (HDR)-BT requires significant coordination and cooperation from multiple departments to complete each task within tight time constraints. Each step in the workflow of HDR-BT is considered a high-risk procedure. Thus, it is crucial to enhance the HDR-BT risk management and workflow efficiency to enable team members to provide more effective care and lead to better patient outcomes.

Failure modes and effects analysis (FMEA) is a step-by-step approach for identifying all possible failure modes (FMs) and mitigating their effects. FMEA is also used to improve process efficiency by implementing preventive measures after prioritizing all possible FMs. The process involves identifying and evaluating potential FMs in a process and their effects, prioritizing FMs using the risk priority number (RPN), and determining actions that could minimize the risk of occurrence of FMs. The RPN value serves as a key metric in FMEA and is calculated by multiplying 3 risk factors: severity (S), occurrence (O), and the probability of not being detected (D). A higher RPN value indicates a higher overall risk.

Several papers on FMEA5, 6, 7, 8 have been published for image-guided HDR-BT. The American Association of Physicists in Medicine commissioned Task Group (AAPM TG)-303 to conduct an FMEA for MRI-based HDR prostate BT, covering a total of 8 steps from patient assessment to posttreatment actions/quality assurance (QA).⁵ The highest-ranked FMs were associated with target contouring and needle misidentification, leading to dose deviations. Mayadev et al⁶ conducted an FMEA study for gynecologic (GYN) BT, identifying a total of 170 FMs from consultation to treatment delivery. The 2 highest-ranked FMs were related to human errors, such as miscommunication and physics plan checks. Wilkinson et al⁷ performed an FMEA for HDR-BT planning. The top-ranked FM was the wrong choice of tip/connector end for reconstruction, followed by the wrong usage of image sets caused by its high S score. Su et al⁸ conducted an FMEA to guide the transition to a different HDR after loader system. The top-scored FMs were missing tests in daily QA and human errors related to incorrect user inputs of source information or indexer length. These FMEA studies were applied to various clinical situations, evaluating both new and existing processes. The high-ranked FMs for each study can differ based on their goals, clinical situations, and the perspective of team members.

However, there are several limitations associated with traditional RPN-based FMEA, some of which are as follows: (1) The 3 risk factors are assumed to have the same importance weight, whereas the importance of each risk factor may be unequal in practical applications. (2) Different combinations of risk factor ratings may result in the same RPN value, making it difficult to precisely evaluate each risk factor. (3) Assigning exact numerical values (on a scale from 1 to 10) to evaluate FMs can be challenging, especially in complex clinical scenarios. (4) The method does not explicitly quantify the degree of interdependence or influence between FMs, limiting its ability to address complex interactions. These limitations highlight the need for more flexible and advanced risk assessment approaches that can better align with clinical workflows and provide a more accurate and comprehensive evaluation of FMs.

Fuzzy theory9, 10 is one of the approaches to resolve the limitations of traditional RPN-based FMEA. Giardina et al¹¹ proposed a fuzzy rule-based FMEA for HDR-BT safety analysis, using fuzzy inference to evaluate FMs and human error. This approach mitigates the drawbacks of traditional RPN values using fuzzy numbers to assess FMs and the relative importance of risk factors. Recently, the integration of fuzzy theory with multiple criteria decision-making methods (MCDMs) has emerged as a potent tool. MCDM techniques aid in decision-making processes involving the assessment of multiple criteria that require simultaneous consideration. The analytical hierarchy process (AHP)12, 13 is one of the well-known MCDMs for prioritizing in the form of a hierarchy of independent elements. Recently, a fuzzy AHP-based RPN approach for risk assessments has been successfully implemented in the commissioning process of a ring gantry low-energy linear accelerator system.¹⁴

In this study, we proposed a fuzzy AHP-based RPN approach, and this approach is specifically designed to enhance both workflow efficiency and risk assessment, prioritizing potential FMs in a manner tailored to our clinical setting and overcoming the limitations of conventional RPN methods. This manuscript provides a detailed identification of potential FMs and their impacts, prioritization using the AHP-based RPN approach, and the development of mitigation strategies for high-priority FMs. Furthermore, we compare the results of the fuzzy AHP method with conventional FMEA, highlighting differences in risk rankings and demonstrating that our assessment aligns accurately with our MRI-based GYN BT workflow.

Methods and Materials

FMEA was conducted by a qualified team involved in the MRI-based BT process. Following AAPM TG-100 guidelines for FMEA,¹⁵ a process map was generated to visually outline the entire process described in Fig. 1. Using this map, a multidisciplinary team comprising nurses, physicians, medical physicists/dosimetrists, MRI technicians, and therapists identified potential FMs. The study divided the entire process into 6 major steps: (1) preprocedure, (2) applicator insertion and simulation, (3) MRI, (4) treatment planning, (5) treatment delivery, and (6) posttreatment.

Figure 1.

Summarized process map. It consists of 6 primary steps, representing 33 sublevel processes, and a total of 82 FMs were identified. The primary job-specialty groups are indicated through a color code for each step.

In the conventional evaluation, 3 physicians and 4 physicists in radiation oncology departments were involved in the evaluation, regardless of experience and knowledge level. Following the AAPM TG-100 guidelines, we instructed evaluators to score the 3 risk factors, ensuring each FM was assessed independently. AAPM TG-100 guidelines provide clear descriptions, with O quantified by frequency percentage, S categorized descriptively, and D estimated by the probability of identifying the FM if it occurs.¹⁵ These standardized criteria allowed evaluators to assess each FM objectively, avoiding speculative or unverified interdependencies based solely on personal experiences or assumptions. This approach was not intended to disregard the principles of FMEA or the importance of interdependencies within the process but rather to ensure a reliable and unbiased initial risk assessment.

In the AHP evaluation, 2 qualified physicists were selected to design and implement the AHP process. These individuals were chosen for their in-depth understanding of the BT process, their expertise in how FMs within substeps contribute to the overall workflow, and their background in risk management or formal training in FMEA. These evaluators score the weights of risk factors using fuzzy numbers. Subsequently, evaluators scored the weights of risk factors using fuzzy numbers, considering the degree of interdependencies among FMs in the 6 major steps and their respective substeps. Details on fuzzy linguistic terms and the AHP methods are described in the Fuzzy Linguistic Terms section, while the final fuzzy RPN calculation is detailed in the Fuzzy AHP-based RPN section.

Fuzzy linguistic terms and AHP

The fuzzy AHP method determines the weights of risk factors for FMs. Qualified members of the physics FMEA team categorized substeps within the steps based on relevant attributes tied to their expertise. A broader group of specialists, comprising nurses, physicians, medical physicists/dosimetrists, MRI technicians, and therapists, was divided into distinct groups based on tasks and responsibilities. Within the same expertise group, we assumed that FMs share identical weight values in the substep because assigning different weights to each FM within the substep is labor-intensive and practically unfeasible. We reduced the number of expert groups within the steps in our process map to ease the burden for evaluators.

The physics FMEA team members established evaluation criteria for AHP. The letter “O” is linked to the causal influence of FMs on the emergence of other FMs in job-specialty groups or their potential impact on work efficiency on occurrence. The ‘S’ reflects the severity degree of FMs and their impact on other job-specialty-related FMs, while detectability is linked to feedback, where the detectability of later FMs leads to an increase in the detectability of earlier FMs. Our goal is to comprehensively understand the degree of interrelationships and influences of various FMs within different job-specialty groups in our clinical environment.

As the evaluation criteria for the AHP method lack quantifiability, we incorporated the fuzzy theory introduced by Zadeh^9,10 into the AHP method, allowing evaluators to manage uncertainty in risk assessments. The evaluators assessed the importance of risk factors using 5 fuzzy linguistic terms. Each term is represented by triangular fuzzy numbers, defined by 3 points, $\stackrel{\sim }{M}=\left(l,m,u\right)$, which are lower value $\left(l\right)$, middle value $\left(m\right)$, and upper value $\left(u\right)$ representing its central value $\left(m\right)$ and the left $\left(l\right)$ and right ends $\left(u\right)$ of the respective triangular shape depicting a probability distribution. The linguistic scales used for AHP include very low (VL), low (L), medium (M), high (H), and very high (VH). Their corresponding triangular fuzzy numbers are (1, 1, and 3), (1, 3, and 5), (3, 5, and 7), (5, 7, and 9), and (7, 9, and 10), respectively.

In this study, the fuzzy extent analysis method introduced by Chang,¹⁶ one of the popular techniques in fuzzy AHP methods, was implemented. Within the AHP framework of our current study, the goal is to determine the relative weights of risk factors for substeps associated with specific job specialties. This is accomplished by using evaluation criteria that emphasize the interrelationships among these job-specialty substeps. Subsequently, evaluators assess the risk factors of each substep based on these criteria.

The fuzzy AHP method is summarized below. First, evaluators’ responses using linguistic terms were converted into triangular fuzzy numbers. These fuzzy numbers were then used to construct AHP pairwise comparison matrices, expressing the relative priorities of risk factors. The pairwise comparison matrix ($\stackrel{\sim }{X_k}$) is illustrated in Equation 1, where ${\stackrel{\sim }{x}}_{ij}^k$ represents the evaluator's assessment of the importance of the i criterion over j, of the evaluator k, expressed using fuzzy triangular numbers. For instance, ${\stackrel{\sim }{x}}_{OS}^1$ denotes the first evaluator's assessment of the importance of O over S. If, in this case, O is moderately more important than S, the fuzzy number would be (3, 5, and 7), while ${\stackrel{\sim }{x}}_{SO}^1$ would be represented as reciprocal values of (3, 5, and 7), which is (1/7, 1/5, 1/3).

$$\stackrel{\sim }{x_k}=\left[\begin{array}{ccc}1& {\stackrel{\sim }{x}}_{12}^k& {\stackrel{\sim }{x}}_{13}^k\\ {\stackrel{\sim }{x}}_{21}^k& 1& {\stackrel{\sim }{x}}_{23}^k\\ {\stackrel{\sim }{x}}_{31}^k& {\stackrel{\sim }{x}}_{32}^k& 1\end{array}\right]$$ (1)

Second, the fuzzy synthetic extent ($S_i$) for i tℎ risk factor can be computed using the algebraic operations on triangular fuzzy numbers described in Equation 2. This synthetic extent is used to assess the relative importance of the i tℎ risk factor. ${\stackrel{\sim }{M}}_{ij}$ represents the fuzzy comparison value for i tℎ risk factor against the j tℎ risk factor. The terms $\sum _{j=1}^3{\stackrel{\sim }{M}}_{ij}$ and $\sum _{i=1}^3\sum _{j=1}^3{\stackrel{\sim }{M}}_{ij}$ denote the sum of fuzzy comparison values for the i tℎ risk factor and the total sum of all fuzzy comparison values, respectively. The operator $\otimes$ represents fuzzy multiplication. Thus, the fuzzy synthetic extent $S_i$ for the i tℎ risk factor is calculated by performing fuzzy multiplication between the sum of fuzzy values for the i tℎ risk factor and the inverse of the total sum.

$$S_i=\sum _{j=1}^3{\stackrel{\sim }{M}}_{ij}\otimes {\left[\sum _{i=1}^3\sum _{j=1}^3{\stackrel{\sim }{M}}_{ij}\right]}^{-1}$$ (2)

Third, the degree of possibility ($V$) between fuzzy numbers is calculated. This process measures the possibility that one set is preferred over another by identifying the intersection point between the 2 fuzzy number sets. The degree of possibility of 2 membership functions, which are ${\stackrel{\sim }{S}}_2=(l_2,m_2,u_2)\ge {\stackrel{\sim }{S}}_1=(l_1,m_1,u_1)$, is defined in Equation 3. The comparison between the degree of probability between fuzzy numbers $S_i\left(i=1,\dots ,k\right)$ can be defined by: $V\left(\stackrel{\sim }{S}\ge {\stackrel{\sim }{S}}_1,\dots ,{\stackrel{\sim }{S}}_k)=minV(\stackrel{\sim }{S}\ge S_i\right),\left(i=1,\dots ,k\right)$.

$$V\left({\stackrel{\sim }{S}}_2\ge {\stackrel{\sim }{S}}_1\right)=\{\begin{array}{c}1\\ 0\\ \frac{l_1-u_2}{\left(m_2-u_2\right)-\left(m_1-l_2\right)}\end{array}\begin{array}{c}if{m}_2\ge m_1\\ if{l}_1\ge u_2\\ \\ ,otherwise\end{array}$$ (3)

Finally, the weight factor is calculated using Equation 4. Each element's weight is then divided by the total weight, resulting in a set of normalized weights totaling 1. With normalization, we get the normalized weight vectors (W) expressed by: $W={(d\left(A_i\right),\dots ,d\left(A_k\right))}^T$

$$\begin{array}{c}{W}^{\prime }={\left(d^{\prime }\left(A_i\right),\dots ,d^{\prime }\left(A_n\right)\right)}^T,fori=1,\dots ,n\hfill \\ d^{\prime }\left(A_i\right)=min{S}_i\ge S_k,fork=1,\dots ,n;k\ne i.\hfill \end{array}$$ (4)

Fuzzy AHP-based RPN

The median values of risk factors for each FM were chosen from the first evaluation sheets provided by evaluators. These values were then used to compute the traditional RPN per AAPM TG-100 evaluation guidelines.¹⁵ FMEA study by Faught et al¹⁷ indicates that despite notable variability between physicists in scoring a particular FM, median scores exhibit relatively little variation across different FMs. In this study, we anticipated broader variation and subjectivity in risk factor assessments caused by variances in the experience and knowledge levels of evaluators.

Meanwhile, because of the limited number of qualified physicist team members, the geometric mean values of risk factors were used to determine the weights of risk factors ($w_O$, $w_S$, and $w_D$) from the second evaluation sheets received from qualified physicist members. The final fuzzy AHP-based RPN, denoted as $FRP{N}_W$, is calculated by multiplying O, S, and D values and their corresponding weights of risk factors ($w_O$, $w_S$, and $w_D$). The numerical expression of $FRP{N}_W$ is described in Equation 5.

$$FRP{N}_w=w_{O}O\times w_{S}S\times w_{D}D$$ (5)

Results

The process map for an MRI-based BT procedure can be found in Fig. 1. It comprises 6 primary steps, representing 33 sublevel processes. Throughout these subprocesses, a total of 82 FMs were identified. Figure 1 also indicates, through a color code, the primary job-specialty groups for AHP group evaluations for each step.

A representative sample evaluation sheet from an evaluator for the fuzzy AHP method is presented in Table 1. For fuzzy AHP analysis, we divided the 13 job-specialty groups within 6 primary steps of our process map as follows: step 1 involved analysis of 3 groups (physician, nurse, and physicist); step 2 involved analysis of 2 groups (therapist and physician); step 3 involved analysis of 2 groups (therapist and MRI technician); step 4 involved analysis of 4 groups (2 for physicist and 2 for physician); step 5 and 6 involved analysis of 2 groups (therapist and physics). We combined the nurse group in steps 1 and 6 and the therapist and physicist groups in steps 5 and 6, because of their overlapping job responsibilities. Additionally, we merged substeps 12 and 16 in step 2 and substep 24 in step 5 because they relate to the same FMs but occur in different steps.

Table 1.

The sample evaluation sheet for the AHP method

	Step	Job-group	Substep	Representative FM	O	S	D
1	1. Preprocedure	Physician	1	Not intended prescription	L	VH	VL
2		Nurse/office staff	2-5 and 33 in Step 6	Incorrect/missing request in EMR	H	VL	VL
3		Physicist	6	Missing QA checklist	M	VL	VL
4	2. Applicator insertion and simulation	Therapist	7-8	Wrong labeling for needles	M	VH	L
5		Physician	9-11	Suboptimal applicator insertion	VH	VH	H
6	3. MR Imaging	Therapist	12, 16, and 24 in Step 5	The patient transferred to the treatment room	M	M	L
7		MRI technician	13-15	Wrong scan area/cut-off MRI scan ranges	VH	H	L
8	4. Treatment planning	Physician	19	Incorrect/missing contouring	VH	VH	L
9		Physicist	17-18, 20	Wrong applicator reconstruction	M	H	H
10		Physician	21	Incorrect EQD2 calculation	L	M	H
11		Physicist	22-23	Physics second check	L	H	VH
12	5. Treatment delivery and 6. Posttreatment	Therapist	25-27, 30-31	Applicator connection	L	M	L
13		Physicist	28-29, 32	Incorrect fractionation plan delivered	VL	M	L

Abbreviations: AHP = analytical hierarchy process; D = probability of not being detected; EMR = electronic medical record; EQD2 = equivalent dose in 2 Gy fractions; FM = failure modes; H = high; L = low; M = medium; MR = magnetic resonance; MRI = magnetic resonance imaging; O = occurence; QA = quality assurance; S = severity; VL = very high.

In step 1, the physician's role in diagnosis and prescription within this substep was given high importance in S because of its substantial impact on the severity level of other FMs on occurrence. The role of the nurse/office staff was assigned high importance in O, potentially affecting work efficiency and leading to entire procedure cancellations at the preprocedure stage, necessitating subsequent rescheduling. Lastly, the physicist's role received moderate importance in O, impacting work efficiency by requiring documentation preparation because of potential violations of the Nuclear Regulatory Commission's/state/university regulations on occurrence. In step 2, during applicator insertion and simulation, the therapist's responsibility in application preparation was considered highly significant for S because of its causal impact, while the physician's role in the applicator insertion process was considered crucial for all risk factors. The therapist's role in transporting the patient to the MRI unit in step 3 and to the treatment room in step 5 received moderate importance in O and S. This is because of the implementation of a mobile air mattress system paired with a dedicated OR bed specifically designed for BT patients, which minimizes potential applicator displacement. The MRI technician's role in the MRI unit was assigned very high importance for O and high S because of its causal effect. In step 4, the physician's first role, contouring, holds very high importance for S and O, while the second role, plan review, was rated as high in D because of the decreased possibility of detection because it serves as the final plan check for the physician. The physicist's first role in image registration and planning holds high importance in O and S because of the potential impact on accurate dose calculation. The second role, physics second check, and plan export received very high importance in D and high in S because it represents the final physics check before treatment. In steps 5 and 6, the therapist's role in handling patient transfers, applicators, and radiation surveys was rated with low importance in O and low in D because it undergoes routine double-checks by physicists. The physicist's role in treatment delivery was also rated very low in O because of no direct causal impact, and low in D because of the need for routine double-checks with a second physicist.

The top 5 FMs ranked by traditional RPN are presented in Table 2. In this analysis, the highest-ranked FM involves incorrect or missing contours in contouring, attributed to high D and S scores. This step solely involves the physician, posing challenges for physicists and dosimetrists in accurately assessing targets, the small bowel, and the sigmoid. The second-ranked FM pertains to incorrect or suboptimal applicator insertion under ultrasound guidance. The third-ranked FM refers to dose-volume histograms (DVHs) that fail to meet the physician's intent because of a high S score. The fourth-ranked FM relates to the applicator or patient movement during patient transfer to the treatment room, potentially causing dosimetric deviations. Finally, the fifth-ranked FM pertains to incorrect applicator reconstruction during the planning process.

Table 2.

Top 5 failure modes (FMs) ranked by traditional risk priority number (RPN) analysis

Rank	Substep	FM	O	S	D	FRPNW
1	20. Contouring	50. Incorrect/missing contours	4	8	9	288
2	10. Applicator insertion	25. Incorrect/suboptimal applicator insertion	5	8	5	200
3	22. Plan review	59. DVH not meeting physician's intent	4	8	6	192
4	25. The patient transferred to the treatment room	66. Applicator/patient moves	5	7	5	175
5	21. Planning	54. Wrong applicator reconstruction	4	7	6	168

Abbreviations: D = detected; DVH = dose-volume histogram; FM = failure modes; FRPN = fuzzy AHP-based RPN; O = occurence; S = severity.

Table 3 presents the top 5 FMs as determined using the fuzzy AHP RPN analysis. In this analysis, the rankings have changed compared with those generated using the traditional RPN method, following the application of weights to risk factors. In the fuzzy AHP-based RPN (FRPN) ranking, the top-ranked FM was “incorrect/suboptimal applicator insertion” in the applicator insertion substep, and this FM ranked second in the traditional RPN method. This particular FM was assigned high weight values for all 3 risk factors, resulting in almost similar weight values for all risk factors. The second-ranked FM in the planning substep, “wrong applicator reconstruction,” held the fifth rank in the traditional RPN method. It received elevated S and D weight values because of its significant impact on the severity level of other FMs and its lower probability of detection even during the physics second check. The third-ranked FM, “DVH not meeting physician's intent” in the plan review substep, maintained the same rank as in the traditional RPN method. This FM received relatively high scores for both S and D because this substep is the final check for contour evaluation before treatment, and there is a very low possibility of detecting this error at later steps in the process. The fourth-ranked FM experienced “incorrect/missing contours” during contouring, dropping from first place in the traditional RPN method. This change is attributed to notably lower D weight values, indicating a higher chance of detection in subsequent planning steps, plan reviews, and physicist's second checks. The fifth-ranked FM was applicator/patient movements during the patient transfer to the treatment room substep, which resulted in high S weight values.

Table 3.

Top 5 failure modes (FMs) based on fuzzy AHP RPN analysis

Rank	Substep	FM	WO	WS	WD	FRPN
1	10. Applicator insertion	25. Incorrect/suboptimal applicator insertion	0.3751	0.3751	0.2498	7.029
2	21. Planning	54. Wrong applicator reconstruction	0.2520	0.4369	0.3114	5.757
3	22. Plan review	59. DVH not meeting physician's intent	0.1267	0.4367	0.4367	4.639
4	20. Contouring	50. Incorrect/missing contours	0.4687	0.4687	0.0625	3.954
5	25. The patient transferred to the treatment room	66. Applicator/patient moves	0.3003	0.5249	0.1750	3.862

Abbreviations: AHP = analytic hierarchy process; D = detected; DVH = dose-volume histogram; FRPN = fuzzy AHP-based RPN; FM = failure modes; H = high; O = occurence; S = severity.

Table 4 highlights the FMs with a rank difference of 10 or more between the conventional RPN and FRPN, with 13 FMs exceeding this threshold. Of these, 9 are associated with MRI imaging tasks performed by MRI technicians. This shift reflects the unique workflow in our clinic, where MRI images are imported and reviewed in near real-time before the patient returns to our department. This process significantly enhances the D of these FMs, enabling earlier and more efficient identification of potential issues. Additionally, 2 FMs are linked to the types and sizes of applicators used during insertion and simulation. These FMs also receive lower D weightings because of recent improvements in documentation practices, which support confirmation and facilitate early detection of potential problems. These tailored adjustments in D weighting within the fuzzy AHP framework result in relatively lower prioritization for these FMs compared with the conventional RPN.

Table 4.

Failure modes (FMs) with a rank difference of 10 or more between conventional risk priority number (RPN) and FRPN

Step	Substep	FM	RPN rank	FRPNW rank
2	7. Applicator type and sizes	Wrong size	14	26
2	7. Applicator type and sizes	Wrong labeling for needles	9	20
2	8. Time out	Wrong patient	25	35
3	13. Waiting time for MRI	Prolonged waiting time because of MRI availability or schedule	6	18
3	14. MRI scan	Wrong protocol used by MRI technician	17	27
3	14. MRI scan	MRI markers were not correctly inserted or no markers were available	12	23
3	14. MRI scan	Poor image set	13	25
3	14. MRI scan	Patient moves/ motion artifact	7	19
3	15. MRI export to Rad. Onc. RTP	MRI tech forgot or delayed exporting images to Rad. Onc. RTP	7	19
3	15. MRI export to Rad. Onc. RTP	Miscommunication or no communication	10	22
4	19. Contouring	Physician did not approve or did not finalize the contours	14	24
4	22. Physics second check	The plan was inadvertently altered during the approval	12	23
4	22. Physics second check	Not done	17	27

Abbreviations: D = detected; FRPN = fuzzy AHP-based RPN; MRI = magnetic resonance imaging; Onc. = oncology; Rad. = radiation; RTP = radiation treatment planning.

Discussion

The top-ranked FM in the conventional RPN method was “incorrect/missing contours,” aligning with the ranking from another FMEA study. However, after applying weights, its rank was adjusted because of our clinic's thorough contour review at the plan review substep, in collaboration with a planning physicist, increasing the possibility of detecting the risk. As a result, this FM was assigned lower D weight values, reducing its impact on the FRPN calculation. Meanwhile, in the FRPN method, the top-ranked FM was “incorrect/suboptimal applicator insertion,” assigned high weight values across all risk factors. The rationale behind the high O weight value is its potential to trigger subsequent FMs, such as applicator movements or suboptimal plans in later substeps. The elevated S weight value is attributed to its significant impact on the severity level of FMs in later process stages. For example, while a minor error in incorrect applicator insertion might be manageable through HDR source weighting in planning, a major error during insertion could lead to a suboptimal plan or failure to achieve the intended DVH. This varying severity level of FMs in later substeps contributed to the assignment of a high S weight value. Additionally, the high D weight value reflects the challenge of rectifying acknowledged errors during later substeps, often attributed to constraints associated with machines, patients, and interdepartmental scheduling. Given the priority of each FM, our primary focus was on devising corrective measures exclusively for the top 3 ranked FMs, directed at mitigating patient risks.

The fuzzy AHP-based method may provide limited additional value to FMEA in image-guided HDR brachytherapy because of the dominance of unchangeable FMs inherent to the procedure, such as incorrect contours and applicator insertion errors. The high-ranked FMs identified in our study closely align with findings from other studies.5, 6, 7 However, the unique contribution of our study lies in the evaluation of medium-priority risks, as detailed in Table 4. Notably, 9 out of 13 FMs with rank differences of 10 or more between conventional methods and the fuzzy AHP method are associated with MRI imaging tasks performed by MRI technicians. While we initially expected MRI-related FMs to rank higher because of their perceived impact, this expectation was influenced by a unique challenge in our clinic, which is the absence of an MRI unit within our department despite managing a large volume of patients. This situation presents challenges such as longer procedure times, MR safety concerns, and increased multidisciplinary efforts, all of which can affect work efficiency. However, our FMEA team prioritized other factors over MR safety and extended waiting times. This decision was shaped by our department's proximity to the MRI suite in the Radiology department, which facilitates efficient patient transfers between departments. Consequently, delays in patient transfers or MRI waiting times were not considered high-priority FMs in our clinical workflow because these factors were deemed less likely to affect applicator movement or cause motion artifacts. These findings highlight the variability in perceived importance and actual FMs based on perspectives and clinical settings across different clinics. They suggest that the fuzzy AHP approach provides a prioritization framework better aligned with the unique workflows and conditions of our clinic.

Corrective measures were carefully revised and implemented within our existing quality management program to specifically address the top 3 ranked FMs. These interventions focus on mitigating patient risks by addressing the critical vulnerabilities identified in our analysis, ensuring resources are strategically allocated to maximize patient safety and clinical outcomes. Addressing the top-ranked FM linked to incorrect/suboptimal applicator insertion, we communicated with relevant multidepartment team members involved in this procedure, emphasizing its criticality because of its subsequent influence on O, S, and D of FMs in later substeps. We requested these team members to allocate ample time for the procedure and to await the completion of a thorough computed tomography (CT) image review after the CT simulation. To address the incorrect applicator reconstruction identified in the planning substep for the second-ranked FM, we are considering the adoption of an MRI marker^18,19 and exploring implementation methods. Our current standard procedure involves using a library/template for reconstructing ring/tandem or tandem/ovoid applicators, specifically for intracavitary procedures. Additionally, we use image fusion with CT scans to verify needles and applicators as needed for hybrid and interstitial Syed procedures. In our effort to enhance seed identification, our ongoing project focuses on optimizing workflow efficiency for patient scheduling and safety in close collaboration with the radiology department. For the third-ranked FM where the DVH does not meet the physician's intent, we modified the plan review process by establishing an equivalent dose in 2 Gy fraction (EQD2) template based on guidelines and references from the External Beam Radiochemotherapy and MRI-based adaptive Brachytherapy in Locally Advanced Cervical Cancer, Groupe Européen de Curiethérapie—European Society for Therapeutic Radiology and Oncology Gynecology, and American Brachytherapy Society DVH (ABS DVH) metrics. During the plan review, both the physicist and physician clarified the EQD2 criteria and objectives, and the physics secondary check involved EQD2 verification following plan approval.

The conventional RPN method is valued for its straightforward and structured approach, making it highly accessible and easy to implement across various clinical settings. Its simplicity allows teams to quickly calculate and prioritize risks, making it particularly effective for standard processes. However, it has several limitations. First, the conventional RPN method assumes equal importance for all risk factors, which may not align with clinical realities where some factors are more influential. In health care, where patient safety is a top priority, S often holds greater importance than other factors. For example, health care failure mode and effects analysis is a method specifically designed to address this principle.²⁰ It introduces a structured hazard scoring matrix based on S and O to evaluate FMs. This hazard score replaces the traditional RPN-based method and has been widely adopted in health care settings, including radiation therapy and quality management initiatives.^21,22 Second, different combinations of S, O, and D values can result in identical RPN scores, leading to the same rankings for diverse FMs and introducing ambiguity in prioritization. While S ranking alone highlights the importance of critical risks, relying solely on S can overlook FMs with high O and D, which may still pose significant risks within the overall workflow. The fuzzy AHP approach introduces a flexible weighting system that balances the importance of all risk factors while ensuring that specific factors, such as S, are appropriately emphasized through the AHP method. Additionally, the use of fuzzy numbers effectively captures the variability and uncertainty inherent in expert evaluations, enabling a broader range of weighted values and generating more diverse RPN scores. Third, assigning exact numerical values (on a scale from 1 to 10) to evaluate FMs is challenging, particularly in complex clinical scenarios, and can lead to large variances among evaluators because of differences in experience and knowledge levels. AAPM TG-275 guidelines guide achieving group consensus and reducing variability using simplified scales, such as a 1-to-3 range for low, medium, and high risk.²³ However, these simplified scales may lack the precision required to fully capture the variability in expert opinions and may not sufficiently address differing perspectives, especially in cases where team members have incomplete information. The fuzzy AHP method uses fuzzy numbers to represent a range of possible values provided by all evaluators, ensuring that the diversity of perspectives and expertise is thoroughly captured. These fuzzy numbers are then aggregated into precise numerical values through a structured analytical process. This integration reflects the collective input of the group while accounting for individual differences in judgment. By addressing variability and uncertainty in evaluators’ assessments, this method provides a robust and effective alternative to traditional group consensus approaches. Finally, the conventional method does not explicitly quantify the degree of interdependence or influence between FMs. While AAPM TG-100 addresses interdependencies using process mapping, FMEA, and fault tree analysis to visualize and evaluate interactions,¹⁵ these methods may oversimplify the complexities of these interdependencies. They neither quantify the degree of influence between FMs nor account for the uncertainty inherent in their relationships. The fuzzy AHP method overcomes these limitations by integrating fuzzy numbers with the AHP methodology to evaluate the degree of interdependencies among subprocesses. This approach offers a robust framework for accurately assessing interdependencies and improving the prioritization of FMs.

The fuzzy AHP method, though more complex and requires specialized training and software tools, provides significant advantages in achieving accurate and reliable risk prioritization. While its initial implementation may require additional effort, once established within a clinic, it integrates seamlessly into clinical workflows, enhancing efficiency and precision in risk management strategies for both established and evolving clinical processes. By incorporating a flexible weighting system and leveraging fuzzy numbers, the fuzzy AHP method effectively addresses variability, uncertainty, and interdependencies. This approach enables a more comprehensive and precise prioritization of FMs, making it well-suited to the specific complexities and dynamic workflows inherent in clinical environments.

Conclusions

We proposed a fuzzy AHP-based RPN approach for failure prioritization to address the limitations of traditional RPN analysis in MRI-based GYN BT. This study demonstrates the feasibility of using the fuzzy AHP-based RPN method for comprehensive analysis and prioritization of FMs.

The use of a 5-level linguistic scale improves evaluators’ ability to assess FMs intuitively, particularly for clinicians. This approach offers a more comfortable and user-friendly alternative to the rigid and less intuitive 10-level numeric scale used in traditional RPN evaluations. By integrating the flexibility of linguistic terms with the precision of numerical analysis, the fuzzy approach creates a robust framework for evaluating interdependencies and refining the prioritization of FMs.

Building on this framework, the quality management program for the top 3 ranked FMs identified through the fuzzy AHP-based RPN incorporates targeted strategies to enhance safety and precision. For incorrect or suboptimal applicator insertion, adequate time is allocated for the procedure, and a thorough review of CT images is performed following the CT simulation, involving all relevant staff to ensure accuracy before proceeding. To address incorrect applicator reconstruction, efforts are directed toward exploring the use of MRI markers for applicator reconstruction and evaluating their feasibility for clinical implementation. For the FM related to the physician's DVH intent not being met, the plan review process has been refined by adopting an EQD2 template aligned with the External Beam Radiochemotherapy and MRI-based adaptive Brachytherapy in Locally Advanced Cervical Cancer, Groupe Européen de Curiethérapie—European Society for Therapeutic Radiology and Oncology Gynecology, and ABS DVH metrics.

The fuzzy AHP-based method serves as a flexible and robust framework for prioritizing risks and refining quality management strategies. It is particularly valuable in addressing challenges associated with complex systems, uncertain conditions, differing expert opinions, or the integration of new and evolving technologies, making it a highly applicable tool in modern clinical settings.

Disclosures

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.