CC BY-NC-ND 4.0 · Indian J Radiol Imaging 2021; 31(02): 265-269
DOI: 10.1055/s-0041-1733814
Original Article

Analysis of Factors Affecting Air Kerma Area Product Obtained during Uterine Artery Embolization Procedures Using Logistic Regression

1  Department of Medical Physics, Prince Sultan Military Medical City, Riyadh, Saudi Arabia
,
Ahmed Almutairi
2  Department of Radiodiagnostic and Medical Imaging, Prince Sultan Military Medical City, Riyadh, Saudi Arabia
,
Murdhi AlHarbi
2  Department of Radiodiagnostic and Medical Imaging, Prince Sultan Military Medical City, Riyadh, Saudi Arabia
,
Khaleel Almutairi
2  Department of Radiodiagnostic and Medical Imaging, Prince Sultan Military Medical City, Riyadh, Saudi Arabia
,
Turky Almutairi
2  Department of Radiodiagnostic and Medical Imaging, Prince Sultan Military Medical City, Riyadh, Saudi Arabia
,
Mousa Bakkari
1  Department of Medical Physics, Prince Sultan Military Medical City, Riyadh, Saudi Arabia
› Author Affiliations
Funding None.
 

Abstract

Purpose Uterine artery embolization (UAE) is a common interventional radiology procedure used in medicine; the procedure is safe but there is always a concern regarding radiation dose received by the patient. The aim of this study was to use multivariable logistic regression analysis (MLRA) to study a certain number of independent prognostic variables believed to provide an estimate of the likelihood of obtaining a high kerma area product (P KA) at the end of the procedure.

Method Radiation dose indices registered by the angiographic system structured dose report, the total fluoroscopy time (FT), the patient’ body mass index (BMI), the number of images taken during the procedures (IMGS), and the performing physician experience (EXPER) were used to drive a logistic regression model (LRM).

Results The LRM found was: Logit (P KA) = −6.1525 + 0.0416 (FT) + 0.1028 (IMGS) + 0.1675 (BMI) – 0.1012 (EXPER). The prediction accuracy of the LRM was assessed using receiver operating characteristic (ROC) curve; by calculating the area under the curve (AUC), we found AUC = 0.7896, with optimal ROC point of 0.3261, 0.8036.

Conclusion The suggested LRM seems to indicate that patients with higher BMI, have taken longer FT, acquired higher IMGS and the procedure done by a less experienced performing physician is more susceptible to receive a higher P KA at the end. The proposed LRM is useful in predicting the occurrence of higher radiation exposure interventions and can be used in patients’ radiation dose optimization strategies during UAE procedures.


#

Introduction

Uterine artery embolization (UAE) is a minimal invasive procedure that requires fluoroscopic and angiographic imaging and this causes a concern regarding the radiation dose received by the patient during the intervention. It is known that angiographic imaging systems can deliver a significant amount of radiation to the patient’s skin; therefore, radiation dose monitoring is required.[1] We have reviewed the radiation dose metrics available from the angiographic system for 102 UAE procedures performed on the system during the past year 2019. Along with the dosimetric metrics we have also included the body mass index (BMI) and the interventional radiologist experience (EXPER) to the variables that will be analyzed.

Multivariable logistic regression (MLR) analysis has been widely used as a method to identify prognostic factors affecting medical treatments outcome.[2] The aim of this study was to analyze radiation dose-related metrics available from the angiographic system and from the radiology information system (RIS) and to use MLR to estimate the occurrence of high radiation dose procedures.


#

Methods

Patient Characteristics

In this study we have retrospectively collected from the angiographic system registered dose report and from the (RIS) data concerning 102 patients who underwent UAE procedures in 2019. Radiation dose indicators such as, the fluoroscopy time (FT) in minutes, the cumulative kerma area product (P KA) in Gy cm2, the cumulative reference air kerma (K a,r) in mGy, the number of images taken during the procedure (IMGS), and the calculated patient body mass index (BMI) were collected. We also noted for each patient the experience of the performing physician in the form of total number of performed UAE procedures in the variable (EXPER). [Table 1] has the summary of the patients’ data used in this work. This retrospective study was approved by the institution’s research ethics committee.

Table 1

Summary of 102 patients’ data and their associated radiation dose metrics

FT (min)

IMGS

BMI (kg/m2)

EXPER

P KA (Gy cm2)

K a,r (mGy)

BMI, body mass index; FT, fluoroscopy time; EXPER, experience; IMGS, images.

Mean

28

14.9

30.0

16.5

480

2,140

Minimum

7.3

6.0

19.3

1

0.46

9.14

Maximum

105.9

37

72.6

24

7,125

10,340

Standard deviation

15.7

5.8

6.9

5.3

991

1,882

Coefficient of variation

0.56

0.39

0.23

0.32

2.06

0.88


#

Imaging Equipment

We used a biplane system C-arm with flat detector angiography, AXIOM Artis dBA (Siemens, Erlangen, Germany).


#

Logistic Regression Analysis

A binary logistic regression model (LRM) was calculated using variables that may predict the level of cumulated P KA at the end of the procedure. It is known that P KA is related to the risk of exposure to radiation.

P KA is the binary outcome variable used in the analysis. High P KA levels are assigned the value of 1 and routine P KA levels will be assigned the value 0. The LRM will have the following form:

Y = ln (odds [event]) = β 0 + β 1 x 1 + β 2 x 2+ …………. + β n x n (1)

In Eq. (1) the variable Y is log (naturel) of the odds of the event under consideration. In our case the event will be the occurrence of high P KA procedure. The βs are the coefficients of the regression calculated by the model and n is the number of predictive variables. The odds ratios are the exponential of the coefficients and will be given by:

Odds ratios = Exp (β)(2)


#

Statistical Analysis

The DAP values for 102 patients were dichotomized into two groups >300 Gy cm2 and ≤300 Gy cm2, the first group is considered as high radiation dose procedure (HRDP) and the second as routine radiation dose procedure.

The choice of 300 Gy cm was based on the American Society of Vascular Interventional Radiology.[3] They recommended alert levels to be displayed on the Angiographic system during the procedure, as to alert the surgeon that a DAP value of 300 Gy cm2 has been reached. Furthermore 300 Gy cm is considered an appropriate indicator for substantial radiation dose levels for most interventional radiology procedures.[4]

One of the objectives of this study was to identify the independent variables which are able to estimate HRDP group during UAE and to propose a LRM for the prediction of HRDP.

[Fig. 2] is showing the data distribution in the form of boxplots.

Zoom Image
Fig. 1 Scatter plots matrix showing K a,r and P KA as function of FT, IMGS, BMI, and EXPER, respectively. The dark blue dots are for the procedures with P KA >300 Gy cm2 and the light green dot for procedures with P KA ≤300 Gy cm2. BMI, body mass index; FT, fluoroscopy time; EXPER, experience; IMGS, images.
Zoom Image
Fig. 2 Boxplots showing the four predicting variables distribution for the two dependent variable categories: P KA >300 Gy cm2 and P KA ≤300 Gy cm2.

The statistical analysis was performed using Matlab (R2016b) statistics and machine learning toolbox (Natick, United States). A p-value of (<0.05) was considered statistically significant.


#
#

Results

As expected, a linear regression relationship was found between the cumulative air kerma measured at the reference point (K a,r) and the cumulated kerma area product (P KA), therefore K a,r was excluded from the list of the predicting independent variables. Fig. 1 contains scatter plot matrix of the Ka,r and PKA as function of FT, IMGS, BMI, and EXPER variables.

Binary Logistic Regression Model

The degree of significance of four variables: FT, IMGS, BMI, and EXPER was examined. [Table 2] has the summary of the results. The obtained LRM with four predictors (variables) FT, IMGS, BMI and EXPER is given by the Eq. (3) below:

Table 2

Results of the multivariable logistic regression analysis for all the variables that may be related to the occurrence of high radiation exposure UAE procedure

p-Value

OR (95% CI)

OR

B

Variables

Abbreviations: B, regression coefficients; OR, odds ratio; p, the p-value.

0.0014

6.1525

Intercept

0.0508

(1.000–1.087)

1.042

0.0416

FT

0.0643

(0.994–1.236)

1.108

0.1028

IMGS

0.0008

(1.072–1.304)

1.182

0.1675

BMI

0.0485

(0.817–0.999)

0.904

0.1012

EXPER

Logit (P KA) = −6.1525 + 0.0416 (FT) + 0.1028 (IMGS) + 0.1675 (BMI) – 0.1012 (EXPER)(3)

These results show that an increase of one unit of FT (1 minute) will have an increase of 4.16% in the cumulative P KA, similarly an increase of one unit in the number of images taken will have an increase of 10.2% in the P KA value, an increase of 1 unit in the BMI (1 kg/m2) will have an increase of 16.75% in the cumulative P KA and an increase of 1 unit in the variable EXPER will have a decrease in the total P KA value by 10.12%; this variable is working in the advantage of lower P KA value as opposed to the other three variables that all working in the advantage of obtaining a larger P KA value.

The model was tested for goodness of fit with the DAP data using the area (AUC) under the ROC. The AUC was equal to 0.7896 with optimal ROC point coordinate of (0.3261, 0.8036). [Fig. 3] shows the ROC for the four variables predictive model. The optimal ROC point seems to indicate a false positive rate of 32.6% and a true positive rate of 70.3% for the predictive model obtained.

Zoom Image
Fig. 3 The ROC curve for the proposed four predicative variables logistic regression model. ROC, receiver operating characteristic.

#
#

Discussion

Comparing radiation dose values and dosimetric quantities among published studies is very difficult because the procedures identification are not standardized and also their complexity varies considerably and there is no classification for procedures in accordance with their respective complexity level yet.[5] Therefore there is always a need to perform regular local clinical dose audits. In this work we have analyzed available patient dose-related metrics with the aim of identifying the metrics or variables that affect the patient radiation exposure the most represented by the kerma area product during UAE.

Kerma Area Products during UAE

There are several published studies reporting radiation dose assessments and dose reduction and optimization techniques.[6] [7] [8] [9] [10] [11] The reported values of P KA are in [Table 3]. In this study we have found a median P KA value of 347 Gy cm2 for 102 UAE procedures conducted in 2019.

Table 3

Some of the reported PKA in the literature

Author

Year

P KA (Gy cm2)

K a,r (Gy)

n

aThe values in parentheses are the values obtained after applying imaging system optimization.

Miller et al

2009

392

2.5

90

Vano et al

2009

236

Ruiz-Cruces et al

2016

214

56

Durrani et al

2016

437 (267)a

Kohlbrenner et al

2017

438 (175)a

Schernthaner et al

2018

527 (146)a

This study

2020

347

2.1

100


#

Suggested Reference Levels

The recommended dose reference level (DRL) for UAE was set at 450 Gy cm2 and the reported 75th percentile P KA from the radiation doses in interventional radiology procedures study (RAD-IR) data was 392 Gy cm.[6]

Procedures with DAP values above 300 Gy cm2 should be optimized if possible. A recent study suggests using a DAP value of 50 Gy cm as target value for UAE procedures. In this study the authors suggested strategies for reducing radiation exposure during UAE; the strategies included: optimized source image and object image distances, avoidance of magnification, use of tight collimation, use of road-mapping, avoidance of oblique projections, use of pulsed fluoroscopy with low images per second, use of low frame rates, use of last-image-hold, and avoid three-dimensional rotational angiography.[12]

The use of optimization strategies will reduce the radiation dose received by the patients as well as the staff performing the procedure especially in cases expected to lead to a higher than usual radiation dose like for obese patients.[13]

A recent study had concluded that during UAE procedures, BMI had the greatest impact on the cumulated K a,r and has a substantial impact on the risk of radiation-induced skin injury even without prolonged FT.[14]


#

Performing Physician Experience

The effect of interventional radiologist experience on FT and K a,r was studied, and the conclusion was: although there was no nonsignificant trend for shorter screening times with experience, technical success and safety were not compromised with appropriate consultant supervision, which illustrates a safe construct for IR training. This is important and reassuring information for patients undergoing a procedure in a training unit.[15]

This conclusion is not in agreement with this study since we have found a statistically significant correlation between the operator experience and the reported P KA (p = 0.0485).


#

Complexity Level of the Procedure

Another study suggested after analyzing 56 UAE procedures, to include procedure complexity levels to facilitate clinical audits and proper use of DRLs in terms of P KA for patient dose optimization in interventional radiology. They recommend DRLs of 167, 214, and 613 Gy cm2 for simple, medium, and complex index UAE procedures, respectively. Statistical analyses (r Pearson and ANOVA test) identified significant correlations between the complexity score and patient dose (KAP) for all of the procedures (F <0.05).[8]


#
#

Limitations

This retrospective study analyzed data from one medical center during 1 year and included 102 patients; the study was on one imaging system only. The study did not include information about the complexity level of the procedure. The level of complexity was reported in the literature to have an effect on the DRLs. Expanding the study to include more than one imaging system, multiple medical centers, and procedure complexity level, when possible, will improve the accuracy of the proposed predictive LRM. Furthermore, a follow-up study aiming at validating the proposed model using other patients’ data is recommended.


#

Conclusion

In conclusion, the results of this retrospective study suggest that a UAE procedure having a cumulative DAP higher than 300 Gy cm2 is likely to occur in procedures having patients with higher BMI values, have taken longer FT, acquired higher IMGS, and were accomplished by a less experienced performing physician.

The proposed LRM is useful in predicting the occurrence of higher radiation exposure interventions and can be used in patients’ radiation dose optimization strategies during UAE procedures.


#
#

Conflict of Interest

None declared.


Address for correspondence

Khaled Soliman, PhD, DABMP
Department of Medical Physics
Prince Sultan Military Medical City, P. O. Box 7897 (B-94), Riyadh 11159
Saudi Arabia   

Publication History

Publication Date:
28 July 2021 (online)

© 2021. Indian Radiological Association. This is an open access article published by Thieme under the terms of the Creative Commons Attribution-NonDerivative-NonCommercial-License, permitting copying and reproduction so long as the original work is given appropriate credit. Contents may not be used for commercial purposes, or adapted, remixed, transformed or built upon. (https://creativecommons.org/licenses/by-nc-nd/4.0/).

Thieme Medical and Scientific Publishers Private Ltd.
A-12, Second Floor, Sector -2, NOIDA -201301, India


Zoom Image
Fig. 1 Scatter plots matrix showing K a,r and P KA as function of FT, IMGS, BMI, and EXPER, respectively. The dark blue dots are for the procedures with P KA >300 Gy cm2 and the light green dot for procedures with P KA ≤300 Gy cm2. BMI, body mass index; FT, fluoroscopy time; EXPER, experience; IMGS, images.
Zoom Image
Fig. 2 Boxplots showing the four predicting variables distribution for the two dependent variable categories: P KA >300 Gy cm2 and P KA ≤300 Gy cm2.
Zoom Image
Fig. 3 The ROC curve for the proposed four predicative variables logistic regression model. ROC, receiver operating characteristic.