External Validation of a Predictive Model for Acute Skin Radiation Toxicity in the REQUITE Breast Cohort

Training and temporally validating an NTCP model of acute toxicity after whole breast radiotherapy, including the impact of advanced delivery techniques

PURPOSE

The aim is to train and validate a multivariable Normal Tissue Complication Probability (NTCP) model predicting acute skin reactions in patients with breast cancer receiving adjuvant Radiotherapy (RT).

METHODS AND MATERIALS

We retrospectively reviewed 1570 single-institute patients with breast cancer treated with whole breast irradiation (40 Gy/15fr). The patients were divided into training (n = 878, treated with 3d-CRT, from 2009 to 2017) and validation cohorts (n = 692, treated from 2017 to 2021, including advanced RT techniques). In the validation cohort, patients were classified according to the delivery techniques into static (n = 404) and arc techniques (n = 288). Several clinical/technical information and DVHs of the "skin" (5 mm inner expansion from the body contour) were available. Skin toxicity was assessed during follow-up using the RTOG scale criteria. A multivariable logistic regression model was generated combining skin DVH and clinical parameters, using cross-validation methods that ensured high internal consistency and robustness. The performance of the model was tested in the validation cohort.

RESULTS

14.0 %/17.4 % of patients developed ≥ G2 toxicity, in the training/validation cohorts, respectively. The resulting multivariable logistic model included axillary lymph node dissection (OR = 1.58, 95 %CI = 1.01-2.48, p = 0.045), hypertension (OR = 1.54, 95 %CI = 1.04-2.27, p = 0.030) and skin V20Gy (OR = 1.008, 95 %CI = 1.004-1.013, p < 0.0001). The AUC of the model was 0.64/0.59 in training/validation, with better performance in the validation cohort if considering only V20Gy (0.62). The model showed satisfactory agreement between predicted and observed toxicity rates: in the validation group, the slope of the calibration plot was 0.96 (R2 = 0.6) with excellent goodness-of-fit (Hosmer-Lemeshow p-value = 0.99). Looking at each of the three predictors individually, only the role of V20Gy was confirmed in the validation group. Results were similar when considering patients treated with static or arc techniques.

CONCLUSION

An NTCP model for acute toxicity after moderately hypofractionated breast RT was trained. The model underwent temporal validation even for patients treated with advanced delivery techniques. Despite clinical differences and techniques, the confirmation of the dosimetry parameter in the validation cohort highlights its robustness and corroborates the hypothesis that skin DVH may assess the risk with the potential for improving plan optimisation.

A dosiomics model for prediction of radiation-induced acute skin toxicity in breast cancer patients: machine learning-based study for a closed bore linac

BACKGROUND

Radiation induced acute skin toxicity (AST) is considered as a common side effect of breast radiation therapy. The goal of this study was to design dosiomics-based machine learning (ML) models for prediction of AST, to enable creating optimized treatment plans for high-risk individuals.

METHODS

Dosiomics features extracted using Pyradiomics tool (v3.0.1), along with treatment plan-derived dose volume histograms (DVHs), and patient-specific treatment-related (PTR) data of breast cancer patients were used for modeling. Clinical scoring was done using the Common Terminology Criteria for Adverse Events (CTCAE) V4.0 criteria for skin-specific symptoms. The 52 breast cancer patients were grouped into AST 2 + (CTCAE ≥ 2) and AST 2 - (CTCAE < 2) toxicity grades to facilitate AST modeling. They were randomly divided into training (70%) and testing (30%) cohorts. Multiple prediction models were assessed through multivariate analysis, incorporating different combinations of feature groups (dosiomics, DVH, and PTR) individually and collectively. In total, seven unique combinations, along with seven classification algorithms, were considered after feature selection. The performance of each model was evaluated on the test group using the area under the receiver operating characteristic curve (AUC) and f1-score. Accuracy, precision, and recall of each model were also studied. Statistical analysis involved features differences between AST 2 - and AST 2 + groups and cutoff value calculations.

RESULTS

Results showed that 44% of the patients developed AST 2 + after Tomotherapy. The dosiomics (DOS) model, developed using dosiomics features, exhibited a noteworthy improvement in AUC (up to 0.78), when spatial information is preserved in the dose distribution, compared to DVH features (up to 0.71). Furthermore, a baseline ML model created using only PTR features for comparison with DOS models showed the significance of dosiomics in early AST prediction. By employing the Extra Tree (ET) classifiers, the DOS + DVH + PTR model achieved a statistically significant improved performance in terms of AUC (0.83; 95% CI 0.71-0.90), accuracy (0.70), precision (0.74) and sensitivity (0.72) compared to other models.

CONCLUSIONS

This study confirmed the benefit of dosiomics-based ML in the prediction of AST. However, the combination of dosiomics, DVH, and PTR yields significant improvement in AST prediction. The results of this study provide the opportunity for timely interventions to prevent the occurrence of radiation induced AST.

Comparing Performances of Predictive Models of Toxicity after Radiotherapy for Breast Cancer Using Different Machine Learning Approaches

PURPOSE

Different ML models were compared to predict toxicity in RT on a large cohort (n = 1314).

METHODS

The endpoint was RTOG G2/G3 acute toxicity, resulting in 204/1314 patients with the event. The dataset, including 25 clinical, anatomical, and dosimetric features, was split into 984 for training and 330 for internal tests. The dataset was standardized; features with a high p-value at univariate LR and with Spearman ρ>0.8 were excluded; synthesized data of the minority were generated to compensate for class imbalance. Twelve ML methods were considered. Model optimization and sequential backward selection were run to choose the best models with a parsimonious feature number. Finally, feature importance was derived for every model.

RESULTS

The model's performance was compared on a training-test dataset over different metrics: the best performance model was LightGBM. Logistic regression with three variables (LR3) selected via bootstrapping showed performances similar to the best-performing models. The AUC of test data is slightly above 0.65 for the best models (highest value: 0.662 with LightGBM).

CONCLUSIONS

No model performed the best for all metrics: more complex ML models had better performances; however, models with just three features showed performances comparable to the best models using many (n = 13-19) features.

Patients' experiences of engaging with electronic Patient Reported Outcome Measures (PROMs) after the completion of radiation therapy for breast cancer: a pilot service evaluation

INTRODUCTION

Over 60 % of people who develop breast cancer will receive radiation therapy (RT) as part of their treatment. Side effects of RT may include inflammation, erythema, desquamation and fatigue. Electronic Patient Reported Outcomes Measures (ePROMs) enable patients to report side effects prior to their scheduled post-RT appointment. This pilot service evaluation aims to explore patients' perceptions regarding the value of the ePROM system, ease of its use and barriers to using the system, after breast irradiation.

METHODS

From July-November 2021, evaluation surveys were posted to 100 people who had received RT to their breast to explore their experience of using the ePROM. Ethical approval was obtained through Ulster University and the Western Health and Social Care Trust (WHSCT), Northern Ireland.

RESULTS

Fifty-two people responded to the survey, of which 27 respondents indicated that they had accessed the ePROM. Despite few participants experiencing significant side effects, the majority of participants recommended the ePROM indicating that it was an important source of support. Those who experienced significant side effects found the system to be prompt and effective. Barriers to accessing the ePROM included technical issues with the link, concerns about confidentiality and forgetting to access the link. Access to the ePROM increased with higher education levels.

CONCLUSIONS

This pilot service evaluation demonstrated that ePROMs are valued by patients and can provide rapid real-time access to support, offering individual care and reassurance. For patients with longer RT schedules (>10 fractions), the introduction of ePROMs during RT was viewed favourably by participants. All patients may benefit from the option of receiving ePROMs post-RT.

Comparison of prone and supine positioning for breast cancer radiotherapy using REQUITE data: dosimetry, acute and two years physician and patient-reported outcomes

OBJECTIVE

Most patients receive whole breast radiotherapy in a supine position. However, two randomised trials showed lower acute toxicity in prone position. Furthermore, in most patients, prone positioning reduced doses to the organs at risk. To confirm these findings, we compared toxicity outcomes, photographic assessment, and dosimetry between both positions using REQUITE data.

METHODS

REQUITE is an international multi-centre prospective observational study that recruited 2069 breast cancer patients receiving radiotherapy. Data on toxicity, health-related quality of life (HRQoL), and dosimetry were collected, as well as a photographic assessment. A matched case control analysis compared patients treated prone (n = 268) versus supine (n = 493). Exact matching was performed for the use of intensity-modulated radiotherapy, boost, lymph node irradiation, chemotherapy and fractionation, and the nearest neighbour for breast volume. Primary endpoints were dermatitis at the end of radiotherapy, and atrophy and cosmetic outcome by photographic assessment at two years.

RESULTS

At the last treatment fraction, there was no significant difference in dermatitis (p = .28) or any HRQoL domain, but prone positioning increased the risk of breast oedema (p < .001). At 2 years, patients treated in prone position had less atrophy (p = .01), and higher body image (p < .001), and social functioning (p < .001) scores. The photographic assessment showed no difference in cosmesis at 2 years (p = .22). In prone position, mean heart dose (MHD) was significantly lower for left-sided patients (1.29 Gy vs 2.10 Gy, p < .001) and ipsilateral mean lung dose (MLD) was significantly lower for all patients (2.77 Gy vs 5.89 Gy, p < .001).

CONCLUSIONS

Prone radiotherapy showed lower MLD and MHD compared to supine position, although the risk of developing breast oedema during radiotherapy was higher. At 2 years the photographic assessment showed no difference in the cosmetic outcome, but less atrophy was seen in prone-treated patients and this seems to have a positive influence on the HRQoL domain of body image.

Longitudinal cosmetic outcome after planned IORT boost with low kV X-rays-monocentric results from the TARGIT BQR registry

Background

Intraoperative radiotherapy can serve as an anticipated boost (IORT boost) in combination with a subsequent external whole breast irradiation in high-risk breast cancer patients and is part of many guidelines. Nevertheless, there are only few prospective data available regarding cosmetic outcome after IORT boost using kV X-rays. The aim of this study was to evaluate the cosmetic outcome of patients treated within the prospective phase IV TARGeted Intraoperative radioTherapy (TARGIT) Boost Quality Registry (BQR) study (NCT01440010) in one center.

Methods

In the context of the TARGIT BQR study standardized photos in three positions (arms down, arms up, from the side) were available for different time points. For this analysis a layperson, a radiation oncologist and a gynecologist evaluated available photos at different time points during follow-up with up to 4 years using the Harvard scale (comparison of treated and the untreated breast; rating: excellent, good, fair, poor). Longitudinal results were compared to preoperative results (baseline).

Results

Seventy-three patients were available for the analysis. Baseline cosmetic assessment was excellent/good in 98.8% (mean value for all three positions). Postoperative cosmetic outcome (median) was good for all positions and remained constant for 4 years. Around 30% of the patients showed a constant or even improved cosmetic outcome compared to baseline. Only few patients showed a poor result at 4 years. The majority of patients showed an excellent or good cosmetic outcome at all time points.

Conclusions

Patients from the prospective TARGIT BQR study treated with IORT boost and additional whole breast irradiation showed good or excellent cosmetic outcomes in most cases during 4 years of follow-up. These results add important information for shared decision making in breast cancer patients.

Predicting acute radiation dermatitis in breast cancer: a prospective cohort study

BACKGROUND

Acute radiation dermatitis (ARD) is one of the most common acute adverse reactions in breast cancer patients during and immediately after radiotherapy. As ARD affects patient quality of life, it is important to conduct individualized risk assessments of patients in order to identify those patients most at risk of developing severe ARD.

METHODS

The data of breast cancer patients who received radiotherapy were prospectively collected and analyzed. Serum ferritin, high-sensitivity C-reactive protein (hs-CRP) levels, and percentages of lymphocyte subsets were measured before radiotherapy. ARD was graded (0-6 grade), according to the Oncology Nursing Society Skin Toxicity Scale. Univariate and multivariate logistic regression analyses were used and the odds ratio (OR) and 95% confidence interval (CI) of each factor were calculated.

RESULTS

This study included 455 breast cancer patients. After radiotherapy, 59.6% and 17.8% of patients developed at least 3 (3+) grade and at least 4 (4+) grade ARD, respectively. Multivariate logistic regression analysis found that body mass index (OR: 1.11, 95% CI: 1.01-1.22), diabetes (OR: 2.70, 95% CI: 1.11-6.60), smoking (OR: 3.04, 95% CI: 1.15-8.02), higher ferritin (OR: 3.31, 95% CI: 1.78-6.17), higher hs-CRP (OR: 1.96, 95% CI: 1.02-3.77), and higher CD3 + T cells (OR: 2.99, 95% CI: 1.10-3.58) were independent risk factors for 4 + grade ARD. Based on these findings, a nomogram model of 4 + grade ARD was further established. The nomogram AUC was 0.80 (95% CI: 0.75-0.86), making it more discriminative than any single factor.

CONCLUSION

BMI, diabetes, smoking history, higher ferritin, higher hs-CRP, and higher CD3 + T cells prior to radiotherapy for breast cancer are all independent risk factors for 4 + grade ARD. The results can provide evidence for clinicians to screen out high-risk patients, take precautions and carefully follow up on these patients before and during radiotherapy.

The Normal, the Radiosensitive, and the Ataxic in the Era of Precision Radiotherapy: A Narrative Review

(1) Background: radiotherapy is a cornerstone of cancer treatment. When delivering a tumoricidal dose, the risk of severe late toxicities is usually kept below 5% using dose-volume constraints. However, individual radiation sensitivity (iRS) is responsible (with other technical factors) for unexpected toxicities after exposure to a dose that induces no toxicity in the general population. Diagnosing iRS before radiotherapy could avoid unnecessary toxicities in patients with a grossly normal phenotype. Thus, we reviewed iRS diagnostic data and their impact on decision-making processes and the RT workflow; (2) Methods: following a description of radiation toxicities, we conducted a critical review of the current state of the knowledge on individual determinants of cellular/tissue radiation; (3) Results: tremendous advances in technology now allow minimally-invasive genomic, epigenetic and functional testing and a better understanding of iRS. Ongoing large translational studies implement various tests and enriched NTCP models designed to improve the prediction of toxicities. iRS testing could better support informed radiotherapy decisions for individuals with a normal phenotype who experience unusual toxicities. Ethics of medical decisions with an accurate prediction of personalized radiotherapy's risk/benefits and its health economics impact are at stake; (4) Conclusions: iRS testing represents a critical unmet need to design personalized radiotherapy protocols relying on extended NTCP models integrating iRS.

Development and optimisation of a machine-learning prediction model for acute desquamation following breast radiotherapy in the multi-centre REQUITE cohort

Purpose

Some patients with breast cancer treated by surgery and radiation therapy experience clinically significant toxicity, which may adversely affect cosmesis and quality of life. There is a paucity of validated clinical prediction models for radiation toxicity. We used machine learning (ML) algorithms to develop and optimise a clinical prediction model for acute breast desquamation after whole breast external beam radiation therapy in the prospective multicenter REQUITE cohort study.

Methods and Materials

Using demographic and treatment-related features (m = 122) from patients (n = 2058) at 26 centers, we trained 8 ML algorithms with 10-fold cross-validation in a 50:50 random-split data set with class stratification to predict acute breast desquamation. Based on performance in the validation data set, the logistic model tree, random forest, and naïve Bayes models were taken forward to cost-sensitive learning optimisation.

Results

One hundred and ninety-two patients experienced acute desquamation. Resampling and cost-sensitive learning optimisation facilitated an improvement in classification performance. Based on maximising sensitivity (true positives), the “hero” model was the cost-sensitive random forest algorithm with a false-negative: false-positive misclassification penalty of 90:1 containing m = 114 predictive features. Model sensitivity and specificity were 0.77 and 0.66, respectively, with an area under the curve of 0.77 in the validation cohort.

Conclusions

ML algorithms with resampling and cost-sensitive learning generated clinically valid prediction models for acute desquamation using patient demographic and treatment features. Further external validation and inclusion of genomic markers in ML prediction models are worthwhile, to identify patients at increased risk of toxicity who may benefit from supportive intervention or even a change in treatment plan.

Risk Factors Related to Acute Radiation Dermatitis in Breast Cancer Patients After Radiotherapy: A Systematic Review and Meta-Analysis

Background

Acute radiation dermatitis (ARD) is the most common acute response after adjuvant radiotherapy in breast cancer patients and negatively affects patients’ quality of life. Some studies have reported several risk factors that can predict breast cancer patients who are at a high risk of ARD. This study aimed to identify patient- and treatment-related risk factors associated with ARD.

Methods

PubMed, Embase, Cochrane Library, China National Knowledge Infrastructure, and WanFang literature databases were searched for studies exploring the risk factors in breast cancer patients. The pooled effect sizes, relative risks (RRs), and 95% CIs were calculated using the random-effects model. Potential heterogeneity and sensitivity analyses by study design, ARD evaluation scale, and regions were also performed.

Results

A total of 38 studies composed of 15,623 breast cancer patients were included in the analysis. Of the seven available patient-related risk factors, four factors were significantly associated with ARD: body mass index (BMI) ≥25 kg/m² (RR = 1.11, 95% CI = 1.06–1.16, I² = 57.1%), large breast volume (RR = 1.02, 95% CI = 1.01–1.03, I² = 93.2%), smoking habits (RR = 1.70, 95% CI = 1.24–2.34, I² = 50.7%), and diabetes (RR = 2.24, 95% CI = 1.53–3.27, I² = 0%). Of the seven treatment-related risk factors, we found that hypofractionated radiotherapy reduced the risk of ARD in patients with breast cancer compared with that in conventional fractionated radiotherapy (RR = 0.28, 95% CI = 0.19–0.43, I² = 84.5%). Sequential boost and bolus use was significantly associated with ARD (boost, RR = 1.91, 95% CI = 1.34–2.72, I² = 92.5%; bolus, RR = 1.94, 95% CI = 1.82–4.76, I² = 23.8%). However, chemotherapy regimen (RR = 1.17, 95% CI = 0.95–1.45, I² = 57.2%), hormone therapy (RR = 1.35, 95% CI = 0.94–1.93, I² = 77.1%), trastuzumab therapy (RR = 1.56, 95% CI = 0.18–1.76, I² = 91.9%), and nodal irradiation (RR = 1.57, 95% CI = 0.98–2.53, I² = 72.5%) were not correlated with ARD. Sensitivity analysis results showed that BMI was consistently associated with ARD, while smoking, breast volume, and boost administration were associated with ARD depending on study design, country of study, and toxicity evaluation scale used. Hypofractionation was consistently shown as protective. The differences between study design, toxicity evaluation scale, and regions might explain a little of the sources of heterogeneity.

Conclusion

The results of this systematic review and meta-analysis indicated that BMI ≥ 25 kg/m² was a significant predictor of ARD and that hypofractionation was consistently protective. Depending on country of study, study design, and toxicity scale used, breast volume, smoking habit, diabetes, and sequential boost and bolus use were also predictive of ARD.