Giovanni Innella

Machine learning prediction of germline BRCA1/2 pathogenic variants in patients with ovarian cancer

Objectives To assess the performance of machine learning (ML) algorithms to predict the presence of germline BRCA1/2 pathogenic variants in ovarian cancer (OC) patients based on clinical–pathological features. Methods Clinical–pathological features of 648 patients with OC tested for BRCA1/2 were analysed using three supervised ML algorithms: random forest, boosting and support vector machine. Results In the ‘test’ sample, boosting proved to be the most effective algorithm (accuracy: 84.5%; precision: 80.0%; recall: 3.1%; area under the curve (AUC): 78.8%), followed by support vector machine (accuracy: 81.4%; precision: 72.7%; recall: 27.6%; AUC: 62.3%) and random forest (accuracy: 74.4%; precision: 55.6%; recall: 14.7%; AUC: 71.3%). In the ‘validation’ sample, accuracy was 79.8% for boosting, 81.7% for support vector machine, 80.8% for random forest. In the most effective algorithm (boosting), family history of OC showed the highest relative influence (52.9), followed by histotype (19.5), personal history of breast cancer (BC) (17.1), age at diagnosis (8.4) and family history of BC (2.2), while Federation of Gynecology and Obstetrics stage had no influence. Discussion We identified the predictive algorithm that best estimates the a priori likelihood of being a carrier of germline BRCA1/2 pathogenic variants in patients with OC. These findings support a role for ML approaches in predicting BRCA1/2 status in patients with OC, but accuracy and precision are still suboptimal for clinical use, suggesting the need for additional research. Conclusions Results support the selection of relevant clinical features for predictive purposes, which could have significant implications for the clinical management of patients with OC.

Cancer Risks Associated With TP53 Pathogenic Variants: Maximum Likelihood Analysis of Extended Pedigrees for Diagnosis of First Cancers Beyond the Li-Fraumeni Syndrome Spectrum

PURPOSE Establishing accurate age-related penetrance figures for the broad range of cancer types that occur in individuals harboring a pathogenic germline variant in the TP53 gene is essential to determine the most effective clinical management strategies. These figures also permit optimal use of cosegregation data for classification of TP53 variants of unknown significance. Penetrance estimation can easily be affected by bias from ascertainment criteria, an issue not commonly addressed by previous studies. MATERIALS AND METHODS We performed a maximum likelihood penetrance estimation using full pedigree data from a multicenter study of 146 TP53-positive families, incorporating adjustment for the effect of ascertainment and population-specific background cancer risks. The analysis included pedigrees from Australia, Spain, and United States, with phenotypic information for 4,028 individuals. RESULTS Core Li-Fraumeni syndrome (LFS) cancers (breast cancer, adrenocortical carcinoma, brain cancer, osteosarcoma, and soft tissue sarcoma) had the highest hazard ratios of all cancers analyzed in this study. The analysis also detected a significantly increased lifetime risk for a range of cancers not previously formally associated with TP53 pathogenic variant status, including colorectal, gastric, lung, pancreatic, and ovarian cancers. The cumulative risk of any cancer type by age 50 years was 92.4% (95% CI, 82.2 to 98.3) for females and 59.7% (95% CI, 39.9 to 81.3) for males. Females had a 63.3% (95% CI, 35.6 to 90.1) cumulative risk of developing breast cancer by age 50 years. CONCLUSION The results from maximum likelihood analysis confirm the known high lifetime risk for the core LFS-associated cancer types providing new risk estimates and indicate significantly increased lifetime risks for several additional cancer types. Accurate cancer risk estimates will help refine clinical recommendations for TP53 pathogenic variant carriers and improve TP53 variant classification.

Clinical implications of VUS reclassification in a single-centre series from application of ACMG/AMP classification rules specified for BRCA1/2

Background BRCA1/2 testing is crucial to guide clinical decisions in patients with hereditary breast/ovarian cancer, but detection of variants of uncertain significance (VUSs) prevents proper management of carriers. The ENIGMA (Evidence-based Network for the Interpretation of Germline Mutant Alleles) BRCA1/2 Variant Curation Expert Panel (VCEP) has recently developed BRCA1/2 variant classification guidelines consistent with ClinGen processes, specified against the ACMG/AMP (American College of Medical Genetics and Genomics/Association for Molecular-Pathology) classification framework. Methods The ClinGen-approved BRCA1/2-specified ACMG/AMP classification guidelines were applied to BRCA1/2 VUSs identified from 2011 to 2022 in a series of patients, retrieving information from the VCEP documentation, public databases, literature and ENIGMA unpublished data. Then, we critically re-evaluated carrier families based on new results and checked consistency of updated classification with main sources for clinical interpretation of BRCA1/2 variants. Results Among 166 VUSs detected in 231 index cases, 135 (81.3%) found in 197 index cases were classified by applying BRCA1/2-specified ACMG/AMP criteria: 128 (94.8%) as Benign/Likely Benign and 7 (5.2%) as Pathogenic/Likely Pathogenic. The average time from the first report as ‘VUS’ to classification using this approach was 49.4 months. Considering that 15 of these variants found in 64 families had already been internally reclassified prior to this work, this study provided 121 new reclassifications among the 151 (80.1%) remaining VUSs, relevant to 133/167 (79.6%) families. Conclusions These results demonstrated the effectiveness of new BRCA1/2 ACMG/AMP classification guidelines for VUS classification within a clinical cohort, and their important clinical impact. Furthermore, they suggested a cadence of no more than 3 years for regular review of VUSs, which however requires time, expertise and resources.

Development of a risk score based on clinical–pathological features to predict the presence of germline BRCA1/2 pathogenic variants in ovarian cancer patients

Identification of germline BRCA1/2 pathogenic variants is crucial for tailoring ovarian cancer treatment and prevention. The purpose of the study was to develop a model to predict BRCA1/2 status in ovarian cancer patients. The association between clinical-pathological features and BRCA1/2 status was analysed in a series of 1009 ovarian cancer patients, using Fisher's exact test. Logistic regression models and a decision tree classification algorithm were used to develop a risk score. Compared with noncarriers, BRCA1/2 carriers (n = 216; 21.4%) presented more frequently with serous histotype non-low-grade (92.3% versus 71.6%, P < 0.001), family history of ovarian cancer (31.6% versus 5.7%, P < 0.001), family history of breast cancer (53.7% versus 31.6%, P < 0.001), previous breast cancer (20.9% versus 8.5%, P < 0.001), advanced stage (78.8% versus 69.5%, P = 0.019) and younger age (56.9 years versus 60.8 years, P < 0.001). Multivariable logistic regression on 648 patients with complete data confirmed histotype, family history of breast/ovarian cancer, previous breast cancer and age as independently and significantly associated with BRCA1/2 status. After refining the categorization of variables according to decision tree classification algorithm results, odds ratios derived from multivariable logistic regression were used to assign weights from 1 to 3 to each feature (non-low-grade serous histotype = 3, low-grade serous/high-grade endometrioid histotype/family history of ovarian cancer = 2, age at diagnosis <50 years/family and personal history of breast cancer = 1) and to develop a score ranging from 0 to 10, associated with increasing risk of BRCA1/2 variants (from 0.6% for score 0 to 88% for score ≥7). The area under the curve of the score was 0.78 (95% confidence interval 0.74-0.82) and the optimal cut-off was ≥4 points with a sensitivity of 81% and a specificity of 62.3%. The proposed risk score may improve assessment and counselling of ovarian cancer patients.

Atypical cancer risk profile in carriers of Italian founder BRCA1 variant p.His1673del: Implications for classification and clinical management

Abstract Background BRCA1 :c.5017_5019del (p.His1673del) is a founder variant relatively frequent in Northern Italy. Despite previous suggestion of pathogenicity, variant classification in public databases is still conflicting, needing additional evidence. Methods Maximum likelihood penetrance of breast/ovarian and other cancer types was estimated using full pedigree data from 53 informative Italian families. The effect of the variant on BRCA1‐ABRAXAS1 interaction was assessed using a GFP‐fragment reassembly‐based PPI assay. Results were combined with additional data from multiple sources to classify the variant according to ACMG/AMP classification rules specified for BRCA1/2 . Results Variant‐carriers displayed increased risk for ovarian cancer (HR = 33.0, 95% CI = 7.0–155.0; cumulative risk at age 70 = 27.6%, 95% CI = 12.6–40.0%) but not for breast cancer (HR = 0.7, 95% CI = 0.2–2.2). An increased risk of uterine cancer (HR = 8.0, 95% CI = 1.03–61.6) emerged, warranting further evaluation. Likelihood‐ratio in favor of pathogenicity was 98898642.82 under assumption of standard BRCA1 breast and ovarian penetrance, and 104240832.84 after excluding breast cancer diagnoses (based on penetrance results). Functional analysis demonstrated that the variant abrogates the BRCA1‐ABRAXAS1 binding, supporting the PS3 code assignment within the ACMG/AMP rule‐based model. Collectively, these findings allowed to classify the variant as pathogenic. Conclusion Pathogenicity of BRCA1 :c.5017_5019del(p.His1673del) has been confirmed; however, breast cancer risk in Italian families is not increased, unlike in families from other countries and in carriers of most BRCA1 pathogenic variants. The knowledge of atypical risk profiles for this and other variants will pave the way for personalized management based on specific genotype.