Predicting response to neoadjuvant chemotherapy for colorectal liver metastasis using deep learning on prechemotherapy cross-sectional imaging.

BACKGROUND AND OBJECTIVES: Deep learning models (DLMs) are applied across domains of health sciences to generate meaningful predictions. DLMs make use of neural networks to generate predictions from discrete data inputs. This study employs DLM on prechemotherapy cross-sectional imaging to predict patients' response to neoadjuvant chemotherapy. METHODS: Adult patients with colorectal liver metastasis who underwent surgery after neoadjuvant chemotherapy were included. A DLM was trained on computed tomography images using attention-based multiple-instance learning. A logistic regression model incorporating clinical parameters of the Fong clinical risk score was used for comparison. Both model performances were benchmarked against the Response Evaluation Criteria in Solid Tumors criteria. A receiver operating curve was created and resulting area under the curve (AUC) was determined. RESULTS: Ninety-five patients were included, with 33,619 images available for study inclusion. Ninety-five percent of patients underwent 5-fluorouracil-based chemotherapy with oxaliplatin and/or irinotecan. Sixty percent of the patients were categorized as chemotherapy responders (30% reduction in tumor diameter). The DLM had an AUC of 0.77. The AUC for the clinical model was 0.41. CONCLUSIONS: Image-based DLM for prediction of response to neoadjuvant chemotherapy in patients with colorectal cancer liver metastases was superior to a clinical-based model. These results demonstrate potential to identify nonresponders to chemotherapy and guide select patients toward earlier curative resection.

Real-World Impact of an In-House Dihydropyrimidine Dehydrogenase (DPYD) Genotype Test on Fluoropyrimidine Dosing, Toxicities, and Hospitalizations at a Multisite Cancer Center.

PURPOSE: Fluoropyrimidine-related toxicity and mortality risk increases significantly in patients carrying certain DPYD genetic variants with standard dosing. We implemented DPYD genotyping at a multisite cancer center and evaluated its impact on dosing, toxicity, and hospitalization. METHODS: In this prospective observational study, patients receiving (reactive) or planning to receive (pretreatment) fluoropyrimidine-based chemotherapy were genotyped for five DPYD variants as standard practice per provider discretion. The primary end point was the proportion of variant carriers receiving fluoropyrimidine modifications. Secondary end points included mean relative dose intensity, fluoropyrimidine-related grade 3+ toxicities, and hospitalizations. Fisher's exact test compared toxicity and hospitalization rates between pretreatment carriers, reactive carriers, and wild-type patients. Univariable and multivariable logistic regression identified factors associated with toxicity and hospitalization risk. Kaplan-Meier methods estimated time to event of first grade 3+ toxicity and hospitalization. RESULTS: Of the 757 patients who received DPYD genotyping (median age 63, 54% male, 74% White, 19% Black, 88% GI malignancy), 45 (5.9%) were heterozygous carriers. Fluoropyrimidine was modified in 93% of carriers who started treatment. In 442 patients with 3-month follow-up, 64%, 31%, and 30% of reactive carriers, pretreatment carriers, and wild-type patients had grade 3+ toxicity, respectively (P = .085); 64%, 25%, and 13% were hospitalized (P < .001). Reactive carriers had 10-fold higher odds of hospitalization compared with wild-type patients (P = .001), whereas no significant difference was noted between pretreatment carriers and wild-type patients. Time-to-event of toxicity and hospitalization were significantly different between genotype groups (P < .001), with reactive carriers having the earliest onset and highest incidence. CONCLUSION: DPYD genotyping prompted fluoropyrimidine modifications in most carriers. Pretreatment testing reduced toxicities and hospitalizations compared with reactive testing, thus normalizing the risk to that of wild-type patients, and should be considered standard practice.

Pertuzumab Plus Trastuzumab in Patients With Biliary Tract Cancer With ERBB2/3 Alterations: Results From the Targeted Agent and Profiling Utilization Registry Study.

PURPOSE: Targeted Agent and Profiling Utilization Registry is a phase II basket trial evaluating the antitumor activity of commercially available targeted agents in patients with advanced cancer and genomic alterations known to be drug targets. Results of a cohort of patients with biliary tract cancer (BTC) with ERBB2/3 amplification, overexpression, or mutation treated with pertuzumab plus trastuzumab are reported. METHODS: Eligible patients had advanced BTC, measurable disease (RECIST v1.1), Eastern Cooperative Oncology Group performance status 0-2, adequate organ function, tumors with ERBB2/3 alterations, and a lack of standard treatment options. Simon's two-stage design was used with a primary end point of disease control (DC), defined as objective response (OR) or stable disease of at least 16+ weeks duration (SD16+) according to RECIST v1.1. Secondary end points included OR, progression-free survival, overall survival, duration of response, duration of stable disease, and safety. RESULTS: Twenty-nine patients were enrolled from February 2017 to January 2022, and all had advanced BTC with an ERBB2/3 alteration. One patient was not evaluable for efficacy. One complete response, eight partial responses, and two SD16+ were observed for DC and OR rates of 40% (90% CI, 27 to 100) and 32% (95% CI, 16 to 52), respectively. The null hypothesis of 15% DC rate was rejected (P = .0015). Four patients had at least one grade 3 adverse event (AE) or serious AE at least possibly related to treatment: anemia, diarrhea, infusion-related reaction, and fatigue. CONCLUSION: Pertuzumab plus trastuzumab met prespecified criteria to declare a signal of activity in patients with BTC and ERBB2/3 amplification, overexpression, or mutation.

Predictive value of CDC37 gene expression for targeted therapy in metastatic colorectal cancer.

BACKGROUND: CDC37 is a key determinant of client kinase recruitment to the HSP90 chaperoning system. We hypothesized that kinase-specific dependency on CDC37 alters the efficacy of targeted therapies for metastatic colorectal cancer (mCRC). MATERIAL AND METHODS: Two independent mCRC cohorts were analyzed to compare the survival outcomes between CDC37-high and CDC37-low patients (stratified by the median cutoff values): the CALGB/SWOG 80405 trial (226 and 207 patients receiving first-line bevacizumab- and cetuximab-containing chemotherapies, respectively) and Japanese retrospective (50 refractory patients receiving regorafenib) cohorts. A dataset of specimens submitted to a commercial CLIA-certified laboratory was utilized to characterize molecular profiles of CDC37-high (top quartile, N = 5055) and CDC37-low (bottom quartile, N = 5055) CRCs. RESULTS: In the bevacizumab-treated group, CDC37-high patients showed significantly better progression-free survival (PFS) (median 13.3 vs 9.6 months, hazard ratio [HR] 0.59, 95% confidence interval [CI] 0.44-0.79, p < 0.01) than CDC37-low patients. In the cetuximab-treated group, CDC37-high and CDC37-low patients had similar outcomes. In the regorafenib-treated group, CDC37-high patients showed significantly better overall survival (median 11.3 vs 6.0 months, HR 0.24, 95% CI 0.11-0.54, p < 0.01) and PFS (median 3.5 vs 1.9 months, HR 0.51, 95% CI 0.28-0.94, p = 0.03). Comprehensive molecular profiling revealed that CDC37-high CRCs were associated with higher VEGFA, FLT1, and KDR expressions and activated hypoxia signature. CONCLUSIONS: CDC37-high mCRC patients derived more benefit from anti-VEGF therapies, including bevacizumab and regorafenib, but not from cetuximab. Molecular profiles suggested that such tumors were dependent on angiogenesis-relating pathways.