PURPOSE Machine learning (ML) algorithms that incorporate routinely collected patient-reported outcomes (PROs) alongside electronic health record (EHR) variables may improve prediction of short-term mortality and facilitate earlier supportive and palliative care for patients with cancer. METHODS We trained and validated two-phase ML algorithms that incorporated standard PRO assessments alongside approximately 200 routinely collected EHR variables, among patients with medical oncology encounters at a tertiary academic oncology and a community oncology practice. RESULTS Among 12,350 patients, 5,870 (47.5%) completed PRO assessments. Compared with EHR- and PRO-only algorithms, the EHR + PRO model improved predictive performance in both tertiary oncology (EHR + PRO v EHR v PRO: area under the curve [AUC] 0.86 [0.85-0.87] v 0.82 [0.81-0.83] v 0.74 [0.74-0.74]) and community oncology (area under the curve 0.89 [0.88-0.90] v 0.86 [0.85-0.88] v 0.77 [0.76-0.79]) practices. CONCLUSION Routinely collected PROs contain added prognostic information not captured by an EHR-based ML mortality risk algorithm. Augmenting an EHR-based algorithm with PROs resulted in a more accurate and clinically relevant model, which can facilitate earlier and targeted supportive care for patients with cancer.

4586 Background: Neoadjuvant cisplatin-based chemotherapy (NAC) prior to radical cystectomy is the standard of care for patients with muscle invasive bladder cancer (MIBC). Three or more cycles of NAC are commonly administered based on outcomes in prior prospective trials, however many patients are intolerant of 3 cycles of NAC due to toxicity. The prognosis of patients receiving fewer than 3 cycles of NAC has yet to be elucidated. Methods: This is a retrospective single-center study to quantify pathologic response, recurrence-free survival (RFS), and overall survival (OS) from time of first NAC in patients treated with < 3 cycles of NAC compared to patients treated with ≥3 cycles of NAC. Inclusion criteria include: diagnosis of MIBC between 2004-2017, ≥5 years of follow up, deceased or had recurrence prior to 5 years follow up but with adequate follow-up data. Patients were excluded from the study if they were found to have metastatic disease prior to initiation of NAC, lost to follow up prior to 5 years with no evidence of recurrence or death, or had progression of disease during NAC. The primary objective of this study was to determine RFS and OS in patients stratified by cycles of NAC. Patient characteristics were compared using chi-square tests, Fisher’s exact tests, and Wilcoxon rank sum tests. Kaplan Meier curves and log-rank tests were used to compare RFS between subgroups, and Cox proportional hazards models were used to compare RFS adjusting for ECOG performance status, baseline GFR, stage, type of NAC regimen, and patient age at first dose of NAC. 5-year OS was calculated as a percentage of patients surviving at 5 years. Results: A total sample of 195 patients were treated with NAC for MIBC, of which 30 (15.3%) patients received < 3 cycles and 165 (84.6%) patients received ≥3 cycles of NAC. 53 patients received gemcitabine and cisplatin (GC) and 142 patients received methotrexate, vinblastine, doxorubicin, and cisplatin (MVAC). Complete pathologic response (ypT0N0) was observed in 7.4% of patients receiving < 3 cycles and in 26.8% of patient receiving ≥3 cycles of NAC (p = 0.024). Significant pathologic response ( < ypT2) was obtained in 22.2% of patients receiving < 3 cycles of NAC and in 41.8% of patients receiving ≥3 cycles (p > 0.05). Median RFS in patients with < 3 cycles of NAC was 8.8 months (95% CI 6.51, 13.4) and 54.5 months with ≥3 cycles of NAC (95% CI 29.8, 111.9). Based on a log-rank test there is a statistically significant difference in unadjusted RFS between the two groups (p < 0.001). The 5-year OS in patients receiving < 3 cycles of NAC was 13.3%, and the 5-year OS in those receiving ≥3 cycles was 53.3%. Conclusions: Early cessation of chemotherapy due to intolerance had significant implications on pathologic response, RFS and OS. Clinicians should prioritize administering at least 3 cycles of cisplatin-based chemotherapy when feasible to optimize outcomes.

ImportanceDelays in starting cancer treatment disproportionately affect vulnerable populations and can influence patients’ experience and outcomes. Machine learning algorithms incorporating electronic health record (EHR) data and neighborhood-level social determinants of health (SDOH) measures may identify at-risk patients.ObjectiveTo develop and validate a machine learning model for estimating the probability of a treatment delay using multilevel data sources.Design, Setting, and ParticipantsThis cohort study evaluated 4 different machine learning approaches for estimating the likelihood of a treatment delay greater than 60 days (group least absolute shrinkage and selection operator [LASSO], bayesian additive regression tree, gradient boosting, and random forest). Criteria for selecting between approaches were discrimination, calibration, and interpretability/simplicity. The multilevel data set included clinical, demographic, and neighborhood-level census data derived from the EHR, cancer registry, and American Community Survey. Patients with invasive breast, lung, colorectal, bladder, or kidney cancer diagnosed from 2013 to 2019 and treated at a comprehensive cancer center were included. Data analysis was performed from January 2022 to June 2023.ExposuresVariables included demographics, cancer characteristics, comorbidities, laboratory values, imaging orders, and neighborhood variables.Main Outcomes and MeasuresThe outcome estimated by machine learning models was likelihood of a delay greater than 60 days between cancer diagnosis and treatment initiation. The primary metric used to evaluate model performance was area under the receiver operating characteristic curve (AUC-ROC).ResultsA total of 6409 patients were included (mean [SD] age, 62.8 [12.5] years; 4321 [67.4%] female; 2576 [40.2%] with breast cancer, 1738 [27.1%] with lung cancer, and 1059 [16.5%] with kidney cancer). A total of 1621 (25.3%) experienced a delay greater than 60 days. The selected group LASSO model had an AUC-ROC of 0.713 (95% CI, 0.679-0.745). Lower likelihood of delay was seen with diagnosis at the treating institution; first malignant neoplasm; Asian or Pacific Islander or White race; private insurance; and lacking comorbidities. Greater likelihood of delay was seen at the extremes of neighborhood deprivation. Model performance (AUC-ROC) was lower in Black patients, patients with race and ethnicity other than non-Hispanic White, and those living in the most disadvantaged neighborhoods. Though the model selected neighborhood SDOH variables as contributing variables, performance was similar when fit with and without these variables.Conclusions and RelevanceIn this cohort study, a machine learning model incorporating EHR and SDOH data was able to estimate the likelihood of delays in starting cancer therapy. Future work should focus on additional ways to incorporate SDOH data to improve model performance, particularly in vulnerable populations.

584 Background: Breast cancer tumor phenotype has prognostic value with triple negative (TN) cancers having higher rates of distant metastases early. While tumor phenotypes are prognostic, published data demonstrates that the choice of local therapy does not affect this predisposition or affect survival. This study was performed to determine whether mastectomy is being performed more frequently for TN or HER2+ phenotypes, relative to hormone receptor positive (HR+) phenotypes despite the lack of benefit this should provide. Methods: Data from the National Cancer Database (NCDB) was analyzed from 2010 through 2019 to assess mastectomy trends and associations with patient and tumor characteristics. Women with invasive breast cancer were included. Women with Stage IV disease were excluded. Patients were categorized as mastectomy or breast conservation surgery. Patient and tumor characteristics were compared across groups using chi-square and Wilcoxon rank sum tests, and a multivariable logistic regression model was fit to assess the association between mastectomy and tumor phenotype controlling for patient and tumor characteristics. Results: 543,590 patients were evaluated. 173,380 (31.9%) patients underwent mastectomy, and 370,210 (68.1%) patients underwent breast conservation surgery. Mean age at diagnosis was 56. There were 425,174 HR+, 64,960 HER2+, and 53,456 TN tumors. The proportion of patients undergoing mastectomy peaked in 2013 at 36.14% before declining. Compared to HR+, HER2+ patients were more likely to undergo mastectomy, OR 1.39 p < 0.0001 (95% CI 1.35 – 1.43); however, there was no significant difference in mastectomy between HR+ patients and TN patients. Compared to whites, black patients were less likely to undergo mastectomy, OR 0.71 p < 0.0001 (95% CI 0.69 – 0.74), and individuals of Hispanic ethnicity less likely to undergo mastectomy, OR 0.92 p < 0.0001 (95% CI 0.89 – 0.95). Compared to private insurance, Medicare had a greater association with mastectomy, OR 1.2 p < 0.0001 (95% CI 1.18 – 1.23). There was no significant difference between other forms of insurance (Medicaid, other government insurance, no insurance) and private insurance. Education and income were not associated with different frequencies of mastectomy. Patients with higher comorbidity scores were more likely to undergo mastectomy. Conclusions: Mastectomy rates have been declining since 2013 at CoC centers. While TN breast cancer is not associated with increased mastectomy percent, mastectomy continues to be performed more frequently for HER2+ positive phenotype when adjusting for tumor and patient characteristics. These data suggest a need for education about HER2 positive phenotype due to a possible lack of understanding about the why such tumors pose a risk, and the role of local therapy in treating them.