Characterizing the Increase in Artificial Intelligence Content Detection in Oncology Scientific Abstracts From 2021 to 2023.

PURPOSE: Artificial intelligence (AI) models can generate scientific abstracts that are difficult to distinguish from the work of human authors. The use of AI in scientific writing and performance of AI detection tools are poorly characterized. METHODS: We extracted text from published scientific abstracts from the ASCO 2021-2023 Annual Meetings. Likelihood of AI content was evaluated by three detectors: GPTZero, Originality.ai, and Sapling. Optimal thresholds for AI content detection were selected using 100 abstracts from before 2020 as negative controls, and 100 produced by OpenAI's GPT-3 and GPT-4 models as positive controls. Logistic regression was used to evaluate the association of predicted AI content with submission year and abstract characteristics, and adjusted odds ratios (aORs) were computed. RESULTS: Fifteen thousand five hundred and fifty-three abstracts met inclusion criteria. Across detectors, abstracts submitted in 2023 were significantly more likely to contain AI content than those in 2021 (aOR range from 1.79 with Originality to 2.37 with Sapling). Online-only publication and lack of clinical trial number were consistently associated with AI content. With optimal thresholds, 99.5%, 96%, and 97% of GPT-3/4-generated abstracts were identified by GPTZero, Originality, and Sapling respectively, and no sampled abstracts from before 2020 were classified as AI generated by the GPTZero and Originality detectors. Correlation between detectors was low to moderate, with Spearman correlation coefficient ranging from 0.14 for Originality and Sapling to 0.47 for Sapling and GPTZero. CONCLUSION: There is an increasing signal of AI content in ASCO abstracts, coinciding with the growing popularity of generative AI models.

Incidence of Severe Acute Respiratory Syndrome Coronavirus 2 and Subsequent Mortality in a Multisite Cohort of Patients With Cancer in the CancerLinQ Discovery Database.

PURPOSE: Understanding risks for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and subsequent mortality among patients with cancer may help inform treatment decisions during the COVID-19 pandemic. METHODS: CancerLinQ is an electronic health record database from US oncology practices. We identified a cohort of patients with malignancy and 2+ encounters at CancerLinQ practices in the 12 months before the study period (January 1, 2020-January 31, 2021). We identified a SARS-CoV-2 subcohort as having a positive SARS-CoV-2 test or International Classification of Diseases, 10th Revision, code. We examined predictors of SARS-CoV-2 infection and mortality including sex, race, ethnicity, age, malignancy type, and prior therapy. Unadjusted and adjusted incidence rate ratios (aIRRs) and 95% CIs were estimated from Poisson regression models for SARS-CoV-2 infections and mortality. RESULTS: The cancer cohort included 629,128 patients, and the SARS-CoV-2 subcohort included 12,300 patients. Higher incidence of SARS-CoV-2 was seen among patients who were male (incidence rate ratio [IRR], 1.14; 95% CI, 1.10 to 1.18), Black (IRR, 1.48; 95% CI, 1.41 to 1.56), Hispanic (IRR, 2.02; 95% CI, 1.91 to 2.14), age < 50 years (IRR, 1.34; 95% CI, 1.26 to 1.42), with hematologic malignancies (IRR, 1.07; 95% CI, 1.02 to 1.12), and with recent chemotherapy (IRR, 1.30, 95% CI, 1.22 to 1.40). In the adjusted analysis, higher incidence was seen in patients who were male (aIRR, 1.17; 95% CI, 1.13 to 1.21), Hispanic (aIRR, 2.01; 95% CI, 1.88 to 2.14), and with recent chemotherapy (aIRR, 1.17; 95% CI, 1.09 to 1.25). There were 182 all-cause deaths within the SARS-CoV-2 subcohort. Higher mortality was seen among patients who were male (IRR, 1.39; 95% CI, 1.04 to 1.86), unknown race (IRR, 2.64; 95% CI, 1.42 to 4.91), other/unknown ethnicity (IRR, 1.99; 95% CI, 1.20 to 3.29), age 60-69 years (IRR, 2.76; 95% CI, 1.23 to 6.19), age 70-79 years (IRR, 5.28; 95% CI, 2.42 to 11.5), age 80+ years (IRR, 7.31; 95% CI, 3.31 to 16.1), or with recent chemotherapy (IRR, 1.52, 95% CI, 1.01 to 2.29). In the adjusted analysis, higher mortality was seen with increased age and receipt of chemotherapy. CONCLUSION: Patients with increased risk of SARS-CoV-2 infection must balance the competing risks of their cancer diagnosis/treatment and SARS-CoV-2 infection.

Summary of the 12 Most Common Cancers in the CancerLinQ Discovery (CLQD) Database.

PURPOSE: In 2014, the ASCO developed CancerLinQ (CLQ), a health technology platform for oncology. The CLQ Discovery (CLQD) database was created to make data available for research and this paper provides a summary of this database. METHODS: This study described the clinical and demographic characteristics of the 12 most common cancers in the CLQD database. We included patients with a new malignant tumor diagnosis between January 1, 2013, and December 31, 2018, of the following cancers: breast, lung and bronchus, prostate, colon and rectum, melanoma of the skin, bladder, non-Hodgkin lymphoma, kidney and renal pelvis, uterus, leukemia, pancreas, and thyroid. Patients with an in-situ diagnosis were excluded. Summary statistics and Kaplan-Meier survival estimates were calculated for each tumor. RESULTS: From 2013 to 2018, 491,360 patients were diagnosed with the study tumors. Breast cancer (139,506) was the most common, followed by lung and bronchus (70,959), prostate (63,303), and colon and rectum (53,504). The median age at diagnosis (years) was 61, 68, 68, and 64 in breast, lung and bronchus, prostate, and colon and rectum cohorts, respectively. Compared to the SEER 5-year overall survival estimates for several tumor types were higher in the CLQD database, possibly because of incomplete mortality capture in electronic health records. CONCLUSION: This paper presents the first description of the CLQD database since its inception. CLQ will continue to evolve over time, and the breadth and depth of this data asset will continue to grow. ASCO and CLQ's long-term goal is to improve the quality of patient care and create a sustainable database for oncology researchers. These results demonstrate that CLQ built a scalable database that can be used for oncology research.

Growth Kinetics of Small Renal Masses on Active Surveillance: Variability and Results from the DISSRM Registry.

PURPOSE: Active surveillance is emerging as a safe and effective strategy for the management of small renal masses (4 cm or less). We characterized the growth rate and its pertinence to clinical outcomes in a prospective multi-institutional study of patients with small renal masses. MATERIALS AND METHODS: Since 2009, the DISSRM (Delayed Intervention and Surveillance for Small Renal Masses) prospective, multi-institutional registry of patients with small renal masses has enrolled patients who elect primary intervention or active surveillance. Patients who elect active surveillance received regularly scheduled imaging and those with 3 or more followup images were included in the current study to evaluate growth rates. RESULTS: We evaluated 318 patients who elected active surveillance, of whom 271 (85.2%) had 3 or more followup images available with a median imaging followup of 1.83 years. The overall mean ± SD small renal mass growth rate was 0.09 ± 1.51 cm per year (median 0.09) with no variables demonstrating statistically significant associations. The growth rate and variability decreased with longer followup (0.54 and 0.07 cm per year at less than 6 months and greater than 1 year, respectively). No patients had metastatic disease or died of kidney cancer. No statistically significant difference was noted in the growth rate in patients with biopsy demonstrated renal cell carcinoma or in those who died. CONCLUSIONS: Small renal mass growth kinetics are highly variable early on active surveillance with growth rates and variability decreasing with time. Early in active surveillance, especially during the initial 6 to 12 months, the growth rate is variable and does not reliably predict death or adverse pathological features in the patient subset with available pathology findings. An elevated growth rate may indicate the need for further assessment with imaging or consideration of biopsy prior to progressing to treatment. Additional followup will inform the best clinical pathway for elevated growth rates.

A Prospective, Comparative Study of Quality of Life among Patients with Small Renal Masses Choosing Active Surveillance and Primary Intervention.

PURPOSE: To our knowledge quality of life has not been evaluated in rigorous fashion in patients undergoing active surveillance for small renal masses. The prospective, multi-institutional DISSRM (Delayed Intervention and Surveillance for Small Renal Masses) Registry was opened on January 1, 2009, enrolling patients with cT1a (4.0 cm or less) small renal masses who elected primary intervention or active surveillance. MATERIALS AND METHODS: Patients were enrolled following a choice of active surveillance or primary intervention. The active surveillance protocol includes imaging every 4 to 6 months for 2 years and every 6 to 12 months thereafter. The SF12® quality of life questionnaire was completed at study enrollment, at 6 and 12 months, and annually thereafter. MCS (Mental Component Summary), PCS (Physical Component Summary) and overall score were evaluated among the groups and with time using ANOVA and linear regression mixed modeling. RESULTS: At 82 months among 3 institutions 539 patients were enrolled with a mean ± SD followup of 1.8 ± 1.7 years. Of the patients 254 were on active surveillance, 285 were on primary intervention and 21 were on active surveillance but crossed over to delayed intervention. A total of 1,497 questionnaires were completed. Total and PCS quality of life scores were better for primary intervention at enrollment through 5 years. There were generally no differences in MCS scores among the groups and there was a trend of improving scores with time. CONCLUSIONS: In a prospective registry of patients undergoing active surveillance or primary intervention for small renal masses those undergoing primary intervention had higher quality of life scores at baseline. This was due to a perceived benefit in the physical health domain, which persisted throughout followup. Mental health, which includes the domains of depression and anxiety, was not adversely affected while on active surveillance, and it improved with time after selecting a management strategy.