Search Results

You are looking at 1 - 4 of 4 items for

  • Author: Kelvin K.W. Chan x
  • Refine by Access: All x
Clear All Modify Search
Full access

Machine Learning–Based Early Warning Systems for Acute Care Utilization During Systemic Therapy for Cancer

Robert C. Grant, Jiang Chen He, Ferhana Khan, Ning Liu, Sho Podolsky, Yosuf Kaliwal, Melanie Powis, Faiyaz Notta, Kelvin K.W. Chan, Marzyeh Ghassemi, Steven Gallinger, and Monika K. Krzyzanowska

Background: Emergency department visits and hospitalizations frequently occur during systemic therapy for cancer. We developed and evaluated a longitudinal warning system for acute care use. Methods: Using a retrospective population-based cohort of patients who started intravenous systemic therapy for nonhematologic cancers between July 1, 2014, and June 30, 2020, we randomly separated patients into cohorts for model training, hyperparameter tuning and model selection, and system testing. Predictive features included static features, such as demographics, cancer type, and treatment regimens, and dynamic features, such as patient-reported symptoms and laboratory values. The longitudinal warning system predicted the probability of acute care utilization within 30 days after each treatment session. Machine learning systems were developed in the training and tuning cohorts and evaluated in the testing cohort. Sensitivity analyses considered feature importance, other acute care endpoints, and performance within subgroups. Results: The cohort included 105,129 patients who received 1,216,385 treatment sessions. Acute care followed 182,444 (15.0%) treatments within 30 days. The ensemble model achieved an area under the receiver operating characteristic curve of 0.742 (95% CI, 0.739–0.745) and was well calibrated in the test cohort. Important predictive features included prior acute care use, treatment regimen, and laboratory tests. If the system was set to alarm approximately once every 15 treatments, 25.5% of acute care events would be preceded by an alarm, and 47.4% of patients would experience acute care after an alarm. The system underestimated risk for some treatment regimens and potentially underserved populations such as females and non-English speakers. Conclusions: Machine learning warning systems can detect patients at risk for acute care utilization, which can aid in preventive intervention and facilitate tailored treatment. Future research should address potential biases and prospectively evaluate impact after system deployment.

Full access

Are Surrogate Endpoints Unbiased Metrics in Clinical Benefit Scores of the ASCO Value Framework?

Sierra Cheng, Matthew C. Cheung, Di Maria Jiang, Erica McDonald, Vanessa S. Arciero, Doreen Anuli Ezeife, Amanda Rahmadian, Alexandra Chambers, Kelley-Anne Sabarre, Ambika Parmar, and Kelvin K.W. Chan

Background: Clinical benefit scores (CBS) are key elements of the ASCO Value Framework (ASCO-VF) and are weighted based on a hierarchy of efficacy endpoints: hazard ratio for death (HR OS), median overall survival (mOS), HR for disease progression (HR PFS), median progression-free survival (mPFS), and response rate (RR). When HR OS is unavailable, the other endpoints serve as “surrogates” to calculate CBS. CBS are computed from PFS or RR in 39.6% of randomized controlled trials. This study examined whether surrogate-derived CBS offer unbiased scoring compared with HR OS–derived CBS. Methods: Using the ASCO-VF, CBS for advanced disease settings were computed for randomized controlled trials of oncology drug approvals by the FDA, European Medicines Agency, and Health Canada in January 2006 through December 2017. Mean differences of surrogate-derived CBS minus HR OS–derived CBS assessed the tendency of surrogate-derived CBS to overestimate or underestimate clinical benefit. Spearman’s correlation evaluated the association between surrogate- and HR OS–derived CBS. Mean absolute error assessed the average difference between surrogate-derived CBS relative to HR OS–derived CBS. Results: CBS derived from mOS, HR PFS, mPFS, and RR overestimated HR OS–derived CBS in 58%, 68%, 77%, and 55% of pairs and overall by an average of 5.62 (n=90), 6.86 (n=110), 29.81 (n=101), and 3.58 (n=108), respectively. Correlation coefficients were 0.80 (95% CI, 0.70–0.86), 0.38 (0.20–0.53), 0.20 (0.00–0.38), and 0.01 (–0.18 to 0.19) for mOS-, HR PFS–, mPFS-, and RR-derived CBS, respectively, and mean absolute errors were 11.32, 12.34, 40.40, and 18.63, respectively. Conclusions: Based on the ASCO-VF algorithm, HR PFS–, mPFS-, and RR-derived CBS are suboptimal surrogates, because they were shown to be biased and poorly correlated to HR OS–derived CBS. Despite lower weighting than OS in the ASCO-VF algorithm, PFS still overestimated CBS. Simple rescaling of surrogate endpoints may not improve their validity within the ASCO-VF given their poor correlations with HR OS–derived CBS.

Full access

Incident Cancer Detection During the COVID-19 Pandemic

Antoine Eskander, Qing Li, Jiayue Yu, Julie Hallet, Natalie G. Coburn, Anna Dare, Kelvin K.W. Chan, Simron Singh, Ambica Parmar, Craig C. Earle, Lauren Lapointe-Shaw, Monika K. Krzyzanowska, Timothy P. Hanna, Antonio Finelli, Alexander V. Louie, Nicole Look Hong, Jonathan C. Irish, Ian J. Witterick, Alyson Mahar, Christopher W. Noel, David R. Urbach, Daniel I. McIsaac, Danny Enepekides, and Rinku Sutradhar

Background: Resource restrictions were established in many jurisdictions to maintain health system capacity during the COVID-19 pandemic. Disrupted healthcare access likely impacted early cancer detection. The objective of this study was to assess the impact of the pandemic on weekly reported cancer incidence. Patients and Methods: This was a population-based study involving individuals diagnosed with cancer from September 25, 2016, to September 26, 2020, in Ontario, Canada. Weekly cancer incidence counts were examined using segmented negative binomial regression models. The weekly estimated backlog during the pandemic was calculated by subtracting the observed volume from the projected/expected volume in that week. Results: The cohort consisted of 358,487 adult patients with cancer. At the start of the pandemic, there was an immediate 34.3% decline in the estimated mean cancer incidence volume (relative rate, 0.66; 95% CI, 0.57–0.75), followed by a 1% increase in cancer incidence volume in each subsequent week (relative rate, 1.009; 95% CI, 1.001–1.017). Similar trends were found for both screening and nonscreening cancers. The largest immediate declines were seen for melanoma and cervical, endocrinologic, and prostate cancers. For hepatobiliary and lung cancers, there continued to be a weekly decline in incidence during the COVID-19 period. Between March 15 and September 26, 2020, 12,601 fewer individuals were diagnosed with cancer, with an estimated weekly backlog of 450. Conclusions: We estimate that there is a large volume of undetected cancer cases related to the COVID-19 pandemic. Incidence rates have not yet returned to prepandemic levels.

Full access

Incident Cancer Detection During Multiple Waves of COVID-19: The Tsunami After the Earthquake

Rui Fu, Rinku Sutradhar, Qing Li, Timothy P. Hanna, Kelvin K.W. Chan, Jonathan C. Irish, Natalie Coburn, Julie Hallet, Anna Dare, Simron Singh, Ambica Parmar, Craig C. Earle, Lauren Lapointe-Shaw, Monika K. Krzyzanowska, Antonio Finelli, Alexander V. Louie, Nicole J. Look Hong, Ian J. Witterick, Alyson Mahar, David Gomez, Daniel I. McIsaac, Danny Enepekides, David R. Urbach, and Antoine Eskander

No population-based study exists to demonstrate the full-spectrum impact of COVID-19 on hindering incident cancer detection in a large cancer system. Building upon our previous publication in JNCCN, we conducted an updated analysis using 12 months of new data accrued in the pandemic era (extending the study period from September 26, 2020, to October 2, 2021) to demonstrate how multiple COVID-19 waves affected the weekly cancer incidence volume in Ontario, Canada, and if we have fully cleared the backlog at the end of each wave.