Swiss Re AG · SIX:SREN
Swiss Re — Technical Presentation (2020)
28 slides from Swiss Re's technical presentation, 2020
Slide 1: Integrated Survival Estimates from Cancer Treatment Delay during the COVID-19 Pandemic
Integrated Survival Estimates from Cancer Treatment Delay during the COVID-19 Pandemic Holly Hartman https://holly-hartman.com/ @holly7holly
Slide 2: Acknowledgments
Acknowledgments UM Radiation Oncology UM Biostatistics Team Penn State Team Team • Yilun Sun • Neil Jairath • Vonn Walter • Theresa Devasia • Bob Dess • Ming Wang • Elizabeth Chase • Will Jackson • Xi Wang • Emily Morris • Daniel Spratt • Nicholas Zaorsky • Pin Li • Kim Hochstedler • Madeline Abbott
Slide 3: Cancer and COVID-19
Cancer patients are at increased risk of COVID-19 mortality Going to the hospital increases the Cancer and risk of contracting COVID-19 (more potentially infectious contacts) COVID-19 Delaying cancer treatment means the cancer can spread and increase mortality rates
Slide 4: Cancer treatments and COVID-19
Cancer treatments and COVID-19 Immunosuppressed Surgery requires Chemotherapy people are at ventilation and also makes people increased risk of increases the risk of immunosuppressed COVID-19 mortality contracting COVID-19
Slide 5: Big question: How to treat cancer patients during the COVID-19 pandemic?
Big question: How to treat cancer patients during the COVID-19 pandemic?
Slide 6: Currently: Making blanket decisions based on cancer type and stage
Currently: Making blanket decisions based on cancer type and stage • Tier systems based on cancer site and stage may: • Ignore comorbidities • Ignore geographical differences in the state of the pandemic • Ignore age • Ignore treatment plans
Slide 7: OncCOVID
OncCOVID • Web app to guide clinicians and help make personalized treatment decisions for patients • 47 customizable inputs • Outputs infection risk, mortality risk, survival curves, and RMST • http://onccovid.med.umich.edu/
Slide 8: 4 Big Pieces
Cancer Cancer mortality mortality under under immediate delayed treatment treatment 4 Big Pieces COVID-19 COVID-19 infection risk mortality risk
Slide 9: Cancer mortality under immediate treatment
Cancer mortality under immediate treatment • Used data from Surveillance, Epidemiology, and End Results (SEER) • Estimate survival based on age, cancer type, and cancer stage • Cox proportional hazards model to estimate overall survival • Used Fine-Grey regression (to account for competing risks) to
Slide 10: Cancer mortality under delayed treatment
Collaborators at Penn State used NCDB to estimate the effect of delaying treatment by cancer disease site and stage Cancer mortality Conducted literature review to find under delayed the impact of delay of treatment by stage and disease site treatment Results from both the lit review and the NCDB an
Slide 11: Cancer mortality under delayed treatment - NCDB
Cancer mortality under delayed treatment - NCDB • Multivariate cox proportional hazard model for overall survival • Adjusted for confounders (age, race, insurance status, facility information, etc). • Ended up with a hazard ratio for a per day increase by disease site and stage.
Slide 12: COVID-19 infection risk
COVID-19 infection risk • Susceptible-Infected-Recovered (SIR) Model • Modeled the health care system separately from the general population due to the increased incidence of COVID-19 among healthcare workers • Assume patients have increased risk when visiting hospital for treatment due to increase
Slide 13: Reproduction rate
• 𝑅! - the number of new infections caused by one person in an entirely susceptible population • 𝑅" - the current number of new infections caused by one person Reproduction • 𝑅" changes based on social rate distancing measures and the number of immune people in the population • China saw the…
Slide 14: SIR Models
SIR Models • Let 𝑁 be the population size, 𝐼! be the number currently infected at time 𝑡, and 𝑀! be the number recovered/removed. Then the number of people susceptible, 𝑆! is 𝑆! = 𝑁 − 𝐼! − 𝑀! . • We denote the average number of people infected from one person on day 𝑡 is 𝑅! . • Let 𝐷…
Slide 15: SIR Model
SIR Model • New infections on day 𝑡 + 1 is "# estimated to be 𝛽 ! ! $ • Number susceptible on day 𝑡 + "# 1 is given by 𝑆!%& = 𝑆! − 𝛽 ! ! $ • Number recovered/removed on day 𝑡 + 1 is 𝑀!%& = 𝑀! + 𝛾𝐼! • The predicted total number infected on day 𝑡 + 1 is 𝐼!%& = "! #! 𝐼! + 𝛽 $ − 𝛾𝐼! .
Slide 16: Multi-setting SIR
• Altered to allow 𝛽 to vary between the general public and the health care setting. • Modeled health care and general population separately assuming a Multi-setting percentage of the general population come in for treatment each day and that a SIR percentage of those are there for surgery.
Slide 17: Multi-setting SIR
• Let 𝑆%,' and 𝑆(,' be the number susceptible on day 𝑡 for the general population and the healthcare population, respectively. • Let 𝐼%,' and 𝐼(,' be the number infected on day 𝑡 for the general population and the healthcare population, respectively. Multi-setting • Let the number of new…
Slide 18: COVID-19 Mortality Risk
We know comorbidities increase the mortality rate of COVID- COVID-19 19, but we also know that age plays an important role. Mortality Risk We also know that age and the number of comorbidities are correlated. Need to find COVID-19 mortality by age and number of comorbidities. Data we have: COVID-19
Slide 19: COVID-19 Mortality Risk
COVID-19 Mortality Risk • Constrained logistic regression (to ensure monotonic relationship between age and mortality and also # comorbidities and mortality) to obtain estimates of mortality by age and number of comorbidities. • Adjusted for the case control design. • Obtained estimates of proportio
Slide 20: Comorbidity results
# Comorbidities Age 0 1 2 3+ Comorbidity 40-50 0.32% 0.40% 0.49% 1.15% results 51-60 0.96% 1.18% 1.45% 3.39% 61-70 2.46% 3.01% 3.69% 8.51% 71-80 5.61% 6.86% 8.42% 19.43% 81+ 11.60% 14.16% 17.34% 39.30% • Also account for the healthcare system being overwhelmed and chemotherapy
Slide 21: Geographic Information
Able to use Johns Hopkins data (publicly available on GitHub) for daily COVID-19 data by county, province, or country Geographic US Census estimates by county for 2019 for population estimates for US Information Counties Used World Bank population estimates for non-US countries, states, or provinces
Slide 22: Restricted Mean Survival Time - RMST
Restricted Mean Survival Time - RMST • The expected time a person will live through a fixed time point. • Area under the survival curve • Alternative measure to HR that works when the assumption of proportional hazards is not valid. Dehbi et al. BMJ 2017;357:j2250 doi: 10.1136/bmj.j2250 (Published 2
Slide 23: Heterogeneity in patients
Heterogeneity in patients RMST- Immediate minus Delayed Treatment (days) 150 120 90 60 Favors Immediate Treatment Cancer Type RMST- Immediate minus Delayed Treatment (days) 150 120 Favors Immediate Treatment LRP ITNT, Favors Delayed Treatment Tier 1 Tier 2 Tier 3 (No Delay of (Short Delay of (Prolon
Slide 24: Heterogeneity by location
Heterogeneity by location II
Slide 25: What is going on with lung?
Immediate ao] a > x a a ) bw | | bse Leo o oO tN > 2 lo ° rey ry 0 3 s Rr 8 N ) = S Ss S S JEANS |}e10A0 an. Q0 e S [~ —_ too W 2 oO — © as) tN > to Cc 6 ©» @ 8 Ss Rn & q iS) oO eb & 8 & JEAIAINS ||2J9AQ 5 . Q0 is 2 oe 3 ~ tN > O S ° ° re) Foy ry 2 s rR & q s = Ss f=) o (—) JEAIAINS ||29AO
Slide 26: Next Steps
Next Steps Update estimates Update model for Improve SIR model Attempt to account of overall survival delay of treatment to automatically for uncertainty in by age. Currently, HR to better incorporate estimates and over estimate account for age. historical trends, develop method to survival for elde
Slide 27: Thank you
Thank you
Slide 28: References
References • 2019 Novel Coronavirus COVID-19 (2019-nCoV) Data Repository by Johns Hopkins CSSE. (Accessed April 15, 2020, at https://github.com/CSSEGISandData/COVID-19.) • US Census 2019 County Estimates. (Accessed April 15, 2020, at https://www2.census.gov/programs-surveys/popest/datasets/2010-2019