Swiss Re AG · SIX:SREN
Swiss Re — Technical Presentation (February 2021)
44 slides from Swiss Re's technical presentation, Feb 16, 2021
Slide 1: Effects of the Covid-19 Pandemic on Health and Mortality in the US
Effects of the Covid-19 Pandemic on Health and Mortality in the US Andrew C. Stokes, PhD Boston University School of Public Health Expert Forum on Secondary Covid-19 Impacts, Swiss Re Institute February 16, 2021
Slide 2: Agenda
Agenda. 1. Background a. What are excess deaths? b. What are the different types of excess deaths? c. How many excess deaths have occured? d. What is known about inequities in excess deaths? e. Gaps in the literature and our approach 2. Research Question and Hypothesis 3. Methods 4. Findings 5. Limi
Slide 3: Background
Background.
Slide 4: What Are Excess Deaths?
What Are Excess Deaths?. Excess deaths are typically defined as the difference between the observed number of deaths in a specific time period and expected number of deaths in the same time period. Image from: https://www.washingtonpost.com/investigations/2020/05/02/excess-deaths-during-covid-19/
Slide 5: What Are the Different Types of Excess Deaths?
What Are the Different Types of Excess Deaths?. 1. Directly assigned deaths 2. Excess deaths not assigned to Covid-19 ○ “Misclassified” deaths ○ Indirect deaths Adapted from tweet from: https://twitter.com/andrewnoymer/status/1241620288825167874?lang=en
Slide 6: Directly Assigned Deaths
Directly Assigned Deaths. 456,900 directly assigned deaths (as of February 5, 2021) Image from: https://coronavirus.jhu.edu/map.html
Slide 7: “Misclassified” Deaths
“Misclassified” Deaths. Chart (top) from: https://www.cdc.gov/nchs/nvss/vsrr/covid19/excess_deaths.htm Figures (bottom) from: https://www.politico.com/news/2020/09/16/dementia-deaths-coronavirus-nursing-homes-416530
Slide 8: Indirect Deaths
Indirect Deaths. Infographic and Table (top) from: https://www.cdc.gov/mmwr/volumes/69/wr/mm6923e1.htm | Graph (bottom) from: https://lernercenter.syr.edu/2020/11/17/ds-30/ (data source: CDC’s Outpatient Influenza-like Illness Surveillance Network)
Slide 9: How Many Excess Deaths Have Occured?
How Many Excess Deaths Have Occured?. % of Excess Deaths Not Assigned to Covid-19: 22% (05/30/2020) % of Excess Deaths Not Assigned to Covid-19: 31% (11/21/2020) Graph from: Weinberger DM, Chen J, Cohen T, et al. Estimation of Excess Deaths Associated With the COVID-19 Pandemic in the United States,
Slide 10: How Many Excess Deaths Have Occured?
How Many Excess Deaths Have Occured?. % of Excess Deaths Not Assigned to Covid-19: 33% (08/01/2020) Graph from: Woolf SH, Chapman DA, Sabo RT, Weinberger DM, Hill L, Taylor DDH. Excess Deaths From COVID-19 and Other Causes, March-July 2020. JAMA. 2020;324(15):1562–1564. doi:10.1001/jama.2020.19545 (
Slide 11: What Is Known About Inequities in Excess Deaths?
What Is Known About Inequities in Excess Deaths?. Graph from: https://covidtracking.com/race (crude direct Covid-19 death rates)
Slide 12: What Is Known About Inequities in Excess Deaths?
What Is Known About Inequities in Excess Deaths?. Graph from: https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/bulletins/deathsinvolving covid19bylocalareasanddeprivation/deathsoccurringbetween1marchand31may2020
Slide 13: Gaps in the Literature and Our Approach
Gaps in the Literature and Our Approach. ● The primary limitation of prior work is that the use of aggregated national and state-level data has made it difficult for researchers to examine inequities in excess deaths. ● Our approach is to use county-level data and area-level measures as a proxy for
Slide 14: Research Question and Hypothesis
Research Question and Hypothesis.
Slide 15: Research Question
Research Question. 1. What are the total number of excess deaths associated with the Covid-19 pandemic to-date? 2. How do excess deaths not directly assigned to Covid-19 differ by sociodemographic and health factors?
Slide 16: Research Question
Research Question. ● Our hypothesis is that Covid-19 mortality and inequities in Covid-19 mortality are underestimated by direct death tallies. ● Why? a. Disparities in testing and higher rate of comorbid conditions may lead to less accurate cause of death assignment b. Structural factors in health
Slide 17: Methods
Methods.
Slide 18: Methods
Methods. ● We take advantage of spatial variation in Covid-19 mortality across US counties to estimate its relationship with all-cause mortality across all counties. ● We anticipate that counties with higher mortality from Covid-19 will also have experienced greater increases in mortality from other
Slide 19: Methods
Methods. ● The value of β1 represents the extent to which past levels of all- cause mortality in a county are replicated in 2020. ● Combinations of α and β1 indicate how mortality changes not associated with Covid-19 vary with all-cause mortality in 2013-18. ● The value of β2 indicates the extent to
Slide 20: Methods
Methods. ● NCHS provisional county-level data on all-cause mortality and direct Covid-19 deaths from February 1 to October 17, 2020 reported before December 15, 2020; dataset limited to counties with more than 20 Covid-19 deaths ● Constructed a historical comparison period using CDC Wonder data from
Slide 21: Findings
Findings.
Slide 22: Findings
Findings. n=787 counties (counties with more than 20 direct Covid-19 deaths as of October 17, 2020)
Slide 23: Findings
All-Cause Death Rate, 2020 - Rate, 2013-2018 De) oO 1 a oa 1 = {=} i Census Region © Midwest © Northeast South ~ West T T T T 1 2 3 4 Direct Covid-19 Death Rate per 1000 Person-Years
Slide 24: Findings
Findings. ● Given the observed range of 2013-18 death rates, the combination of α and β1 implies that crude all-cause mortality rose on average across counties between 2013-18 and 2020. The value of β2 suggests that, for every 100 deaths assigned to Covid-19, the number of all-cause deaths rose by 1
Slide 25: Findings
Findings. For every 100 deaths assigned to Covid-19, the number of all-cause deaths rose by 144 (95% CI, 130 to 159). 31% [95% CI, 24% to 38%] of all excess deaths were not directly assigned to Covid-19 on death certificates. 199,124 + 88,142 = 287,266 Directly Assigned Excess Deaths Not Total Exces
Slide 26: Findings
Findings. Fully Stratified Analysis: ● We examined how county-level characteristics modified the β2 coefficient (relationship between the rate of direct Covid-19 deaths and the rate of excess deaths from all causes). ● We compared counties in the upper and lower quartiles for various sociodemographi
Slide 27: Findings
Findings. Decomposition Analysis: ● We calculated mean directly assigned Covid- 19 death rates and predicted excess death rates not assigned to Covid-19 using the β2 coefficient. ● Counties with more non- Hispanic Black residents, who are already at high risk of Covid-19 death based on direct counts
Slide 28: Next Steps
Next Steps. Article: https://www.statnews.com/2021/01/25/undercounting-covid-19-deaths-greatest-in-pro-trump-areas-analysis-shows/
Slide 29: Next Steps
Deaths by death investigation system Centralized state medical examiner tecaimedies fy Mix of local medical examiner and coroner Local coroner fo) 0.5 1 1.5 2 2.5 Deaths per 1,000 person-years {0 Deaths assigned to Covid-19 Excess deaths not assigned to Covid-19
Slide 30: Limitations, Conclusions and Next Steps
Limitations, Conclusions and Next Steps.
Slide 31: Limitations
Limitations. ● Provisional all-cause and direct Covid-19 data for 2020 ● Some counties that are currently reporting lower all-cause mortality in 2020 than in the historical period likely have incomplete data. ● Another challenge was that the majority of counties in the study had fewer than 50 direct
Slide 32: Conclusions
Conclusions. ● Direct Covid-19 death counts in 2020 substantially underestimated total excess mortality attributable to Covid-19. ● Inequities in Covid-19 mortality also increased when excess deaths not assigned to Covid-19 were considered. ● Our results highlight the significant role of social dete
Slide 33: Next Steps with the MarketScan Database
Next Steps with the MarketScan Database.
Slide 34: MarketScan
MarketScan. ● In February 2021, we entered an agreement with IBM Watson for a limited edition dataset for Covid-19 research (MarketScan). ● The data provide information on medical claims from hospitals and practices across the US, including records related to billing for procedures, diagnostic codes
Slide 35: MarketScan
MarketScan. ● The contract with IBM Watson includes 5 data components to be shared with Boston University ● Respiratory subset (9 million patients, 338 GB, 42 files) ● Circulatory subset (10 million patients, 403 GB, 42 files) ● Metabolic subset (11 million patients, 400 GB, 42 files) ● Infections s
Slide 36: MarketScan
MarketScan. ● The Inpatient Admissions Table contains records that summarize information about a hospital admission. IBM Watson Health constructs this table after identifying all of the service records associated with an admission (e.g., the hospital claims, physician claims, surgeon claims, and cla
Slide 37: MarketScan
MarketScan Preliminary timeline Exposure: - Covid-19 diagnosis End of study - Covid-19 Recovery and/or (through Q3 2020) hospitalization hospital discharge or loss to follow-up - Covid-19 ICU following original due to change in admission event enrollment uh re 2 se > Cohort follow-up for: - Health s
Slide 38: Acknowledgments
Acknowledgments. ● I thank Swiss Re for the generous funding. A special thank you to Dr. Christoph Nabholz, Dr. Florian Rechfeld, and Dr. Daniel Meier for their extensive scientific and technical input.
Slide 39: Questions?
Questions?. This research is in-progress and I would be grateful for your feedback acstokes@bu.edu Link to US Excess Deaths Pre-Print: https://www.medrxiv.org/content/10.1101/2020.08.31.20184036v4
Slide 40: Technical Appendix
Technical Appendix.
Slide 41: Table 1. Characteristics of Counties in the Sample Compared to All Counties in the U.S.
Table 1. Characteristics of Counties in the Sample Compared to All Counties in the U.S. Counties in the Sample (n=787) All Counties in the U.S. (n=3142) Characteristics Mean SD Mean SD % 65 or Older 15.4% 3.8% 16.0% 4.0% % Rural 10.3% 13.9% 18.6% 24.0% % Hispanic 21.0% 17.7% 18.3% 17.2% % Non-Hispan
Slide 42: Figure S1: Flowchart Detailing Sample Exclusions
Figure S1: Flowchart Detailing Sample Exclusions Counties in the NCHS Database with 10+ Directly Coded Covid-19 Deaths as of October 17, 2020 (n=1,239) —EE——EEw v Final County Analytic Sample* (n=787) Limited to counties with more than 20 directly coded Covid-19 Deaths (n=378) Excluded states that t
Slide 43: Figure S3. Relationship Between Indirectly Age Standardized All-Cause and Direct Covid-19 Mortality
Figure S3. Relationship Between Indirectly Age Standardized All-Cause and Direct Covid-19 Mortality across Strata of Sociodemographic and Health Factors*»< Factors | : Coefficients (95% Cl) Overall D Estimate + 1.50 (1.36, 1.64) 1 ‘Age (65 or Over) s Lower 25% of Values —+— 1.51 (1.17, 1.85) Upper 2
Slide 44: Figure S4. Decomposition of 2020 Indirectly Age Standardized Excess Death Rates
Figure S4. Decomposition of 2020 Indirectly Age Standardized Excess Death Rates across Strata of Sociodemographic and Health Factors*4| Overall Age (65 or Over) Rural Non-Hispanic Black Non-Hispanic White Hispanic Median Household Income Some College or More Education Income Inequality Homeownership