key: cord-0697392-7hers03k
authors: Rosenbaum, J. E.; Stillo, M.; Graves, N.; Rivera, R.
title: Timeliness of U.S. mortality data releases during the COVID-19 pandemic: delays are associated with electronic death registration system and elevated weekly mortality
date: 2021-01-08
journal: nan
DOI: 10.1101/2021.01.07.21249401
sha: 52d94283bf382f9aca3f597c30a93eb3a1fa61e5
doc_id: 697392
cord_uid: 7hers03k

All-cause mortality counts allow public health authorities to identify populations experiencing excess deaths from pandemics, natural disasters, and other emergencies. Further, delays in mortality reporting may contribute to misinformation because death counts take weeks to become accurate. We estimate the timeliness of all-cause mortality releases during the Covid-19 pandemic, and identify potential reasons for reporting delays, using 35 weeks of provisional mortality counts between April 3 and December 4, 2020 for 52 states/jurisdictions. On average, states' mortality counts are delayed by 5.6 weeks (standard deviation 1.74), with a range of 8.8 weeks between the fastest state and the slowest state. States that hadn't adopted the electronic death registration system were about 4 weeks slower, and 100 additional weekly deaths per million were associated with 0.4 weeks delays, but the residual standard deviation was 0.9 weeks, suggesting other sources of delay. Disaster planning should include improving the timeliness of mortality data.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint

During natural disasters, pandemics, and other emergencies, policymakers can use estimates of excess mortality to assess the number of deaths that resulted from the emergency and identify populations at continuing risk. During the Covid-19 pandemic, the U.S. Centers for Disease

Control and Prevention has found that excess mortality exceeds the official Covid-19 mortality count (1) , which may be due to underdiagnosed Covid-19 due to low test access, atypical Covid-19 presentation, reduced healthcare seeking for acute non-Covid-19 conditions (2), or etiologically nonspecific death reporting (3) . Despite excess deaths, the infodemic accompanying Covid-19 included misinformation that Covid-19 mortality was exaggerated (4) (5), and belief in misinformation is cross-sectionally associated with adherence to Covid-19 prevention measures such as mask-wearing (6).

Accurate and timely estimation of excess mortality allows policymakers and clinicians to formulate appropriate responses to mitigate excess mortality, such as providing appropriate guidance about seeking health care for acute conditions and encouraging the population to adhere to non-pharmaceutical interventions such as social distancing and wearing masks (2) .

Public health statisticians often estimate excess mortality from weekly provisional all-cause mortality data from the Mortality Surveillance System, which exclude deaths not yet reported and are updated in successive weekly releases (7) . States vary in the timeliness of death reporting varies in part because states vary in the extent of adoption of the Electronic Death Registration System (8) . States have improved the timeliness of death reporting in recent years: within 13 weeks, all-cause deaths were 84% complete in 2015 (7) and 95% complete in 2017 (9). This . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint study estimates delays in all-count mortality counts for each state and estimates improvements in timeliness associated with the extent of electronic death registration adoption.

Using the Mortality Surveillance System, we archived 35 weeks of provisional mortality counts by state between April 3 and December 4, 2020. The National Center for Health Statistics stratified the provisional counts into 52 jurisdictions: all 50 states, with New York City (NYC) and non-NYC New York State separated, and the District of Columbia included.

We used adoption of the electronic death registration prior to the starting point of this data as the primary predictor variable. Four states did not use electronic death registration (CT, NC, RI, WV), 9 states and New York City (NYC) had fewer than 75% death certificates filed with electronic death registration (AR, CO, MD, MI, MS, NY, PA, TN, VA), and the 38 remaining jurisdictions (37 states and DC) filed more than 75% of death certificates with electronic death registration (8) . We don't have data about the percentage of death certificates filed electronically, but it is reasonable to believe that the closer jurisdictions get to 100% of death certificates filed electronically, the smaller the delay. We confirmed electronic death registration implementation with each state's public health vital statistics website.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint

We hypothesized that weeks with more all-cause deaths would have greater delays, due to the resources needed for processing additional deaths, so we tested whether total deaths or deaths per million were associated with delay.

We hypothesized that states with more economic resources would have faster death certificate processing. We used the Bureau of Economic Analysis's 2018 per capita GDP for the 50 states and the District of Columbia from 2018 as a measure of economic resources; although New York State's delay excludes NYC death certificates, the tax base of New York State includes NYC.

To assess whether this delay measure (weeks of delay) is associated with a prior measure of data completeness, we used a 2017 measure of the percent of death certificates available within 13 weeks as a covariate (9).

We estimated the mortality reporting delay for each of the 23 weeks from April 3-September 5, 2020, based on the weekly provisional mortality releases April 3-December 4, 2020. We chose the ending date 12 weeks before the most recent data release. We estimated the mortality reporting delay as the number of provisional releases until the increases in provisional mortality count were less than 1% for all subsequent weekly releases. All 52 jurisdictions in the data achieved this standard, yielding 1196 mortality delay observations from 52 jurisdictions. We assessed the face validity of these mortality reporting delay estimates by comparison with a . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. We used linear regression with weeks of delay as the outcome variable, with varying intercept by states; we plotted these varying intercepts for the null model ( Figure 1) . We used the linear model because it provided more interpretable results than Poisson models because intercepts represented weeks of delay. We judged that the residuals appeared homoskedastic and a quantilequantile plot of the residuals appeared to fit the normal distribution with few outliers.

We estimated the delay due to paper-based systems using fixed slope and varying intercept regression models: the first model used only a binary indicator for no electronic death registration adoption and the second used a categorical variable for no adoption, less than 75% adoption, and more than 75% adoption. Both models' results yielded similar answers.

We also fitted models controlling for state gross domestic product (GDP) per capita, population, and the 2017 completeness measure, but these variables did not improve the fit of the model according to a log-likelihood test. We included mortality per million population per year as a covariate.

This study is an analysis of publicly available data from United States federal sources in broad categories such that individuals cannot be identified, so it is not human subjects research and is exempt from requiring human subjects board review. We have made the raw data and code publicly available.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint All analyses were performed in R 4.0.3 between April and December 2020.

On average, we found that all-cause mortality counts take 5.6 weeks to become accurate with less than 1% increases subsequently. We display the average number of weeks of delay for all 52 jurisdictions in Figure 1 Table 1 shows the regression results predicting delay in mortality reporting with varying intercept by state. Compared with full electronic death registration adoption (greater than 75% of death certificates reported electronically), states without electronic death registration adoption had a 4.1 week delay in reporting all-cause . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint mortality counts on average (4.13, 95% confidence interval (CI) (2.23, 6.03)). States with partial electronic death registration did not differ from states with full electronic death registration adoption in mortality reporting delay (-0.90, 95% CI (-2.18, 0.39)). States with greater mortality had greater delay in reporting mortality; each week's delay in mortality reporting was associated with that week's mortality; each additional 100 weekly deaths per million population per year was associated with 0.69 more weeks of delay (95% CI (0.58, 0.80)). We found that electronic death registration modified the effect of deaths: greater deaths are associated with more delays for states that didn't adopt electronic death registration. States with greater per capita GDP, allcause mortality completeness within 13 weeks in 2017, or population did not have greater delay, based on likelihood ratio tests of nested models that include these variables.

We found that mortality reporting delay is associated with greater mortality, suggesting that states take longer to report accurate mortality counts during times of excess death, when these mortality counts are most needed. Policy makers use all-cause mortality for estimating deaths due to natural disasters and health emergencies when deaths may not be coded accurately.

Delays in reporting mortality result in provisional counts lower than actual mortality. In the case of the Covid-19 pandemic, pandemic misinformation about mortality could use low provisional mortality counts (4, 5) , and other studies found that pandemic misinformation reduced public adherence to non-pharmaceutical interventions (6).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint These mortality reporting delays are not attributable to state resources: high-resource states are no faster than low-resource states. The three slowest states, North Carolina, Connecticut, and Alaska, are the 33rd, 4th, and 8th richest states, and the three fastest states, Maine, Vermont, and New Hampshire, are the 43rd, 36th, and 18th richest states.

This measure of completeness captures delays not captured by previous measures of completeness. Alaska is considered to be a full adopter of electronic death registration (8) with 95% completeness within 13 weeks in 2017 (9), but Alaska was among the slowest states by our measure of number of weeks of delay.

Connecticut and North Carolina began to pilot electronic systems respectively in July 2020 (10) and October 2020 (11). However, our results suggest that substantial delays in all-cause death counts occur even in states that fully implemented electronic death registration. Further, Connecticut's delays decreased in mid-May when mortality decreased, rather than in July when the electronic system began implementation; among Connecticut's 5 weeks with the largest delays (12+ weeks), 4 weeks were also the highest mortality weeks.

As suggested after earlier pandemics (12), policy makers need to increase resources to improve the timeliness of mortality data as part of pandemic planning. Improving mortality data timeliness will also benefit natural disaster planning, when excess deaths can be used for mortality estimation. The vital statistics infrastructure is under-funded (13). State and federal . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint pandemic planners should seek resolution for delays in mortality reporting so that all-cause deaths can be used to estimate excess deaths to identify areas and populations in need of additional intervention.

Funeral directors, who enter demographic information, adopted electronic death registration quickly, but medical examiners have lagged (8) . California and Arizona allowed electronic death registration submissions by fax machine (8) , and our analysis found that these states were both faster than average. States that consider unconventional approaches for electronic death registration submission that meet the needs of all stakeholders may have similar success.

The CDC includes percent completeness metrics in the Mortality Surveillance System, defined as the number of deaths divided by the average number of deaths from the most recent 4 years.

This completeness measure cannot measure completeness accurately during a period of excess deaths, which is when these measures are most crucial and subject to the most public scrutiny.

Data completeness measures that can remain accurate during periods of high mortality may reduce misinformation, such as claims that mortality counts are exaggerated.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint Ref.

Weekly deaths per million 0.00069 0.001 0.6 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint

Excess Deaths Associated with COVID-19 Provisional Death Counts for Coronavirus Disease (COVID-19)

Excess mortality in the United States during the first three months of the COVID-19 pandemic

The Importance of Proper Death Certification During the COVID-19

COVID-19-Related Infodemic and Its Impact on Public Health: A Global Social Media Analysis

Who do you trust? The digital destruction of shared situational awareness and the COVID-19 infodemic

Misinformation with Face Mask Wearing and Social Distancing in a Nationally Representative US Sample

Timeliness of Death Certificate Data for Mortality Surveillance and Provisional Estimates

National Vital Statistics System: Vital Statistics Rapid Release)

National Vital Statistics System: transitioning into the 21st century

It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint Figure 1 : Weeks of delay for reporting all-cause mortality within 1% for April 3-September 5, 2020, centered at the mean.. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted January 8, 2021. Positive values represent slower than average and negative values represent faster.. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted January 8, 2021. ; https://doi.org/10.1101/2021.01.07.21249401 doi: medRxiv preprint