Changes in Cause-of-Death Attribution During the Covid-19 Pandemic: Association with Hospital Quality Metrics and Implications for Future Research

Abstract

AbstractBackgroundSevere acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is often comorbid with conditions subject to quality metrics (QM) used for hospital performance assessment and rate-setting. Although diagnostic coding change in response to financial incentives is well documented, no study has examined the association of QM with SARS-CoV-2 cause-of-death attribution (CODA). Calculations of excess all-cause deaths overlook the importance of accurate CODA and of distinguishing policy-related from virus-related mortality.ObjectiveExamine CODA, overall and for QM and non-QM diagnoses, in 3 pandemic periods: awareness (January 19-March 14), height (March 15-May 16), and late (May 17-June 20).MethodsRetrospective analysis of publicly available national weekly COD data, adjusted for population growth and reporting lags, October 2014-June 20, 2020. CODA in 5 pre-pandemic influenza seasons was compared with 2019-20. Suitability of the data to distinguish policy-related from virus-related effects was assessed.ResultsFollowing federal guidance permitting SARS-CoV-2 CODA without laboratory testing, mortality from the QM diagnoses cancer and chronic lower respiratory disease declined steadily relative to prior-season means, reaching 4.4% less and 12.1% less, respectively, in late pandemic. Deaths for non-QM diagnoses increased, by 21.0% for Alzheimer’s disease and 29.0% for diabetes during pandemic height. Increases in competing CODs over historical experience, suggesting SARS-CoV-2 underreporting, more than offset declines during pandemic height. However, in the late-pandemic period, declines slightly numerically exceeded increases, suggesting SARS-CoV-2 overreporting. In pandemic-height and late-pandemic periods, respectively, only 83.5% and 69.7% of increases in all-cause deaths were explained by changes in the reported CODs, including SARS-CoV-2, preventing assessment of policy-related mortality or of factors contributing to increased all-cause deaths.ConclusionsSubstitution of SARS-CoV-2 for competing CODs may have occurred, particularly for QM diagnoses and late in the pandemic. Continued monitoring of these trends, qualitative research on pandemic CODA, and the addition of place-of-death data and psychiatric CODs to the file would facilitate assessment of policy-related and virus-related effects on mortality.Ascertainment of the number of deaths from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is foundational to understanding the severity, scope, and spread of the infection. Despite its importance, estimation of SARS-CoV-2 deaths is challenging because advanced age, genetic polymorphisms, and obesity-related comorbidities that predispose to inflammatory states increase the likelihood of dysregulated immunological function, severe respiratory distress, and mortality from infectious respiratory illness.1,2 These host factors represent competing potential causes of death (COD). For example, 98.8% of Italy’s SARS-CoV-2 deaths occurred in persons with >1 comorbidity, 48.6% with >3 comorbidities, and median decedent age was 80 years.3 Similarly, of U.S. SARS-CoV-2 deaths reported as of May 28, 2020, 93% involved other CODs (mean 2.5 additional causes), and 60% occurred in persons aged >75 years.4 This pattern of multiple contributing CODs is common in respiratory infection-related mortality.5In death certificate issuance during the pandemic, methods to account for this pattern varied, as no single standard for SARS-CoV-2-attributable death exists. In Italy, all deaths in patients testing positive for SARS-CoV-2 were attributed to the infection despite high prevalence rates for comorbid conditions, measured in early deaths: ischemic heart disease (30%), diabetes (36%), cancer (20%), and atrial fibrillation (25%).6 The U.S. National Center for Health Statistics (NCHS) issued death-certification guidance on March 4, 2020, indicating that SARS-CoV-2 should be reported if “the disease caused or is assumed to have caused or contributed to death.”7 Follow-up guidance issued on April 3 indicated that it was “acceptable to report COVID-19 on a death certificate without [laboratory test] confirmation” if circumstances indicating likely infection were “compelling within a reasonable degree of certainty.”8This nonspecific guidance should be interpreted in light of previous research findings that COD attribution (CODA) errors are common on death certificates, particularly in infectious disease and septic shock.9,10 In one survey of New York City (NYC) resident physicians in 2010, 49% indicated they had knowingly reported an inaccurate COD on one or more certificates, often (54%) at the behest of hospital staff, and 70% reported they had at least once been unable to report septic shock “as an accepted cause of death” and had been “forced to list an alternate cause.”9 In an audit of NYC data from 2010-2014, 67% of pneumonia death certificates contained >1 error, compared with 46% for cancer and 32% for diabetes.10Such CODA ambiguities are often addressed by calculating “excess deaths,” defined as all-cause deaths exceeding those projected from historical experience.11 This method, recently used to estimate that official tallies of SARS-CoV-2 deaths represented only about 66%-78% of the disease’s true mortality impact,12,13 is potentially advantageous in estimating SARS-CoV-2 impact by accounting for deaths that may not have been explicitly coded as infection-related.14 Examples include deaths from cardiac events to which undetected SARS-CoV-2 may have contributed15 or out-of-hospital deaths occurring without medical care because of health-system overcrowding.16 Despite these advantages, the method is compromised by 3 considerations when applied to SARS-CoV-2 CODA, which should be quantified to inform future policy.First, the method should distinguish natural from societal causes to account for possible consequences of policy decisions and fears that, although prompted by anticipated effects of SARS-CoV-2, were not direct or inevitable viral sequelae. Examples include suicides from stay-at-home order-related labor market contraction17 and social isolation,18 increases in domestic violence,19 overdoses due to interruptions in substance use disorder treatment,20 and delays in emergency care for life-threatening conditions21–23 in geographic areas where health-system overcrowding was expected but not realized.24,25 To promote evidence-based public health policy, population-level disease-mitigation strategies that go beyond traditional practices of isolating the sick and quarantining those exposed to disease merit empirical investigation.26,27Second, the method should reflect the effects that financial incentives around hospital quality metrics (QM), which are commonly associated with provider coding practices, may have on CODA.28–30 For example, in United Kingdom hospitals, increases in coding for palliative-care admissions produced a severity-adjusted mortality-rate decline of 50% over the 5-year period ending in 2009, while the crude death rate remained unchanged.31 Although we are not aware of studies linking QMs to CODA, it is known that CODA errors are more likely to occur in hospitals than elsewhere,32 with an 85% error rate reported in comparisons of death certificates with autopsy findings at one regional academic institution.33 The potential effect of financial incentives on CODA is particularly important for SARS-CoV-2 because several competing CODs, including chronic lower respiratory disease (CLRD), acute myocardial infarction, heart failure, pneumonia, and stroke, are included in Medicare 30-day mortality measures used to calculate prospective payment rates.34 All but one of these (CLRD) is included in Agency for Healthcare Research and Quality inpatient quality indicators.35 Sepsis and cancer, other competing causes of death, are also the target of QM.36–38 Although not affecting all-cause death counts, the incentive to substitute SARS-CoV-2 for another COD could affect the accuracy of the SARS-CoV-2-attributed count.Third, the method should account for baseline life expectancies among those whose deaths were reported as caused by SARS-CoV-2. For example, at age 80 years, the 1-year probability of death is 5.8% for males and 4.3% for females, higher in those with cardiovascular comorbidities.39,40 In that age group, the population-level risk of a SARS-CoV-2 death in New York City, a pandemic epicenter, was 1.5% in about 3 pandemic months through June 17, 2020.41 Thus, deaths from competing CODs would be expected to decline late in the pandemic and in subsequent months. From a policy perspective, quantifying this effect is consistent with the quality-adjusted life year approach in evidence-based medicine, which considers future life expectancy in assessing the effects of disease and disease-mitigation interventions.42To permit assessments of SARS-CoV-2-related mortality, publicly available NCHS data include weekly aggregated totals for all-cause deaths, natural-cause deaths, and selected categories of CODs, reported as final data for 2014-2018 and provisional data for 2019-2020.4 These data, which are updated weekly, have important limitations. First, International Classification of Diseases (ICD)-10 diagnosis codes are grouped into broad categories, rather than the individual ICD-10 codes available in full COD files (Appendix 1). Second, only 11 selected diagnostic categories are reported. Third, although 63% of deaths are reported within 10 days, reporting lags vary by state.4 Reporting delays for injurious deaths are greater because they require investigation (e.g., forensic toxicology).43 Pending investigation, these deaths are often assigned ICD-10 code R99, “ill-defined and unknown cause of mortality.”43In this exploratory study, we used these files to provide preliminary evidence on the following: (1) change in CODA compared with historical experience; (2) association of CODA with QM; and (3) suitability of the files to distinguish policy-related from virus-related effects. All analyses were adjusted for population and reporting lags and based on comparisons of 2020 with equivalent weeks in the 5 most recent years. We hypothesized that if substitution of SARS-CoV-2 for alternative CODs occurred, death counts for competing diagnoses would decline relative to historical experience during the pandemic, especially after issuance of the NCHS death-certification guidance; these declines would be greater for QM than for other conditions; and they would accelerate late in the pandemic as earlier SARS-CoV-2 deaths offset later deaths from competing causes.