Hospital quality and deaths from COVID-19 in US counties
To protect citizens from the effects of COVID-19, governments across the world have made heavy use of non-pharmaceutical interventions (NPIs), such as mask-wearing, lockdowns, and stay-at-home orders. The impact of these interventions at a local level on cases and death rates has been subject to rapid assessment (Bargain and Aminjonov 2020, Rothert et al. 2020, Cherif and Hasanov 2020, McLaren 2020). However, while such NPIs may limit the spread of the virus – as well as having an impact on the popularity of politicians (Giommoni and Loumeau 2020) – individuals who have severe COVID-19 symptoms need to be treated in a hospital. Thus, the quality of local hospital care might be expected to affect death rates. However, to date most studies of the effect of local healthcare on COVID-19 deaths have examined the association between death rates and the quantity of local resources such as protective equipment and shortages thereof, and availability of hospital and/or ICU beds (Kanter et. al. 2020, Knittel and Ozaltun 2020) or ventilators (Ranney et al. 2020, Branas et al. 2020). Much less evidence exists on the relationship between the quality of the healthcare system and death rates at the local area level.
In a recent paper we focus on this issue, analysing the relationship at the county level in the US between deaths from COVID-19 and quality of hospital care (Kunz and Propper 2020). To construct a measure of the hospital quality that is available to residents of a county, we use a measure derived from a flagship US healthcare programme that penalises hospitals for poor quality. The Hospital Readmission Reduction Program (Ody and Cutler 2019, Gupta 2017) is a nationwide pay-for-performance initiative that uses readmission rates for a small number of conditions as a measure of quality. We focus on readmission rates for patients admitted with pneumonia, as COVID-19 is a respiratory illness. We adjust the published excess (risk-adjusted) pre-COVID readmission rates to allow for their potential association with a wide set of area-level socioeconomic factors, for the excess influence of low volumes and to allow for the fact that some hospitals are never penalised whilst others always are (Kunz et al. 2020). To allow for the fact that patients cross county lines to access hospital care, we allocate to each county an average quality based on the hospitals that residents of that county use pre-pandemic, whether those hospitals lie within the county or not.
Figure 1 is at the county level for the US. Panel A shows the spatial variation in our measure of the quality of hospital care. Panel B shows the number of deaths from COVID-19 between 22 January and 28 June 2020. It is clear that there is substantial variation in both the measure of hospital quality and death rates across counties, but comparison of the two panels also shows a considerable overlap between lower hospital quality (darker shading in Panel A) and higher death rates (darker shading in Panel B).
Figure 1 County-level spatial distribution of HRR quality exposure and the cumulative number of deaths per 10,000, 22 January to 28 June 2020
Source: Kunz and Propper (2020).
Figure 2 examines this in more detail. Panel A shows the cumulative deaths from COVID-19 per 10,000 population, separating counties into those with access to high-quality and to low-quality care (i.e. above and below the median of our quality measure distribution). Death rates in counties where residents have access to lower quality hospitals diverge from those where residents have access to higher quality care around 60 days after the first case was recorded. By late June, the cumulative number of deaths was approximately twice that of counties with higher quality care. Panel B plots the association between quality (in hundred percentile bins) and deaths per 10,000 residents over the period. This shows a similar association between lower quality and higher death rates.
Figure 2 The association between quality exposure and mortality rate at the county level
While suggestive of a relationship, this correspondence could be driven by many other factors. Our regression analysis controls for a large set of these, at both the county and the hospital market (HRR) level. We find that, after controlling for this large number of possible confounders and conditioning on state fixed effects, our measure of hospital quality based on excess risk-adjusted readmissions from pneumonia admissions in the hospital referral network correlates strongly with higher death rates at the county level. We show that the measures of quality of care are not predictive of the number of COVID-19 cases after accounting for the possible county-level confounders and state fixed effects. This suggests that our findings are not driven by unobservables which may drive both the recording of cases and the recording of deaths, and that unobserved factors associated with health at the local level – which could lead to both more cases and more deaths – are not driving our findings. We also find a broader quality measure based on readmissions for the three penalised conditions covered by the HRRP programme – pneumonia, heart attack, and heart failure – is associated with deaths, but this association is less strong. This suggests that it is not overall quality of the hospital that matters as much as the quality of care for patients with respiratory diseases. We also find no association between the health of the population (measured by the population mortality rate for heart attacks, heart failure and pneumonia), hospital availability or hospital market concentration, and COVID-19 deaths. This suggests that what we isolate is a local hospital quality exposure effect.
Finally, we find important heterogeneity in the association between the quality of locally available hospital and the mortality rate. The association between health care quality and deaths from COVID depends on the share of minority populations in the county. In the highest minority areas, health care quality has no association with death rates. However, the protective impact of hospital quality increases monotonically as the share of minorities in the county population falls. It is not clear why this is the case, but our findings underline the need to focus efforts to understand the relationships between ethnicity and COVID-19 and, more generally, between access to high-quality healthcare and ethnicity.
Implications of our findings
Our results suggest that the preparedness of the healthcare system an individual is exposed to is important in the early stages of a pandemic. This echoes arguments (e.g. Hooper et al. 2020, Yancy 2020) that the early response to COVID-19 was another failed test of the US healthcare system’s capacity to equalise inequalities in society, as it was unable to prevent the emergence of racial differences in outcomes. Our results also suggest that it is important to strive for a high-quality healthcare system in non-pandemic times to insure against such shocks and support policy efforts to reward healthcare providers for the provision of higher quality.
Bargain, O and U Aminjonov (2020), “Trust and compliance with public health policies in the time of COVID-19”, VoxEU.org, 23 October.
Branas, C C, A Rundle, S Pei et al. (2020), “Flattening the curve before it flattens us: hospital critical care capacity limits and mortality from novel coronavirus (SARS-CoV2) cases in US counties”, medRxiv.
Cherif, R and F Hasanov (2020), “Universal testing: An overlooked Covid-19 policy response”, VoxEU.org, 15 August.
Engle, S, J Stromme and A Zhou (2020), “Staying at home: The mobility effects of COVID-19”, VoxEU.org, 12 May.
Giommoni, T and G Loumeau G. (2020), “Lockdown and voting behaviour: A natural experiment on postponed elections during the COVID-19 pandemic”, Covid Economics 41.
Gupta, A (2017), “Impacts of performance pay for hospitals: The readmissions reduction program”, Becker Friedman Institute for Research in Economics Working Paper 2017-07
Hooper, M W, A M Nápoles and E J Pérez-Stable (2020), “COVID-19 and racial/ethnic disparities”, JAMA.
Kanter, G P, A G Segal, and P W Groeneveld (2020), “Income disparities in access to critical care services”, Health Affairs 39(8): 1362.
Knittel, C R and B Ozaltun (2020), “What does and does not correlate with COVID-19 death rates”, NBER Working Paper 27391.
Kunz, J S and C Propper (2020), “Does Higher Hospital Quality Save Lives? The Association between COVID-19 Deaths and Hospital Quality in the USA”, CEPR Discussion Paper 15105.
Kunz, J S, C Propper, K E Staub and R Winkelmann (2020), “Assessing the Quality of Public Services: Does Hospital Competition Crowd Out the For-Profit Quality Gap?” CEPR Discussion Paper 15045
McLaren, J (2020), “Racial disparity in COVID-19 deaths: Seeking economic roots in census data”, VoxEu.org, 11 August.
Ody, C and D Cutler (2019), “The Medical Hospital Readmission Reduction Program: does it do any good?”, JAMA Internal Medicine 179(9): 1174-1175.
Ranney, M L, V Griffeth and A K Jha (2020), “Critical supply shortages—the need for ventilators and personal protective equipment during the Covid-19 pandemic”, New England Journal of Medicine 382(18): e41.
Rothert, J, R Brady and M Insler (2020), “The fragmented US: Local COVID-19 policies impact the rest of the country”, VoxEU.org, 22 September.
Yancy, C W (2020), “COVID-19 and African Americans”, Journal of the American Medical Association 323(19): 1891–1892.
1 Our county measures include the local economic situation (all ages percent poverty and median household income); measures of community health (including the population share in poor or fair health and adult smoking), population life expectancy and premature deaths; share without health insurance(access); share vaccinated (awareness of risks of contagious diseases); air pollution (argued to exacerbate risk of poor outcomes from COVID-19); commuting patterns (driving alone to work and having a long commute to capture public transportation); residential racial segregation; measures of local social capital; population measures including population density, average household size, households 65 and older living alone, share age 65 and older, shares of different education compositions, and share foreign-born. Our HHR level measures include access to local primary health care (number of admissions for ambulatory care sensitive conditions), and hospital market concentration.