The spread of infectious diseases, caused by viruses such as the corona, influenza, noro or respiratory syncytial virus (RSV), is a feature of human civilization. As most recently evidenced by the COVID-19 pandemic, infection waves can create widespread disruption to daily life around the globe and pose a major burden for healthcare systems, with the possibility of large secondary effects (Moss et al., 2020). Most existing research has primarily focused on quantifying the direct health and economic burden that waves of infectious diseases exert on hospitals and healthcare systems (see e.g. de Courville et al. (2022)). However, it has typically not been possible to identify and measure the extent of any indirect spillover effects within the healthcare system. Spillover effects arise when infectious disease waves reduce the ability of the healthcare system to provide high-quality care for patients seeking medical help for reasons unrelated to the infectious disease itself. It is an empirical question whether such spillover effects exist and how far-reaching they are. This paper presents ample evidence suggesting that the COVID-19 pandemic produced negative spillover effects in English National Health Service (henceforth, NHS) hospitals and that these were a significant driver of non-COVID-19 excess mortality.

Past research has been unable to quantify spillovers from seasonal infectious diseases mostly due to data limitations. Being able to measure, for example, whether an influenza outbreak is affecting the quality of care for non-influenza patients necessitates that all patients are routinely tested for influenza upon admission and during their hospital stay. Further, qualifying the extent to which a hospital faces pressures due to an unusually pronounced seasonal disease, possibly a new virus strain, requires a good understanding of the population prevalence. For most seasonal diseases, no population testing programs exist to establish this. Lastly, in most instances, it is hard to track health outcomes of patients whose quality of care may have been compromised by the impact of seasonal diseases. Therefore, it is often simply not possible to determine to what extent health outcomes among patients seeking medical care for unrelated reasons may have been adversely affected by a seasonal disease.

The COVID-19 pandemic provides a unique natural experiment that can help to cast light on the extent to which infectious disease waves may cause such spillover effects. Many of the data constraints that previously made this kind of quantification exercise impossible have been relaxed. In this paper, we leverage data that is derived from the population of all hospital episodes within the English NHS during the first year of the pandemic.1 This individual-level data has been linked to the population of COVID-19 testing data, allowing those patients in hospital with a COVID-19 diagnosis to be distinguished from those without a COVID-19 diagnosis. This is possible because the NHS adopted a standard operating procedure from the start of the pandemic whereby all hospital patients were routinely screened for COVID-19 upon admission and during their stay. Therefore, we can study the spillover effects of COVID-19 pressures measured at the healthcare provider level by analysing health outcomes among patients that were in hospital for reasons unrelated to COVID-19. In particular, we focus on excess mortality among non-COVID-19 patients.

For each hospital episode, the NHS uses a statistical model to predict the probability of the admitted patient’s death in the time window between admission and the 30th day after discharge. The mortality risk model is trained on historic patient-level NHS data, which includes a broad range of patient characteristics across all NHS providers. Since the statistical model is trained off historic data mostly from before the pandemic, it captures the expected mortality risk of a patient based on their individual characteristics and medical diagnosis assuming pre-pandemic normal operating circumstances across hospitals. We use the output from the NHS’ mortality prediction model to construct a measure of the difference between the actually observed deaths and the expected number of deaths. Crucially, the data excludes all patients with a positive COVID-19 test in the time between their admission to hospital and the 30th day after discharge as well as all deaths that mention a COVID-19 infection as a cause or contributing factor on the death certificate. This ensures that our analysis is not confounded by COVID-19 patients and that we can focus fully on consequences of the pandemic for patients that were seeking medical help for reasons unrelated to COVID-19.

Fig. 1 presents the overall time-series patterns in our measure of non-COVID-19 excess deaths among patients who were in hospital for reasons unrelated to COVID-19 and whose deaths are not linked in any way to a COVID-19 infection. Our estimates capture the evolution of excess deaths up until February 2021 across NHS providers. On average, prior to the pandemic, the excess mortality measure is approximately centered around zero, suggesting that the individual-level mortality risk model has good out-of-sample predictive power. With the start of the COVID-19 pandemic in March 2020, there is a sharp upwards jump in the excess deaths measure to between 420 and 550 excess deaths in each of the first three months of the pandemic among patients in hospital for reasons unrelated to COVID-19. In the following months, there continue to appear systematic deviations in observed deaths from expected deaths, with observed deaths being significantly higher than expected deaths among non-COVID-19 patients. This suggests that the individual-level patient mortality risk estimate trained off historic data may produce expected mortality estimates that are downward biased due to a significant omitted variable: the impact of the pandemic on the ability of the healthcare system to provide high-quality care to non-COVID-19 patients.

Taking the sum of the red dots in Fig. 1, we estimate that, for the first twelve months of the pandemic from March 2020 to February 2021 alone, there were at least 3,058 (with a 90% confidence interval of [2,572, 3,543]) excess deaths of non-COVID-19 hospital patients in England who, if it were not for the pandemic disruptions, may not have died. This number stands significant in the context of COVID-19 deaths in the population: for every 42 deaths which mention COVID-19 on the death certificate, there was one excess death of a non-COVID-19 patient in hospital. Our estimates also indicate that the 3,058 excess deaths of non-COVID-19 patients between March 2020 and February 2021 make up a non-negligible 3.0% of all excess deaths in the population.

We document that a key omitted variable that can explain the systematic variation in non-COVID-19 excess mortality across healthcare providers is the extent to which different hospitals were exposed to pressures arising from COVID-19. We find that the number of excess deaths among non-COVID-19 patients rises sharply with the number of hospitalized COVID-19 patients in a given month. For every 100 new COVID-19 admissions, there are an additional 1.3 to 1.8 non-COVID-19 excess deaths among patients seeking medical help for reasons unrelated to COVID-19 in a month. Further, we find significant heterogeneity in these effects: the spillover effects from COVID-19 affecting excess mortality among non-COVID-19 hospital patients are increasing in hospitals serving catchment areas which are larger, have a higher share of ethnic minorities and have a lower share of old people. This result could reflect the fact that, prior to the pandemic, areas that structurally had higher levels of demand for healthcare (due to having an older, less healthy population) also had, on the margin, higher levels of resources devoted to them (Barr et al., 2014), potentially enabling them to cope better with COVID-19 disruptions. It also suggests that the indirect health burden of COVID-19 crowding out care may have been borne especially by younger and ethnic minority populations.

Our research contributes to a wider literature on the health-related effects and economic consequences of seasonal diseases. Much of this work has focused on studying the impact of influenza (see e.g. Bellia et al., 2013, de Courville et al., 2022, He et al., 2023). For seasonal diseases like influenza, data is often severely limited, making it impossible to distinguish between direct and indirect impacts of the disease. However, in this paper, we can actually study spillover effects on the health outcomes of non-COVID-19 patients by virtue of widespread COVID-19 testing in the population and in hospitals, and the centralised collection of rich patient-level data in the NHS. This provides a vital estimate of one aspect of the significant burden COVID-19 imposed on the healthcare system: excess mortality among non-COVID-19 patients. Our results are relevant to the issue of seasonal diseases more broadly and can contribute to debates on optimal healthcare policy such as, for example, vaccination provision. These debates are regularly held around seasonal waves of the flu, with those opposing vaccination mandates pointing to the lack of rigorous evidence of spillover effects (see e.g. Tilburt et al., 2008, Prematunge et al., 2012). We argue that, although the strain under which healthcare systems came during the first waves of the COVID-19 pandemic (de Oliveira Andrade, 2020, Mahase, 2021) was extraordinary, some parallels can, for example, be drawn to what may be expected in the case of a new potent influenza strain, with intensive-care units (ICU) and healthcare workers being forced to work at and above capacity (Mehta et al., 2021), raising concerns about the quality of healthcare that patients can receive (Mira et al., 2020).

Our paper also relates to previous work on attempting to measure the death toll that arose from the pandemic. This literature, similarly to that on other seasonal diseases, has typically also not been able to decompose excess deaths into the underlying drivers. Rather, it provides an aggregate estimate of excess mortality. For example, for India (Adam, 2022, Jha et al., 2020), the UK (Laliotis et al., 2023) and the US (Ruhm, 2021), numerous studies estimate that the true number of COVID-19 deaths may be notably larger than initially reported. Cronin and Evans (2021) also show for the US that non-COVID-19 excess mortality increased particularly among Black men. By quantifying excess deaths of patients seeking healthcare for reasons unrelated to COVID-19, this paper is able to document that there are notable negative spillovers to non-COVID-19-related care. This approach, combined with the administrative data we use, provides a quantification of spillover effects that are widely discussed around other seasonal diseases as well. We are able to provide a lower bound estimate on the likely number of deaths that may have been caused by the deterioration of care that patients could receive in hospital under COVID-19 stress. Our approach contrasts with existing work on excess deaths which relies on modelling studies of the likely increases, e.g. due to undetected or delayed treatment of cardiovascular diseases (Banerjee et al., 2021), cancer (Lai et al., 2020b), or lacking access to insurance (Galvani et al., 2022).

Finally, we contribute to an emerging literature which uses forecast errors for causal inference (Mueller and Rauh, 2024, Fetzer and Yotzov, 2023, Valente et al., 2023). As prediction models become more accurate, thanks to increasing computing power and the availability of better data, forecasts serve as benchmarks against which to compare actual outcomes. By providing a counterfactual scenario, they play a crucial role in detecting abnormal periods, especially when the treatment is omitted from the prediction model. In Fig. 1, we observe that, before the onset of the COVID-19 pandemic, the NHS prediction model exhibited errors that were centered around zero, indicating a good model fit. However, we can attribute the substantial and systematic forecast errors during the COVID-19 pandemic to the heightened pressures associated with the treatment of COVID-19 patients.

The rest of this paper is structured as follows. Section 2 describes our data and how we measure non-COVID-19 excess mortality and provider-level exposure to COVID-19. Section 3 details our empirical strategy for analysing whether COVID-19 pressures are a driver of non-COVID-19 excess mortality, as well as for investigating heterogeneity across different diagnoses and trust population characteristics. Section 4 presents our results, and Section 5 provides a discussion.