CHJ and FG are joint first authors.
For people with symptomatic COVID-19, the relative risks of hospital admission, death without hospital admission and recovery without admission, and the times to those events, are not well understood. We describe how these quantities varied with individual characteristics, and through the first wave of the pandemic, in Milan, Italy.
A cohort study of 27 598 people with known COVID-19 symptom onset date in Milan, Italy, testing positive between February and June 2020 and followed up until 17 July 2020. The probabilities of different events, and the times to events, were estimated using a mixture multistate model.
The risk of death without hospital admission was higher in March and April (for non-care home residents, 6%–8% compared with 2%–3% in other months) and substantially higher for care home residents (22%–29% in March). For all groups, the probabilities of hospitalisation decreased from February to June. The probabilities of hospitalisation also increased with age, and were higher for men, substantially lower for healthcare workers and care home residents, and higher for people with comorbidities. Times to hospitalisation and confirmed recovery also decreased throughout the first wave. Combining these results with our previously developed model for events following hospitalisation, the overall symptomatic case fatality risk was 15.8% (15.4%–16.2%).
The highest risks of death before hospital admission coincided with periods of severe burden on the healthcare system in Lombardy. Outcomes for care home residents were particularly poor. Outcomes improved as the first wave waned, community healthcare resources were reinforced and testing became more widely available.
An analysis of a database of all COVID-19 cases in Milan between February and June 2020.
Uses multistate modelling to estimate relative risks of hospital admission, death without admission and recovery without admission, jointly with the times to those events, for different groups of people diagnosed with COVID-19 in the community.
The model for events following onset is combined with a previous model for analysis of outcomes following hospital admission, to enable predictions of final outcomes for people of different groups diagnosed in the community.
Changes through time in outcomes are hard to attribute confidently to causal effects of changing healthcare burden and improvements in healthcare resources, due to possible selection biases from changes through time in testing availability and policy.
There is now an extensive literature on the risks of COVID-19 hospitalisation and fatality.
The majority of studies of mortality have either estimated the risk in hospital cohorts
In Lombardy, the first person in Italy was diagnosed with COVID-19 in mid-February 2020. In Milan, confirmed cases rose from 403 on the 27 February 2020 to 22 264 at the first wave peak on 20 March 2020. This abrupt rise, not only in Milan but across Lombardy, put a large strain on the region’s healthcare system.
An integrated database was created to collate detailed pseudo-anonymised data from the LHAs’ combined symptom and laboratory surveillance, and from hospitals, on each PCR-confirmed case of COVID-19 in Lombardy. These data have been described in full elsewhere,
The dataset represents all individuals in Lombardy diagnosed with COVID-19 during the first wave from February to June 2020,
To make use of the information on relative risks of events from all individuals, including the 7% who had none of the events recoded in the data period, a multistate model is implemented. This represents how a person transitions from an ‘initial’ state representing current COVID-19 infection, to three alternative states, representing hospital admission, confirmed recovery from infection without having been hospitalised or death without having been hospitalised (
The multistate model is implemented as a ‘mixture model’,
We consider models that include the following covariates: gender, age group (three categories: age 45 and below, age 46–65, and age 66 and above), month of onset (five categories: February, March, April, May and June), whether the individual is a healthcare worker, or had any comorbidities, or is a care home resident. Covariates were considered to affect both the probability of each event occurring (through multinomial logistic regression) and the expected time to the event (through a generalised gamma accelerated failure time model). Full details of the regression model specification are given in the
The multistate model in the
Although patients were not directly involved in the study design, the experiences of clinicians and public health officials interacting with patients were crucial to the design of the data collection.
The 27 598 people in the cohort had a median age of 64 years (IQR 48–82), 45% were male and 55% were female, 13% were healthcare workers, 21% were care home residents and 47% had at least one comorbidity. Overall, 43% of people were admitted to hospital, while 5% died without being admitted, 45% recovered without being admitted and 7% had none of these events recorded.
As estimates were substantively similar whether or not missing outcomes were assumed to represent censoring (
There were only 28 deaths (<1%) observed from people under 65 years, so we report mortality results for the 65+ years age group only. The probability of death without hospital admission (
Probability of death before hospital admission by month of COVID-19 onset, gender, and whether a person is a care home resident or had comorbidities. Only considers people over 65 years of age who were not healthcare workers. Estimates and 95% CIs.
The probabilities of hospitalisation for various subgroups are compared in
Probability of hospital admission following COVID-19 onset, by month of onset, age group and gender, for care home residents, people with comorbidities and healthcare workers, compared with a baseline individual with none of these risk factors. Estimates and 95% CIs.
Times from COVID-19 onset to hospital admission, to death before hospital admission and to recovery before hospital admission (median and range containing 95% of individuals), by month of onset, age and gender, for people without comorbidities who were neither care home residents nor healthcare workers. Estimates and intervals including 95% of individuals.
Combining the hospital model of Boëlle
Estimated symptomatic case fatality risk with 95% CIs, by gender, age and month of symptom onset.
The probabilities of taking each pathway, by gender, age group and month are displayed in
Estimated probability (95% CI) of each pathway defined by sequences of different events, by gender, age and calendar month of symptom onset.
Without taking covariates into account, the median time to death from symptom onset, averaged over pathway, is 14.5 days (95% CI 13.3 to 14.8), but there is substantial variation between individuals, with a 95% quantile interval of 1.6–65.0 days. The distribution of time to confirmed recovery from symptom onset, averaged over pathway, is also highly heterogeneous, with median 34.5 days (95% CI 33.9 to 35.3) and 95% of individuals having times between 4.6 and 107 days.
Further summaries of the distributions of times from symptom onset to final events, overall and by pathway, are given in the
Our analysis has resulted in three key findings concerning risks of severe events in the community following symptom onset among confirmed COVID-19 cases in Milan: (1) groups at elevated risk of death without hospital admission include older age groups, men, care home residents and those with comorbidities, whereas those at higher risk of hospital admission include older age groups, men and those with comorbidities, with care home residents and healthcare workers having substantially lower risk than others; (2) risks of hospital admission and death without hospital admission peaked early in the first wave; (3) these risks steadily decreased with month of symptom onset after their peak.
The first result confirms previous findings of which groups of people with COVID-19 are most at risk of severe events.
Hospital admission risk peaked for cases with onset in February, while non-hospital death risk peaked for cases with onset in March for care home residents, and April for all others, while the overall sCFR peaked in February and March. These results suggest that the epidemic spread initially in the general population from February, then spread to care homes only from March and April onwards.
Following these peak months, risks of severe events steadily declined. The decreasing trend in probabilities of hospital admission among all groups and increasing probability of confirmed recovery without needing hospital care may reflect a combination of factors. The LHAs’ roles in the pandemic response may have been important. With the increased resources assigned to these providers by Regione Lombardia over the first few months, as well as the decreasing number of confirmed cases as the first wave waned, it has been noted that the Lombardy health system had greater ability to diagnose and treat patients with COVID-19.
As noted above, care home residents were at a very high estimated risk of death without admission and low risk of hospital admission following COVID-19 onset in March–April. Furthermore, as estimated by Presanis
To obtain estimates of absolute risks from censored data, for combinations of risk factors, we used statistical models with parametric assumptions. While the substantive findings are not affected, the exact values of some of our risk estimates may be sensitive to these assumptions for particular combinations of risk factors for which the data were weak, as detailed in the
The trends we have estimated in the risks of outcomes following symptom onset may also be affected by what has been termed ‘epidemic phase bias’.
Despite the constraints of interpretation due to testing availability and policy and its change over time, we have nevertheless illustrated important estimates of sCFR and risks of hospital admission and death in the community, and their changes through time. While our study is based on a population without immunity, infected with the original strain of SARS-CoV-2, our estimates still contribute to the overall picture of how outcomes changed through the pandemic, and might still be used in models that combine different sources of evidence on absolute and relative risks, to inform future policymaking.
@cjackstats
DC, DDA and AP designed the study. FG and AP searched the literature. MG, DC, FG and MT collected the data. CHJ, AP and KK performed the statistical modelling. DC, FG and MT contributed to interpreting the results. CHJ, AP and FG wrote the original draft of the manuscript. AC, MG, MT, SC and AC contributed to revisions of the manuscript. CHJ is the guarantor, accepting full responsibility for the work and publication.
This work was supported by the Medical Research Council, programme number MRC_MC_UU_00002/11.
None declared.
Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Not commissioned; externally peer reviewed.
This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
Data are not publicly available. This study is based on a database maintained by the Prevention Unit of the General Directorate of Welfare of Regione Lombardia. The authors can forward inquiries relating to use of this database.
Not required.
Data collection and analysis were part of outbreak investigations during a public health emergency. Processing of COVID-19 data is necessary for reasons of public interest in the area of public health, such as protecting against serious cross-border threats to health or ensuring high standards of quality and safety of healthcare, and therefore is exempted from institutional review board approval (regulation EU 2016/679 GDPR). The use of the data for this retrospective cohort study was approved by the COVID-19 Research Committee of Regione Lombardia. The research question, design and data collection were motivated by the public health emergency response to the COVID-19 epidemic early in March 2020. Regione Lombardia collected data during the emergency using the existing information system for infectious diseases, regulated by a 2004 regional decree, that was then updated in May 2020 to regulate the COVID-19 data flows. Data are pseudo-anonymised. Informed consent from all subjects and legal guardians of minor patients for this data collection at patient level is given prior to the data collection in hospitals and laboratories, which send the data to Regione Lombardia. Although patients were not directly involved in the study design, the experiences of clinicians and public health officials interacting with patients were crucial to the design of the data collection.