Pressure on the Health-Care System and Intensive Care Utilization During the COVID-19 Outbreak in the Lombardy Region of Italy: A Retrospective Observational Study in 43,538 Hospitalized Patients

Abstract

Abbreviations

 
• CI

  confidence interval

 
• COVID-19

  coronavirus disease 2019

 
• ICU

  intensive care unit

 
• IRR

  incidence rate ratio

 
• OR

  odds ratio

 
• SARS-CoV-2

  severe acute respiratory syndrome coronavirus 2

Italy was the first country in the Western Hemisphere to be affected by a widespread epidemic of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection (1, 2), and it still has one of the highest cumulative burdens of coronavirus disease 2019 (COVID-19) hospitalization and death worldwide (3). Lombardy, in particular, was the first-hit and by far the hardest-hit region in Italy, accounting alone for over half of COVID-19 hospital admissions in Italy (4) despite having about one sixth of the country’s population. The explosive spread of SARS-CoV-2 in the region, coupled with the high COVID-19 morbidity, threatened to collapse even one of the most advanced health systems in the country and resulted in the rapid adoption of unprecedented control measures. Despite a rate of critical-care beds per inhabitant above the European average (5) and the drastic actions taken, culminating in a national lockdown on March 11, 2020, hospitals in Lombardy were put under severe pressure. By mid-March 2020, the bed occupancy due to COVID-19 in intensive care units (ICUs) in the region exceeded the precrisis total capacity of about 720 beds (6). The rapid saturation of hospital capacity was predicted by mathematical models in the early phase of the epidemic (7), prompting the emergency expansion of ICU and hospital beds dedicated to COVID-19 (8, 9), similarly to what was previously experienced during the epidemic in Wuhan, China (10). Hospital task forces were created following previously established guidelines for preparedness against disastrous influenza epidemics (8, 11), with the aim of increasing the hospital surge capacity (space, staff, and supplies) and safely admitting a larger number of critically ill patients with COVID-19 (8, 9). Ethics recommendations issued at the beginning of the COVID-19 epidemic in Italy (12) suggested that selective criteria for admission to an ICU should be applied in order to save resources (mainly ICU beds and staff) when these became scarce, to maximize the benefits for the largest number of people. Careful evaluation of the functional status of any critically ill patient was recommended in order to prioritize for ICU admission those patients with a greater probability of survival and life expectancy (12).

By April 3, 2020, total ICU capacity had been increased to 1,761 beds, of which 1,381 (78%) were occupied by COVID-19 patients. Starting from the beginning of April 2020, ICU bed occupancy started to decrease, and by the end of the month it had fallen below the precrisis capacity of 720 beds. The case of Italy was later taken as a benchmark to provide indications for hospital surge capacity in European countries (13). Saturation of health-care resources had been demonstrated to worsen clinical outcomes for inpatients in prepandemic times (14) and was later evaluated with respect to the impact of COVID-19 using different perspectives (15–17). Here, we retrospectively investigate the impact of saturation of intensive-care resources during the first wave of the COVID-19 epidemic in Lombardy to elucidate trends in ICU admission probabilities, ICU admission delays, and ICU mortality across different epidemic periods and age groups.

METHODS

Our study was based on retrospective data collected on the complete set of 46,554 patients admitted with a laboratory-confirmed SARS-CoV-2 infection to one of the 73 hospitals in Lombardy between February 21 and July 12, 2020. Laboratory confirmation of SARS-CoV-2 was defined as a positive result on a real-time reverse transcriptase polymerase chain reaction assay of nasal and pharyngeal swabs (or, occasionally, lower respiratory tract aspirates).

The institutional ethics board of Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico in Milan approved this study and, because of the nature of the retrospective chart review, waived the need for informed consent from individual patients.

Study population

The analysis focused on a subset of 43,538 patients, obtained by considering patients who had symptom onset before hospital admission and excluding those with inconsistencies in dates due to data entry (Figure 1).

Figure 1

Selection of hospitalized coronavirus disease 2019 (COVID-19) patients for a study of the impact of saturation of intensive-care resources during the first wave of the COVID-19 epidemic in Lombardy, Italy, February 21–July 12, 2020. ICU, intensive care unit.

Open in new tabDownload slide

More than 99.5% of the patients in our study were followed from hospitalization through either death or discharge. Among ICU patients, only 10 (0.2%) were still in hospital at the end of our study period, but all of them had already been discharged from the ICU.

Data were collected at hospital admission (age, sex, province of residence, date of symptom onset, date of SARS-CoV-2 diagnosis, date of admission) and throughout the course of the patient’s stay (dates of admission and discharge from the ICU if any, date of discharge from the hospital, clinical outcome (i.e., recovery or death)). Age was grouped into 3 classes (≤55 years, 56–69 years, and ≥70 years), based on the first and third quartiles of age at ICU admission, and was treated as a categorical variable. Age cutoffs were set to the first and third quartiles of age at ICU admission in order to guarantee a sample size balanced by age group. Comorbid conditions were aggregated into 3 major groups: cardiovascular diseases, respiratory diseases, and metabolic disorders (see Web Appendix 1, available at https://doi.org/10.1093/aje/kwab252).

We stratified patients according to the period in which they were hospitalized (see Figure 2): 1) the early phase of the epidemic, from February 21 to March 13, 2020, when occupancy of ICU beds by COVID-19 patients was increasing but was still below the precrisis capacity of 720 beds (period 1); 2) the period of highest pressure on the health-care system, from March 14 to April 25, 2020, when the COVID-19 ICU occupancy ranged between 720 and 1,381 (period 2); and 3) the declining phase, from April 26 to July 12, when COVID-19 ICU occupancy again fell below 720 beds (period 3).

Figure 2

Daily number of intensive care unit (ICU) beds occupied in Lombardy, Italy, between February and July 2020, and definition of periods for the statistical analysis. Lighter gray bars refer to period 1 (early phase of the epidemic), medium gray bars to period 2 (period of greatest strain on the health-care system) and darker gray to period 3 (declining phase of the epidemic). The dashed black line refers to the precrisis ICU capacity of 720 beds.

Open in new tabDownload slide

Statistical analysis

The outcomes of our analyses were the probability of being admitted to an ICU, the probability of death among ICU patients, and duration of time between hospitalization and ICU admission. Time to ICU admission was computed as the time interval (in days) between hospitalization and admission to the ICU among patients admitted to the ICU.

In descriptive analyses, the probabilities of ICU admission and death were summarized by sample proportions, while the time between hospitalization and ICU admission was characterized by sample means. A sufficient sample size in each category allowed us to compute 95% confidence intervals (CIs) by assuming that the 2 statistics followed a normal distribution. To assess differences across multiple groups, we used 1-way analysis of variance, followed by post hoc analysis. Estimated 95% CIs and P values were obtained using Tukey’s “honest significant difference” method, based on the Studentized range statistic.

In multivariate analyses, we estimated odds ratios (ORs) for the odds of being admitted to the ICU and the odds of dying in the ICU during periods 2 and 3 versus period 1. We also estimated incidence rate ratios (IRRs) to compare the delays between hospital admission and ICU admission between different periods. ORs were estimated via logistic regression models adjusting for sex, comorbidity, and province of residence. Negative binomial regression analyses were used to estimate IRRs while adjusting for sex and the presence of comorbid conditions. ORs and IRRs were computed by exponentiating the estimated models’ coefficients. Negative binomial regression was preferred to Poisson regression based on the likelihood ratio test. All analyses were performed separately for subgroups of patients in the 3 different age classes to account for the interaction between age and both the presence of comorbidity and the period of hospitalization. Sensitivity analyses were conducted to assess the robustness of our results with respect to the definition of the 3 periods of interest. In particular, we defined the 3 periods using alternative ICU bed occupancy thresholds of 650 and 800. The statistical significance of the parameters of the logistic and negative binomial regressions was assessed through the Wald test. Statistical analysis was performed using R, version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria), and the R packages “boot,” “MASS,” “multcomp,” “lmtest,” and “aod.”

RESULTS

Among the 43,538 selected patients hospitalized in Lombardy with COVID-19 by July 12, 2020, a total of 3,997 (9.2%) were admitted to an ICU. The median age of hospitalized patients across the 3 different study periods was 68 years (interquartile range, 55–79), and the majority of patients were male (59.6%; Table 1). Among COVID-19 patients admitted to an ICU, the median age was 63 years (interquartile range, 56–70), 78.7% were male, and 55.8% had at least 1 comorbid condition.

There were 11,906 (27.3%) deaths related to COVID-19 among hospitalized patients (see Web Table 1 in Web Appendix 2), of which 1,863 occurred among patients admitted to an ICU (46.6% of all ICU admissions and 15.6% of all deaths). Times between key events for patients hospitalized between February 21 and July 12, 2020, are presented in Web Table 2 (Web Appendix 3).

The overall proportion of hospitalized patients who were admitted to an ICU showed a decreasing trend over the course of the epidemic (Figure 3A and Table 1), from 13.2% (95% CI: 12.6, 13.9) for patients hospitalized during period 1 to 8.4% (95% CI: 8.1, 8.7) for those hospitalized in period 2 and 3.1% (95% CI: 2.5, 3.7) for those hospitalized in period 3 (post hoc Tukey test for the difference in mean values: P < 0.001). A progressive decrease was observed for overall ICU mortality as well, from 51.9% (95% CI: 49.3, 54.5) for patients hospitalized during period 1 to 44.3% (95% CI: 43.7, 44.9) and 27.6% (95% CI: 26.1, 29.1) for those hospitalized during periods 2 and 3, respectively (post hoc Tukey test for the difference in means: P < 0.001) (Figure 3B and Table 1). The delay between hospital and ICU admission, however, was longer for patients hospitalized during period 2, the period of highest pressure on the health-care system (about 5.8 days (95% CI: 5.55, 6.14)—1.18 days (95% CI: 0.53, 1.83) longer compared with period 1 and 2.6 days (95% CI: 065, 4.53) longer compared with period 3 (post hoc Tukey test: P < 0.001 and P = 0.005, respectively; see Figure 3C). Similar qualitative trends over the 3 periods could be observed for the same quantities within a given age group.

Figure 3

Impact of period of hospital admission (period 1, February 21–March 13; period 2, March 14–April 25; period 3, April 26–July 12) on the probability of admission to an intensive care unit (ICU), the probability of death in the ICU, and time between hospitalization and ICU admission among coronavirus disease 2019 patients, by age group, Lombardy, Italy, February 21–July 12, 2020. A) Proportion of hospitalized patients admitted to the ICU, by period of hospital admission. B) Proportion of nonsurvivors among ICU patients, by period of hospital admission. C) Time (days) between hospital admission and ICU admission, by period of hospital admission. In panels A–C, colored columns represent sample means, and gray bars show 95% confidence intervals. D) Adjusted odds ratio for ICU admission, by period of hospital admission. E) Adjusted odds ratio for dying in the ICU, by period of hospital admission. F) Adjusted incidence rate ratio for time between hospitalization and ICU admission, by period of hospital admission. In panels D–F, colored squares represent mean estimates, and colored bars show 95% confidence intervals; the reference period is period 1. In all panels, the gray shaded area highlights the period of greatest strain on the health-care system (period 2).

Open in new tabDownload slide

However, a comparison of the relative variations within each age group highlighted important quantitative differences (see Web Table 3 in Web Appendix 4, Web Table 4 in Web Appendix 5, and Web Table 5 in Web Appendix 6). Therefore, in Figures 3D–3F we show the OR for being admitted to the ICU, the OR for dying in the ICU, and the IRR for the delay between hospitalization and ICU admission for patients hospitalized in periods 2 and 3 with respect to those hospitalized during the early phase of the epidemic (period 1).

Compared with period 1, during period 2 (the period of highest pressure on the health-care system) the odds of being admitted to the ICU decreased much more briskly for patients aged 70 years or more (OR = 0.47, 95% CI: 0.41, 0.54) than for younger patients (about 0.64 for those aged ≤55 years or 56–69 years) (Figure 3D) after adjustment for sex, province of residence, and the presence of comorbidity. The odds of dying in the ICU for patients under 56 years of age increased for those hospitalized during the period of highest pressure on the health-care system with respect to period 1 (OR = 1.43 (95% CI: 1.00, 2.07); see Figure 3E and Web Table 4 in Web Appendix 5); conversely, the risk for patients aged ≥70 years decreased (OR = 0.47 (95% CI: 0.34, 0.64); see Figure 3E and Web Table 4 in Web Appendix 5). The increase in the delay between hospitalization and ICU admission during period 2 was much more important for patients aged ≥70 years (IRR = 1.82, 95% CI: 1.51, 2.18) compared with younger patients (IRR = 1.16 (95% CI: 0.95, 1.42) for patients aged ≤55 years and 1.11 (95% CI: 0.98, 1.27) for patients aged 56–69 years; see Figure 3F and Web Table 5 in Web Appendix 6).

A similar pattern could be observed when considering disaggregation by comorbidity rather than by age group (Figure 4). For patients hospitalized during period 2 compared with period 1, the odds of being admitted to an ICU and the odds of dying in an ICU tended to decrease more sharply when comorbidity was present (Figures 4D and 4E), and the delay in admission to an ICU tended to increase more (Figure 4F) after adjustment for age, sex, and province of residence.

Figure 4

Impact of period of hospital admission (period 1, February 21–March 13; period 2, March 14–April 25; period 3, April 26–July 12) on the probability of admission to an intensive care unit (ICU), the probability of death in the ICU, and time between hospitalization and ICU admission among coronavirus disease 2019 patients, according to the presence of any comorbid conditions, Lombardy, Italy, February 21–July 12, 2020. A) Proportion of hospitalized patients admitted to ICU, by period of hospital admission. B) Proportion of nonsurvivors among ICU patients, by period of hospital admission. C) Time (days) between hospital admission and ICU admission, by period of hospital admission. In panels A–C, colored columns represent sample means, and gray bars show 95% confidence intervals. D) Adjusted odds ratio for ICU admission, by period of hospital admission. E) Adjusted odds ratio for dying in the ICU, by period of hospital admission. F) Adjusted incidence rate ratio for time between hospitalization and ICU admission, by period of hospital admission. In panels D–F, colored squares represent mean estimates, and colored bars show 95% confidence intervals; the reference period is period 1. In all panels, the gray shaded area highlights the period of greatest strain on the health-care system (period 2).

Open in new tabDownload slide

All findings were robust when we considered different ICU bed occupancy thresholds to define the 3 periods of interest (see Web Figure 1 and Web Tables 6–11 in Web Appendix 7).

DISCUSSION

In the spring of 2020, the health-care system of Lombardy was under intense pressure due to the COVID-19 epidemic, with hospital and ICU bed capacity being saturated by large numbers of COVID-19 patients. During this period, high-risk patients admitted to resource-limited hospitals were transferred to designated hub hospitals in the network with available ICU beds and highly skilled intensive-care staff (18). Even after a coordinated regional effort to massively increase the number of available ICU beds, ICU occupation reached levels of around 80% during the peak of the outbreak at the beginning of April 2020. In addition to the limited availability of beds, the pressure on the health-care system was aggravated by a shortage of hospital workers due to the large number of infections occurring among physicians, nurses, and other health-care professionals (7) and to the precautionary quarantines needed to limit hospital transmission.

In this study, we evaluated the dynamics of ICU admission and mortality over the course of 3 time periods representing different levels of health-care resource utilization, as measured by the overall number of ICU beds occupied by COVID-19 patients.

We found that the overall probability of admission to an ICU and the probability of death in an ICU decreased continuously over the 3 periods considered, and this was consistent with similar declines in the overall mortality in hospitals (see Web Figure 2 in Web Appendix 8). These declines are consistent with a recent study of a large cohort of patients admitted to US hospitals with COVID-19, in which the authors attributed the decline to increasing clinical experience specific to SARS-CoV-2 infection (19). However, age-specific trends showed more subtle patterns. Between the early phase of the epidemic (period 1) and the period of highest pressure on the health-care system (period 2), the probability of admission declined by over one-half for patients aged 70 years or older and by only about one-third for younger patients. At the same time, the mortality of patients aged 55 years or younger increased by 43% (95% CI: 0, 107), while that of patients aged 70 years or more declined by 53% (95% CI: 36, 66). These trends, together with the observation of a significantly larger increase in waiting times for an ICU bed for elderly patients, may reflect the prioritization of admission criteria in ICUs during period 2 of patients with higher probabilities of survival and higher life expectancy. In fact, such adjustments of admission criteria had been recommended, according to ethical principles of disaster medicine, to face scarcity of resources during the COVID-19 pandemic (12, 20–22). Stricter criteria for admission of older patients based on their assessed chances of survival probably reduced their observed mortality within ICUs, at a time when the chances of survival for younger patients were worsening. We could reproduce a similar pattern when aggregating patients by the presence of comorbidity, which can be interpreted as another proxy for the frailty of patients as an alternative to age. Just like older patients, during the period of highest pressure on the health-care system, patients with comorbid conditions tended to have greater reductions in their probability of admission to an ICU and extended delays, as well as a higher increase in survival rates compared with patients without comorbidity.

The analysis of ICU admission patterns is made complex by the superposition of multiple dynamic factors, such as the varying level of pressure on the health-care system and the progressive improvement of clinical and pharmaceutical practices. A limitation of this study was the definition of pressure on the health-care system’s being based solely on ICU bed occupancy (Figure 2). Further data with which to better characterize different aspects of pressure on hospital management (e.g., the availability of ventilators, health-care staff, drugs used for COVID-19 therapy, and personal protective equipment, or the progressive expansion of ICU bed capacity over time and the percentage of existing beds occupied by COVID-19 patients) were not available. In addition, we did not have information about interhospital referrals, which might have provided further insight into the dynamics of pressure on the health-care system for individual hospitals (18).

We note that the observed patterns may have been influenced, at least in part, by mechanisms other than the changing admission criteria. For example, it is possible that the health conditions of patients appearing at the hospital changed over time, affecting waiting times in the ICU. In the cohort under study, individual data on the severity of patients’ illness at hospital admission and during their hospital stay were available for only a very small fraction of patients and could not be included in the analysis. Similarly, we did not have granular information on individual therapeutic course, on possible changes in therapeutic approaches over time, or on the use of life-support measures. Notably, as of April 22, 2020, about 87% of ICU patients in Lombardy had received invasive mechanical ventilation, while the remainder were assisted with noninvasive respiratory support (23). We acknowledge that experience in the treatment of COVID-19 patients in ICUs and the physician-to-patient and nurse-to-patient ratios varied widely among hospitals (23) and over the course of the epidemic—for example, there was increasing caution about early intubation among critically ill COVID-19 patients (24)—and this represents a possible confounder of our analysis. Information on individual comorbidity was coarsely represented by the use of 3 macro-level categories (cardiovascular diseases, respiratory diseases, and metabolic disorders) that contained diverse conditions with a heterogeneous level of prognostic relevance. For this reason, only binary information on the presence or absence of a given type of comorbidity was used in the regression models. In Web Figures 3 and 4 (Web Appendix 9), we present the results of a sensitivity analysis showing that the observed trends were confirmed when individual categories of comorbidity were considered.

All results pertaining to mortality in the ICU refer to deaths among persons with a SARS-CoV-2 diagnosis, but we did have data on the specific cause of death; however, the majority of these fatal outcomes were attributable to COVID-19 complications (25). Hospital outbreaks of SARS-CoV-2, especially in the early days, may have played a role in altering transmission dynamics, but we believe that this was not a major source of bias for ICUs, on which our study was focused. Finally, to estimate the probability of ICU admission and the delay between hospitalization and ICU admission, we considered only those patients who were admitted to an ICU. As such, we acknowledge a possible bias in these statistics, as we were not accounting for patients who died in a hospital without being admitted to an ICU. Indeed, our data did not allow us to assess whether or not these patients would have been admitted to an ICU under different circumstances. The pervasive use of personal protective equipment and strict infection prevention protocols in ICUs probably reduced the transmission of SARS-CoV-2 in ICUs; indeed, a serological study on health-care workers in Lombardy found significantly lower SARS-CoV2 seroprevalence among ICU physicians and nurses compared with staff in other hospital wards (26).

Many studies have focused on risk factors associated with COVID-19–related mortality (27–30) and ICU admission (30) among hospitalized patients. A positive association between the proportion of occupied ICU beds and the number of COVID-19 deaths was identified previously, but in a setting that was quite far from saturation of resources (average ICU occupation 20%) (31). Our work adds to the findings of recent studies which identified negative associations between pressure on the health-care system during the COVID-19 epidemic and improvements in patients’ outcomes (15–17).

The use of complete hospitalization data from 43,538 COVID-19 patients in Lombardy, the largest and hardest-hit region of Italy, allowed us sufficient statistical power to characterize subtle age-specific trends in ICU admission criteria and mortality during the COVID-19 pandemic. The saturation of available health-care resources following the rapid upsurge of cases and its impact on ICU utilization may have played a role in the high mortality risk observed during the first wave of the COVID-19 epidemic in Lombardy (32) and in the excess mortality observed, especially among adults over 75 years of age, in municipalities of northern Italy (33). Our analysis stresses the importance of epidemiologic surveillance and modeling (7) to support the prompt implementation of interventions and social distancing measures to limit the transmission of a newly emerging virus and avoid the saturation of health-care resources, which can ultimately result in greater loss of life.

ACKNOWLEDGMENTS

Author affiliations: Center for Health Emergencies, Bruno Kessler Foundation, Trento, Italy (Filippo Trentini, Valentina Marziano, Giorgio Guzzetta, Piero Poletti, Stefano Merler); Directorate General for Health, Lombardy Region, Milan, Italy (Marcello Tirani, Danilo Cereda, Alessandra Piatti, Aida Andreassi, Maria Gramegna); Health Protection Agency of Milan, Milan, Italy (Marcello Tirani); Dondena Centre for Research on Social Dynamics and Public Policy, Bocconi University, Milan, Italy (Filippo Trentini, Raffaella Piccarreta, Alessia Melegaro); Department of Decision Sciences, Bocconi University, Milan, Italy (Raffaella Piccarreta); Regional Agency for Innovation and Procurement, Milan, Italy (Antonio Barone, Giuseppe Preziosi, Fabio Arduini); Department of Public Health, Experimental and Forensic Medicine, Faculty of Medicine, University of Pavia, Pavia, Italy (Petra Giulia Della Valle); Department of Pathophysiology and Transplantation, Faculty of Medicine, University of Milan, Milan, Italy (Alberto Zanella, Giacomo Grasselli); Department of Public Health, Faculty of Medicine, University of Milan, Milan, Italy (Francesca Grosso, Gabriele del Castillo, Ambra Castrofino); Department of Anesthesia, Intensive Care and Emergency, Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Milan, Italy (Giacomo Grasselli); Department of Social and Political Sciences, Bocconi University, Milan, Italy (Alessia Melegaro); Laboratory for Computational Epidemiology and Public Health, Department of Epidemiology and Biostatistics, School of Public Health, Indiana University, Bloomington, Indiana, United States (Marco Ajelli); Laboratory for the Modeling of Biological and Socio-technical Systems, Northeastern University, Boston, Massachusetts, United States (Marco Ajelli). F.T. is now affiliated with the Dondena Centre for Research on Social Dynamics and Public Policy at Bocconi University (Milan, Italy).

This work was supported by European Union grant 874850 MOOD (catalogued as MOOD 000), the VRT Foundation Trento project “Epidemiologia e transmissione di COVID-19 in Trentino,” and the Italian Ministry of Education Progetto di Rilevante Interesse Nazionale (grant 20177BRJXS).

The data underlying this article will be shared upon reasonable request to the corresponding author.

M.A. reports receiving funding for research not related to COVID-19 from Seqirus S.r.l. (San Martino, Italy). None of the other authors reported any potential conflicts of interest.

REFERENCES