A nationwide retrospective study was conducted using the HDRs on health care services in 2020, provided by more than 1300 public and private Italian hospitals. Hospital discharge data are routinely collected by the Italian MoH and contain patient demographic information (eg, gender and age), admission and discharge dates, up to 6 discharge diagnoses (ICD-9-CM), up to 11 medical procedures or surgical interventions, and status at discharge (eg, alive, deceased, or transferred to another hospital). In addition, the National Tax Registry was used to determine vital status or death following hospitalization. HDRs were linked with National Tax Registry records using deterministic record linkage.

This study is conducted in accordance with the Ministerial Decree, “Evaluation in terms of quality, safety and appropriateness of activities provided for accreditation and contractual agreements with health facilities,” released on December 19, 2022, Article 4 [15]. This decree grants a mandate, inter alia, to the Italian National Agency for Regional Healthcare Services to monitor hospital activity and publish reports based on anonymized Italian hospitalizations. We did not include any identifiable data of patients’ personal information, including name, identity information, address, and telephone number; therefore, ethics approval was not required.

Inpatient hospitalizations were detected by implementing an algorithm based on specific ICD-9-CM code combinations. This was carried out through the integration of 2 distinct approaches. One approach, deductive (ie, from the general rule to practical use), was based on the ICD-9-CM codes reported in the official documents of the MoH for COVID-19 case identifications. In addition, we used codes referring to SARS (though not specifically due to SARS-CoV-2), assuming that these codes, mainly at the beginning of the pandemic and before the MoH guidelines publication, were used to indicate COVID-19 hospitalizations. These codes are presented in Table 1.

The other approach, inductive (ie, from an empirical observation to a general rule) was used to confirm and integrate the algorithm. In detail, comparing the number of hospitalizations between 2019 and 2020, a dramatic increase was found in the principal diagnosis with the ICD-9-CM code 484.8 for “pneumonia in other infectious diseases classified elsewhere.” This code was presumed to be associated with COVID-19 and was integrated into the algorithm.

The hospitalizations were then grouped into clinical profiles according to a hierarchical approach based on 3 main criteria, as follows:

All diagnosis codes and their combined use (algorithm) in the principal and secondary diagnoses used to select and classify COVID-19 hospitalizations are shown in Table S1 in Multimedia Appendix 1.

Hospitalizations were analyzed with respect to the distribution of different clinical presentations associated with diagnoses of COVID-19.

A further analysis was performed only among patients hospitalized “due to COVID-19” from January 1 to December 31, 2020, to investigate potential risk factors associated with 30-day mortality. The cohort included all patients with at least 1 “due to COVID-19” hospitalization, excluding repeated admissions due to transfers to other hospitals or readmissions.

Time-to-event techniques were used to analyze survival. The follow-up of patients began from the hospital admission date until either the ascertainment of death (outcome) or up to 30 days after admission (censoring).

The 30-day death rate was calculated as the total number of all-cause deaths up to 30 days after admission divided by the total follow-up days of hospitalized patients in the cohort and was reported per 1000 patient-days. To investigate the association of one or more risk factors with the 30-day death rate among patients hospitalized “due to COVID-19,” a multivariable Cox proportional-hazards regression model (Cox, 1972) was used, after having checked the proportional hazard assumption.

Sociodemographic data (eg, age, gender, area of residence, or nationality) were considered in the analysis. Patients were classified according to the following geographic areas of nationality: Italy or other high-income countries; migrants from high migration pressure countries (HMPC), which include Central and Eastern European countries (including those belonging to the European Union) and Malta; Northern African countries; other African countries; Central and South American countries, Asian countries (excluding Japan, Israel, and South Korea); and Oceania (excluding Australia and New Zealand) [16].

The transfer to ICU has been used as a proxy for the clinical severity of the disease.

The Charlson Comorbidity Index (CCI) [17,18] was used to account for current comorbidities or comorbidities over the previous 4 years in patients that could affect the death rate. To calculate the CCI, 17 medical conditions were considered [19,20]. The CCI was categorized into the following scores: CCI=0, CCI=1 or 2, and CCI≥3.

To detect differences in the 30-day death rate during the year 2020, three pandemic periods were considered, as follows:

Two further multivariable Cox regression models were performed to assess temporal changes over the course of the pandemic; in particular, the 30-day death rates during the first and second waves were analyzed across 3 main geographical areas (ie, North, Center, and South and Islands) and by discharge wards (ie, ordinary and intensive care).

The selection of patients and hospital admissions was performed using SAS Studio (version 3.81; SAS Institute; Enterprise edition). The Cox proportional-hazards regression was performed using STATA (version 14.0; StataCorp). Tests were 2-sided, and statistical significance was set at P<.05.

We identified a total of 325,810 hospitalizations with COVID-19-related diagnosis codes within the public and private accredited Italian hospitals throughout the year 2020 (Table 2).

A total of 239,114 (73.4%) out of the initial HDR selection were classified as “due to COVID-19” hospitalizations (in most cases, with a diagnosis of pneumonia and acute respiratory distress syndrome), and 47,416 (14.5%) were classified as “SARS-CoV-2 positive, but not due to COVID-19” hospitalizations, mainly in association with diseases of the circulatory system, injuries, chronic diseases of respiratory system, and complication of pregnancy, childbirth, and the puerperium.

The remaining 12.1% (n=39,280) were “suspected COVID-19 hospitalizations.” A high proportion of such hospitalizations (n=29,327, 9%) were also found in association with other pathological conditions reported in principal diagnosis.

There was a significant increase in the length of ICU stay out of the overall length of hospital stay (calculated in days), compared with the prepandemic period (1,031,037/54,170,360, 1.9% in 2019 vs 1,204,216/46,113,945, 2.6% in 2020; P<.001). In addition, this proportion was 2.6% (33,509/1,295,932) in 2020 for the “SARS-CoV-2 positive, but not due to COVID-19” and “suspected COVID-19” hospitalizations, while it increased to 8.4% (281,783/3,355,386) among the “due to COVID-19” hospitalizations (data are not shown in the Tables).

Table 3 shows the baseline demographic and clinical characteristics of patients hospitalized “due to COVID-19.”

The cohort included 205,048 patients, with a median age of 72 years. Of these patients, 147,266 (71.8%) were older than 60 years, and there was a higher prevalence of male patients (n=124,181, 60.6%). Additionally, 6.1% (n=12,364) were HMPC citizens coming from Central and Eastern Europe (n=3721, 1.8%), Asia (n=2842, 1.4%), Central and South America (n=2637, 1.3%), Northern Africa (n=1786, 0.9%), and other African countries (n=1378, 0.7%). More than two-thirds of patients resided in the northern Italian regions, while 31.5% (n=64,491) were in Central and Southern Italy. Over the 3 periods of analysis, 39.4% (n=80,843) of patients were hospitalized in the first wave, 53.9% (n=110,446) were hospitalized in the second wave, and only 6.7% (n=13,759) were hospitalized in the postlockdown period. A total of 199,283 (97.2%) patients reported a Charlson Comorbidity Index score lower than or equal to 2.

The overall 30-day death rate in patients hospitalized “due to COVID-19” was 9.9 per 1000 person-days. After adjustment, mortality was lower for women (HR=0.83; P<.001) and for patients coming from HMPC. Among the latter, North African patients (HR=0.65; P<.001) and patients from Central and Eastern Europe (HR=0.66; P<.001) were around 35% less likely to die within 30 days, compared to patients coming from Italy and other high-income countries; Asian patients as well as Central and Southern American patients were 17% less likely to die within 30 days.

In addition, we found a higher 30-day survival in the central regions compared with the northern regions, which were the most affected by COVID-19 (HR=0.90; P<.001), while the southern regions and the Islands showed a higher risk of 30-day mortality (HR=1.17; P<.001).

Transfers to ICUs involved 13.8% (28,247/205,048) of hospitalized patients.

Patients referred to ICUs were two and a half times more likely to die within 30 days, compared to patients hospitalized in ordinary wards (HR=2.55; P<.001).

Table 4 shows the survival analysis by geographic area and wards during 2 pandemic waves.

Over the first wave, among patients in ordinary wards, we found a higher 30-day death rate in the northern regions (10.9 per 1000) compared to the central (7.7 per 1000) and southern islands regions (5.6 per 1000). On the contrary, among patients with a transfer to ICUs, the southern regions and the Islands showed a higher 30-day death rate (28.4 per 1000) compared to the northern regions (15.2 per 1000).

Throughout the second wave, among patients in ordinary wards, there was a significant decrease in 30-day death rates in the northern regions (8.7 per 1000), whereas such rates increased up to 7.4 per 1000 person-days in the southern regions and the Islands. On the other hand, among patients with a transfer to ICUs, the 30-day death rate increased in all geographic areas, particularly in the southern regions and the Islands, where the number of ICU patients more than tripled from the first to the second wave; this resulted in a significantly higher mortality risk in the southern regions and the Islands compared to the northern ones (HR=2.52; P<.001).

In Italian hospitals, due to the lack of COVID-19 codes in the ICD-9-CM system, multiple codes were used for COVID-19 patient admissions throughout the pandemic outbreak; this made it difficult to identify COVID-19 hospitalizations. This study enabled the definition of a specific combination of codes to identify COVID-19 hospitalizations at the national level within the HIS. Additionally, it facilitated the investigation of risk factors associated with mortality due to COVID-19 among patients admitted to Italian hospitals in 2020. The selection algorithm made it possible to identify 239,114 hospitalizations “due to COVID-19” from HDR. This number is very similar to the data on hospital admissions gathered from the COVID-19 Integrated Surveillance System, which amounted to 240,542 [11]. The monthly trend (data are not shown in Tables) is also very similar in the 2 data sources. Only in a few regions, the number of cases derived from HDRs is lower than the number obtained from the Surveillance System, especially in the first wave of the pandemic, likely due to a miscoding of data at the beginning of the epidemiological crisis. Nevertheless, the use of HDR data for COVID-19 impact analyses remains preferable to Surveillance System data for several reasons. The Surveillance System is based on data provided by the regions to monitor the evolution of the pandemic with timely data. Nevertheless, HDR data, although less timely, have higher quality, as they are collected following criteria and rules established by national legislation and guidelines, which guarantee greater homogeneity and comparability across different periods and territories. Moreover, HDR contains more information than the Surveillance System, especially on comorbidities, and allows comparisons with non–COVID-19 hospitalizations as well.

Overall, we observed a decrease in the 30-day death rate among patients hospitalized “due to COVID-19” between the 2 waves of the pandemic. This cannot be explained by the introduction of new pharmacological therapies [21]; however, it could be due to a better use of existing therapies associated with gained experience in managing the disease by health care providers and a more effective preparedness of the health care system, including reorganization of hospitals and increasing number of ICU beds for patients with COVID-19 [22].

In particular, in the second wave, mortality was reduced in the northern regions among patients treated in ordinary wards, with a lower proportion of cases referred to ICUs (from 9144/65,993, 13.9.% to 7990/67,406, 11.9%); this may have led to a higher concentration of patients with severe symptoms in ICUs, consistent with a small increase in 30-day death rates (from 15.2 to 17.0 per 1000 person-days). On the contrary, in the southern regions and the Islands, mortality increased between the 2 waves, both in the ordinary wards and in the ICUs, and this might be due to a higher burden on hospital care compared to the first wave, which had affected Southern Italy to a lesser extent. Furthermore, the South reported a decrease in non–COVID-19 hospitalizations, which was greater than the national mean [23], a sign indicating that the hospital system was struggling.

Our study also highlighted the lower 30-day death rate for women compared to men, regardless of age. The observed disparity may be attributed to sex-based differences in immunological responses, comorbidities prevalence, or differences in behaviours [24,25]. Consistent with several studies, this analysis showed that critically ill older patients were at high risk of disease severity and mortality [26,27]. Some studies also highlighted important post–COVID-19 sequelae in patients with cancer, which adversely affect survival and oncological outcomes after recovery [28].

Finally, our findings, consistent with the findings of a study carried out in the city of Milan [29], confirmed that mortality among migrants in hospitals due to COVID-19 was lower than that of Italians and people coming from high-income countries. The lower 30-day death rate observed among HMPC migrants can be explained as the effect of specific selection processes (especially in the early and late stages of the migration path), which tend to maintain the overall health status of the foreign population high. One of them is the so-called “healthy migrant effect,” a dynamic for which only people in good health tend to emigrate [30]. Another selection dynamic, known as the “salmon effect,” is due to the habit of older migrants, especially if ill, to go back to their country of origin [31].

Moreover, an Italian study on data retrieved from the COVID-19 Integrated Surveillance System found an interaction between nationality and the epidemic phases; the analysis stratified by year showed that compared to Italians, mortality and other severe clinical outcomes in non-Italian nationals slightly increased from the last months of 2020 through 2021. The authors hypothesize that reduced access to vaccination by the immigrated population could partly explain such an increase [32], as a high risk of missed vaccination was observed among non–Italian-born people living in Italy, both overall and for individuals ≥50 years of age [33].

This study has several limitations. First, results may be affected by the information quality of the HDR or by missing or incorrect coding of some diagnoses, leading to possible underestimation of comorbidities. Therefore, to minimize the effect of underreporting bias, comorbidities over the previous 4 years were also retrieved.

Second, the study is based on the HDR, which collects administrative data and does not contain clinical information of patients, which is useful to better characterize COVID-19 severity. For this reason, the transfer to the ICU has been used in this study as a proxy for the clinical severity of the disease.

Third, due to the lack of specified COVID-19 codes from the onset of the pandemic, the overall hospitalizations detected through the algorithm might be underestimated.

Finally, the issue of patients hospitalized due to “suspected COVID-19,” which cannot be defined as COVID-19 cases, remains unsolved.

To our knowledge, the algorithm represents the first attempt to define, at the national level, selection criteria for identifying COVID-19 hospitalizations within the HIS.

While awaiting the adoption of more updated International Classification of Diseases coding systems for hospital diagnoses and procedures in Italy, consistent with the World Health Organization recommendations, the implemented algorithm will be used to monitor the pandemic over time, and the patients selected in 2020 will be followed up in the subsequent years to assess the long-term effects of COVID-19.

Further analyses will be useful to assess the impact of the anti–COVID-19 vaccination campaign on the severity and mortality of the disease and to investigate possible inequalities between population subgroups.