A repeated cross-sectional analysis on the economic impact of SARS-CoV-2 pandemic at the hospital level in Italy

We conducted a repeated cross-sectional analysis on hospitalization episodes occurred in a large research hospital in Milan, Lombardy (Humanitas Research Hospital). In order to cover sufficiently long time period before and after the pandemic outbreak, we analyze discharges occurred between 1st January, 2018 and 31st December 2021.

The Independent ethical committee of IRCCS Humanitas Research Hospital reviewed and approved this study. They confirmed the analyses on retrospective data herein conducted were in line with the informed consent regularly obtained by patients. All methods were carried out in accordance with relevant guidelines and regulations.

Humanitas Research hospital is a highly specialized clinical, research and teaching hospital located south of Milan. It is a private for-profit hospital accredited by the Italian National Health Care System. The hospital works with 759 beds, 38 operating rooms and manages more than 45.000 inpatient admissions, around 57.000 Emergency department accesses and more than 2 million outpatients on an annual basis.

In the period between March 2020 and December 2021 the hospital treated as inpatient 2.986 patients positive for SARS-CoV-2 through 4 waves of disease. The peak of beds needed to treat the patients has been 270 ordinary and 40 intensive care beds.

Among all hospitalizations we selected those associated with the following Diagnosis Related Groups (DRGs): respiratory infections and inflammations with (079) or without (080) complication or comorbidity (CC), simple pneumonia and pleurisy with (089) and without (090) CC, pulmonary edema and respiratory failure (087), respiratory system diagnosis with ventilator support >  = 96 (565) or < 96 h (566), Intracranial Hemorrhage or Cerebral Infarction (014) and kidney and urinary tract infections with CC (320).

Costing is time and effort consuming and, depending on the objective of the analysis, researchers need to decide how accurate and precise cost estimates need to be. In this study we followed a costing approach that can be classified as top-down micro-costing¹⁹. Micro-costing due to the significant degree of disaggregation used in the identification and measurement of resource and cost components, top-down due to the method for the valuation of resource and cost components that relied on Humanitas Research Hospital detailed internal accounting system. Resources used during the inpatient episodes are routinely collected by the hospital data warehouse where all clinical and administrative events are gathered: diagnostic exams, blood transfused, drugs administered, implants. The monetary value of these resources is estimated using the internal actual cost reported by the hospital. Activity-based costing has been used to account for surgery/operating room usage, ICU stay, and CCU stay. Time was the driver chosen to apportion personnel and PPEs use. Clinical time (spent by physicians, nurses and healthcare professionals) per day of stay for patients with a positive test for SARS-CoV-2 was on average 25% longer than an ordinary inpatient episode according to the specific organization of the shifts. The monetary value is estimated using the average hospital cost per type of employee per day of patient stay. The estimation of unit costs took into consideration the specific bonuses and incentives guaranteed by the hospital during the pandemic. In March 2020, specific cost centers were created to precisely determine the consumption in terms of PPE (Personal Protection Equipment) and other materials connected with the COVID-19 patient care. These costs are added to hospitalization episodes according to the length of stay in hospital.

Data on age, sex and type of hospitalization (emergency room or elective) were recorded at hospital admission, while data on the total length of hospital stay, duration of surgery, intensive care admission (ICU) and length of ICU stay, coronary care unit (UCC) admission and length of UCC stay, unit and date of discharge, and the DRG code were recorded at hospital discharge. If hospitalized patients tested positive for SARS-CoV-2 either at admission or during their stay a flag indicating whether the patient was positive for SARS-CoV-2 was recorded.

The primary outcome of our analysis is the full hospital costs of each hospitalization, computed across different cost items: personnel (including physicians, nurses and health care workers), diagnostics, drugs and treatments, implants, blood transfusion and all other consumables (i.e. medical gases, PPEs, etc.), operating room, ICU and CCU time.

We compared these costs across different DRG-homogenous groups of patients based on their discharge date and SARS-CoV-2 status both in univariate and multivariate analyses. More specifically, we defined three different types of hospitalization: inpatients discharged before 21st February 2020, i.e. before the first SARS-CoV-2 case was notified by Italian Health authorities (reference group), inpatients discharged after this date with a negative SARS-CoV-2 test, and inpatients discharged after this date with at least one positive SARS-CoV-2 test during the hospital admission.

Firstly, costs associated with different DRG codes were compared across the three groups using a stochastic dominance test. Graphical inspection of empirical distribution functions and a one-sided Kolmogorov–Smirnov (KS) test were adopted to assess the stochastic dominance of the distribution function of the hospitalization costs for patients discharged before the pandemic and for those discharged after without a SARS-CoV-2 diagnosis over that of SARS-CoV-2 positive patients.

To better disentangle the association between sustained costs and the different groups of patients defined by their discharge date and SARS-CoV-2 status, we run generalized linear models on the total costs considering the type of patient as the exposure and adjusting for relevant available covariates.

In particular, we modeled the outcome through generalized linear model assuming costs are distributed according to a Gamma distribution and using a logarithmic link function.

Let Y be our response variable and X the matrix with the considered covariates, the model can be written as

with Y ~ Gamma.

Running the model on all the observed hospitalizations allowed to compare costs before and after the pandemic, differentiating between patients with and without SARS-Cov2. We also adjusted for a series of covariates, i.e. age of the patients, sex of patient, the type of admission (elective vs. emergency), the type of discharge (death, transfer to another hospital vs. other), the total length of stay in hospital and three dummy variables indicating whether the hospitalization required surgery, intensive care and UCC admission. A quadratic and cubic terms for the length of stay in hospital were introduced due to the non-linear relation of this variable with the total costs. Following Jones et al.²⁰, we also performed additional approaches in modelling the distribution of healthcare costs to check the robustness of our results. Statistical analyses were performed using R version 3.6.2.