Frequency, risk factors, and outcomes of hospital readmissions of COVID-19 patients

To determine the proportion of patients with COVID-19 who were readmitted to the hospital and the most common causes and the factors associated with readmission. Multicenter nationwide cohort study in Spain. Patients included in the study were admitted to 147 hospitals from March 1 to April 30, 2020. Readmission was defined as a new hospital admission during the 30 days after discharge. Emergency department visits after discharge were not considered readmission. During the study period 8392 patients were admitted to hospitals participating in the SEMI-COVID-19 network. 298 patients (4.2%) out of 7137 patients were readmitted after being discharged. 1541 (17.7%) died during the index admission and 35 died during hospital readmission (11.7%, p = 0.007). The median time from discharge to readmission was 7 days (IQR 3–15 days). The most frequent causes of hospital readmission were worsening of previous pneumonia (54%), bacterial infection (13%), venous thromboembolism (5%), and heart failure (5%). Age [odds ratio (OR): 1.02; 95% confident interval (95% CI): 1.01–1.03], age-adjusted Charlson comorbidity index score (OR: 1.13; 95% CI: 1.06–1.21), chronic obstructive pulmonary disease (OR: 1.84; 95% CI: 1.26–2.69), asthma (OR: 1.52; 95% CI: 1.04–2.22), hemoglobin level at admission (OR: 0.92; 95% CI: 0.86–0.99), ground-glass opacification at admission (OR: 0.86; 95% CI:0.76–0.98) and glucocorticoid treatment (OR: 1.29; 95% CI: 1.00–1.66) were independently associated with hospital readmission. The rate of readmission after hospital discharge for COVID-19 was low. Advanced age and comorbidity were associated with increased risk of readmission.

were independently associated with hospital readmission. The rate of readmission after hospital discharge for COVID-19 was low. Advanced age and comorbidity were associated with increased risk of readmission.
The novel coronavirus disease 2019 (COVID-19) pandemic has dramatically impacted many hospitals around the world. For several weeks, the demand for hospital beds in Spain surpassed the capacity to admit patients, hindering the ability to treat other serious illnesses such as neoplasms or cardiovascular disease [1][2][3] . After the highest incidence of this infection during the months of March and April 2020, outbreaks of varying magnitudes are being observed in different regions throughout the world that could eventually compromise hospitals' capacity again 4 . Therefore, improving knowledge on the course of the disease could contribute to the appropriate use of the health resources available within the epidemiological framework.
The aforementioned potential shortage of hospital beds and the lack of firmly established discharge recommendations could result in the hasty, risky discharge of admitted patients. Conversely, physicians may act in an overly cautious manner, unnecessarily prolonging hospital stays. Although the evolution of microbiological tests after the acute phase of the disease [polymerase chain reaction (PCR) of nasal exudate)] has been the subject of several studies, the clinical characteristics of patients at increased risk of readmission have been analyzed in a limited number of studies [5][6][7][8][9][10] . Increased knowledge of the magnitude and characteristics of this issue could help in decision-making related to the initial hospital stay, time of discharge, and clinical follow-up after discharge.
The SEMI-COVID-19 Network has arisen as an initiative of the Spanish Society of Internal Medicine (SEMI, for its initials in Spanish) to improve the management of COVID-19. The main objective of the registry is to generate, in a short period of time, a large, multicenter cohort with detailed information on the epidemiology, clinical progress, and treatment received by patients 11 . Using this information, we aimed to determine the proportion of COVID-19 patients who were readmitted after discharge, the causes of readmission, and factors associated with this poor outcome.

Methods
Study design. This is an observational study base on the SEMI-COVID-19 Registry, which is a retrospective cohort comprising consecutive patients admitted in 147 hospitals in Spain from March 1, 2020 discharged with confirmed COVID-19 disease. The aim of the study was to analyze the clinical characteristics of patients with COVID-19 who were readmitted to the hospital during the first 30 days after being discharged. Patients included in the SEMI-COVID-19 Registry from March 1, 2020 to April 30, 2020 were included in this study, representing approximately 10% of the patients admitted in Spain during this time period 1 . Comparative group was constituted by patients who were discharged alive from the primary admission and were not readmitted during the 30 days following discharge. Patients who were attended in the emergency department after hospital discharge but that were not admitted, were not considered as readmitted patient. Study population and participants. All consecutive patients discharged after hospital admission with confirmed SARS-CoV-2 infection were eligible for inclusion in the SEMI-COVID-19 Registry 11 . COVID-19 was confirmed by a positive result on real-time polymerase chain reaction (RT-PCR) testing of a nasopharyngeal or sputum sample. Patients were treated at their attending physician's discretion, according to local protocols and clinical judgement. Patients included in open-label clinical trials could be included in the registry, provided all information about treatment was available.
Data collection. Data were entered into the database by retrospective review of medical records by resident or staff physicians of the internal medicine departments of the participating hospitals. An online electronic data capture system was developed, which includes a database manager along with procedures for the verification of data and contrasting of information against the original medical record in order to ensure the best possible quality of data collection. Patient identifiable data were dissociated and pseudonymized. Collected data and included epidemiological data, RT-PCR and serology data, medical and medication history, symptoms and physical examination findings at admission, laboratory and diagnostic imaging tests, treatment, complications during the hospitalization, and hospital readmission. Obesity was defined as a body mass index greater than 30 kg/m2. Acute cardiac injury was defined as the detection of acute myocardial infarction, heart failure, arrhythmia or myocarditis. Acute kidney injury was defined as a 50% increase in the baseline creatinine level or a creatinine level greater than 1.5 mg/dl if the previous value was unknown.
Immunocompromised patients included those with solid organ, hematopoietic stem cell transplantation, glucocorticoid treatment (equivalent dose of prednisone ≥ 15 mg/day) or immunosuppressive drugs. Patients with HIV were reported in a separate group. Reasons for readmission, including respiratory symptoms, venous or arterial thrombosis, exacerbation of chronic diseases, organ failure, and bacterial infection were registered. Information on readmissions and death after admission was collected by retrospective review of the clinical  . We attempted to obtain informed consent from all patients. When this was not possible due to biosafety issues, informed consent was requested verbally and noted in the medical record. The STROBE statement guidelines were followed in the conduct and reporting of this study. The information contained in the database may be accessible after contacting the corresponding author and after its reasoned justification.

Results
Over the course of the study period, 8678 patients were included in the registry. Of them, 7137 patients (82.2%) were discharged alive. In the following days, 298 patients (4.2%) were readmitted. The median time from discharge to readmission was 7 days (IQR 3-15). 1541 patients (17.7%) died during the index admission and 35 died during hospital readmission (11.7%, p = 0.007) (Fig. 1). Of the 6839 patients who were discharged alive and were not readmitted, 50 died during the first month after discharge (0.73%). Among the patients who received macrolides, azithromycin was the most frequently administered (99.1%).
Clinical characteristics of the patients who are readmitted. Table 1 shows the clinical characteristic of patients who were readmitted compared to those who were not. The univariate analysis showed that readmitted patients were older (74 vs 65 years, p < 0.001) and had more severe comorbidity. The duration of symptoms before the index admission was shorter in patients who were subsequently readmitted (p < 0.001).
A higher percentage of patients who were readmitted had received steroid treatment compared to those who were not readmitted (38.3% vs 29.7%, p = 0.002, Table 2). The proportion of patients who received prophylactic low-molecular-weight heparin (LMWH) during the primary admission and were readmitted for DVT (87.5%) was similar to patients who were readmitted for another reason (86.2%, p = 0.885) and to patients who were not readmitted (82.7%, p = 0.645). Age, Charlson Comorbidity Index score, diabetes, chronic obstructive pulmonary disease (COPD), asthma, solid neoplasia, hypertension, dementia, duration of symptoms before admission, hemoglobin level and platelets count at admission, ground-glass infiltrate at admission, acute cardiac injury, acute renal failure and steroid treatment during admission were included in the final model ( The characteristics of patients who were not readmitted versus those who were readmitted because of worsening of SARS-CoV-2 pneumonia are shown in the supplementary material. Most of the variables associated with all-cause readmission were also related to readmission for pneumonia (Tables 1, 2, 4 and Table 3S in supplementary material). Patients who were readmitted for pneumonia had a lower prescription of prophylactic heparin at discharge. The rate of readmission for pneumonia was lower in patients who had been admitted to the ICU during the index admission (2%, 3 patients) than in those who had not been admitted to the ICU (5.7%, 390 patients; p = 0.041). (Table 3S in supplementary material). The specific multivariate analysis aggregated by type of variable is presented in the supplementary material. Table 5 shows the reasons for the readmission. The most frequent causes were worsening of previous pneumonia (54%), bacterial infection (13%), venous thromboembolism (5%), and heart failure (5%). Acute kidney failure, neurological complications, and severe hemorrhage were less common as a cause of readmission.

Discussion
We present the first multicenter study on hospital readmissions of COVID-19 patients carried out in Spain. The main finding was that the rate of hospital readmission was relatively low with half of the readmissions occurring during the first week after discharge from the hospital. Most of them were due to respiratory worsening. Age, comorbidity (especially, asthma and COPD) were associated with an increased risk of readmission. Patients with higher hemoglobin levels and ground glass opacification at admission had a lower risk of readmission. Our readmission rate was similar to those found by other investigators 7,10,12 , but lower than those observed in other studies 6,9,13,14 . The higher readmission rate observed in some studies can be related to the longer followup period in some of them 6,14 and to the higher incidence of diseases such as obesity, hypertension, diabetes, asthma and chronic renal failure 6,9,14 . On the other hand, we cannot rule out that the severity of infection in the patients included in the study was lower than in other series, considering the uncertainties in clinical management, which were more prominent during the first wave of the disease. Our results indicate that, after an initial clinical improvement, a small but significant number of patients worsen and require readmission to hospital. The readmission rate in the present study appears to be lower than that observed in patients admitted for other reasons to internal medicine wards in Spain. This observation could be related to the older age and higher comorbidity of patients hospitalized for non-COVID-19-related reasons in our country 15 .
Although the average hospital occupancy per COVID-19 in Spain was around 10% during the first wave, it does not appear that there were many precipitous discharges, in view of the modest rate of readmissions 1,16 . The criteria usually followed to evaluate hospital discharge in patients with respiratory infection, such as disappearance of fever and improvement of respiratory failure, may be effective parameters for discharge ensuring a low readmission rate. We have the impression that the recommendations established in China at the beginning of the pandemic could be overly conservative (no fever for 3 days in addition to clinical, analytical, and radiological improvement) and could lead to an unnecessary prolongation of the hospital stay 5 .
Another characteristic of readmissions is their lower mortality compared to the index admission 6,13 . It cannot be ruled out that this finding could be influenced by the possible death of the most fragile patients during the index admission. The differences detected in the mortality of hospital readmissions in patients with COVID-19 Clinical characteristics of the patients who are readmitted. Older age and comorbidity were associated with an increased risk of readmission, as has been reported in previous studies 5,17 . Both factors often predispose patients to complications that may require readmission 6,7 . Not unexpectedly, readmissions for an infection that primarily affects the lungs are more frequent in patients with chronic lung disease such as asthma or COPD 7,18,19 . Adjusting the treatment of chronic lung disease (corticosteroids, bronchodilators or oxygen flow) www.nature.com/scientificreports/ appropriately and determining the frequency of contacts with their physicians could be beneficial. Dementia was associated with an increased risk of readmission in the univariate analysis. Interestingly, patients transferred to a skilled nursing facility also showed an increased risk of readmission in previous studies 6 . The prevention and treatment of delirium during admission to reduce the risk of readmission is also one of the aspects to be considered 17 . The duration of symptoms in patients who were readmitted (median 5 days) was shorter than in those who were not readmitted (IQR) (7 days, < 0.001, Table 2). The prolonged clinical course with frequent worsening that  www.nature.com/scientificreports/ characterizes moderate and severe forms of the disease suggests that the time of onset of the patient's symptoms should be taken into account when considering hospital discharge 20,21 . A shorter duration of index admission has been associated with a higher risk of readmission 6,7 ; a result that has not been observed in our series. Admission to the ICU has also been associated with a lower risk of readmission due to SARS-CoV_2 pneumonia in our patients, which may be related to a longer hospital stay of index admission and possible better medical care 6 . Serum hypoalbuminemia at admission has been associated with a worse prognosis including the risk of hospital readmission 22 . Yeo et al., observed a direct correlation between the peak creatinine concentration during admission and the risk of readmission 10 . In our study, acute renal failure was associated with an increased risk of readmission but only in the univariate analysis. Hypoalbuminemia and renal dysfunction could be markers of more severe disease and multiorgan failure increasing the risk of readmission. The presence of ground-glass opacities (GGO) showed to be a protective factor for hospital readmission. This result may be explained by the fact that the presence of ground-glass opacities, both in unilateral and bilateral involvement, is associated with an early stage of the disease, which could be related to the risk of hospital readmission 23 . The efficacy of the combination of certain clinical variables and analytical markers in the prediction of hospital readmission could be a line of research that could improve our knowledge of the evolution of this disease 16,24 .
In our study the association between glucocorticoid treatment and hospital readmission was remarkable. Since there was no generalized prescription of corticosteroids during the first wave, we cannot rule out that this www.nature.com/scientificreports/ treatment was a surrogate marker of greater pulmonary involvement predisposing to hospital readmission 25 . In addition, a positive association was found in the univariant analysis between macrolide treatment and a reduction in risk of hospital readmission. These results are surprising considering the results of several clinical studies that examined the role of azithromycin (administered together with hydroxychloroquine) in the prognosis of COVID-19 and found no benefit in terms of mortality or significant clinical improvement 26,27 . Macrolide treatment was not included in the multivariate analysis because we are not sure that its association with the risk of readmission was due to variables not analyzed.
Description of the cause for admission at initial admission and readmission. As in other similar studies, worsening SARS-CoV-2 pneumonia, bacterial infection, venous thromboembolism, heart failure, acute renal failure, and neurological or psychiatric conditions were common causes of hospital readmission 5,7,13 . The administration of more effective and less toxic treatments, determining the dose and rate of decrease of steroids, and the development of prognostic scores could lead to fewer or milder exacerbations of SARS-CoV-2 pneumonia 28,29 . Bacterial infection was a common cause of readmission. Early removal of invasive devices and close monitoring of infection symptoms are important measures, given the late onset of some nosocomial infections in these patients and the masking effect of steroids or interleukin inhibitors on clinical presentation 8,30 .
Venous thromboembolism was also a relatively frequent cause of readmission that usually appeared in the four weeks following the initial hospital admission 31 . This disease is related to coagulopathy, viral endothelial damage due to COVID-19, and immobilization during hospitalization 31 . Although our study is limited in its ability to evaluate this disease, use of LMWH was not associated with a lower risk of readmission due to DVT 32 . Likewise, there is still a need to better define the risk estimate and the dose and duration of prophylactic LMWH in COVID-19 patients 31,32 .
Worsening of cardiac, digestive, endocrine or urinary diseases due to COVID-19 their sequelae appear to be frequent reasons for readmission in our and other studies 6 . Therapy during index admission such as the infusion of IV fluids, diuretics, nephrotoxic drugs, should be given with great caution, given its possible relationship with heart and kidney failure that could precipitate a readmission 33 . Preventive strategies according to comorbidity of the patient may reduce the risk of organ failure and readmission in covid-19 patients. A close contact with the patient, provided with an app., a thermometer and a pulse oximeter, after discharge has resulted in a lower risk of readmission 34 . Limitations. Several limitations of the study should be pointed out. First, as a multicenter study, it is likely that the discharge criteria at each of the participating hospitals differed, so quite different patients may have been analyzed all together. Second, relevant variables such as persistence of fever in the final hours before discharge were not collected and as such, we are unable to determine their relationship to readmission risk or compare them to other variables. Third, the learning curve for this new disease and changes in the demand for hospital admissions during the course of the epidemic may have increased disparity in patient management. Fourth, since the patients included in this study corresponded to the months of March and April 2020, the COVID-19 treatment they received was different from that subsequently recommended. Specifically, there has been a decrease in the use of drugs that we now know do not reduce mortality (hydroxychloroquine, azithromycin, lopinavir-ritonavir, interferon) and an increase in the use of steroids and interleukin inhibitors. Fifth, another limitation is the inability to analyze whether some patients remained hospitalized longer solely due to social and epidemiological reasons (i.e. being unable to take precautions to reduce transmission of the infection at home). Sixth, since we have not analyzed the number of patients who returned to the emergency department with persistent symptoms but were not readmitted during the peak of the pandemic, we must acknowledge some uncertainty in the assertion that the reduced proportion of readmissions was due to a favorable clinical outcome. Lastly, some variables were missing results for a large number of patients, especially in the case of procalcitonin, interleukin-6, D-dimer, lactate dehydrogenase and ferritin. In any case, only variables with a number of failures lower than 10% were included in the multivariate analysis. Despite this, we believe that the information obtained is useful for the management of patients admitted with COVID-19.

Conclusions
In conclusion, we observed a low readmission rate after discharge from the hospital for COVID-19. Worsening of previous pneumonia, bacterial infection, venous thromboembolism, and heart failure were common causes of hospital readmission. Advanced age and comorbidity were associated with an increased risk of readmission. Our study could help to recognize patients at high risk of readmission, which would allow the establishment of strategies during the index admission and after discharge to reduce the frequency of these episodes.