Sequelae, persistent symptomatology and outcomes after COVID-19 hospitalization: the ANCOHVID multicentre 6-month follow-up study

A 6-month retrospective longitudinal observational follow-up study was designed, according to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Additional file 1). The study sample consisted of the cohort of patients admitted to the selected hospitals in four cities in Andalusia, Spain (Córdoba, Granada, Jaén and Puerto Real) from March 1 to April 15, 2020. Inclusion criteria included confirmed polymerase chain reaction (PCR) to SARS-CoV-2 (confirmed cases according to the Spanish Ministry of Health). Therefore, patients with compatible clinical symptoms and imaging test (suspected cases) but negative PCR were not included. In addition, only patients who required hospitalization were considered (patients identified in the Emergency Services who did not require hospitalization were excluded). All cases were followed up for 6 months after discharge. Therefore, the follow-up of the study concluded on January 5, 2021.

Hospitalization medical records were consulted to collect data on sociodemographic, clinical, therapeutic and evolution factors of COVID-19 patients. To collect information on SPS after discharge, primary care records were consulted. Follow-up consultation records and periodic telephonic reports scheduled from primary care standardized in Andalusia were consulted. Data from hospital specialties (pneumology and infectious diseases services) were also consulted to obtain more information on SPS. The SPS collected occurred at any time after discharge and before the outcomes. Therefore, SPS registered on return to Emergency Services or readmission were not considered as source of SPS information, in an attempt to avoid potential selection bias. Similarly, when an outcome occurred more than once (e.g. return to the Emergency Services several times), only the SPS reported before the first return were considered for the analysis to avoid reverse causality. Finally, data on dependency and residential care centres were obtained by accessing the Andalusian Epidemiological Surveillance System (SVEA). Obesity and smoking were not considered for the analyses because most of the records did not include these variables. The variables included in the study were:

A detailed description and definitions of all these SPS are presented in Additional file 2: Table S1.

A descriptive analysis of the main sociodemographic and clinical variables of the study was conducted, stratified by outcomes (return to Emergency Services, readmission to hospital and death after discharge). The frequency of SPS in surviving patients was described in detail and stratified by sex. Hierarchical cluster analyses of the SPS were made in order to provide evidence of the validity of the information collected and to explore possible unknown associations between different SPS. This analysis was applied to cluster variables rather than observations, using a dissimilarity measure based on Euclidean distances.

Bivariant analyses were then conducted to identify differences in SPS between men and women, and according to the three outcomes. T-test was applied for quantitative variables (mean differences) and chi-squared tests for qualitative variables. When the application conditions were not met, Mann-Whitney and Fisher exact tests were conducted, respectively.

Multiple logistic regression models were applied. The three outcomes were analysed as dependent variables in three models, and SPS were analysed as independent variables. Crude and adjusted odds ratios (cOR and aOR, respectively) were calculated. The models were adjusted for sex, age and sociodemographic and clinical variables, when possible. All statistical analyses were performed using Stata (StataCorp)®, version 15.0.

The ethical implications of the study were considered according to the principles of the Declaration of Helsinki Declaration. The study was approved by the Provincial Research Ethical Committee of Granada on October 1, 2020. As this was a retrospective observational study, the informed consent was not possible. The database was anonymized, and no identification data was used in the analyses.

Table 1 shows the distribution of the main sociodemographic variables recorded in the study, stratified by outcome. Of the total cohort (n = 969), the subgroup of patients who were discharged alive (n = 797) constituted the sample followed. The rest (17.8%) represented in-hospital mortality from COVID-19 in our cohort. The 8 patients who died after discharge showed older age (82.1 years) and higher frequency of comorbidities and dependence.

Of the cohort followed, 509 patients (63.9%) presented any SPS during the follow-up. The prevalence of the different SPS, classified by organs and systems, is presented in Table 2. It can be seen that the frequency of SPS was high and varied. The most frequent groups of SPS were respiratory (42.0%), systemic or general (36.1%), digestive (26.2%), neurological (20.8%), mental health (12.2%), dermatological (9.3%), infectious (7.9%), cardiovascular (5.8%), ophthalmological (4.6%), nephrological (4.5%), haematological (4.4%) and urological (4.3%).

In order to provide evidence of the validity of the information collected, a cluster analysis was performed for all SPS (Fig. 1). The dendrogram for specific SPS showed an association between dyspnoea and fatigue, depressive and anxiety symptoms, exanthema and pruritus, vomiting and nausea, nephrological and urological SPS, and both with urinary tract infection. Figure 1 shows the SPS associations in the vertical (ordered) edge: the stronger the association between the SPS, the lower the line linking them. Therefore, “vomiting” and “nausea” should be clearly associated if the data have been properly collected, and the same for the other logical associations shown. Therefore, these cluster associations support the validity of the data collected.

We analysed outcomes after 6 months of follow-up in the patients who survived the first COVID-19 hospitalization (n = 797): return to Emergency Services, hospital readmission and death after discharge. The main SPS associated with the outcomes in the bivariant analyses are shown in Additional file 2: Table S2. There were several SPS that showed association (P < 0.05) with all outcomes analysed, such as persistent fever, pressure ulcers, headache, nephrological SPS, hypotension or syncope and superinfection, especially pneumonia (see Additional file 2 for full details).

The main factors associated with return to the Emergency Services (n = 160) in the adjusted logistic regression models were persistent fever, thoracic pain, dermatological SPS, arrythmia or palpitations, superinfection and pneumonia (Additional file 2: Table S3).

For hospital readmissions (n = 35), the main SPS associated with this outcome in our analysis were persistent fever, any nephrological SPS, superinfection and pneumonia. However, we observed large confidence intervals indicating unstable standard errors (Additional file 2: Table S4).

As only 8 post-discharge deaths were observed, no associations were calculated for this outcome due to limited external validity. The main sociodemographic or clinical factor associated with mortality after discharge was older age. After adjusting for age, most associations disappeared (Additional file 2: Table S5).

The main SPS associated with the outcomes studied in the adjusted models are summarized and illustrated in Fig. 2.

The characteristics of our sample are consistent with other cohorts of patients hospitalized during the first wave of the COVID-19 pandemic in different countries. Therefore, the mean age of our sample (63.0) and the sex distribution (53.7% of men) are similar to previous studies such as the recently published Mediterranean cohort study [18], studies conducted in Italy [21] and Brazil [22] and other Spanish multicentre studies [23], although slightly different from other national reports [24]. For example, Carfi et al. showed 56.5 mean age and 63% of men [20] and Garrigues et al. showed 58 mean age and 63% of women [12].

Our higher mean age could be partially explained by the high life expectancy in our country and by an ageing reference population in South Spain [25]. Also, during the first weeks of the pandemic in Spain, a high number of elderly patients from residential care homes were hospitalized [26], given that exhaustive prevention and control measures had not yet been established. The in-hospital mortality of our cohort was 17.8%, data similar to those reported in other studies in the same period performed in New York [27] but lower than in other studies performed in Wuhan [28] and Spain [23]. Similarly, the proportion of patients requiring intensive care was 12.1%, which is consistent with large patient cohorts in the USA (14.2%) [27].

Since the sample of our study is composed of the most severe COVID-19 cases (those requiring hospitalization), they presented a high frequency of previous diseases, polymedication and dependency for daily living activities, as shown in Table 1.

We showed 63.9% of SPS during the 6 months after discharge, similar to other studies. A study conducted in Spain with 14 weeks of follow-up showed 50% post-acute COVID-19 syndrome [18]. Another study conducted in France showed 55% persistent symptoms after a mean of 110 days of follow-up [12] and 62.5% of SPS was present 50 days after discharge [19]. However, other series have reported up to 68% at 2 months [29], 87% at 2 months [20] and 87% at 3 months [17].

Table 2 lists all reported SPS and their prevalence stratified by sex. The most relevant were as follows.

This outcome has not yet been analysed in depth in previous studies. However, given the increasing number of patients discharged from COVID-19, the healthcare costs and the current saturation of the Emergency Services, it is important to detect which SPS cause patients to return to these services. Although we found a high frequency of patients who returned to the Emergency Services (20.3%), it is possible that these data are even underestimated, given the high concern for seeking healthcare at the onset of the pandemic. In our cohort, it was interesting to find that previous diseases, dependency, living in residential care homes or prognostic scores showed no association with this return. However, visible or easily identifiable symptoms like dermatological SPS and persistent anosmia or dysgeusia (which did not increase the risk of mortality or hospital readmission in our cohort) were strongly associated with return to the Emergency Services. It is possible that these SPS could be appropriately addressed by primary care telephone consultations. SPS that showed association with other negative outcomes, such as thoracic pain, arrhythmia, fever and pneumonia, which are undoubtedly alarming symptoms, were also associated with necessary return to the Emergency Services. Recognition of the severity of COVID-19 symptoms is crucial for correct therapeutic management [11], which should also improve follow-up treatment. We believe that more information on the potential alarming symptoms of patients discharged from COVID-19 could be provided to patients and their caregivers to improve follow-up and preventive measures in primary care.

We present the prevalence of SPS during the 6 months after discharge of patients hospitalized for COVID-19. No distinctions were made between early or long-term SPS and any interpretation of the results of this study should take into account that our exposures are SPS during the 6 months after discharge, at any time point. Since the data assessment was collected retrospectively and no prior standardization of SPS was designed, analytical associations through OR are of limited value and should be interpreted with caution. However, we present these results as potential risk factors to be considered in future longitudinal studies. Also, we cannot confirm that these SPS are a consequence of COVID-19, as there is no available comparison group (prevalence of all SPS in patients that did not suffer from COVID-19). We attempted to minimize this limitation by suggesting only potential risk factors to be taken into account during patient follow-up and analysing their relationship with potential negative outcomes in adjusted models. However, the adjusted models could not be well optimized for two outcomes: hospital readmission and death after discharge, given their small sample size (35 and 8, respectively). Therefore, both models were adjusted for a small number of variables and the validation criteria were not always optimal. Given this limitation, associations on post-discharge death are only presented as Additional file 2: Table S5. In addition, obesity and smoking could not be adequately collected and, therefore, models were not adjusted for these important variables. The aim of this study, however, is to describe SPS and propose (but not corroborate or conclude) possible risk factors on negative outcomes.

Our study is also conducted in a sample of high-risk COVID-19 patients (i.e. those who required hospitalization). Therefore, the results regarding SPS frequencies and association with outcomes could not be extrapolated to all COVID-19 patients, but only to those who required hospital care. Also, the SPS reported by this cohort could be attributable to COVID-19 or could be partially explained by the hospitalization process and the treatment received.

The study was conducted during the first wave of COVID-19, when diagnostic and therapeutic protocols were different from the current ones. This bias was addressed by adjusting all associations for diagnostic and therapeutic variables.

Finally, our study was conducted in Spain. External validity may be biased given that the life expectancy, health care and baseline characteristics of Spanish patients cannot be extrapolated to other countries. We tried to minimize this effect by designing a multicentre study including 4 different hospitals and 969 patients. The results of this study could be applied to the Spanish population but should be included in systematic reviews and compared with international samples to optimize accurate conclusions.