Data source

The data for this nationwide retrospective study was obtained from the Estonian Health Insurance Fund (EHIF), Estonian Health Board (EHB) and Estonian Causes of Death Registry (ECDR). The EHIF, EHB and ECDR are all population-based and nation-wide health/administration organizations. Unique personal identification codes which are assigned to all Estonian residents at birth or at the time of immigration were used for data linkage. Confirmed COVID-19 cases were those based on real-time polymerase chain reaction (PCR) testing on nasopharyngeal specimens by an accredited molecular diagnostics laboratory using certified methods [8]. Data on the date of SARS-CoV-2 RNA test and results were obtained from EHB. Personal information on gender and date of birth and health care utilization with dates of service, diagnostic codes based on International Statistical Classification of Diseases and Related Health Problems 10^th Revision (ICD-10), were obtained from EHIF. As of December 2021, the EHIF contained information on 1 265 601 individuals or 94% of the Estonian population with insurance coverage.

Population and setting

We included data on all people aged 18 and over who tested positive for SARS-CoV-2 during the first year of the COVID-19 epidemic (26 February 2020 to 28 February 2021) and were hospitalized during the acute phase of COVID-19. For the current study we refer to the date of the first positive SARS-CoV-2 test as the index date. Individuals with COVID-19 entered the cohort on the index date. In our study we focused on the pre-vaccination period of the pandemic, so we included individuals with the index dates before mass vaccination in February 2021. Participants with COVID-19 who were admitted to a hospital within 30 days after or 7 days before their positive SARS-CoV-2 test were considered as cases. Those cases who died during acute phase of disease meaning during 30 days of the hospitalization were excluded.

Using EHIF listing as a matching source up to four controls for each case were selected. Eligible for matching were those without evidence of SARS-CoV-2 infection at the index date of each case, were alive and without evidence of SARS-CoV-2 at and during 30 days of case hospitalization. Controls were matched to COVID-19 cases based on a combination of age, gender, Charlson Comorbidity Index score (CCI) and index date. Patients aged 18 and older were matched within 10 years and those aged 60 and older within 5 years from individual case. There was a random selection without replacement among eligible controls for each case done. Index date was divided on 3-month periods (until May 2020, June-August 2020, September-November 2020, December-February 2021). The CCI was calculated using medical claims one-year prior to index date containing 16 comorbidities and classified according to Charlson et al [9]. There was a random selection without replacement among eligible controls for each case. We followed all study subjects for up to 12 months starting from 30 days after discharge until the study end (November 15th, 2021), death, diagnosis of the relevant outcome in each analysis or SARS-CoV-2 positive test in control group.

Outcomes

We aimed to estimate the burden of post-acute COVID-19 sequalae in a cohort of hospitalized SARS-coV-2 positive patients in comparison to infection-free controls from general population. We also explored whether the severity of COVID-19 infection, as proxied by hospitalization and intensive care unit (ICU) admission, affects the risks. The main outcomes were all-cause mortality and all-cause hospital (re)admission. All-cause mortality was defined as death from all causes during study follow-up; all-cause hospital readmission was defined as hospital admission after discharge for COVID-19 for cases, which was compared to hospital admissions in controls. Secondary outcomes were new-onset (incident episodes) of selected clinical conditions identified based on primary ICD-10 codes a (respiratory diseases, heart diseases, stroke, neurological illness, diabetes, chronic kidney, liver disease or mental disorder) appearing at least once on the medical bill. Table including ICD-10 codes used to define an outcomes are presented in (S1 Table). Incident outcome refers to the occurrence of new case of any of the selected health conditions over a follow-up period for an individual who had not had this condition before (the index date) (recall period of 3 years). Intensive care episode during follow up was captured by admission to intensive care unit on medical bill (health insurance claim).

Covariates

Covariates for analyses included age, preexisting comorbidities, and indicators of pre-COVID-19 healthcare utilization, such as number of outpatient encounters and number of hospital admissions. Pre-COVID-19 comorbidities were captured using ICD-10 codes for the 3*365-day period prior to the index date for case patients and their matched controls from medical bills. Comorbidities were defined as primary or other diagnoses coded at the claim and/or diagnoses of any type on hospital or outpatient health care claims during the three years preceding the index. We applied a restriction to outpatient claims, such that a comorbid condition could be flagged during the preceding period only if it appeared two or more times at least 7 days apart [10]. In the final model we accounted for comorbidities with the prevalence higher than 5% (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline descriptive characteristics of patients who were hospitalized with COVID-19 and matched general population controls during 2020–2021 in Estonia.

https://doi.org/10.1371/journal.pone.0278057.t001

Statistical analysis

Distribution of baseline characteristics were compared between cases and controls; and by care type (general ward, ICU) of cases. Descriptive statistics were presented as absolute numbers with proportions or means and standard deviation (SD). The rates of the occurrence of all cause death, readmission, and health outcomes per 10 000 person years in cases and controls were calculated. We estimated 95% confidence intervals (CI) with a Poisson distribution. If the control participant tested positive for SARS-CoV-2 during follow-up they were censored at date of SARS-CoV-2 positive result. The risks of incident outcome were estimated from a Cox survival model. In the survival analysis case-control matching structure was accounted for by including a random effect for each case-controls set in a Cox frailty model. The final Cox survival model was adjusted for comorbidities with prevalence of ≥5%, age, healthcare utilization and previous hospitalization. We produced Kaplan-Meier curves to graphically display the estimates of the significant outcomes. Hazard ratios for each of the outcomes including incident diagnoses were estimated from cause-specific hazard models. All analyses and visualization were done using STATA 17 [11]. To assess heterogeneity related to the matching variables we conducted subgroup analyses stratified by sex and age group (under and over 60 years). This was a retrospective observational cohort study, using claims data. The study was approved by the Research Ethics Committee of the University of Tartu (351/M-8), and no additional requirements were needed regarding gathering informed consent of study participants due to retrospective design of the study.

All-cause mortality

During the follow-up period 399 (10%) cases and 561 (3.6%) control group subjects died (Table 2). All-cause mortality was over two times higher among cases with rate per 10 000 person-years of 1332.4 (95% CI 1207.8–1469.7) in comparison to the rate in controls of 467.7 (95%CI 430.6–508.1) (Table 2). Significantly increased risk was observed in cases compared to controls for all-cause mortality aHR 2.57 (95%CI 2.23–2.96) (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Adjusted HRs comparing COVID-19 hospitalised patients and controls for risk of post-acute COVID-19 sequalae in Estonia 2020–2021.

Circles represent the hazard ratio, and the horizontal bars extend from the lower limit to the upper limit of the 95% confidence interval. Abbreviations LRD lower respiratory diseases; HD heart diseases.

https://doi.org/10.1371/journal.pone.0278057.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Rates of incident outcomes (events per 10 000 person years) among patients who were hospitalized with COVID-19 and a matched general population control in Estonia 2020–2021.

https://doi.org/10.1371/journal.pone.0278057.t002

Among those required ICU admission 13.1% died during study period (risk of all-cause mortality compared to controls aHR = 3.38 95%CI (2.59–4.40)) (S3 Table).

All cause hospital (re)admission

Of 3949 cases over the study period 824 (20.9%) were readmitted for any reason. During the follow-up, hospital admission occurred at rate of 3153.9 (95% CI 2945.8–3376.8) for cases and 1482.5 (95% CI 1413.4–1555.0) per 10 000 person-years for controls. We report markedly increased risk of all cause readmission aHR 1.73 (95%CI 1.58–1.90) in cases compared to age, sex and CCI matched controls.

Cases of new selected clinical sequelae

The data on incident cases of clinical sequelae are provided in Table 2. The occurrence of newly diagnosed hypothyroidism (p = 0.13), disorder of lipoprotein (p = 0.06), and substance abuse (p = 0.42) did not differ between cases and controls. Higher risk for all other clinical sequelae was observed in COVID-19 hospitalized patients compared to the matched general population group.

After adjustment for age, healthcare utilization, previous hospitalization and 11 clinical sequelae cases had significantly increased risk for all outcomes except hypothyroidism, substance abuse and disorder of lipoprotein metabolism compared to matched general population group (Figs 1 and 2 and S2 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Kaplan-Meier curves for the comparison of the significant outcomes during 12 months comparing hospitalized COVID-19 patients (red curves) and matched general population (blue curves).

Shaded zones indicate 95% CI.

https://doi.org/10.1371/journal.pone.0278057.g002

The largest increases in risk were observed for dementia aHR = 4.50 (95% CI 2.35–8.64), chronic lower respiratory disease aHR = 4.39 (95% CI 3.09–6.22), liver disease aHR 4.20 (95% CI 2.01–8.77) and other (than ischemic) forms of heart diseases (ICD-10 I30-52) aHR = 3.39 (95%CI 2.58–4.44). Chronic kidney disease was not included in the final model due to a small number of cases.

Risks of outcomes across subgroups

To assess heterogeneity related to a matching variable we conducted subgroup analysis. Significantly increased risk was universal and was observed for the majority of adverse events across both age and gender groups. Biggest differences in hazard ratios were observed for all-cause mortality in individuals aged <60 years aHR 5.99 (95%CI 3.45–10.38) versus aged ≥60 years aHR 2.76 (95%CI 2.42–3.15) and readmission <60 years with aHR 2.72 (95% CI 2.27–3.27) versus aged ≥60 years aHR 2.0 (1.82–2.19). No substantial differences in aHRs for cases vs controls were observed between males and females across the outcomes analysed.

All-cause mortality and re-admission

Rates for all cause readmission (21%) and death (10%) reported in our study are consistent with the Donelly et al. study from USA, which reported rates within 60 days of discharge of these outcomes prior to mass vaccination of 20% and 9% respectively among COVID-19 hospitalization survivors [13]. Patients hospitalised due to COVID-19 and surviving at least 30 days had more than 2.5-fold higher risk of all-cause mortality and almost double risk for subsequent hospitalization than controls from general population. Our findings of relative risks are consistent with study results from England showing that the overall risk of hospitalisation or death was higher in the COVID-19 group than general population controls [14].

The severity of COVID-19 had a clear effect on all-cause mortality, readmission and clinical conditions, risks were higher in those COVID-19 patients requiring ICU admission. The inclusion of COVID-19 patients requiring ICU in our analysis is important because the current literature is limited regarding long term complications in those patients.

We found that 40.3% of individuals hospitalized due to COVID-19 developed at least one new clinical condition with in one year after the acute phase of SARS-CoV-2 infection, and in comparison, to age/gender and CCI matched peers, this is at least doubling of the risk.

New clinical conditions after acute COVID-19

Individuals surviving after COVID-19 hospitalization are at increased risk for several new clinical conditions; especially for chronic lower respiratory disease, other forms of heart diseases (ICD-10 I30-52), dementia and chronic liver disease.

In our study the largest increase in risk of 4.5 times was observed for dementia. Adequate post-discharge care is vital for those vulnerable group. Two-fold higher risk for neuropsychiatric sequelae as mood disorder and anxiety in our study is in line with the results reported in previously mentioned studies investigating neurological and psychiatric outcomes in COVID-19 hospitalized individuals [15].

Increased risk for chronic lower respiratory diseases can be explained by the finding from China reporting restrictive pulmonary physiology in COVID-19 hospitalized patients [16]. A study from USA with shorter follow-up reported that 65% of severe cases survivors experience fibrotic changes after COVID-19 [17]. Incidence of pulmonary sequelae is the aftermath of both virus-dependent (invasion and alveolar damage by the virus) and virus-independent mechanisms that cause persistent lung damage [18].

Furthermore, we report risk for chronic liver disease was four times higher in hospitalized COVID-19 patients, which is supported by the results reported in the study from UK [19]. In our study the risk of incident type 2 diabetes among COVID-19 patients of being 3 times higher than in general population cohort which is similar to that reported previously [20]. Al-Aly et al reported elevated risk of other cardiovascular disorders compared to general population without evidence of COVID-19 [21].

Our results are in line with reported evidence of biological mechanisms associated with respiratory [22], cardiovascular [23], metabolic [24] and neuropsychiatric [15] post-acute COVID-19 sequelae. Mechanisms surrounding post-acute COVID-19 complications are not entirely clear, but direct effects including with long-term tissue damage or strong immune system reaction leading to autoimmunity or immune dysregulation [25]; direct viral effect (viral invasion affecting heart function [26]; along with indirect effects as loneliness, isolation, economic situation has been described [27,28].

It is important to put these findings into the context of long-term effects of other infectious (viral) diseases as COVID-19 has brought to attention previously understudied phenomenon of post-acute infection syndrome. Post-infectious sequelae have been documented following infection by other viral infections (Ebola, influenza, dengue) including coronaviruses. Studies conducted up to 15 years reported effects on respiratory system [29], mental health [30], lipid metabolism [31] after acute SARS. In the population-based study from England most risks were similar to those observed after influenza hospitalizations, but COVID-19 patients had higher risks of all-cause mortality, readmission due to the initial infection and dementia, highlighting the importance of post-discharge monitoring [14]. A study from China has shown that patients who survive influenza A (H7N9) virus infection are at risk of physical and psychological complications of lung injury and multi-organ dysfunction [32].

To inform prevention and treatment strategy persistent follow-up of survivors is necessary. More emphasis has to be put on those experiencing severe initial COVID-19, as study from Denmark found, that the risk of severe post-acute sequelae after SARS-CoV-2 infection without hospital admission is low [33].

Strengths and limitations

To our knowledge this is one of the largest cohort studies with the longest follow-up time assessing the post-acute COVID-19 consequences in adult COVID-19 hospitalised patients in Europe. We believe our findings are robust given the sample size, the age, sex and CCI matching and the completeness of the data. Using nationwide data and long term follow-up allowed us to obtain precise estimates of assessed outcomes.

However, the study does have some limitations. Due to the nature of study design, we lack important information on subjective symptoms and other self-reported health characteristics that can be obtained from COVID-19 survivors (fatigue, poor attention, altered sense of taste). Accordingly, we are potentially missing a subset of post-COVID-19 health problems. Outcomes were presented in broad categories and more specific and granular presentation would be of a future interest.

Furthermore, there could be a number of patients with unrecorded post-acute COVID-19 features due to health care service provision limitation making our incidence estimates lower. It is difficult to gauge the completeness of electronic health records, or to validate diagnoses, and these factors could influence incidence estimates with the risk of measurement bias. We hypothesise that these deficiencies would affect cases and controls similarly leading to non-differential misclassification. The hospital admission thresholds could be lower in post-COVID-19 individuals than in the general population, leading to an overestimation of the impact on health. Due to lockdowns and inadequate access to medical services, rates of diagnoses might have decreased in individuals without hospital admission, which could lead to an underestimation of associations. Also, there could be individuals among the matched general population cohort that were untested for COVID-19, leading to an underestimation of effects. Based on the nature of the study design we were unable to match for all potentially relevant factors [34], for example, we did not have data on sociodemographic factors. Moreover, we showed that COVID-19 hospitalized patients were more likely to have baseline comorbidities than the general population controls, reflecting known associations between comorbidities and risks of severe COVID-19 outcomes. We have accounted for this by matching our cohorts for age, sex, Charlson Comorbidity Index and adjusting the final Cox survival model for a rich set of covariates to remove confounding as imbalance existed after matching. The adjustment was made for preexisting comorbidities with a prevalence of ≥5%, health care utilization, and previous hospitalization. We do agree that the propensity score matching would be also another appropriate way to balance exposed groups on the confounding factors. In addition, we do not know which SARS-CoV-2 virus-type individuals were infected. Wild type was overwhelmingly prevalent in Estonia during the current study period [35] It has been reported that some SARS-CoV-2 variant types are overrepresented [36] in severe cases and could possibly modify the incidence of post-acute COVID-19 manifestations [37].

Regarding the true impact of SARS-CoV-2, in medical literature the term of post-intensive care syndrome (PICS) has been described [38]. PICS manifestation involves physical, mental and cognitive health impairments after ICU discharge, which are partially similar to post-acute COVID-19 sequelae. In this regard, more studies are required in order to separate the impact of PICS and a true effect of SARS-CoV-2 leading to the consequences after acute phase of COVID-19. Finally, observational epidemiological studies cannot identify mechanisms of effect and causality; complementary studies are needed.

It is important to note that the current study was conducted before mass vaccination was introduced. There are a limited number of studies looking at whether vaccination reduces the risk of post-acute COVID-19 clinical complications. In the large cohort study by Taquet et al. analysis showed that the risk of several important outcomes including diabetes, mood disorder, anxiety appear to be unaffected by vaccination suggesting that other biological mechanisms support those outcomes [39]. More nationwide studies with standardized design, longer follow-up time and involving vaccinated individuals are therefore needed.