Unveiling the Unexpected: Why Doctors Should Look beyond the Lungs when Predicting COVID-19 Mortality

The COVID-19 pandemic, caused by the SARS-CoV-2 virus, broke out globally in 2019, originating from Wuhan, China, and quickly spreading worldwide. The virus presented a wide range of symptoms and severities, with some individuals being asymptomatic while others developed severe pneumonia and long-term consequences, leading to a high mortality rate and infectivity rate [1]. In response, the World Health Organization declared COVID-19 a global pandemic on March 11th, and many countries imposed various types of quarantines [2].

As the pandemic progressed, vaccines have played a crucial role in preventing severe illness and reducing the mortality rate. Despite this progress, COVID-19 remains a persistent threat, making it essential to predict its mortality rate. Accurate prediction methods can help evaluate patients on admission to the hospital, allowing for closer monitoring of patients with the potential for rapid deterioration.

Several studies have explored the factors that increase the risk of mortality from COVID-19, but only a few have attempted to find the best combination of factors that may predict it [3, 4], particularly when a patient comes to the emergency unit and his health status requires hospitalization. Diseases such as chronic obstructive pulmonary disease (COPD), asthma bronchiale, chronic kidney disease (CKD), acute kidney injury (AKI), hypertension (HTN), and coronary artery disease (CAD) have been shown to increase the risk of mortality from COVID-19 [5]. Other studies have used parameters such as age, SpO₂ saturation, laboratory values (CRP, platelets, lactate dehydrogenase [LDH]), and radiological analysis to predict mortality [3]. However, none have compared or integrated both sets of pulmonary and renal parameters for the best mortality prediction. Furthermore, many of the parameters investigated in these studies are not typically collected during routine hospital admission, a crucial time for determining a patient’s prognosis.

To address these gaps in the literature, our study aimed to compare and integrate pulmonary and renal parameters to predict mortality in COVID-19 patients. Furthermore, we sought to investigate the utility of these parameters at the time of emergency department arrival when a patient’s health status necessitates hospital admission. By examining these important clinical factors, our study aimed to enhance the ability of healthcare professionals to predict patient outcomes and implement appropriate interventions accurately. We endeavored to be the first to emphasize this comparison. We used a larger patient sample and a more extensive range of variables to identify the most accurate predictors of COVID-19 mortality. Our selection of variables was carefully chosen and included standard assessments typically performed during hospital admission, such as chest X-ray (CXR) imaging and laboratories. Furthermore, we sought to determine the most useful predictors for application across diverse healthcare settings, including Emergency Medical Service (EMS), Emergency Department, and Internal Medicine Floor.

Our study is a retrospective analysis that examined all patients over the age of 18 who were consecutively admitted to FNKV (a tertiary University Hospital Královské Vinohrady) between September 1st, 2020, and December 31st, 2020. Initially, our study identified patients suspected of having COVID-19. After reviewing the electronic medical system, we identified patients who tested positive for COVID-19 using a PCR (polymerase chain reaction) test on throat swabs. We excluded asymptomatic patients and patients with missing data (laboratories, history, CXR) or with poor quality of the CXR.

To ensure that we did not miss any COVID-19-positive patients, we checked the National Information System of Infectious Diseases (ISIN) documentation of COVID-19-positive patients. This system was created by the Czech Ministry of Health and strictly regulated to document all COVID-19-positive patients nationwide.

After finalizing the number of patients in our cohort, we collected various data from the electronic medical record (EMR) on patient admission, including age; gender; mortality status during hospitalization; hospitalization length; comorbidities (such as arterial HTN, CAD, diabetes mellitus [DM], obesity [defined by BMI >30], dyslipidemia, COPD, asthma bronchiale); laboratory values on admission (such as s-Cr [creatinine], estimated glomerular filtration rate [eGFR], ALT, AST, LDH, ferritin, procalcitonin, platelets, white blood cell count, D-dimer, CRP, peripheral oxygen saturation [SpO₂] measured by pulse oximeter); and CXR images.

The cohort of hypertensive patients was defined as those receiving pharmacological intervention, whereas the cohort of CAD patients was defined as those who underwent a coronary procedure or received any form of pharmacological treatment. The diabetes mellitus cohort was comprised of individuals receiving anti-diabetic medication.

Given that COVID-19 predominantly affects the respiratory system, we were intrigued by the potential role of chest imaging in mortality prediction. Given that chest computed tomography (CT) scans were not routinely conducted for most patients, we instead utilized chest X-rays (CXRs) in our analysis. To standardize our evaluation of CXRs, we adopted the 18-point scoring system developed by Borghesi A. et al. [6]. This system involves dividing each lung image into six equally sized quadrants and assigning a score ranging from 0 to 3 to each quadrant based on the following criteria: 0 indicated the absence of any abnormality, 1 indicated the presence of interstitial infiltration, 2 indicated the presence of both interstitial and alveolar infiltrates with interstitial infiltration being predominant, and 3 indicated the presence of both interstitial and alveolar infiltrates with alveolar infiltration being predominant.

Our evaluation of CXRs was conducted as follows: every image was evaluated by two independent trained physicians who were blinded to the patients’ mortality status. Following evaluation, each image was assigned a score ranging from 0 to 18 points. The scores were then compared, and if the difference between the two scores was less than three points, the final score for the image was calculated as the average of the two. In cases where the difference was greater than three points, an expert radiologist who was also blinded to the patients’ mortality status provided the final score. Patients who lacked CXR images or had technically unsatisfactory images were excluded from our analysis. Furthermore, patients with CXR images that could not be appropriately scored by the expert radiologist (primarily due to pre-existing interstitial changes, severe pleural effusion, or cardiac disease) were also excluded from our cohort.

We created 2 groups of patients with kidney disease based on their s-Cr and eGFR values. We used the CKD-EPI equation (Chronic Kidney Disease – Epidemiology Collaboration) to evaluate the eGFR [7]. The RIFLE and KDIGO criteria were used to define AKI and CKD, respectively [8]. To further classify the patients, we followed the methodology outlined in a recently published article [9]. The patients were divided into the “AKI&CKD” and “eGFR” groups, with further subdivisions based on the severity of renal damage.

The “AKI&CKD” group was divided into six subgroups using a step-by-step process. In the first step, patients whose creatinine rose by at least 50% within 7 days or by at least 26.5 [μmol/L] within 48 h during hospitalization were assigned to group E, called “AKI-RISE.” In the second step, patients whose creatinine dropped by more than 33.3% during hospitalization were assigned to group C, called “AKI-DROP.” In the third step, patients who were not part of the AKI groups and had an eGFR of 0.75–1.5 [mL/s/1.73 m²] during hospitalization were assigned to group B, called “CKD grade 2 and 3a.” In the fourth step, patients with an eGFR of 0.5–0.75 [mL/s/1.73 m²] during hospitalization were assigned to group D, called “CKD 3b.” In the fifth step, patients with an eGFR of less than 0.5 [mL/s/1.73 m²] during hospitalization were assigned to group F, called “CKD grades 4 and 5.” In the final step, patients with an eGFR of at least 1.5 [mL/s/1.73 m²], those who did not fulfill any of the above criteria, and those who had at least one normal eGFR were assigned to group A, called “normal renal function group.”

The “eGFR” group also contained six subgroups, with patients classified based on their eGFR values. Group 1, called “normal,” included patients with CKD grade 1 and an eGFR of at least 1.5 [mL/s/1.73 m²]. Group 2, called “mild,” included patients with CKD grade 2+3a and an eGFR of 0.75–1.5 [mL/s/1.73 m²]. Group 3, called “moderate,” included patients with CKD grade 3b and an eGFR of 0.5–0.75 [mL/s/1.73 m²]. Group 4, called “severe,” included patients with CKD grade 4 + 5 and an eGFR of less than 0.5 [mL/s/1.73 m²].

The study divided both groups based on the severity of renal damage. During hospitalization, the CKD groups in both categories relied solely on eGFR and did not consider eGFR data from the 3 months leading up to hospitalization. This decision was made because some patients lacked sufficient renal data in the EMR during the 3 months before hospitalization. Therefore, our definition ensured that no CKD patients were missed in these groups.

It is important to note that there is a difference in the classification of CKD subgroups between the “AKI&CKD” and “eGFR” groups. In the former, patients who did not meet the criteria for AKI could still be categorized as part of the CKD groups. In the latter, patients in the “eGFR” group were categorized based strictly on their eGFR upon admission, regardless of their AKI status.

Initially, this study sought to compare the efficacy of pulmonary and renal parameters in predicting mortality from COVID-19. The pulmonary parameters consisted of CXR radiological score and SpO₂, while renal parameters were represented by two groups, namely, “AKI&CKD” and “eGFR.” Subsequently, we combined both sets of parameters to identify the most effective predictor of mortality. To evaluate the predictive value of each parameter, the study utilized the area under the curve (AUC) metric.

To improve the ability to predict mortality from COVID-19, the study also examined other factors that could influence an individual’s overall health and increase the risk of mortality. A comprehensive review of the existing literature was conducted, identifying and analyzing factors previously reported to be associated with mortality in COVID-19 patients. These factors included comorbidities such as age, gender, CAD, diabetes mellitus, HTN, dyslipidemias, COPD, and asthma, as documented in previous studies [10, 12]. We also evaluated several laboratory values such as white blood cell, platelets, liver enzymes (AST, ALT), LDH, inflammatory markers (CRP, ferritin, procalcitonin), and D-dimers, all of which have been reported to be linked with increased mortality from COVID-19 [10, 13, 16]. To create the ultimate mortality predictor, we combined these additional factors with the pulmonary and renal parameters.

Using the collected data, the study determined the most practical combination of factors to create the best predictor of mortality in three different medical settings: Emergency Medical Service (EMS), the Emergency Department, and the Internal Medicine floor. This division was made to consider the differing levels of available data in each setting, ranging from limited data in the ambulance to more extensive data in the internal medicine floor.

Since the onset of the COVID-19 outbreak, guidelines have been updated repeatedly. During the course of our study, we adhered to the WHO guidelines at our hospital [17]. Patients who experienced hypo-saturation were administered oxygen via nasal cannula, nonrebreather masks, high-flow nasal oxygen (HFNO), or mechanical ventilation if necessary. While all patients in the hospital were eligible to receive oxygen through nasal cannula or HFNO, mechanical ventilation was subject to certain limitations, and the Ethics Committee determined which patients would receive it. At the time of the study, most patients with hypo-saturation received dexamethasone as a potential mortality benefit had been reported possible [18].

Moreover, patients with severe COVID-19 infections received convalescent plasma (CCP) within the first 3 days of their hospitalization, alongside other therapeutic modalities and symptomatic management in accordance with the hospital’s standard medical guidelines (such as antipyretics, antibiotics, and diuretics) [19].

We utilized the STATISTICA 12 software to conduct statistical analyses. Continuous variables were reported as means and standard deviations and compared using either a two-sample t test or ANOVA test. Categorical variables were presented as proportions and compared using Pearson’s χ² test. Multivariate logistic regression was utilized to determine the adjusted impact of variables on the outcome. The results of the multivariate analyses were expressed as odds ratios with 95% confidence intervals and p values. We evaluated the logistic regression model’s prediction performance by employing the ROC curve and the AUC value. All statistical tests were performed at a 5% significance level.

During the initial screening phase, we identified 808 patients who presented as possible cases of COVID-19 out of the numerous individuals who were admitted to our hospital. However, only patients exhibiting symptoms and receiving a positive PCR test for COVID-19 were included, resulting in the exclusion of 128 patients. An additional 2 patients were excluded due to incomplete data in the electronic medical system, while 41 patients were excluded due to the absence of CXR images, and 6 patients were excluded due to technical issues with the images at the time of admission. After the evaluation of 82 CXR images by an experienced radiologist, 14 patients were excluded due to severe pathologies that rendered their CXR images unsuitable for assessment. These pathologies included severe pleural effusion in 4 patients, severe cardiac pathologies in 8 patients, and severe interstitial changes in 2 patients. Ultimately, the final cohort consisted of 619 patients.

The patients in the cohort had a median age of 72 years and were of Caucasian ethnicity. Among them, 45% were females, and 55% were males. The distribution of comorbidities was as follows: 66% had HTN, 32% had diabetes mellitus type 2 (DM), 32% had dyslipidemia, 19% had CAD, 12% had obesity, 8% had COPD, and 8% had bronchial asthma (Table 1).

The cohort was further divided into two primary groups based on renal parameters. The “AKI&CKD” group was divided into six subgroups, namely, group A (normal renal function) comprising 231 patients, group B (CKD G2+G3a) comprising 138 patients, group C (“AKI-DROP”) comprising 112 patients, group D (CKD G3b) comprising 31 patients, group E (“AKI-RISE”) comprising 78 patients, and group F (CKD G4+G5) comprising 29 patients. The eGFR group was divided into four groups based on their eGFR, with group 1 (normal renal function) comprising 147 patients, group 2 (CKD G2+G3a) comprising 324 patients, group 3 (CKD G3B) comprising 67 patients, and group 4 (CKD G4) comprising 81 patients (Table 1).

Initially, we determined that the mean age at death was 80.1 years compared to 69.3 years for those who survived. We then analyzed mortality rates according to different variables of interest. Specifically, when examining mortality rates based on various comorbidities, the following percentages were observed: dyslipidemia (22%), CAD (34%), COPD (16%), asthma (12%), diabetes mellitus (30%), obesity (24%), and HTN (26%) (Table 1, 2).

Furthermore, we assigned a radiological score to each chest X-ray image and calculated the mean score for patients who survived COVID-19 and those who died from it. Our findings revealed that the mean score for deceased patients was 8.6 ± 1.5, while that for surviving patients was 7.1 ± 1.2 (p < 0.01). Therefore, we conclude that the radiological images of patients who did not survive were more severe than those of patients who survived.

Lastly, we calculated mortality rates within subgroups of our “AKI&CKD” and “eGFR” groups. Within the “AKI&CKD” group, we first compared mortality rates in patients with normal renal function to those with AKI (AKI-DOWN and AKI-RISE combined) and those with CKD (grades 2–5 combined). Mortality rates were 7.8% for patients with normal renal function, compared to 35.7% and 30.3% for AKI and CKD groups, respectively (p < 0.01). Mortality rates for subgroups were as follows: group A – 7.8%, group B – 19.6%, group C – 22.3%, group D – 42.0%, group E – 55.1%, and group F – 69.0% (p < 0.01). In the “eGFR” group, mortality rates for subgroups were as follows: group 1–7.5%, group 2–21.0%, group 3–37.3%, and group 4–51.9% (p < 0.01) (Table 1).

Upon evaluating numerous combinations of the gathered parameters, our findings indicate that renal parameters exhibit superior predictive power for mortality compared to pulmonary factors. Specifically, the area under the curve (AUC) of the receiver operating characteristic (ROC) curve, which includes age, “AKI&CKD” group, and “EGFR” as renal parameters, was found to be 0.8068 (Table 3). Meanwhile, the AUC of the ROC curve, which factored in age, radiological score, SpO₂, CRP, and D-dimer as pulmonary parameters, was 0.7667 (Table 3). Even with the inclusion of additional factors such as multiple comorbidities (CAD, DM, HTN, dyslipidemias, COPD, asthma) and laboratory results (CRP, AST, platelets) in the respiratory group, the AUC remained slightly higher (AUC = 0.8145) than the renal group without these additions.

However, when both renal and respiratory parameters were combined (Table 3), the ability to predict mortality improved considerably. Initially, the AUC increased to 0.861 when adding pulmonary scoring, SpO₂, and the “EGFR” group. Furthermore, when the “AKI&CKD” group was added, requiring knowledge of whether the patient’s creatinine level increased or decreased during hospitalization, the AUC rose even further to 0.8813.

Following a thorough analysis of multiple parameters for COVID-19 mortality prediction, we sought to identify the optimal combination that could be utilized in routine clinical practice. To this end, we assessed the predictive capabilities of various parameters in three distinct settings: Emergency Medical Services (EMS), the Emergency Department (ED), and the Internal Medicine floor.

At the outset, the EMS team typically has access to limited patient data, necessitating the utilization of readily available information such as gender, age, and SpO₂ for rapid and effective mortality prediction. The resulting AUC in this scenario was 0.7874 (Table 4 – “EMS”).

As patients progress to the ED, additional vital information becomes available, including basic laboratory tests, medical history, eGFR, and CXR imaging. The application of a radiological scoring system, coupled with the aforementioned laboratory parameters (AST, LDH, CRP, D-dimer, procalcitonin, platelets), the eGFR group, and initial EMS team evaluation (SpO₂, age, gender), can significantly enhance mortality prediction. This approach yielded an AUC of 0.8614 (Table 4 – “Emergency Department”).

Upon admission to the internal medicine floor, the medical team can further stratify patients based on AKI&CKD groups. Clear documentation of CKD history, as well as the assessment of creatinine levels and fluctuations (AKI-RISE vs. AKI-DROP), can be accomplished. Consequently, the AKI&CKD group can be integrated into the parameters used to predict COVID-19 mortality. By combining the parameters employed in the ED with the AKI&CKD group, the most effective prediction of COVID-19 mortality can be achieved. The resulting AUC, in this case, was 0.8813 (Table 4 – “Medicine Floor”).

Since the emergence of the COVID-19 pandemic in 2020, numerous aspects of daily life have been impacted. This viral outbreak has not only adversely affected the personal health of billions of individuals but has also drastically altered healthcare systems and had a significant impact on the global economy. During the pandemic, COVID-19 has caused the deaths of over six million people, and millions of others have been afflicted with chronic conditions [20, 21]. It is believed that accurate mortality prediction for COVID-19 patients is critical, especially in clinical settings. Such predictions would aid physicians in identifying high-risk patients who require more vigilant care. For instance, patients with a high mortality risk may require admission to the ICU, those with moderate risk may be admitted to the floor, while low-risk patients may be discharged. This approach would enhance the efficiency of the healthcare system, thereby saving more lives.

Radiographic imaging is an essential tool for assessing the severity of pneumonia, and we included chest X-ray (CXR) as a pulmonary parameter. While CT is more accurate in evaluating pneumonia and its severity, we opted not to use it due to several reasons. First, CXR is more widely available and used globally, whereas conducting CT scans for all suspected COVID-19 pneumonia patients on admission would strain healthcare resources and potentially compromise overall efficiency. Nonetheless, we believe that combining CT scans with renal parameters may enhance the accuracy of mortality prediction. Second, a limited number of patients in our cohort had CT images on admission. To maintain an adequate sample size and minimize exclusion of potential participants, we chose CXR as the diagnostic tool for mortality prediction.

Borghesi et al. [6] developed an 18-point scoring system for assessing CXR images, which has been shown to be a significant predictor of mortality. Due to the advantages of this scoring system, we decided to use it for the evaluation of CXR images as it provides a broader range of scores (0–18), allowing for a more comprehensive assessment of pulmonary involvement in COVID-19 pneumonia. However, it is important to note that comorbidities can impact the CXR score. For example, a patient with congestive heart failure or interstitial lung disease may have a more severe CXR image than a patient with healthy lungs and heart. To address these potential discrepancies, the CXR images were evaluated by two experienced physicians and an expert radiologist, ensuring the most precise score possible.

Many studies have traditionally focused on laboratory parameters and respiratory factors, such as images, respiratory rate, and oxygen saturation as these have been traditionally linked with COVID-19’s pulmonary nature [6, 22, 23]. However, recent research has identified renal parameters, including AKI and CKD, as factors that contribute to increased mortality from COVID-19 [9, 24, 27]. Additionally, recent research indicates that eGFR may predict the mortality form COVID-19 [9]. As COVID-19 primarily affects the respiratory system, we sought to investigate the potential of renal parameters as predictors of mortality in comparison to pulmonary parameters.

In our study, CKD was defined as the absence of AKI and abnormal eGFR during hospitalization. The 3-month period of elevated eGFR, which is commonly used in CKD definitions, was not utilized in our research due to insufficient data in the electronic medical system. We did it to prevent the loss of patients with CKD who lacked medical data in the EMR. This approach increased the possibility of including patients who lacked CKD in the CKD group. Therefore, patients who had abnormal eGFR throughout hospitalization but normal eGFR during the 3 months before hospitalization were also included in the CKD group. Despite this limitation, we demonstrated that patients with renal damage (both AKI and CKD) had a mortality rate more than three times higher than that of patients with normal renal function.

In this investigation, we defined CKD as the absence of AKI and abnormal eGFR during the hospitalization period. Owing to insufficient medical data in the EMR, we refrained from utilizing the common 3-month duration of elevated eGFR in CKD definitions. This decision was made to prevent any loss of patients with CKD who had inadequate medical records in the EMR. Consequently, we potentially included patients without CKD in the CKD group, such as individuals with abnormal eGFR during hospitalization but normal eGFR 3 months before hospitalization. Nevertheless, despite this constraint, our research highlighted that patients with renal damage, comprising both AKI and CKD, had a mortality rate over 3 times higher than patients with normal renal function.

The “AKI&CKD” classification utilized in this study included two subgroups, namely, “AKI-RISE” and “AKI-DROP,” which enabled more precise prediction of mortality [9]. Our hypothesis was that patients with rising creatinine levels during hospitalization (i.e., the “AKI-RISE” group) would be in a more severe condition compared to those with declining creatinine levels (i.e., the “AKI-DROP” group). Our findings confirmed this hypothesis as the mortality rate was higher in the “AKI-RISE” group than in the “AKI-DROP” group. Moreover, this classification allows physicians to enhance the accuracy of mortality prediction by accounting for creatinine fluctuations during hospitalization, in contrast to other factors that are determined only at admission.

One of the main limitations of our study was the homogeneity of our study population, which comprised only Caucasian individuals as African American and Asian ethnicities are rare in Prague. As ethnicity can influence eGFR values, it is possible that the ability of the eGFR calculator to predict mortality from COVID-19 may differ in other ethnic groups. Therefore, the generalizability of our findings to other ethnicities is uncertain.

It is important to acknowledge that our study is based on the treatment management of COVID-19 during the initial wave of the pandemic in 2020, and therefore may not accurately reflect the most recent treatment protocols. Consequently, the disease severity and duration of hospitalization described in our study may be more severe than what is currently observed with the use of newly developed vaccines and pharmacological treatments. An updated study, which focuses on patients who have been vaccinated, may provide more up-to-date insights into predicting mortality during the current phase of the pandemic. Furthermore, it should be noted that since the time of our data collection, several variants of the COVID-19 virus have emerged, and the ability to predict mortality may differ depending on the specific viral strain.

Our study’s key contribution is demonstrating the significance of renal factors in predicting mortality. This highlights the need to consider renal damage in discussions regarding mortality prediction, as patients with impaired renal function experience more severe illness and higher mortality rates. Our findings suggest that examining how renal parameters influence mortality in other respiratory infections could be of interest. Combining these parameters with other predictors may improve mortality prediction, and our study is the first to emphasize the crucial role of renal parameters in predicting mortality from COVID-19. Although COVID-19 is primarily a respiratory disease, we recommend that renal function be given more extensive consideration when evaluating patients’ mortality risk. We also identified which combination of factors could be used in different healthcare settings (EMS, emergency departments, internal medicine) to predict mortality.

The authors express their gratitude to RNDr. Marian Rybář, MHA, for providing the statistical analysis for this study.

Before conducting the study, the complete study protocol was reviewed and approved by the Ethical Committee of the University Hospital, Královské Vinohrady (“Eticka Komise Fakultni Nemocnice Královské Vinohrady”), with the reference number EK-VP/70/0/2020. As this was a retrospective study using anonymized data, written consent was not required and was waived by the Ethical Committee (EK-VP/70/0/2020).

The authors have no conflicts of interest to declare.

This study was partially supported by the Charles University research program COOPERATIO 34 – Internal disciplines.

E.Z. and A.S. contributed equally; E.Z., A.S., M.H., and I.R. substantially contributed to the conception and design of the study, performed the analysis and interpretation of data, and drafted and revised the work; M.R. and D.G. contributed to the conception of the study, collected data, and revised critically the work; U.H. and K.K. made contributions to the data collection and provided critical feedback for both the content and writing of the article.

The data that support the findings of this study are not publicly available due to information that could compromise the privacy of research participants, but they are available from Prof. Ivan Rychlík, MD, PhD (Corresponding author).

Eli Zolotov and Anat Sigal contributed equally to this work.