Haematological and biochemical pathology markers for a predictive model for ITU admission and death from COVID‐19: A retrospective study

Abstract Coronavirus disease (COVID‐19) caused by SARS‐CoV‐2 has affected over 227 countries. Changes in haematological and biochemical characteristics in patients with COVID‐19 are emerging as important features of the disease. This study aims to identify the pathological findings of COVID‐19 patients at Bedford Hospital by analysing laboratory parameters that were identified as significant potential markers of COVID‐19. Patients who were admitted to Bedford Hospital from March–July 2020 and had a positive swab for COVID were selected for this study. Clinical and laboratory data were collected using ICE system. Multiple haematological and biochemistry biomarkers were analysed using univariate and multivariate logistic regression to predict intensive therapy unit (ITU) admission and/or survival based on admission tests. Neutrophil‐to‐lymphocyte ratio (NLR) and C‐reactive protein were elevated in most patients, irrespective of ITU status, representing a common outcome of COVID‐19. This was driven by lymphopenia in 80% and neutrophilia in 42% of all patients. Multivariate logistic regression identified an increase in mortality associated with greater age, elevated NLR, alkaline phosphatase activity and hyperkalaemia. With the area under the receiver operating characteristic (ROC) curve of 0.706 +/− 0.04117, negative predictive value (NPV) 66.7% and positive predictive value (PPV) 64.9%. Analysis also revealed an association between increases in serum albumin and potassium concentrations and decreases in serum calcium, sodium and in prothrombin time, with admission to ITU. The area under the ROC curve of 0.8162 +/− 0.0403, NPV 63.3% and PPV 80.5%. These data suggest that using admission (within 4 days) measurements for haematological and biochemical markers, that we are able to predict outcome, whether that is survival or ITU admission.


INTRODUCTION
On 31 December 2019, Wuhan Municipal Health Commission, China, reported a cluster of pneumonia cases of unknown aetiology to the World Health Organisation (WHO) [1,2]. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was identified as the causative agent of this infection [3]. WHO director declared the SARS-CoV-2 as a public health emergency of international concern on 30 January 2020, which is the WHO's highest level of alarm regarding the emer- Coronaviruses (CoVs) are a large viruses group belonging to the Coronaviridae family, which causes severe respiratory disease called COVID-19 [4]. The clinical presentation of symptomatic patients of COVID-19 is fever, which is defined as an axillary temperature of 37.5 • C or higher, dry cough, shortness of breath, dyspnoea, loss of smell and/or taste, fatigue, muscle pain and pneumonia [5,6]. Respiratory droplets and human-to-human contact are the main routes of transmission of the virus [3,7].
Symptomatic patients can require hospitalisation due to the acceleration of the infection and may subsequently require admission to an intensive therapy unit (ITU). A minority of patients develop severe pneumonia, acute respiratory distress syndrome, acute respiratory failure, refractory metabolic acidosis, coagulopathy, septic shock, multiple organ failure and consequently death [3,[8][9][10][11]. Patients who are above the age of 65 with previous co-morbidities such as diabetes, cancer, cardiovascular diseases, respiratory disease, leukaemia and myelodysplastic syndrome are more likely to develop serious clinical complications, and they constitute between 50% and 70% of deaths [12,13].
As of 24 January 2022, 228 countries across the world have positive COVID cases: 349,641,119 confirmed cases and 5,592,266 deaths worldwide (2). Whilst in the United Kingdom there were 15,859,292 positive cases and 153,862 deaths within 28 days of a positive test.

METHODS
Retrospective data were collected between 21 March 2020 and 19 July 2020, which has been classified as the first wave at Bedford

Clinical laboratory data
Clinical and laboratory data required for the study were collected during routine clinical consultations and using Integrated Clinical Environment (ICE), which is Bedford Hospital's Pathology reporting system.
Blood was collected within 4 days of admission to Bedford Hospital.

Statistical analysis
The parameters were expressed as median with IQR. Differences in the parameters between groups were determined by an independent t-test ROC analysis was performed, and AUC, positive predictive and negative predictive power were calculated. Both Hosmer-Lemeshow and Log-likelihood ratio (G squared) were used to assess goodness of fit for the logistic regression models used.

RESULTS
In this study we investigated patients that were admitted with a positive polymerase chain reaction (PCR) test for SARS-Cov-2 between 21/03/20 and 19/07/20 into Bedford Hospital trust; this resulted in 806 records. Figure 1 shows the PRISMA diagram generated to classify patients into non-ITU/ITU and survived/non-survivors. From these 806, 58 (7%) patients were admitted to ITU, 26 (45%) were discharged (ITU -survived), and 32 (55%) passed away (ITU -passed away). Again from these 806, 267 (33%) met our criteria and were not admitted to ITU; of these 78 (29%) passed away, and 50 were randomly selected for inclusion in our study (non-ITU -passed away). Of the 189 that survived, 146 (77%) were not readmitted or passed away 6 months post-discharge; of these 50 were randomly selected for inclusion in our study (non-ITU -survived). Clinical characteristics are presented in  Figure 2A). Using parameters from the univariate analysis, which had a p value <0.1, we were able to generate a model of survival using multivariate logistic regression analysis ( Table 3). The AUC was 0.706 +/− 0.04117 (p < 0.001 - Figure 2B), the model had a negative predictive power  p < 0.05) were significantly associated with being in ITU (Table 4). Again using parameters that had a p value <0.1 in the univariate analysis, we generated a model that looked for an association between admission data and admission to ITU ( Table 5). The area under the ROC was 0.8169 +/− 0.0403 (p < 0.001 - Figure 3) with a negative predictive power of 63.33%, a positive predictive value (PPV) of 80.46%

Modelling risk of admission to ITU
and an accuracy of 75.2%. This analysis again revealed that being male increased the chances of being in ITU, but that ITU was associated with younger patients. Increased serum albumin and potassium, whilst a decrease in PT, adjusted calcium, ALT and sodium concentrations were associated with a greater risk of being admitted to ITU.

DISCUSSION
As of January 2022, SARS-Cov-2 has infected close to 350 million people worldwide and led to 5,592,266 deaths worldwide (2). Whilst more (2020) who demonstrated that increased potassium was associated with a great chance of death in ITU patients [20]. Furthermore, Ravioli and colleagues demonstrated that hyponatraemia and hyperkaliaemia are related to an increase in admission to ITU and death, due to community-acquired pneumonia [21]. Therefore these risk factors might be an indication of the development of pneumonia or an indication of the extent of damage by COVID-19. Both of these studies are consistent with our modelling for survival and ITU admissions, which both pointed to higher serum potassium as a risk factor for ITU admission and death.
On their own, most of the biomarkers recorded in this study are unable to significantly differentiate between the four different cohorts. The power of the multivariate logistic analysis is that it pools data on multiple different risk factors and can better differentiate groups and potentially predict outcomes, whether that is survival or ITU status. In the future, the addition of further clinical markers, such as oxygen satu-ration, heart rate, etc., would further improve this multivariate analysis and drive a great predictive potential.

Limitations of our study
Logistic analysis revealed that if you were younger, you were more likely to end up in ITU. This however might be influenced by the fact that there is a greater age of patients in the non-ITU group who scored high on the clinical frailty scale or passed away, meaning that they did not meet the criteria to be admitted to ITU or died before being transferred to ITU. Another limitation of this study is that this is only carried out in a single hospital setting during the first wave, and there is a lack of a control group to compare to. This is due to the rapid emerging nature of the first wave, and future studies would use equally weighted groups, along with a control group for comparison of haematological and biochemical parameters. Our data here were based on blood tests from within the first 4 days of admissions, where testing within the first day might provide more reliable indicators of disease. Furthermore, due to the nature of COVID-19 infections, there was a slight bias towards more men within the study; this means it is more difficult to generalise specifically to both sexes. Future analysis will require more patient data inclusion and a need for more than one centre. Other future analysis will have to take into account whether quicker testing would have altered the results, and whether the predictive effects are equally useful for both sexes. All of these points would help to demonstrate the effectiveness of the model in its predictions and the ability to predict COVID-19 outcomes.

CONCLUSION
It is vital in the ongoing COVID-19 pandemic that there is an ability to tightly monitor patients that are admitted and predict the chance that they will die or need ITU. Using multivariate logistic analysis we have developed a model for predicting outcomes. Whilst differences in severity and mode of cellular damage might be different with various strains, the findings in the study underpin the underlying damage to the lungs and possible pneumonia, from COVID-19. In November and December 2021, The UK approved two new therapeutics targeted towards COVID-19: 1) RNA-dependent RNA polymerase inhibitor: molnupiravir and 2) neutralizing antibody: sotrovimab [22,23]. The effectiveness of both of these new therapeutics is in delivery to high-risk patients to maximise survival, whilst delivering costeffective treatments. Our model could allow stratification into patients that would most benefit from these treatments, highlighting at-risk patients, based on admission haematological and biochemical profiles.
For example, in the Sotrovimab trial, it was shown that in the placebo group 13 of the 21 patients that died, died due to covid19-related pneumonia, which might have been linked to changes in NLR, CRP and potassium as shown here [23]. Data from our study would also help to potential monitor other experimental or repurposed therapeutics such as nafamostat mesylate [24,25]. For example, In four consecutive SARS-Cov-2 positive critically ill patients, administration of nafamostat mesylate was associated with hyperkaliaemia. We have identified hyperkaliaemia as a risk factor for death in this study [26,27], potentially highlighting that this therapeutic might not offer the same level of protection as the others mentioned here.