Introduction
Early in the SARS-CoV-2 pandemic, Siddiqi et al. proposed a model on the natural history of SARS-CoV-2 infection that comprises a viral and an inflammatory phase and three stages: early infection, which presents with systemic, mild symptoms, a pulmonary stage, when pneumonia and local inflammation develops, and systemic hyperinflammation that involves only the minority of patients who progress to severe illness with multiorgan involvement [
1].
Several markers have been proposed to identify the progression towards this last stage, and C-reactive protein (CRP), D-dimer, lactate dehydrogenase (LDH), troponin I, leucocyte alterations and thrombocytopenia have already been confirmed to be of prognostic significance [
2,
3]. Leucocyte alterations involve both the lymphocyte and the neutrophil populations. Neutrophils, in fact, possess several inflammatory and pro-thrombotic properties such as the production of neutrophil extracellular traps (NETs) and reactive oxygen species (ROS), which can lead to sustained inflammation during SARS-CoV-2 infection [
4‐
6]. Lymphocytes, on the other hand, mainly exhibit the hallmarks of immune exhaustion with CD4 and CD8 T-cell loss, which has been linked to an amplified inflammatory response due to the persistent viral load and the consequent neutrophil stimulation [
4,
5].
In this context, the neutrophil-to-lymphocyte ratio (NLR) and the platelet-to-lymphocyte ratio (PLR) could be used as prognostic factors. Thanks to their availability, low cost and reproducibility, they already have been proposed as inflammatory and prognostic markers in a variety of specialties (traumatic brain injury, cardiovascular diseases, hepatic and pancreatic diseases, solid cancer, rheumatic diseases and chronic obstructive pulmonary diseases) [
7‐
12].
Several studies have shown a similar trend in COVID-19. Three small Chinese studies performed early in the pandemic demonstrated that a high NLR correlates with severity and risk of progression [
13‐
15]. Subsequent studies confirmed this finding, adding that an upward trend correlates with mortality and that peak NLR is associated with mechanical ventilation [
16‐
18]. A large meta-analysis including 30 studies and 5570 patients confirmed that NLR accurately determines COVID-19 severity [
19].
Similarly, several studies have determined that a baseline low platelet count and a downward trend are associated with severe COVID-19 [
20,
21]. Studies that focus specifically on PLR are scant, but demonstrate that a high PLR and a big PLR increase during treatment is associated with a longer hospital stay and more severe disease [
22,
23].
Our aims are:
1.
To describe NLR and PLR trends at different timepoints during the hospital stay of patients with COVID-19;
2.
To compute for each timepoint an optimal cutoff that predicts the risk of four outcomes: use of mechanical non-invasive ventilation (in the form of continuous positive airways pressure—CPAP), ICU admission, invasive ventilation and death. For each cutoff, we will also produce a receiver operating characteristic (ROC) curve and we will compute the area under the receiver operating characteristic curve (AUROC) as well as the sensitivity, specificity, accuracy, positive predictive value (PPV) and negative predictive value (NPV);
3.
To compare NLR and PLR performance as predictors.
Methods
Design and Setting
This is a cohort retrospective study, conducted at the Fondazione IRCCS Policlinico San Matteo (Pavia, Italy), an academic hospital. All procedures were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.
Eligibility
All adult patients (older than 18 years) with pneumonia and a positive RT-PCR SARS-CoV-2 nasopharyngeal swab who were admitted to the hospital from 21 February 2020 to 18 May 2021 were included in the study. Patients without at least one value of neutrophils, lymphocytes and platelets were excluded. No patients were excluded on the basis of comorbidities that could impact their complete blood count, in an effort to improve generalisation of our results.
Ethics Compliance
All patients provided informed consent for the use of clinical data for scientific purpose according to hospital policy. Administrative data about our hospital’s patients with COVID-19 are collected in a registry (SMACORE), approved by the Fondazione IRCCS Policlinico San Matteo’s ethics committee with protocol number 20200046877, within the framework of the PERISCOPE (Pan-European Response to the ImpactS of COVID-19 and future Pandemics and Epidemics) European project, funded by Horizon 2020 (grant agreement 101016233).
Data and Variables
Data extracted from the SMACORE registry include epidemiological data (age and gender), duration of hospital stay and wards of stay. Comorbidities data were extracted using ICD-9-CM codes. We selected diabetes mellitus (250), chronic obstructive pulmonary disease (from 490 to 496), hypertension (from 401 to 405) and ischaemic heart disease (from 410 to 414).
Outcome data include ICU stay, death, use of intubation/tracheostomy/invasive ventilation and non-invasive mechanical ventilation (CPAP). ICD-9-PCS codes were used to identify patients who underwent intubation (96.04, 96.05), tracheostomy (31.1), invasive ventilation (96.70, 96.71, 96.72) and non-invasive mechanical ventilation (93.90). In the context of this study, intubation, tracheostomy and invasive ventilation are considered under a single outcome called “invasive ventilation”. No missing data were detected in all the extracted variables.
Laboratory data, extracted from the patients’ records, include neutrophils, lymphocytes and platelet count. Laboratory data were extracted at four timepoints: admission (0–24 h), 48 h (24–72 h), 7 days (120–216 h) and discharge (within the last 48 h before discharge). If more than one value of neutrophils, lymphocytes and platelets was available at each timepoint, the mean value was used to compute NLR and PLR.
Data were anonymised before use.
Statistical Analyses
Categorical data are presented as absolute frequencies and proportions. Continuous data are presented by mean and standard deviation if normally distributed or by median and interquartile range if not (Shapiro test is used to assess the normality hypothesis).
To study the ability of NLR and PLR to distinguish the patients’ outcomes at each timepoint, a non-parametric (Mann–Whitney U) test was used (univariate analysis). We performed a subset of univariate analyses to assess the statistical difference in NLR and PLR values according to the comorbidities.
ROC curves were built for NLR and PLR at each timepoint (minus discharge) for males and females in order to identify a cutoff to distinguish severe and non-severe disease. We defined our optimal cutoff as the NLR or PLR value with the highest sensibility and a minimum specificity of 0.5, to have the smallest possible number of false negatives. The statistical significance of the cutoff values was assessed with the chi-square test, applied to 2 × 2 adjacency tables from which classification metrics, such as sensitivity and specificity, were extracted.
We also compared NLR and PLR ROC curves with DeLon’s test to define which is the best predictor of each outcome at the four timepoints.
All statistical tests were two-sided, and a p value of 0.05 was used.
Data pre-processing, descriptive analysis and univariate analyses were handled with Python 3.7.
ROC curves and cutoff analyses were performed on R 4.0.5.
Discussion
We showed that NLR and PLR are able to distinguish between patients with mild/moderate COVID-19 and patients with severe disease (with need of invasive/non-invasive ventilation or ICU admission and patients who died). In fact, both ratio show higher values in the population with severe disease. An NLR that is continually increasing at successive timepoints, in particular, is able to predict death.
We also computed cutoffs for males and females for the four outcomes at each timepoint. Males’ NLR cutoffs are higher than females’ of at least 1 point. Cutoffs’ sensitivity is also higher in males. At each timepoint an “easy to remember” value (e.g. 6.3 and 7.0 at admission for females and males, respectively, or 5.3 and 7.3 at 48 h for males and females, respectively) could be quite good at predicting all the outcomes, with a sensitivity that increases progressively at successive timepoints (the minimum is 0.61 for females at admission for ICU admission; the maximum 0.91 for males at 7 days for invasive ventilation). The positive predictive values are low, but the negative predictive values range from 0.73 to 0.98, making our cutoffs efficacious in
excluding patients at risk if their NLR is lower. The same can be said for PLR cutoffs, even though, differently from NLR, they start out similar for males and females. Moreover, at a single timepoint, PLR cutoffs can be quite different depending on the outcome (e.g. 250 versus 293 for males at 48 h for CPAP and death, respectively), possibly making it easier to distinguish different grades of severity.
Finally, we showed that NLR is generally better than PLR at predicting severity.
A Chinese study aimed at finding NLR and PLR reference values for healthy subjects showed that both are sex and age dependent [
24]; successive studies confirmed that this distinction is maintained in disease [
25], in line with our findings.
Four Chinese studies plotted NLR and PLR from admission to post-discharge and obtained trends similar to ours, with stable and lower values for patients with non-severe disease and higher and increasing values for patients with severe disease [
16,
22,
26,
27].
Regarding the predictive value of the ratios and their cutoffs, several studies plotted ROC curves for NLR and PLR, finding that they perform well in predicting severity and death, with NLR performing better than PLR in those studies that compared them [
14,
16,
17,
23,
27‐
29]. A Turkish study on 306 patients found that at admission NLR (but not PLR) significantly predicted COVID-19 pneumonia, and they computed a cutoff of 1.73 [
28]. A similar study on 69 patients found a cutoff for severe disease of 3.3 for NLR and 180 for PLR, and another on 61 patients used 3.13 for NLR [
13,
23]. All of these studies, however, did not aim at distinguishing patients with severe disease from those with non-severe disease, but patients with or without pneumonia, which could explain the difference between their cutoffs and ours, which are much higher. In fact, a study which tried to distinguish different grades of severity (common cases, severe-non-ICU patients and severe-ICU patients) found values of NLR and PLR that increase with severity, although they did not compute any cutoffs [
27]. A cutoff was actually computed by Ye et al., who found a NLR cutoff of 7.13 for death and 7.28 for invasive ventilation, both values that are very similar to our own [
16]. An Indian study limited to patients with severe disease found a cutoff of 5.1 for mortality, but it did not specify at which point during hospitalisation the NLR was extracted, and two Iranian ones have the same limitation and, moreover, are focused only on patients with specific characteristics [
17,
30,
31]. Finally, Yldiz et al. computed a 5.94 NLR cutoff for mortality on a derivation cohort and subsequently validated it in a successive cohort [
29].
All these studies have the limitation of being retrospective and limited to a single centre with no more than a few hundred patients. Our study presents the same limitations, but dealt with more than 2000 patients and avoided stringent exclusion criteria, thus limiting the confounding factors and outliers and expanding the generalisation of our results as much as possible. We also divided our results for males and females, which we consider a strength as it takes into consideration the physiologic differences of NLR and PLR and adds a further level of stratification. Moreover, our cutoffs not only deal with severity and mortality but allow for a more tailored approach, dividing patients with severe disease in those who can still be treated in an ordinary ward, possibly with the implementation of CPAP, and those who need early referral to the ICU.
Another limitation is selection bias, since we selected our patients and stratified them in severity outcomes based on ICD9 codes, which can be inaccurate.
Finally, another confounder could be the use of anti-inflammatory or myelotoxic drugs, but this effect should be corrected by the number and the periodicity of observations. In fact, the greatest usefulness of the cutoffs lies in the admission and 48 h timepoints, where the steroid-induced neutrophilia is not yet present.
Acknowledgements
We thank the participants of the study.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.