Common hematological values predict unfavorable outcomes in hospitalized COVID-19 patients.

COVID-19 can range from asymptomatic to life-threatening. Early identification of patients who will develop severe disease is crucial. A number of scores and indexes have been developed to predict severity. However, most rely on measurements not readily available. We evaluated hematological and biochemical markers taken on admission and determined how predictive they were of development of critical illness or death. We observed that higher values of readily available tests, including neutrophil:lymphocyte ratio; derived neutrophil index; and troponin I were associated with a higher risk of death or critical care admission (P < 0.001). We show that common hematological tests can be helpful in determining early in the course of illness which patients are likely to develop severe forms, as well as allocating resources to those patients early, while avoiding overuse of limited resources in patients with reduced risk of progression to severe disease.

Introduction

In December of 2019 in Wuhan, China, a respiratory illness caused by a new coronavirus (SARS-CoV-2) was described, giving place to a new syndrome known as COVID-19 (COronaVIrus Disease 2019) [1], spreading rapidly and reaching pandemic dimensions by March 2020 [2]. As of 1 January 2021, at least 81.9 million cases of (and 1.9 million deaths attributed to) COVID-19 had been reported globally [2]. In Mexico, 1.4 million confirmed cases and 126,851 deaths attributed to COVID-19 have occurred [3]. Most cases of COVID-19 are mild and self-limited, but progression to respiratory failure and death occur in up to a third of hospitalized patients [4,5].

COVID-19 has strained healthcare availability, particularly in the developing world and underserved areas, pushing resources to the brink of saturation. A number of tests and stratifying scores have been proposed; however, most rely on laboratory tests that are not readily available in non-academic settings. Moreover, there is no consensus on which can predict unfavorable outcomes early after hospital admission [6]. In that context, inexpensive and readily available tools to predict progression are crucial for clinical decision-making and allocation of limited resources. We aimed to evaluate the usefulness of blood biomarkers taken at the time of hospital admission as early predictors of eventual need of ICU, mechanical ventilation, or in-hospital death.

Here, we show preliminary but compelling evidence suggesting that simple hematological and serum tests determined on admission accurately predict progression to critical illness or death.

Methods

Study design

An ambispective cohort study evaluating the strength of prediction by laboratory variables of unfavorable outcomes among hospital-admitted COVID-19 patients.

Cohort description

Patients from a single COVID-19-designated reference center in Mexico City admitted with a confirmatory real time polymerase chain reaction (RT-PCR) nasopharyngeal test for SARS-CoV-2 and a non-contrast chest CT scan compatible with COVID-19 pneumonia. All patients were tested on admission for circulating levels of C-reactive protein (CRP), ferritin, D-dimer, lactate dehydrogenase (LDH), creatinine kinase (CK), troponin I, complete blood counts (CBC), arterial blood gases, as well as liver and kidney function tests. All patients had a new electronic health record created upon arrival (unless one already existed for other medical reasons) which was updated on a daily basis until discharge, transfer to a different facility, or death, and was the source of information shown hereby.

Ethics and data management

The study was approved by the Institutional Review Board (NER-3632-21-21-1). All data was de-identified, such that data analyses used only individual identifier codes for each patient. All authors had access to the complete dataset.

Study definitions and outcomes

We included every admitted patient between 21 March and 30 April 2020 that had a positive SARS-CoV-2 RT-PCR test. Thereafter, outcomes were updated up to 15 May. The primary outcome was a composite of the following: 1) mechanical ventilation; 2) admission to a critical care area; or, 3) death (whichever occurred first). Clinical and demographic variables registered on admission included: age; sex; dates of symptom onset and hospital arrival; vital signs (heart rate, temperature, respiratory rate, blood pressure, and oxygen saturation); presence of dyspnea, chest pain, and cough; body mass index (BMI); past medical history of diabetes and hypertension; percentage of lung parenchyma affected by CT scan as reported by the Department of Radiology [7,8]. For analysis, inflammatory hematological indexes were calculated with values obtained from CBCs, including: absolute neutrophil count (cells/mm3); absolute lymphocyte count (cells/mm3); neutrophil:lymphocyte ratio (NLR: absolute neutrophil count/absolute lymphocyte count); derived-neutrophil index (absolute neutrophil count/[total white blood cell count – absolute neutrophil count]); and platelet:lymphocyte ratio (absolute platelet count/absolute lymphocyte count). We also included D-dimer, ferritin, LDH, CRP, CK, and troponin I. All were treated as continuous variables.

Statistical analyses

We calculated crude incidence of the primary outcome, as well as of its composite parts. Standardized differences were calculated for clinical and laboratory variables between patients that had a primary outcome and those who did not.

We used a multivariate Cox proportional hazard's model with each of the laboratory variables to estimate the hazard ratio of the primary outcome. All Cox models were adjusted for age, sex, BMI, presence of diabetes, presence of hypertension, and oxygen saturation at hospital arrival, as we hypothesized these were the variables that could have a causal effect or be surrogates of causal variables on the measured laboratory values. Pneumonia severity scores were not included in the risk models as most include variables that we are already controlled for. Laboratory variables were not adjusted for each other. Cubic splines with three nodes were used to transform laboratory continuous variables and include them in each model. We did not input missing data and thus each model had a slightly different number of patients included. A P value ≤0.05 was considered significant.

Data analysis was performed with R software version 4.0.0. The script used for the analysis is available at https://github.com/isaac-nunez/hematol_val_covid19.

Results

Two hundred and eighty-two patients were hospitalized with a diagnosis of COVID-19 during the study period. Clinical, demographic and laboratory characteristics are described in Table 1 . All patients had symptoms on arrival to the Emergency Room. Patients who developed the primary outcome were significantly older and had a higher BMI, but had no differences in prevalence of diabetes or hypertension. Dyspnea, cough, and chest pain were also more common among patients who developed an outcome. Respiratory rate and other vital signs were also modestly higher in the outcome group. Oxygen saturation was lower among those that developed an outcome (Table 1).

Table 1

Baseline clinical and laboratory parameters.

Variable	Whole cohort (n = 282)	No primary outcome (n = 202)	Primary outcome (ICU/IMV/death, n = 80)	P valuea
Median follow up (days)	8 (6–11)	9 (7–12)	5 (3–7)	<0.001
Clinical characteristics
Female	105 (37.2%)	87 (40.3%)	18 (27.3%)	0.077
Age	50 (41–60)	52 (39.5–60)	52 (43–60)	<0.001
BMI (n = 280)	29 (26.6–31.4)	29 (26.6–31.2)	29 (27.5–32.1)	<0.001
Comorbidities
Diabetes	69 (24.5%)	49 (22.7%)	20 (30.3%)	0.273
Hypertension	79 (28%)	59 (27.3%)	20 (30.3%)	0.752
Symptoms on admission
Dyspnea	218 (77.3%)	159 (73.6%)	59 (89.4%)	0.012
Fever (n = 279)	254 (90.1%)	196 (90.7%)	58 (87.9%)	0.738
Cough (n = 281)	249 (88.3%)	185 (85.6%)	64 (97%)	0.009
Chest pain (n = 281)	66 (23.4%)	44 (20.4%)	22 (33.3%)	0.038
Headache (n = 281)	178 (63.1%)	133 (61.6%)	45 (68.2%)	0.329
Myalgia (n = 280)	204 (72.3%)	151 (69.9%)	53 (80.3%)	0.102
Diarrhea (n = 281)	52 (18.4%)	38 (17.6%)	14 (21.2%)	0.592
Vital signs
Heart rate	102 (89–115)	102 (88–114)	104 (91–119)	<0.001
Respiratory rate (n = 281)	24 (20−30)	24 (20–28)	28 (24–34)	<0.001
Temperature (n = 277)	37.1 (36.5–38)	37 (36.5–38)	37.2 (36.5–37.8)	<0.001
Mean blood pressure	92 (84–99)	92 (84–100)	92.5 (83–99)	<0.001
Pneumonia severity scales
PSI (n = 267)	2 (2–3)	2 (2–3)	3 (2–3)	<0.001
CURB65 n = 268)	1 (0–1)	0 (0–1)	1 (1–1)	<0.001
qSOFA (n = 278)	1 (0–1)	1 (0–1)	1 (1–1)	<0.001
NEWS (n = 276)	7 (5–8)	6.5 (5–8)	8 (6–9)	<0.001
Blood count and indexes
Hemoglobin (g/dL, n = 265))	17 (16.01–18.04)	17 (16.01–18.04)	17 (16.01–18.04)	<0.001
Platelets (x103/mm3, n = 265)	207 (164–264)	205 (165–259)	216 (156–267)	<0.001
Leucocytes (x103/ mm3, n = 265)	8 (7.01–11)	8 (7–10.05)	9.5 (7.04–12.05)	<0.001
Neutrophils (x103/ mm3, n = 265)	6.42 (4.81–8.89)	6.14 (4.69–8.43)	8.16 (5.47–10.73)	<0.001
Lymphocytes (x103/ mm3, n = 265)	1.1 (0.82–1.4)	1.2 (0.9–1.4)	0.93 (0.64–1.33)	<0.001
Monocytes (x103/ mm3, n = 264)	0.6 (0.47–0.8)	0.62 (0.48–0.81)	0.58 (0.45–0.77)	<0.001
Neutrophil:lymphocyte ratio (n = 265)	6.18 (3.88–9.76)	5.5 (3.59–8.36)	8.98 (6.145–11.745)	<0.001
Derived neutrophil index (n = 265)	4.32 (2.75–6.69)	3.76 (2.68–5.71)	5.92 (4.31–8.01)	<0.001
Platelet:lymphocyte ratio (n = 265)	187.46 (130.75–280.91)	180.66 (127.12–271.76)	222.435 (141.48–330.255)	<0.001
Lymphocyte:monocyte ratio (n = 264)	1.79 (1.335–2.45)	1.86 (1.395–2.625)	1.61 (1.1–2)	<0.001
Inflammatory markers on hospital admission
C-reactive protein (mg/L, n = 264)	12.01 (5.835–19.295)	10.62 (5.05–17.12)	16.02 (9.84–25.38)	<0.001
Ferritin (ng/mL, n = 262)	546.4 (275.3–931.9)	486.95 (250–877)	689 (380.7–1225.45)	<0.001
D-dimer (μg/mL, n = 251)	650 (425–1090)	632.5 (408–1050)	705 (467–1302)	<0.001
Lactate dehydrogenase (U/L, n = 253)	339 (265–432)	321 (252–395)	420 (302–571.5)	<0.001
Creatine kinase (IU/L, n = 244)	102 (57–234.5)	92 (55–223)	135 (72.5–238.5)	<0.001
Troponin I (ng/mL, n = 226)	6.06 (4.09–9.04)	6.025 (4.08–8.03)	9.035 (5.555–18.065)	<0.001
Arterial blood gases
PaO2 (mmHg, n = 249)	62.6 (54.6–76)	60.8 (53.7–75.8)	67.05 (57.85–81.8)	<0.001
PaCO2 (mmHg, n = 249)	31.9 (27.05–34.5)	32.1 (28–34.4)	31.015 (26.54–35.05)	<0.001
SaO2 without supplemental oxygen with pulse oximeter (%, n = 277)	88 (84–90)	88 (86–90)	85 (77–88)	<0.001

Unless specified otherwise, n = 282. Continuous variables were compared using Mann Whitney U test; categorical variables with χ2 test, and follow-up time with a median test. Abbreviations: ICU: admission to an intensive care unit; IMV: invasive mechanical ventilation; BMI: body mass index; PSI: pneumonia score index; CURB65: CURB-65 Score for Pneumonia Severity; qSOFA: Quick SOFA Score for Sepsis; NEWS: National Early Warning Score; PaO2: arterial partial pressure of oxygen; PaCO2: arterial partial pressure of carbon dioxide; SaO2: arterial oxygen saturation; FiO2: fractional inspired oxygen.

Continuous variables are expressed as medians and IQR (continuous clinical variables were rounded to the nearest integer for clarity, laboratory values were not) and categorical variables as absolute counts and proportions.

Among blood count values, total platelets, total leucocytes, total neutrophils, neutrophil-to-lymphocyte ratio, derived neutrophil index, and platelet-to-lymphocyte ratio were higher in the outcome group, while hemoglobin, total lymphocytes, total monocytes, and lymphocyte to monocyte ratio were lower in the outcome group. Inflammatory markers also were higher among those that developed an outcome. Of note, pneumonia score indexes were higher among patients who reached a primary outcome (Table 1).

All patients completed follow up by 15 May, with 80 (28.3%) developing a primary outcome, of which 66 were admitted to the ICU and 14 died without having been admitted to an ICU. Median time to primary outcome was five days (IQR 3–7). Two hundred (71%) patients were discharged from our center during the study period after a median of 9 days of hospitalization (IQR 7–12).

Univariate and multivariate Cox proportional hazard models are shown in Table 2 . The adjusted models are also presented graphically in Fig. 1 . Lower NLR, derived neutrophil index, LDH, and CRP, as well as higher total lymphocyte counts showed a protective hazard ratio of achieving the primary outcome.

Table 2

Univariate and multivariate Cox proportional Hazard models of complete blood count and inflammatory markers as predictors of the primary outcome in hospitalized patients with severe COVID-19.

Variable	Univariate cox model (HR, 95% CI)	Multivariate cox model (HR, 95% CI)
Neutrophil:lymphocyte ratio	n = 265	n = 258
2.5	0.59 (0.36–0.98)	0.66 (0.38–1.14)
5	1 (ref)	1 (ref)
10	1.79 (1.26–2.54)	1.6 (1.07–2.38)
15	1.96 (1.34–2.88)	1.75 (1.13–2.7)
20	1.96 (1.33–2.9)	1.78 (1.15–2.76)
Derived neutrophil index	n = 265	n = 258
1	0.3 (0.14–0.64)	0.43 (0.19–0.96)
3	0.46 (0.38–0.56)	0.54 (0.44–0.65)
5	0.66 (0.51–0.85)	0.66 (0.5–0.86)
10	1 (ref)	1 (ref)
15	1.39 (0.82–2.35)	1.48 (0.85–2.58)
Platelet:lymphocyte ratio	n = 265	n = 258
40	0.9 (0.44–1.86)	1.08 (0.49–2.37)
120	0.96 (0.73–1.26)	1.03 (0.77–1.39)
180	1 (ref)	1 (ref)
240	1.02 (0.83–1.27)	0.98 (0.77–1.26)
300	1.04 (0.78–1.38)	0.98 (0.72–1.35)
Lymphocyte:monocyte ratio	n = 264	n = 257
0.5	1.63 (0.92–2.88)	1.99 (1.08–3.7)
1	1.26 (0.94–1.69)	1.39 (1.01–1.91)
1.5	1 (ref)	1 (ref)
2	0.89 (0.71–1.1)	0.85 (0.69–1.06)
2.5	0.91 (0.75–1.11)	0.92 (0.74–1.13)
Total neutrophils (x103/mm3)	n = 265	n = 258
2000	1.15 (0.48–2.72)	1.71 (0.68–4.28)
5000	1 (ref)	1 (ref)
10,000	1.15 (0.85–1.55)	0.89 (0.64–1.24)
15,000	1.79 (1.19–2.7)	1.54 (0.97–2.44)
20,000	2.81 (1.33–5.95)	2.68 (1.18–6.06)
Total lymphocytes (x103/mm3)	n = 265	n = 258
300	2.53 (1.48–4.3)	2.96 (1.6–5.46)
500	1.87 (1.31–2.66)	2.07 (1.38–3.11)
1000	1 (ref)	1 (ref)
1500	0.93 (0.78–1.1)	0.97 (0.81–1.17)
2000	1.22 (0.67–2.21)	1.46 (0.82–2.6)
Creatine kinase (IU/L)	n = 244	n = 237
20	0.73 (0.5–1.07)	0.81 (0.52–1.25)
60	0.86 (0.7–1.07)	0.9 (0.7–1.16)
100	1 (ref)	1 (ref)
200	1.27 (0.97–1.68)	1.17 (0.86–1.61)
500	1.27 (0.86–1.87)	1.1 (0.7–1.73)
Lactate dehydrogenase (U/L)	n = 253	n = 246
150	0.49 (0.2–1.17)	0.49 (0.2–1.17)
250	0.72 (0.52–0.99)	0.72 (0.52–0.99)
350	1 (ref)	1 (ref)
500	1.21 (0.87–1.71)	1.21 (0.87–1.71)
700	1.32 (0.94–1.85)	1.32 (0.94–1.85)
Ferritin (ng/mL)	n = 262	n = 255
100	0.94 (0.6–1.47)	1.19 (0.69–2.06)
200	0.96 (0.7–1.32)	1.13 (0.77–1.67)
400	0.99 (0.9–1.08)	1.04 (0.94–1.16)
500	1 (ref)	1 (ref)
1000	1.03 (0.78–1.37)	0.91 (0.64–1.29)
C-reactive protein (mg/L, n = 264, 257)
0.5	0.74 (0.37–1.47)	0.74 (0.37–1.47)
1	0.75 (0.39–1.43)	0.75 (0.4–1.42)
10	1 (0.83–1.2)	1 (0.83–1.2)
20	1.26 (0.93–1.71)	1.26 (0.93–1.71)
30	1.51 (0.89–2.54)	1.51 (0.89–2.55)
D-dimer (μg/mL)	n = 251	n = 244
200	1.06 (0.69–1.63)	1.33 (0.78–2.25)
500	1 (ref)	1 (ref)
2500	0.95 (0.66–1.39)	0.62 (0.39–1.98)
5000	1.04 (0.73–1.49)	0.67 (0.42–1.05)
10,000	1.3 (0.9–1.89)	0.82 (0.5–1.33)
Troponin I (ng/mL)	n = 226	n = 220
2	0.56 (0.31–1.01)	0.59 (0.29–1.21)
5	1 (ref)	1 (ref)
10	1.95 (1.45–2.62)	1.85 (1.28–2.66)
15	2.47 (1.58–3.85)	2.28 (1.32–3.96)
20	2.56 (1.63–4.02)	2.36 (1.35–4.12)

‘n = x, y’ indicates the number of patients included in the univariate and multivariate model, respectively. These numbers differ in some cases because of missing values of the variables included in the multivariate model.

Fig. 1

Risk of ICU admission or death for hematological indexes and inflammatory markers (multivariate Cox models). *Adjusted for age, sex, BMI, hypertension, diabetes, and oxygen saturation at the moment of patient admission to the hospital. Abbreviation: NLR: neutrophil:lymphocyte ratio.

Discussion

Several inflammatory markers have been evaluated as predictors of mortality among hospitalized patients with severe [[9], [10], [11], [12]], as well as non-severe COVID-19 [[13], [14], [15]]. Circulating levels of cytokines and inflammatory biomarkers have been shown to successfully predict severity [16], but unfortunately these are not readily available outside of tertiary-care medical centers.

Here, we observed that NLR, the derived neutrophil ratio, lymphocyte-monocyte ratio, total neutrophil count, as well as total lymphocyte count successfully predict outcomes at the moment of hospital admission for COVID-19.

The NLR, lymphocyte-monocyte ratio, and platelet-lymphocyte ratio are thought to reflect physiologic stress. Stress leads to increased circulating cortisol, in turn triggering an increase in circulating neutrophils and a reduction in lymphocytes. Higher values of NLR, lymphocyte-monocyte ratio, and platelet-lymphocyte ratio are commonly seen in critically ill patients. As they are not specific to a particular disease, those indexes are not useful for diagnosing, but can guide in weighting the severity of a known an inflammatory illness [[17], [18], [19]].

NLR has been previously shown to predict in-hospital critical illness and mortality among COVID-19 patients [[20], [21], [22]]; NLR has also been incorporated into prediction scores of critical illness or mortality [12,23]. Lymphopenia has been described as a predictor of mortality in some models [[24], [25], [26]]; accordingly, we observed that lymphopenia is a good predictor of the primary outcome even in the absence of other laboratory biomarkers. To our knowledge, the utility of the derived neutrophil index, as well as the lymphocyte-monocyte ratio -two easily computed indexes obtainable from a simple and widely available CBC- have not been previously shown to predict unfavorable outcomes.

We observed that troponin I predicts worse outcome, probably reflecting myocardial injury. Myocardial injury has been linked to worse outcomes in other studies [26,27]. Nonetheless, troponin I measurement should be judicious and limited to patients that have clinical characteristics compatible with myocarditis. Interestingly, D-dimer showed a U-shaped curve with lower risk of outcome in the range of 1000–2000 μg/mL compared with lower and higher values. CRP and LDH were not associated with the development of primary outcomes, with the exception of the extremely high (but rarely observed) values.

We found a discrepancy between PaO2 and SaO2, with PaO2 being paradoxically higher in the primary outcome group, while SaO2 was higher in the group that did not develop an outcome. That may be explained by the moment in which those values were measured. Patients were frequently given supplementary O2 prior to the arterial blood gas analysis. A PaO2/FiO2 ratio analysis would be useful to confirm this, but unfortunately we lack accurate information on FiO2 values on admission. In our context, ambient air SaO2 is a more accurate value and thus is the one we included in the risk model.

Our study has limitations. First, it is partly retrospective. Also, we aimed to evaluate the usefulness of markers individually, and we did not incorporate them into a score with other laboratory values. Additionally, we lack information on treatment prior to hospital admission, which could have modified certain laboratory values. We are underpowered to determine if individual markers convey the same information. This would require an additional analysis, such as principal component analysis, which is beyond the scope of the present work.

This study also has strengths. Our sample is representative of Hispanic patients; also, ours is a public hospital; all patients had access to the best medical care regardless of their financial or insurance status. We had a very low rate of missing individual data; between-hospital transfers were exceedingly rare during the study period, such that if patients were discharged it is highly unlikely that they were sent to another hospital for critical care or further treatment. Also, discharges in general occurred after 72 h of in-hospital clinical stability, and while clinical worsening after discharge could be possible, it is unlikely. This kind of analysis allows for easier comparisons between patients that present different values of a certain variable.

In conclusion, here, we show that simple hematological markers measured on admission to the Emergency Room can accurately predict progression to critical illness and mortality in hospitalized patients with COVID-19.