The neutrophil to lymphocyte ratio is an independent predictor for severe COVID-19

In total, 213 patients (3 patients were excluded due to missing NLR data at admission) with confirmed COVID-19 from 21 hospitals in Sichuan Province between 16 January 2020 and 15 March 2020 were included in the analysis: among them, 81 patients were defined as having severe COVID-19. The definitions of COVID-19 and severe COVID-19 were described in a previously published article [14]. The protocol was approved by the Ethics Committee of the West China Hospital of Sichuan University and the participating hospitals, and informed consent was obtained. The study was registered and the registration number was ChiCTR2000029758.

We conducted the study based on the electronic data capture and analysis system for each included patient. Demographic characteristics, comorbidities, symptoms, signs and laboratory findings at admission were collected. The NLR is the ratio of neutrophil count to lymphocyte count, and platelet to lymphocyte ratio (PLR) is the ratio of platelet count to lymphocyte count.

Categorical variables are expressed as numbers and percentages, continuous normally distributed variables are expressed as the mean and standard deviation, and continuous skewness-distributed variables are expressed as the median and interquartile range. The χ²-test, t‑test and Mann–Whitney U test were used to compare the differences between severe and non-severe COVID-19 groups. A 2-tailed P value lower than 0.05 was considered to be statistically significant. Variables found to be statistically significant in univariate analysis were included in the multivariate logistic regression analysis to identify the independent risk factors for severe COVID-19. The odds ratio (OR) and its 95% confidence interval (CI) were used as binary parameters. The Cochran-Armitage test for trend and the area under the receiver operating characteristic (ROC) curve were used to analyze the linear relationship of NLR level with the rate of severe COVID-19 and the ability of NLR to predict severe COVID-19, respectively.

PubMed, EMBASE, Web of Science, MedRxiv and BioRxiv, the WanFang database and the China National Knowledge Infrastructure (CNKI) were searched through inception to May 2020. The search terms of ‘neutrophil-to-lymphocyte ratio’ OR ‘NLR’ and ‘COVID-19’ were used to identify studies that reported the relationship between NRL and the severity of COVID-19. The reference lists of the included studies were examined to identify any other potentially qualified studies.

Two reviewers reviewed the titles and abstracts independently to determine whether they needed full-text review according to the following inclusion and exclusion criteria. After that the full-text review of the studies was completed to identify the articles included by the same two reviewers. When the opinions were inconsistent, the question was discussed and determined by a third reviewer.

Data were extracted using the predesigned data collection table. Two reviewers extracted data from all included studies independently. The collected data included author, year, study design, study location, study samples, NLR values, age and sex. When the opinions were inconsistent, the question was discussed and determined by a third reviewer.

Studies were evaluated by the Newcastle–Ottawa assessment tool, for which the maximum score was 9 points and a score of 7 or greater was considered to be a low risk of bias [16]. Two reviewers assessed the risk of bias of all of the included studies independently; when opinions were inconsistent, the question was discussed and determined by a third reviewer.

The software Stata version 16.0 (STATA Corporation, College Station, TX, USA) and Meta-Disc, version 1.4 (Unit of Clinical Biostatistics team of the Ramón y Cajal Hospital, Madrid, Spain) were used to conduct the statistical analysis. The median and interquartile range were transformed into mean and standard deviation based on a previously proven formula; [17, 18]. Standardized mean differences (SMDs) and corresponding 95% confidence intervals (CIs) were used to evaluate the differences in NLR between the severe COVID-19 and non-severe COVID-19 groups. Heterogeneity among the studies was evaluated using Cochran’s Q test and the I² statistic. An I² index of 50% or greater and a p-value less than 0.1 were taken to indicate significant statistical heterogeneity. When statistical heterogeneity existed, a random effects model was used. Publication bias was assessed by Egger’s and Begg’s tests [19]. Subgroup analyses were performed according to the mean age (> 50 years or ≤ 50 years) and study location (the study included patients in Hubei or not).To further analyze the predictive power of the NLR in severe COVID-19, 2 × 2 data points were obtained from the studies. The area under summary receiver operating characteristic (SROC), Q index, sensitivity and specificity with 95% CI were calculated.

Fig. 3 shows the work flow of the citations reviewed and the studies included. Our search strategy yielded 327 citations after de-duplication, and 35 potential citations were full-text reviewed. Finally, 16 citations were included. The study from Yabing Guo and Wenhua Liang reported 2 sets of data. Therefore, 19 datasets were included in the quantitative synthesis (18 datasets from 16 studies combined with our case-control study) [6, 20‐34].

A total of 7049 (6836 patients in the previous study and 213 patients in our study) patients were included, among whom 1211 (1130 patients in the previous study and 81 patients in our study) patients had severe COVID-19. The sample size of the individual studies ranged from 63 subjects to 1590 subjects. Only one study was from Britain, and all other studies were from China, with 8 studies included patients in Hubei. One of these studies was prospectively conducted, and all of the others were retrospectively. All studies reported patient ages, and the mean age of the patients in 5 studies was greater than 50 years old (see Table 2).

The Newcastle–Ottawa assessment tool was used to evaluate the quality of the included studies. All studies were considered to have a low risk of bias (Additional file 3). Among the included studies, no significant publication bias was found (Egger’s test: P = 0.924; Begg’s test: P = 0.889; Additional file 4).

A total of 7049 patients from 19 datasets were analyzed. The pooled results showed that NLR were higher in patients with severe COVID-19 than in those with non-severe COVID-19 (SMD = 1.10, 95% CI: 0.90–1.31, P < 0.001). High heterogeneity existed (I² = 87.1%, P < 0.001, a random-effects model was used) (see Fig. 4).

To determine other parameters that might affect the predictive power of NLR for severe COVID-19, subgroup analyses were conducted according to the mean age (> 50 years or ≤ 50 years) and study location (the study included patients from Hubei or not). The results showed that the mean age and the study location did not affect the predictive power of the NLR for severe COVID-19 (see Table 3 and Additional file 5).

The 2 × 2 data were obtained from 6 studies. In total, 1698 patients, including 278 severe COVID-19 patients, were included in the final analysis (see Additional file 6). The area under the SROC was 0.802, and the Q index was 0.738, with a sensitivity of 0.67 (95% CI: 0.61–0.72) and a specificity of 0.75 (95% CI: 0.73–0.78) (see Fig. 5).