Utility of the New Early Warning Score (NEWS) in combination with the neutrophil–lymphocyte ratio for the prediction of prognosis in older patients with pneumonia

Study design and setting

We performed a retrospective study of the medical records of patients who were stated to have pneumonia at the time of admission.

Selection of study sample

Data were obtained for older patients (age ≥65 years) admitted to the General Medicine Department of our hospital between 1 April 2018 and 31 March 2020 who had been diagnosed with pneumonia. We identified the pneumonia to be CAP or nursing and healthcare-associated pneumonia.19 We diagnosed pneumonia comprehensively using respiratory symptoms (cough, sputum, dyspnoea and tachypnoea), other symptoms and signs (fever, tachycardia, anorexia, decreases of activities of daily living, impaired consciousness and incontinence), laboratory findings (markers of an inflammatory reaction, leucocytosis or leucopenia) and imaging findings (new or the worsening of pre-existing infiltrates on chest X-ray or CT).20 The diagnosis at admission was coded by the attending physician using this information. The participants were treated routinely, according to the recommendations of the major Japanese clinical guidelines,21 and we followed them until they were discharged from the hospital.

The exclusion criteria were as follows: (1) transfer from another hospital, discharge from the hospital within 10 days or a diagnosis of hospital-acquired pneumonia; (2) the presence of empyema, pulmonary tuberculosis, pulmonary oedema, pulmonary thromboembolism or non-infectious interstitial pneumonia; (3) previous administration of antibiotics by another physician; (4) immunosuppression, treatment with a corticosteroids, or chemotherapy within the preceding 90 days or radiotherapy; (5) the presence of liver disease or a haematological disorder that might affect the incidences of inflammation measured22; and (6) missing baseline data regarding any element of the NEWS or NLR.

For older patients with pneumonia, with an effect size of 0.25, an α level of 0.05 (5%), a β level of 0.20, a power of 80% and a two-sided analysis, the required sample size was calculated to be 53 per group. The power analysis was performed using G*Power V.3.1.7.9 for Windows (Franz Faul, University of Kiel, Germany).

Outcome measures

The primary outcome was 30-day mortality following a diagnosis of pneumonia and the secondary outcome was the length of hospital stay.

Data collection and processing

We collected demographic data, data regarding comorbidities and the laboratory findings at baseline. Specifically, the age, sex, body mass index (BMI), vital signs, comorbidities (dementia, cerebrovascular disease, congestive heart failure, chronic respiratory disease, chronic kidney disease and malignant disease), medication (for hypertension, diabetes mellitus and/or dyslipidaemia), Charlson comorbidity index,23 number of prescriptions, presence of polypharmacy (use of ≥5 medications), place of residence before admission and level of care of the participants were collected. The laboratory data collected were the serum albumin, urea nitrogen and C reactive protein (CRP) concentrations; the haemoglobin concentration; and the white cell, neutrophil and lymphocyte counts. These variables were used to calculate the NEWS and NLR. We treated these variables as potential confounders of the analysis of mortality owing to pneumonia in the older participants, as described in previous reports.7–13

The NEWS was calculated using data regarding six physiological parameters that were collected during hospitalisation: respiratory rate, oxygen saturation, body temperature, systolic blood pressure, pulse rate and the level of consciousness.6 Every continuous variable was awarded a maximum score of 3 points, whereas the need for supplemental oxygen and the level of consciousness were awarded 0 points if absent/normal and 2 or 3 points if present/altered. Participants with a NEWS of 0–4 points were classified as low risk, those with a NEWS of 5–6 points were classified as medium risk and those with a NEWS of 7+ points were classified as high risk. However, when a single physiological parameter received a score of 3 points, the participant was categorised as being at medium risk, instead of low risk. To rate any confusion, we reviewed the medical records of the participants at the time of admission and recorded the presence of any abnormalities in the Alertness, response to Voice, Pain, Unresponsiveness score, a Glasgow Coma Scale score of ≤13, abnormalities identified during the mental status examination, and any confusion or delirium.

Total leucocyte and leucocyte fraction counts were obtained by fluorescence flow cytometry and hydrodynamic focusing (forward and side scatter) using a Sysmex XT-2000i automated haematology analyser (Sysmex, Kobe, Japan) and peripheral blood diluted in EDTA. Platelet counts were performed using sheath flow direct-current detection, and the circulating CRP concentration was measured by immunoturbidimetry using a TBA-2000FR instrument (Canon Medical Systems, Tochigi, Japan). NLR was calculated using the results obtained during routine haematological analyses. The values for each of these markers of inflammation were compared among the NEWS categories.

Statistical analysis

The Kolmogorov-Smirnov test was used to analyse the normality of the collected datasets. Categorical datasets are presented as frequency (%), and continuous datasets are presented as mean (SD) for parametric data or median (IQR) for non-parametric data. The Χ² test was used to compare the proportions for the categorical data. Analysis of variance and the Kruskal-Wallis test were used to analyse normally and non-normally distributed continuous datasets, respectively, among the three groups. Time-to-event data were evaluated using Kaplan-Meier estimates. A Cox proportional hazards model was used to perform a sensitivity analysis to ascertain whether the NEWS is a useful predictor, even after the exclusion of factors associated with pneumonia. Variables for which p<0.20 that were obtained during the univariate analysis of the baseline data were included in a multivariate-adjusted Cox proportional hazards analysis of 30-day mortality. The discriminability of each index for 30-day mortality was assessed using the area under the receiver operating characteristic curve (AUC) and 95% CI.24 The optimal cut-off value, sensitivity and specificity were determined using the Youden index,25 and differences in the AUC were identified using DeLong’s test.26 Positive (PPVs) and negative predictive values (NPVs) were calculated to assess the accuracy of the NEWS alone and the combinations of NEWS and NLR. To compare the sensitivity, specificity, accuracy, PPV and NPV, McNemar’s test was used.27 28 Statistical significance was defined as a two-sided p<0.05. Analyses were conducted using SPSS V.28 (IBM). We did not include participants with missing baseline data, and in particular those with missing laboratory data.

Patient and public involvement

Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Participants

During the present study, 446 people of ≥65 years of age were diagnosed with pneumonia, of whom 164 were excluded (76 because of treatment with antibiotics, 17 because of corticosteroid use, 44 because of the presence of a haematological disease, 7 because of the presence of a liver disease and 20 because of missing data), leaving 282 for inclusion in the final analysis (online supplemental figure 1). Because all the participants were hospitalised, we were able to readily follow them until they were discharged.

Characteristics of the cohort

Table 1 shows the baseline characteristics of the study sample (age=85.3 (7.9); 56.7% male), categorised according to the NEWS classification. The low NEWS group included five participants with 0 points. The 30-day mortality rate for the entire group of participants was 9.2% (n=26). With respect to the outcomes, both the 30-day and in-hospital mortality rates and the duration of the hospital stay significantly increased with the NEWS category. Figure 1 shows the Kaplan-Meier curves for the 30-day mortalities of the low, medium and high NEWS risk groups (log-rank test, p<0.03). Dementia was the most frequent comorbidity (65.2%), followed by cerebrovascular disease (34.8%) and congestive heart failure (21.6%). With the exception of sex, there were no significant demographic differences between the participants in each NEWS category. With the exception of systolic blood pressure, all the vital signs of the participants were significantly worse in the high NEWS group. There were no significant differences in the albumin, haemoglobin or urea nitrogen concentrations among the groups. The white cell count, neutrophil count and NLR were higher in the high NEWS group than in the other groups.

Request permission

Relationships of the outcomes with potential predictor

Table 2 shows the results of the univariate and multivariate analyses of the relationships between 30-day mortality and parameters at baseline. On univariate analysis, there were parameters showing a significant relationship with 30-day mortality for age, BMI, chronic kidney disease, neoplastic disease, level of care, albumin, urea nitrogen and the NEWS. Then, NEWS and urea nitrogen were entered into the multivariate analysis. Because chronic kidney disease and neoplastic disease had less than 10 deaths, the variables were not used in the multivariate analysis to avoid the possibility of overfitting and the distortion of the model. Table 3 shows the HRs for 30-day mortality by the Cox proportional hazards model in the participants as categorised according to the NEWS category. We obtained a crude HR, an HR adjusted for age, and an HR adjusted for age and urea nitrogen, identified as risk parameters for mortality in table 2 versus the low NEWS group. The age-adjusted and multivariate-adjusted HRs for the high NEWS group were 8.87 (95% CI: 1.19 to 66.20, p=0.03) and 8.69 (95% CI: 1.17 to 64.81, p=0.04), indicating significant associations with high 30-day mortality.

Evaluation of model performance

Figure 2 shows the predictive ability of the NEWS alone, NEWS×NLR and NEWS+NLR. The AUC analysis showed that the NEWS was a moderately good predictor, with an AUC of 0.73. The AUCs for NEWS×NLR and NEWS+NLR were 0.84 and 0.83, respectively, implying significantly better prognostic ability than the NEWS alone (p=0.007 and 0.04, respectively). The optimal cut-off values, sensitivities and specificities for the AUCs were 7.5, 73.1% and 67.6% for the NEWS alone; 65.1, 96.2% and 62.1% for NEWS×NLR; and 17.4, 92.3% and 60.2% for NEWS+NLR, respectively (table 4). The PPV, NPV and accuracy for each predictor of 30-day mortality, compared with NEWS alone, are also shown in table 4. NEWS×NLR and NEWS+NLR tended to have better sensitivity, accuracy, PPV and NPV than NEWS alone, but these differences were not statistically significant (p=0.06).

Request permission