Neutrophilia independently predicts death in tuberculosis

To the Editor:

Experimental animal work indicates that neutrophils play a key role in the immune response to mycobacteria [1, 2]. They appear protective against early infection [3] but in established disease, neutrophilia associates with pathology [1, 4]. In humans, higher neutrophil counts at tuberculosis diagnosis predict slower sputum conversion to negative during therapy [5, 6], but the overall prognostic significance of neutrophilia in human tuberculosis remains elusive. We therefore aimed to analyse this phenomenon in a study powered to detect an independent relationship with mortality.

Tuberculosis patients were identified by database/case-note review at Newham University Hospital Trust and King’s College Hospital, London, UK. All patients diagnosed between 1999 and 2006 were eligible for inclusion in an analysis of neutrophilia at baseline; those with a recorded outcome of successfully completing treatment or death were included in an analysis of determinants of mortality. Healthy contacts of tuberculosis cases were recruited from the same hospitals.

Data were extracted on patient age, sex, ethnicity, comorbidity, use of immunosuppressive medication, HIV status and site of disease. Laboratory data were collected from samples taken on the date of tuberculosis diagnosis: serum sodium, bilirubin and albumin concentrations; peripheral blood haemoglobin concentration; and peripheral blood neutrophil, monocyte, lymphocyte and platelet counts. Blood culture results were recorded where performed. Protocols were approved by the Barking and Havering NHS Research Ethics Committee (REC 08/H0702/25) and North East London Research Ethics Committee (REC P/02/146).

We calculated that 584 patients (34 deaths and 550 survivors) would be required to detect a three-fold difference in mortality in the presence of neutrophilia with 80% power (5% significance level), assuming a 15% prevalence of neutrophilia and a death/survival ratio of 1/16 (parameters derived from preliminary analysis). We aimed to include twice this number to allow for missing data.

Comparisons of proportions used Chi-squared tests. Comparisons of non-Gaussian continuous data (neutrophil counts and age) used Mann–Whitney tests. To investigate demographic or clinical associations with neutrophilia in tuberculosis patients, neutrophil counts were treated as a categorical dependent variable for binary logistic regression: ≥7.5×10⁹ or <7.5×10⁹ cells per litre. Analysis of predictors of mortality used death/survival as the binary dependent variable. Multivariate regression was performed using all significant (p<0.05) predictors from univariate analysis. Age was divided into strata pre-analysis. Laboratory parameters were also assigned pre-analysis into categorical predictors, as relationships were not anticipated to be linear, and both high and low values are usually pathological. A distinction was made between pathological neutropenia (<1×10⁹ cells per litre) and mild, usually benign ethnic neutropenia. Predictors with >25% missing values (HIV status and comorbidity) were assigned a separate group (“unknown”) to enable inclusion of patients with missing data in multivariate analyses. Bootstrapping analysis used simple (nonstratified) sample selection. Analyses were performed using SPSS versions 18–21 (IBM, Armonk, NY, USA).

1236 tuberculosis patients were identified; 855 had recorded neutrophil counts and data for all demographic variables except HIV status and comorbidity (see earlier). There was no difference in age (p=0.29), sex distribution (p=0.80), ethnic distribution (p=0.07) or neutrophil count (p=0.55) between included patients and excluded patients for whom this information was available. 49 patients were transferred or lost to follow-up and 88 patients lacked data for one or more laboratory parameters, resulting in 718 patients entering case fatality analysis.

Pulmonary tuberculosis was the commonest disease site (49.4%), and HIV infection was known to be present in 13.5% of the 855 patients. 214 contacts were also analysed. Cases and contacts did not differ in sex distribution (57.3% versus 50.5% male, respectively; p=0.07) but did differ in age (median age 33 versus 30 years, respectively; p=0.002) and ethnic distribution (11.6% versus 16.4%, respectively, were white; 19.3% versus 33.6% South Asian; 59.8% versus 40.7% black; and 9.4% versus 9.3% other ethnic origin; p<0.001).

Neutrophilia (peripheral blood neutrophil count ≥7.5×10⁹ cells per litre) was commoner in patients with active tuberculosis disease than in healthy contacts (158 (18.5%) or of 855 versus eight (3.7%) out of 214). The adjusted odds ratio (aOR) for neutrophilia among cases versus contacts, controlling for age and ethnicity, was 6.13 (95% CI 2.94–12.82; p<0.001). Median (interquartile range) neutrophil count was also higher in cases than contacts (4.65 (3.17–6.75) versus 3.66 (2.78–4.78) ×10⁹ cells per litre; p<0.0001).

Analysis of 297 blood cultures performed on tuberculosis patients revealed three pathogenic bacteria other than Mycobacterium tuberculosis (one culture of methicillin-resistant Staphylococcus aureus, one of Proteus vulgaris and one of an unidentified coliform). Only the patient with P. vulgaris had concomitant neutrophilia.

We next sought to identify any associations with neutrophilia at tuberculosis diagnosis (table 1). In multivariate analysis, white ethnicity increased the odds of neutrophilia compared to black ethnicity (aOR 1.75, 95% CI 1.03–3.03; p=0.036). Pulmonary disease was associated with increased the odds of neutrophilia compared to peripheral lymph node tuberculosis (aOR 2.56, 95% CI 1.25–5.26; p=0.011). HIV infection reduced the odds of neutrophilia compared to HIV-uninfected subjects (aOR 0.50, 95% CI 0.26–0.97; p=0.039); however, this result is confounded by pathological neutropenia (five (38.5%) out of 13 patients with neutrophil count <1×10⁹ cells per litre were HIV infected, and removing all pathologically neutropenic patients from the analysis negates the association between HIV positivity and the absence of neutrophilia).

table 1 also summarises results from a logistic regression analysis of predictors of mortality (n=718). Neutrophilia was present in 16 (38.1%) out of 42 patients who died and 120 (17.8%) out of 676 survivors, and was an independent risk for case fatality in multivariate analysis (aOR 2.93, 95% CI 1.17–7.34; p=0.022). Bootstrapping analysis (1000 samples) confirmed the result’s robustness (aOR 2.93, 95% CI 1.16–12.03; p=0.018). Further laboratory parameters predicting fatality were hypernatraemia, hypoalbuminaemia, thrombocytopenia and lymphopenia. Increased age and the presence of comorbidity other than HIV were also associated with increased risk of death, but receiving immunosuppressive medication was not.

Our study yielded some important new findings. A modest neutrophilic response was common in tuberculosis: even survivors with active tuberculosis had higher median neutrophil counts than healthy contacts. Others have reported that higher blood neutrophil counts correlate with sputum M. tuberculosis PCR positivity and, especially, smear positivity [7], while separate studies discovered higher neutrophil counts associated with slower conversion of sputum culture to negative [5, 6]. Together with the higher prevalence of neutrophilia in patients who die (reported here), these results suggest that, broadly speaking, the neutrophil count in tuberculosis positively correlates with bacillary load.

It is therefore important to know which other factors associate with neutrophilia in human tuberculosis. We found no convincing evidence that nontuberculous bacteraemia explained this phenomenon, as only one out of 158 instances of neutrophilia was associated with a pathogenic nonmycobacterial species in blood culture. Lower risk of neutrophilia with isolated peripheral lymph node disease probably reflects lower mycobacterial load and less systemic inflammation. The finding that white ethnicity independently predicts neutrophilia may be biologically important in tuberculosis and help to explain the previous finding that European ethnic origin is a risk factor for death independently of age [8]. Indeed, the lowest case fatality in our study was seen with neutrophil counts in the range 1–1.99×10⁹ cells per litre, which is likely to largely reflect benign ethnic neutropenia. The apparent effect of HIV in reducing risk of neutrophilia is explained by pathological neutropenia, a well-described complication of HIV [9]. Indeed, pathological neutropenia (<1×10⁹ cells per litre) was associated with higher case fatality as compared to a normal-range neutrophil count. In addition to the association with HIV infection, this can be seen in the context of severe, disseminated tuberculosis [10].

Our study has some limitations. 381 potentially eligible patients were excluded, but their demographics and neutrophil counts were similar to included patients. Comorbidity and HIV status were poorly documented, necessitating an “unknown” coding category. Tuberculosis cases and contacts were not formally matched; in particular, they differed in age and ethnic distribution, but the odds of neutrophilia were much higher in the former even after adjustment for these factors.

In summary, we have demonstrated that neutrophilia in tuberculosis independently associates with increased risk of mortality. Interestingly, abrogating the immunopathological neutrophil response in some animal models improves outcome in acute infection [4]. Similar strategies might therefore have therapeutic application in humans with severe tuberculosis.

• Received June 17, 2013.
• Accepted September 5, 2013.

• ©ERS 2013