Diagnostic Accuracy of Cerebrospinal Fluid Procalcitonin in Bacterial Meningitis Patients with Empiric Antibiotic Pretreatment

ABSTRACT

Accurate diagnosis of bacterial meningitis (BM) relies on cerebrospinal fluid (CSF) Gram staining and bacterial culture, which often present high false-negative rates because of antibiotic abuse. Thus, a novel and reliable diagnostic biomarker is required. Procalcitonin (PCT) has been well demonstrated to be specifically produced from peripheral tissues by bacterial infection, which makes it a potential diagnostic biomarker candidate. Here, we performed a prospective clinical study comprising a total of 143 patients to investigate the diagnostic value of CSF PCT, serum PCT, and other conventional biomarkers for BM. Patients were assigned to the BM (n = 49), tuberculous meningitis (TBM) (n = 25), viral meningitis/encephalitis (VM/E) (n = 34), autoimmune encephalitis (AIE) (n = 15), or noninflammatory nervous system diseases (NINSD) group (n = 20). Empirical antibiotic pretreatment was not an exclusion criterion. Our results show that the CSF PCT level was significantly (P < 0.01) higher in patients with BM (median, 0.22 ng/ml; range, 0.13 to 0.54 ng/ml) than in those with TBM (median, 0.12 ng/ml; range, 0.07 to 0.16 ng/ml), VM/E (median, 0.09 ng/ml; range, 0.07 to 0.11 ng/ml), AIE (median, 0.06 ng/ml; range, 0.05 to 0.10 ng/ml), or NINSD (median, 0.07 ng/ml; range, 0.06 to 0.08 ng/ml). Among the assessed biomarkers, CSF PCT exhibited the largest area under the receiver operating characteristic curve (0.881; 95% confidence interval, 0.810 to 0.932; cutoff value, 0.15 ng/ml; sensitivity, 69.39%; specificity, 91.49%). Our study sheds light upon the diagnostic dilemma of BM due to antibiotic abuse. (This study has been registered at ClinicalTrials.gov under registration no. NCT02278016.)

INTRODUCTION

Despite advances in antibiotic therapy and vaccine application, bacterial meningitis (BM) remains a significant cause of morbidity and mortality worldwide, especially in developing countries (1). Rapid and accurate diagnosis is of paramount significance for efficient treatment and favorable outcome. However, it is still a formidable challenge to differentiate BM from other central nervous system (CNS) infections, especially tuberculous meningitis (TBM) or viral meningitis/encephalitis (VM/E), for which disparate medical interventions and prognoses are involved (2). Cerebrospinal fluid (CSF) examination is the gold standard test for BM (3), with accuracy rates of 50 to 90% with Gram staining and 70 to 90% with CSF culture among untreated patients (4, 5). Nevertheless, in clinical practice, CSF cultures are time-consuming. The trickier problem is that the diagnostic yields of CSF Gram staining and subsequent culture in tertiary hospitals decrease significantly for referred patients who have received empirical antibiotic pretreatment, including insufficient or nonspecific applications, in a primary care setting or by self-medication (2), which is common, particularly in developing countries with severe antibiotic abuse (6, 7). Therefore, more convenient and reliable tools are needed urgently to address the differential diagnosis dilemma in such patients with suspected BM.

Recently, serum procalcitonin (PCT), a polypeptide of 116 amino acids, was proposed as a promising biomarker for differentiating bacterial infections from nonbacterial infections or other inflammatory conditions (8). In healthy humans, the serum PCT concentration is very low (e.g., <0.1 ng/ml) since nearly all of it is produced and cleaved to calcitonin (CT) in the thyroid (9, 10). However, when Escherichia coli was injected into hamsters to induce sepsis, almost all the tested tissues throughout their bodies produced PCT (11), resulting in a distinct increase in the serum PCT level at the early stage of infection (9, 12). Accordingly, it is widely applied as a biomarker for the early diagnosis of sepsis in the clinical setting.

Given its great value in diagnosing sepsis, more and more attention is being paid to the role of serum PCT in CNS infection. Two recent meta-analyses showed that the serum PCT level is a highly accurate diagnostic biomarker for BM (13, 14). However, it should be noted that most of the included studies excluded patients who had received antibiotic treatment prior to serum PCT testing. The serum PCT level was demonstrated to decrease in patients who had received effective antibiotic therapy (15). So, for BM patients who have received antibiotic pretreatment, serum PCT may not be a reliable biomarker, while CSF PCT is theoretically indicative of a bacterial cause of meningitis. Here, we performed a prospective clinical study to compare the diagnostic accuracies of CSF PCT, serum PCT, and other conventional biomarkers for BM among patients with suspected meningitis without excluding empirical antibiotic pretreatment.

RESULTS

Patient characteristics.

One hundred thirty-five consecutive patients with suspected meningitis or encephalitis were screened for entry into the clinical study (Fig. 1). Twelve patients were excluded due to their unclear diagnosis. Eligible patients were divided into four groups, a BM group (n = 49), a TBM group (n = 25), a VM/E group (n = 34), and an AIE group (n = 15). In addition, we screened another 20 patients with NINSD to be used as controls.

FIG 1.

Diagram of inclusion and exclusion of participants in the study. BM, bacterial meningitis; TBM, tuberculous meningitis; VM/E, viral meningitis/encephalitis; AIE, autoimmune encephalitis; NINSD, noninflammatory nervous system diseases.

The baseline characteristics of the patients are shown in Table 1. There were no significant differences among the five groups in terms of demographic characteristics (P > 0.05) or Glasgow coma scale (GCS) scores (P > 0.05). The most common symptoms presented in patients with BM at admission were fever (91.8%), headache (65.3%), seizure (32.7%), mental symptoms (30.6%), consciousness impairment (69.4%), and vomiting (69.4%). On physical examination, neck stiffness and focal neurological symptoms were present in 67.3% and 18.4% of the BM patients, respectively. Since our patients were all in a neurocritical care unit, some of them had complications with infections of peripheral tissues, including pneumonia (69.4%) and sepsis (38.8%). Prior to PCT detection, all of the BM patients had antibiotic treatment, and 20.4% of them had received corticosteroid treatment. The percentages of antibiotic and corticosteroid treatment were 92.0% and 24.0% in the TBM group, 76.4% and 17.6% in the VM/E group, 86.6% and 6.7% in the AIE group, and 0% and 5.0% in the NINSD group, respectively. Notably, the median time of PCT detection (from the appearance of symptoms related to respective diseases to the moment when CSF and serum PCT were collected for the first time in our hospital) was more than 10 days in every group, because most of our subjects were transferred from a primary care setting, where they had accepted preliminary treatment, such as empirical antibiotic administration, but showed no significant improvement. In association with this, the median duration of antibiotic treatment (regardless of the type, dosage, or administration route) prior to PCT detection in all the enrolled patients was 8 days.

TABLE 1.

Baseline characteristics of patients^a

Characteristic	All patients (n = 143)	BM group (n = 49)	TBM group (n = 25)	VM/E group (n = 34)	AIE group (n = 15)	NINSD group (n = 20)	P
Age (yr)	41.00 (25.00–52.00)	43.00 (28.00–52.00)	42.00 (25.00–55.50)	39.50 (24.75–54.75)	27.00 (23.00–39.00)	43.00 (32.25–45.75)	0.366
Sex, male	93 (65)	36 (73.5)	15 (60.0)	25 (73.5)	6 (40.0)	11 (55.0)	0.095
Fever	109 (76.2)	45 (91.8)	24 (96.0)	23 (67.6)	13 (86.7)	4 (20.0)	<0.001
Headache	82 (57.3)	32 (65.3)	22 (88.0)	16 (47.1)	11 (73.3)	1 (5.0)	<0.001
Seizure	52 (36.4)	16 (32.7)	4 (16.0)	14 (41.2)	12 (80.0)	6 (30.0)	0.001
Neck stiffness	77 (53.8)	33 (67.3)	18 (72.0)	16 (47.1)	8 (53.3)	2 (10.0)	<0.001
Mental symptoms	46 (32.2)	15 (30.6)	8 (32.0)	11 (32.4)	12 (80.0)	0 (0.0)	<0.001
Consciousness impairment	91 (63.6)	34 (69.4)	15 (60.0)	20 (58.8)	15 (100.0)	7 (35.0)	0.002
Vomiting	83 (58.0)	34 (69.4)	21 (84.0)	18 (52.9)	7 (46.7)	3 (15.0)	<0.001
Focal neurological symptoms	47 (32.9)	9 (18.4)	8 (32.0)	11 (32.4)	2 (13.3)	17 (85.0)	<0.001
Pneumonia	84 (58.7)	34 (69.4)	16 (64.0)	18 (52.9)	11 (73.3)	5 (25.0)	0.008
Sepsis	39 (27.3)	19 (38.8)	9 (36.0)	5 (14.7)	5 (33.3)	1 (5.0)	0.016
Glasgow coma scale score							0.836
9–15	105 (73.4)	36 (73.5)	19 (76.0)	27 (79.4)	9 (60.0)	14 (70.0)
3–8	38 (26.6)	13 (26.5)	6 (24.0)	7 (20.6)	6 (40.0)	6 (30.0)
Corticosteroid treatment	24 (16.8)	10 (20.4)	6 (24.0)	6 (17.6)	1 (6.7)	1 (5.0)	<0.001
Antibiotic treatment	127 (88.8)	49 (100.0)	23 (92.0)	26 (76.4)	13 (86.6)	0 (0.0)	<0.001
Duration of antibiotic treatment (days)b	8.00 (3.00–16.00)	11.00 (6.00–16.5)	15.00 (8.00–25.5)	6.00 (1.50–13.00)	16.00 (5.00–22.00)	0.00 (0–4.50)	2.274
PCT time (days)c	16.00 (9.00–28.00)	12.00 (7.00–17.00)	21.00 (15.50–32.50)	12.50 (7.00–21.25)	20.00 (15.00–35.00)	57.50 (16.00–567.50)	<0.001

^aData are median (interquartile range) for continuous variables (differences were tested using the Kruskal-Wallis H test) and number (percentage) for categorical variables (differences were tested using the χ² test). BM, bacterial meningitis; TBM, tuberculous meningitis; VM/E, viral meningitis/encephalitis; AIE, autoimmune encephalitis; NINSD, noninflammatory nervous system diseases.

^bDuration of antibiotic use refers to the duration of antibiotic administration before the CSF and serum were collected for the first time in our hospital.

^cProcalcitonin (PCT) time was calculated from the appearance of symptoms of the respective disease to CSF and serum collection for the first time in our hospital; it is equal to the duration of symptoms.

In the BM group (n = 49), there were 29 patients with acute community-acquired bacterial meningitis and 20 patients with postoperative bacterial meningitis after neurosurgery. Seventeen patients (34.7%) had a positive CSF culture, five (10.2%) had a positive CSF Gram stain, one (2.0%) had a positive blood culture, and 26 (53.1%) had a CSF white cell count of greater than 500/mm³ and rapid improvement after antibacterial therapy. The bacterial etiologies were Streptococcus pneumoniae (n = 6), Klebsiella pneumoniae (n = 3), Staphylococcus haemolyticus (n = 2), and one each of Bacteroides fragilis, Enterococcus faecium, Pseudomonas aeruginosa, Listeria monocytogenes, Staphylococcus aureus, Burkholderia cepacia, and Acinetobacter junii. In the TBM group, 14 patients had a definite diagnosis, and the other 11 patients had a probable diagnosis. In the VM/E group, there were two patients with positive CSF PCR and four with IgM antibodies in their CSF or serum. The remaining 28 patients were given a probable diagnosis. In the AIE group, 12 patients were diagnosed with anti–N-methyl-d-aspartate receptor encephalitis, one patient with anti-LGI1 receptor encephalitis, and two patients with a probable diagnosis.

CSF and serum characteristics.

Comparisons of common laboratory parameters for diagnosing BM revealed that there were significant differences in CSF PCT, serum PCT, CSF leukocyte, CSF neutrophil, and CSF protein levels, CSF/serum glucose ratios, and serum C-reactive protein (CRP) levels among the five groups (Table 2). Further multiple comparisons indicated that the level of CSF PCT was substantially higher (P < 0.01) in patients with BM (median, 0.22 ng/ml; range, 0.13 to 0.54 ng/ml) than in patients with TBM (median, 0.12 ng/ml; range, 0.07 to 0.16 ng/ml), VM/E (median, 0.09 ng/ml; range, 0.07 to 0.11 ng/ml), AIE (median, 0.06 ng/ml; range, 0.05 to 0.10 ng/ml), or NINSD (median, 0.07 ng/ml; range, 0.06 to 0.08 ng/ml) (Fig. 2A). However, the other tested parameters between the BM and at least one other group were not pronouncedly different (Fig. 2B through G).

TABLE 2.

CSF and serum data^a

Datum typeb	All patients (n = 143)	BM group (n = 49)	TBM group (n = 25)	VM/E group (n = 34)	AIE group (n = 15)	NINSD group (n = 20)	P
CSF
PCT (ng/ml)	0.10 (0.07–0.18)	0.22 (0.13–0.54)	0.12 (0.07–0.16)	0.09(0.07–0.11)	0.06 (0.05–0.10)	0.07 (0.06–0.08)	<0.001
Leukocyte (cells/mm3)	38.00 (6.00–183.0)	199.00 (40.00–663.50)	102.00 (49.50–250.00)	21.00 (2.00–88.25)	12.00 (4.00–21.00)	1.00 (0–6.50)	<0.001
Neutrophils (cells/mm3)	0.32 (0.00–31.85)	104.50 (1.02–439.58)	4.84 (0.08–31.41)	0.02 (0–0.74)	0 (0–0.16)	0 (0–0)	<0.001
Lymphocyte (cells/mm3)	20.72 (2.72–81.60)	46.98 (12.35–117.49)	73.08 (29.46–229.90)	17.14 (0.10–77.60)	11.76 (1.85–17.28)	0.80 (0–1.88)	<0.001
Protein (g/L)	0.60 (0.30–1.10)	1.00 (0.60–1.95)	1.00 (0.60–2.65)	0.35 (0.28–0.63)	0.30 (0.20–0.50)	0.40 (0.30–0.50)	<0.001
Glucose (mmol/L)	56.34 (44.46–68.40)	53.82 (33.39–64.35)	39.60 (26.55–58.77)	62.82 (50.13–70.20)	64.80 (55.80–71.10)	63.81 (54.90–74.75)	<0.001
First pressure (cm H2O)	172.50 (135.00–245.00)	160.00 (135.00–232.50)	220.00 (150.00–330.00)	165.00 (128.75–233.75)	200.00 (150.00–255.00)	157.50 (130.00–180.00)	0.126
Last pressure (cm H2O)	100.00 (74.00–145.00)	87.50 (69.50–132.50)	100.00 (76.25–150.00)	100.00 (63.75–152.50)	120.00 (95.00–150.00)	88.50 (80.00–130.00)	0.560
Serum
PCT (ng/ml)	0.13 (0.05–0.44)	0.44 (0.13–1.71)	0.12 (0.06–0.34)	0.07 (0.04–0.16)	0.16 (0.07–0.34)	0.04 (0.02–0.07)	<0.001
Leukocyte (cells/mm3)	9.13 (6.94–12.73)	11.50 (7.69–14.89)	7.83 (5.89–10.56)	9.07 (7.24–12.54)	10.07 (8.10–15.01)	7.05 (5.14–8.85)	<0.001
Glucose (mg/dl)	113.40 (91.80–142.20)	127.80 (108.36–153.90)	109.80 (79.20–150.30)	111.60 (88.20–135.45)	102.6 (91.80–115.20)	90.81 (82.80–126.63)	0.017
CRP (mg/L)	10.20 (1.19–38.90)	20.90 (4.75–55.25)	10.60 (1.25–36.70)	10.20 (1.05–26.63)	9.82 (0.55–39.90)	1.95 (0.17–17.80)	0.047
CSF and serum
CSF/serum PCT ratio	0.89 (0.40–2.00)	0.49 (0.19–1.35)	0.89 (0.36–1.24)	1.33 (0.47–2.36)	0.69 (0.20–0.90)	1.88 (1.14–2.93)	0.001
CSF/serum glucose ratio	0.48 (0.36–0.64)	0.41 (0.27–0.52)	0.32 (0.21–0.43)	0.59 (0.46–0.73)	0.60 (0.46–0.67)	0.64 (0.53–0.75)	<0.001

^aData are median (interquartile range) for continuous variables; differences were tested using the Kruskal-Wallis H test. BM, bacterial meningitis; TBM, tuberculous meningitis; VM/E, viral meningitis/encephalitis; AIE, autoimmune encephalitis; NINSD, noninflammatory nervous system diseases.

^bCSF, cerebrospinal fluid; PCT, procalcitonin; CRP, C-reactive protein; first pressure and last pressure, intracranial pressure in lumbar puncture.

FIG 2.

Comparisons of CSF PCT (A), serum PCT (B), CSF leukocytes (C), CSF neutrophils (D), CSF protein (E), CSF/serum glucose ratio (F), and serum CRP (G) among the five groups. The red lines and blue squares correspond to the group interquartile ranges and averages, respectively. *, P < 0.05; **, P < 0.01; ***, P < 0.001 (versus the respective BM group).

In the BM group, CSF PCT levels did not differ (P > 0.05) between those who received corticosteroid treatment prior to PCT detection (median, 0.13 ng/ml; range, 0.09 to 0.21 ng/ml) and those who did not (median, 0.09 ng/ml; range, 0.07 to 0.16 ng/ml). Interestingly, the mean level of CSF PCT was higher in the BM patients with Gram-negative bacteria than in those with Gram-positive bacteria, although the difference was not significant (Fig. 3A). Among all five groups, there was a significant positive correlation between the number of CSF leukocytes and the level of CSF PCT (Spearman r = 0.467; P < 0.0001) (Fig. 3B).

FIG 3.

Levels of CSF PCT and their correlation with bacterial etiology (A) and the numbers of CSF leukocytes (B). The red lines and blue squares correspond to the group interquartile ranges and averages, respectively.

Discriminatory ability was then calculated using the area under the receiver operating characteristic curve (AUROC). Among the six parameters assessed, CSF PCT exhibited the largest AUROC (0.881; 95% confidence interval [CI], 0.882 to 0.940), with a cutoff value of 0.15 ng/ml (sensitivity, 69.39%; specificity, 91.49%; positive predictive value, 81.0%; negative predictive value, 85.1%) (Fig. 4; Table 3). In addition, the AUROC for CSF PCT combined with the other five parameters assessed (0.880; 95% CI, 0.819 to 0.942) was significantly (P < 0.05) higher than the AUROC for only those five parameters together without CSF PCT (0.846; 95% CI, 0.777 to 0.914).

FIG 4.

ROC curves of CSF PCT, serum PCT, CSF leukocyte, CSF neutrophil, CSF protein, and serum CRP levels for predicting the diagnosis of BM using the data of all enrolled patients.

TABLE 3.

Discrimination of CSF PCT and five compared parameters with regard to BM

Datum typea	AUROCb	95% CIc	Cutoff	Sensitivity (%)	Specificity (%)	PPVd (%)	NPVe (%)
CSF PCT	0.881	0.882–0.940	0.15 ng/ml	69.39	91.49	81.0	85.1
Serum PCT	0.759	0.669–0.849	0.19 ng/ml	67.35	75.53	58.9	81.6
CSF leukocytes	0.801	0.721–0.881	164.00 cells/mm3	57.14	90.32	75.7	80.0
CSF neutrophils	0.830	0.750–0.910	77.90 cells/mm3	56.25	100.00	100.0	81.4
CSF protein	0.754	0.665–0.842	0.66 g/L	71.43	73.40	58.3	83.1
Serum CRP	0.651	0.551–0.752	5.49 mg/L	82.22	42.68	44.0	81.4

^aCSF, cerebrospinal fluid; CRP, C-reactive protein.

^bAUROC, area under the receiver operating characteristic curve.

^cCI, confidence interval.

^dPPV, positive predictive value.

^eNPV, negative predictive value.

DISCUSSION

For the first time, we found that CSF PCT level was significantly higher in patients with BM than in those with TBM, VM/E, AIE, or NINSD, and we showed that CSF PCT was superior to other biomarkers in terms of diagnostic accuracy for BM in patients who have received antibiotic pretreatment, which sheds light upon the diagnostic dilemmas of BM, especially when compared with TBM or VM/E. In addition, a positive correlation between the number of CSF leukocytes and the level of CSF PCT was found, suggesting that CSF leukocytes might be a source of CSF PCT.

Diagnostic dilemmas are common in patients with suspected BM, especially regarding the differential diagnosis between BM and TBM or VM/E. It can become even more difficult for patients who have received empirical antibiotic administration. Among the patients with BM in our clinical trial, 100% had received antibiotic treatment before PCT detection, which is typical because of the high rate of antibiotic usage in inpatients and outpatients in China (7, 16). After their enrollment, the commonly used laboratory parameters for BM, such as CSF/serum glucose ratio, are not particularly abnormal for these patients due to the effect of empirical antibiotic pretreatment, leading to failure in distinguishing BM from other CNS infections. However, CSF PCT was the only index in our study that was pronouncedly higher in the BM group than in any other CNS infection group, including the TBM group. Our clinical study also revealed that CSF PCT exhibited the largest AUROC among all the CSF parameters assessed, and the AUROC of CSF PCT combined with the other five parameters assessed showed better performance than the other five parameters together without CSF PCT, which further confirmed the usefulness of CSF PCT in BM diagnosis when there has been antibiotic pretreatment. It was interesting to find that serum PCT as a biomarker failed to perform well in our study, which seems to be discrepant from previous reports (17, 18). Possible explanations lie in the sample size and experimental design we used. The strengths of our study are the relatively large study sample and intensive grouping. A total of 143 enrolled patients were divided into groups of patients with BM, TBM, VM/E, ATE, or NINSD, while previous studies involving limited numbers of patients grouped them as only BM and non-BM patients (17). Moreover, we included patients who had received antibiotic pretreatment, while previous studies excluded them (17). The serum PCT level may decrease under antibiotic pretreatment because of its close relation to peripheral antibiotic treatment (15). However, in our study, CSF PCT levels remained high, possibly because the blood-brain barrier (BBB) prevented the antibiotics from partly leaking into the CSF (19). Therefore, we have managed to provide compelling evidence that CSF PCT is a much better biomarker for BM among patients with suspected meningitis without excluding those with a history of empirical antibiotic pretreatment.

Besides serum PCT, previous studies have reported several other diagnostic biomarkers with high sensitivity and specificity for diagnosing BM. Two meta-analyses concluded that CSF lactate was a reliable biomarker for differentiating BM from aseptic meningitis, with a sensitivity of 0.93 (95% CI, 0.89 to 0.96) and specificity of 0.96 (95% CI, 0.93 to 0.98) (20, 21). However, for patients who had received antibiotic pretreatment, the sensitivity dropped to 0.49 (95% CI, 0.23 to 0.75), which tremendously limited the application of CFS lactate as a biomarker in those patients (20). CSF heparin-binding protein (HBP), released from activated neutrophils, is another reported biomarker for BM. Linder et al. presented an AUROC of 0.994 for CSF HBP levels, a value even higher than that for CSF lactate levels, which were tested simultaneously. In addition, CSF HBP levels also rose in patients who had received antibiotic treatment less than 48 h prior to CSF sample collection (22). We did not restrict the dosage or duration of antibiotic pretreatment, which conforms more to the clinical situation. Besides, use of the CSF PCT level as a biomarker is quite readily incorporated into clinical practice, since the serum PCT level has been widely applied to diagnose sepsis in the clinical setting.

It was interesting to find a positive correlation between the number of CSF leukocytes and the level of CSF PCT. Blood leukocytes have been well demonstrated to be one of the cellular origins of serum PCT during sepsis. The expression of PCT in monocytes, granulocytes, and lymphocytes, three major types of leukocytes in human peripheral blood, were confirmed by flow cytometric analysis (23), immunoluminometric assay (24), Western blot analysis (25), and reverse transcription-PCR (26). Additionally, molecules related to bacteria, such as lipopolysaccharides (LPS) and various proinflammatory cytokines (interleukin-1β [IL-1β], IL-6, tumor necrosis factor α [TNF-α], IL-2), significantly stimulate the expression of PCT in peripheral blood monocytes (26). Indeed, there is a strong correlation between the PCT released from peripheral blood monocytes and the serum PCT level (r = 0.76; P < 0.001) (27). Thus, it is tempting to speculate that the leukocytes in CSF contribute, at least partially, to the level of CSF PCT. Besides CSF leukocytes, other possible origins of elevated CSF PCT levels in BM patients are passive leakage from peripheral blood and local production in the CNS. Notably, serum PCT levels were also significantly increased in the BM group, resulting in decreased CSF/serum PCT ratios in the BM group compared with those in the other groups. This may be because 69.4% and 38.8% of BM patients had pneumonia or sepsis, respectively, both of which were well demonstrated to elevate serum PCT (12, 28). A previous study indicated that the penetration of CT from blood to CSF was restricted by the BBB in patients with medullary thyroid carcinoma, suggesting even lower penetration of serum PCT because its molecules are larger than those of CT (29). However, disruption of the BBB was found in BM patients and in experimental models (30, 31). As a resultant, passive leakage from the serum to CSF possibly contributes to the elevated CSF PCT level in BM patients, especially those whose conditions are complicated with pneumonia or sepsis. In the CNS, four types of brain cells are also likely to be the origins of CSF PCT, neurons, astrocytes, and microglia in the parenchyma and meningeal cells. Recently, amino procalcitonin (N-PCT), a peptide derived from PCT, was reported to be expressed in neurons and astrocytes in the paraventricular nucleus (PVN) of the rat hypothalamus (32). Microglia and meningeal cells express the responding receptors (Toll-like receptors [TLRs]) to the invading bacteria (33, 34). Whether PCT can be synthesized and secreted in various brain cells during BM should be investigated further. The mean level of CSF PCT was found to be higher in BM patients with Gram-negative bacteria than in those with Gram-positive bacteria, albeit without significant difference. This is consistent with a previous report that serum PCT is associated more with Gram-negative bacteria than with Gram-positive bacteria during sepsis (35). Thus, the CSF PCT level might provide some reference for the selection of first-line antibiotic treatment.

Further studies are required to determine whether elevated PCT levels in the CSF of BM patients has a physiological or pathological role or is just a diagnostic marker. The function of PCT in sepsis hints that it is a potential amplifier of the inflammation cascade by increasing leukocyte-derived cytokines and augmenting reactive oxygen species (ROS) during sepsis (15). Accumulating evidence suggests that the survivors of BM are at a high risk of long-term disabling sequelae such as cognitive impairment (1), which is partially caused by the neurotoxic effect of the inflammatory response in CSF (36). In this case, the elevated CSF PCT level may play a critical role in the regulation of the inflammatory response in CSF. Clarifying the function of CSF PCT will open new avenues in understanding the pathogenesis of BM as well as the therapy for BM and its sequelae, which is under investigation in our laboratory.

The following limitations of this study should be noted. First, due to the presently limited diagnostic technologies, not all the patients had a definite etiological diagnosis, especially in the VM/E group, which might lead to a grouping bias in some cases. Second, it is difficult to collect more details about previous antibiotic therapy, such as the specific dosage, since many BM patients were from rural and impoverished areas, where antibiotic abuse is much more serious. Third, CSF PCT was indicative only of a bacterial cause of meningitis, while it did not identify a specific bacterium. Although proven to be helpful in etiological diagnoses during the past decade, PCR, even multiplex PCR, is capable only of detecting the pathogens that are already suspected and included in the primer mix (5). As with many unknown pathogens, we need to develop novel diagnostic approaches, such as next-generation sequencing and bioinformatic analysis (37).

In conclusion, our study showed that CSF PCT is a reliable biomarker for the diagnosis of BM, especially when the diagnostic accuracies of Gram staining, bacterial culture, and other conventional biomarkers are significantly decreased due to early empirical antibiotic pretreatment.

MATERIALS AND METHODS

Patients.

This prospective clinical study was performed in the neurocritical care unit at Xijing Hospital, a 3,200-bed tertiary hospital in Xi'an, China. It was approved by the ethics committee of Xijing Hospital (approval no. KY20140916-3) and registered with ClinicalTrials.gov (under registration no. NCT02278016). Consecutive patients with clinically suspected meningitis/encephalitis between October 2013 and July 2015 were screened for inclusion in this study. The inclusion criteria were (i) age 13 years or older with (ii) at least two of the following clinical characteristics: headache, fever, neck stiffness, mental symptoms, consciousness impairment, or seizures (38), and (iii) receipt of signed informed consent from the patient or his or her proxy. Exclusion criteria were (i) complication with active malignancy and (ii) any contraindication to lumbar puncture. It should be noted that empirical antibiotic treatment, prior to the first lumbar puncture in our hospital, regardless of the type, dosage, or administration route of antibiotics, was not an exclusion criterion. In addition, we recruited 20 patients with a noninflammatory nervous system disease (NINSD) as controls. The final diagnosis of each patient at discharge was made by at least two experienced neurologists based on the following clinical diagnostic criteria, and a consensus was reached.

The diagnosis of BM was based on compatible clinical features and one of three criteria, (i) positive CSF culture, (ii) negative CSF culture but with identification of bacteria by either CSF Gram staining or blood culture, or (iii) CSF white cell count of 500/mm³ or higher and rapid improvement after antibacterial therapy despite negative CSF and blood culture results (39, 40).

A definite diagnosis of tuberculous meningitis (TBM) was made if results of (i) smear microscopy for acid-fast bacilli, (ii) culture for Mycobacterium tuberculosis, or (iii) an Xpert MTB/RIF assay of CSF were positive. In the absence of proven tuberculous etiology, probable TBM was determined when patients scored 10 points or more on the diagnostic scoring system when cerebral imaging was not available or 12 points or more when cerebral imaging was available and improved significantly after anti-TB therapy (41).

Viral meningitis/encephalitis (VM/E) was diagnosed based on acute compatible symptoms, the absence of any bacterial meningitis criteria, and either a positive PCR result for viral DNA or RNA in the CSF or specific antiviral IgM antibodies in CSF and serum (22). If a viral etiology was not proven, the criteria for probable viral meningitis or encephalitis were clear CSF with inflammatory reaction, negative results on Gram staining and sterile bacterial culture, and favorable evolution of the disease not requiring intervention with extensive antibiotic treatment, apart from antiviral therapy (42, 43).

Autoimmune encephalitis (AIE) was defined as encephalitis associated with antibodies against neuronal cell surface or synaptic proteins (44) or negative antibodies but with typical manifestations and/or findings of a related tumor. NINSD included idiopathic epilepsy (n = 3), cerebral infarction (n = 3), migraine (n = 2), cerebellar ataxia (n = 2), peripheral neuropathy (n = 2), hypoxic-ischemic encephalopathy (n = 2), venous sinus thrombosis (n = 2), delayed encephalopathy after acute carbon monoxide poisoning (n = 2), dystonia (n = 1), and Charcot-Marie-Tooth disease (n = 1).

Measurements.

All the enrolled patients underwent a lumbar puncture to collect CSF immediately after admission to our hospital. PCT levels (ng/ml), leukocyte counts (cells/mm³), neutrophil counts (cells/mm³), lymphocyte counts (cells/mm³), protein levels (g/L), and glucose levels (mmol/L) in the CSF were immediately determined. We also examined the PCT levels (ng/ml), leukocyte counts (cells/mm³), glucose levels (mmol/L), and C-reactive protein (CRP) levels (mg/L) in the serum samples collected at the same time or within the same day of CSF sample collection. The PCT levels in the CSF and serum samples were measured using a Cobas E601 electrochemiluminescence immunoassay analyzer (Roche) (minimum detection level, 0.02 ng/ml). Relevant clinical data were collected on admission, including (i) Glasgow coma scale (GCS) score, dichotomized as scores 3 through 8 or 9 through 15, and (ii) infection conditions of peripheral tissues (pneumonia was diagnosed using British Thoracic Society criteria [45], and sepsis was diagnosed using diagnostic criteria proposed in 1992 [46]).

Statistics.

Continuous variables were expressed as medians and interquartile ranges, and categorical variables were expressed as counts and percentages. The differences were tested using the Kruskal-Wallis H test with Dunn test post hoc analysis for continuous data (47) and the χ² test for categorical data (sex). Through the receiver operating characteristic (ROC) curve, sensitivity and specificity were determined using the best cutoff values. We examined the correlation between the number of CSF leukocytes and the level of CSF PCT using the Pearson correlation test. The level of significance was set at P < 0.05. All statistical analyses were performed using SPSS 21.0.