Retrospective validation of a β-trace protein interpretation algorithm for the diagnosis of cerebrospinal fluid leakage

Introduction

Cerebrospinal fluid (CSF), leakage can occur after head trauma, surgery, congenital malformation, or tumor invasion and potentially lead to infectious complications such as intracranial abscess or bacterial meningitis [1, 2]. The cumulative risk of developing bacterial meningitis after acute traumatic CSF leak without dural repair has been estimated to be higher than 85% at 10 years [3]. Hence, the prompt diagnosis of CSF leakage is crucial since its treatment can potentially prevent the development of life-threatening consequences. The accurate confirmation of CSF leakage and its precise localization are required for a successful surgical intervention. Imaging techniques (i.e. CT cisternography, high-resolution CT, MRI, detection of intratecal administrated radioactive tracer, or fluorescin dye) allow the localization of the leak but in some cases lack sensitivity, are expensive, and expose patients to potential risks [4, 5]. Therefore, the diagnostic approach should first include biochemical testing for CSF-specific proteins in secretion to select for patient with high probability of CSF leakage before imaging techniques are applied [6, 7]. Currently, two CSF rhinorrhea biomarkers with outstanding diagnostic characteristics are available routinely in clinical laboratory [8]. β₂-Transferrin, the desialated isoform of transferrin, was the first specific and sensitive biochemical test available to the clinic [9]. The most common technique used to detect this molecule is immunofixation electrophoresis, which tends to be time-consuming, expensive, and not readily available to every clinical laboratory [10, 11]. The detection of β-trace protein (β-TP) is a rapid and more accessible nephelometric alternative for the diagnosis of CSF fistulas. β-TP (prostaglandin D synthase) is one of the most abundant proteins in CSF and has the highest CSF/serum ratio of all CSF-specific proteins [12]. Several groups investigated the use of β-TP measurements in secretion for the detection of CSF-leakage and different cut-offs values have been proposed [13]. This study aimed to validate the β-TP interpretation algorithm currently used at our institute by the mean of a retrospective analysis of clinical records of patients in which β-TP measurements have been routinely performed.

Materials and methods

Algorithm for the interpretation of β-TP results

At our institute, routine β-TP results are interpreted on the basis of the algorithm depicted in Figure 1. Briefly, the presence of CSF in secretion can be excluded when β-TP values are <0.7 mg/L, whereas β-TP values ≥1.3 mg/L indicate the presence of CSF in secretion. β-TP values between 0.7 and 1.29 mg/L indicate the presence of CSF if the β-TP ratio (β-TP secretion/β-TP serum) is ≥2. Similar to other low-molecular-weight proteins, markedly increased serum levels of β-TP had been described in patient with impaired renal function or undergoing haemo- and peritoneal-dialysis [14, 15]. These conditions could lead to an elevation of the β-TP concentration in secretion by passive transfer or blood contamination. Therefore, samples showing β-TP values in secretion ≥1.3 mg/L and a β-TP ratio ≤1 cannot be interpreted and should be commented appropriately. On the contrary patients with bacterial meningitis, a feared complication of CSF leakage, show lower β-TP concentration in CSF, potentially leading to false-negative results [16].

Figure 1:

Flowchart of β-TP interpretation algorithm.

^aSamples with both β-TP≥1.3 mg/L and ratio≤1 should not be considered indicative for CSF leakage.

Results

The diagnostic performance of the β-TP interpretation algorithm currently used at our institute was compared with that of absolute cut-offs (either 1.3 mg/L or 0.7 mg/L), of β-TP secretion/serum ratio cut-off of 2 and with the algorithm previously proposed by Risch et al. [7]. To do this, we performed a retrospective evaluation of 236 samples originating from 121 patients with suspect CSF leakage. Fifty-nine samples from patients with clinically confirmed CSF leakage and 177 samples from patients without CSF leakage were included in the study. The demographic characteristics of the study population as well as the sample origin are summarized in Table 1. β-TP measurements were primarily performed in nose secretion (98%) and were mainly requested by neurosurgery/surgery and otorhinolaryngology departments (52% and 35% of the requests, respectively).

Strategy 1: β-TP interpretation using absolute cut-offs of either 1.3 mg/L or 0.7 mg/L

Patients with CSF leakage had significantly higher β-TP values in secretion (median 6.8 mg/L; range 0.5–51.1 mg/L) than patients without CSF leakage (median 0.3 mg/L; range 0.1–24.2 mg/L; p<0.001). The ROC generated using β-TP values yielded an AUC of 0.98 (Figure 2A and B). Applying a cut-off of 1.3 mg/L resulted in sensitivity and specificity of 91.5% and 97.7%, respectively. The PPV, NPV, +LR, and −LR at this cut-off were 93.1%, 97.2%, 40.5, and 0.09, respectively. Meanwhile, using the lower cut-off of 0.7 mg/L the sensitivity increased to 98.3% at the cost of a lower specificity of 80.2%, generating PPV, NPV, +LR, and −LR of 62.4%, 99.3%, 5, and 0.02, respectively (Table 2).

Figure 2:

(A, C) Dot plots comparing of β-TP values in secretion (A) and β-TP secretion to serum ratios (C) in patients with or without CSF leakage. Dashed lines delineate the gray zone (0.7–1.3 mg/L) and the cut-off ratio of 2. (B, D) ROC analysis of β-TP measurements in secretion (B) and of β-TP secretion to serum ratios (D).

Table 2:

Diagnostic performance of different β-TP interpretation approaches.

Interpretation	Sensitivity	Specificity	PPV	NPV	+LR	−LR	Odds ratio
β-TP≥1.3 mg/L	91.5	97.7	93.1	97.2	40.5	0.09	467
β-TP≥0.7 mg/L	98.3	80.2	62.4	99.3	5	0.02	235
Ratio≥2	96.5	95.3	87.3	98.8	20.5	0.04	557
Algorithm Risch et al. [7]	93.2	96	88.7	97.7	23.4	0.07	332
Algorithma	98.3	96	89.2	99.4	24.4	0.02	1384

^aβ-TP interpretation as described in Figure 1.

Discussion

The accurate diagnosis of CSF leaks is crucial to avoid life-threatening complications such as fulminant meningitis derived from ascending infections. The β-TP determination in nasal and ear fluid allows the rapid and accurate detection of CSF otorhinorrhea bypassing the drawbacks of β₂-transferrin measurement, which is based on a labor-intensive and time-consuming technique with delayed results availability. Even though the β-TP cut-off has been a matter of debate during the last decade, the optimal decision limit for the diagnosis of CSF leakage is still controversial. In a previous prospective study, we demonstrated the excellent diagnostic performance of β-TP in the clinical context of CSF otorhinorrhea [7]. The ROC analysis of 105 patients with suspect CSF leakage generated an optimal β-TP cut-off of 1.11 mg/L with excellent sensitivity (93%) and specificity (100%). This cut-off is in the range of that proposed by Schnabel et al. (1 mg/L) [17] and Arrer et al. (1.31 mg/L) [11] and was integrated in the diagnostic approach of CSF fistula suggested by Bachmann-Harildstad [11]. For the use in our clinical practice, the β-TP interpretation algorithm proposed by Risch et al. has been slightly adapted (Figure 1). The current cut-off values were defined using ROC analysis and represent the best compromise between published data and over 10-years’ experience of β-TP measurements in our daily routine. The present study aimed to retrospectively validate our routine β-TP interpretation algorithm using the largest patient collective assessed for β-TP in the context of a suspect CSF otorhinorrhea. For this purpose, we compared the diagnostic performance of three β-TP interpretation strategies.

The best diagnostic performance was obtained when β-TP results were interpreted according to the proposed algorithm (Figure 1) resulting in the highest odds ratio of 1384 (Table 2). Using this approach, eight samples collected from seven patients showed discrepant results between β-TP measurements and clinical interpretation (Table 3). Four samples out of seven had values above 1.3 mg/L but clinically no confirmed CSF leakage (sample 3, 4, 6 and 8). Samples 3 and 4 originated from one patient and were collected simultaneously from the left and right nostrils 2 days after trauma. Both samples showed strongly elevated β-TP results suggesting a possible clinical misinterpretation of this case. Sample 6 was collected immediately after a traumatic injury and showed a β-TP value slightly over the upper cut-off (β-TP: 1.34 mg/L). Five days later, nasal secretion from the same patient had a β-TP value under the lower cut-off (β-TP: 0.38 mg/L). Similarly, sample 8 was obtained from a patient 2 days after traumatic event and had β-TP of 1.99 mg/L. A further sample of the same patient obtained 5 days later had a β-TP value in the gray zone (β-TP: 1.13 mg/L). Unfortunately, the β-TP ratio could not be calculated because a simultaneously drawn serum sample was not available. These discrepant results might also be explained by the presence of a subclinical CSF leakage at presentation. Both follow-up samples 6 and 8 were derived from patients with conservative management. Depending on the clinical situation, waiting for spontaneous closure of a CSF leak is a valuable treatment strategy and negative or decreased β-TP results in the follow-up samples would indicate clinical improvement.

One single sample was classified as false-negative (sample 1, β-TP 0.49 mg/L). In this case β-TP was measured in nasal secretion of a patient 22 days after surgical repair of a skull base defect. A sample of the same patient collected before surgery had a β-TP of 1.32 mg/L and was clinically classified in the CSF leakage group. Even if the negative β-TP measurement after surgical repair would indicate a successful intervention, the follow-up sample was clinically categorized into the CSF leakage group.

Three additional samples (2, 5, and 7) without clinically confirmed CSF leakage had β-TP values in the gray zone and ratios ≥2. Applying our interpretation algorithm, these results indicate the probable presence of CSF in secretion and were therefore considered as false positive. Interestingly, sample 7 had a β-TP value in secretion of 0.71 mg/L and a very low β-TP serum concentration (β-TP serum: 0.22 mg/L). Rarely, such extremely low β-TP values in serum might be due to a pipetting error and consequently lead to artificially elevated ratios influencing β-TP interpretation. In such circumstances, we suggest serum β-TP values approaching the detection limit of the method (0.2 mg/L) to be confirmed by repetition. On the contrary, elevated serum β-TP values can be found in patients with impaired renal function and potentially lead to false-positive β-TP results in secretion. As described in Figure 1, samples with serum β-TP≥1.3 and ratios≤1 should not be interpreted as indicating CSF leakage and appropriately commented. In our experience, including a written commentary with the interpretation of the β-TP results according to the algorithm represents an added value to the laboratory report which is appreciated by clinicians and might avoid misinterpretation of the results.

Compared to our interpretation algorithm, setting an absolute cut-off of 1.3 mg/L resulted in a slightly increased specificity (97.7%, +1.7%), at a cost of substantially reduced sensitivity (91.5%, −7.2%), whereas applying a ratio of two as unique cut-off resulted in both lower sensitivity (96.5%, −1.8%) and specificity (95.3%, −1.3%). Applying the algorithm proposed by Risch at al. to this patient collective resulted in an equal specificity (96%) but reduced sensitivity (93.2%, −5.1%) and an odds ratio of 332 (Table 2).

One main limitation of the current study results from its retrospective design. Indeed, β-TP measurements were readily available to the clinicians and this might have influenced the clinical interpretation and case management. In addition to this, clinically “uncertain or questionable” CSF leakage cases could not be categorized and were excluded from the analysis (28 samples from 17 patients). These cases represent a clinical gray zone which is frequently encountered in the clinical reality and might have a relevant impact on the laboratory diagnostic performance.

Taken together, these data show that the use of an interpretation algorithm as the one proposed here can improve the performance of β-TP measurements and is a valuable tool that can easily implemented in the context of CSF leakage diagnostics.