Abstract
Abstract
Background
Spectrophotometry of cerebrospinal fluid (CSF) for bilirubin is the recommended method for investigation in suspected cases of subarachnoid haemorrhage (SAH), when a computed tomography (CT) of the head is negative for blood. There is a potential need for a simpler alternative. Measurement of CSF ferritin might fulfil this need.
Method
We have measured ferritin in the CSF from 252 patients with suspected SAH who were negative on a CT of the head for blood, recruited on a consecutive intention to recruit basis from four centres. CSF spectrophotometry was performed on all samples. A positive outcome was taken as an aneurysm found on angiography that was treated or a discharge diagnosis of non-aneurysmal SAH.
Results
A final diagnosis of aneurysmal SAH was made in six patients, an arteriovenous malformation in one and non-aneurysmal SAH in nine. Receiver operating characteristic (ROC) analysis showed that at 6.4 μg/L, sensitivity, specificity, positive and negative predictive values were 1.0, 0.48, 0.12 and 1.0, respectively. At 12 μg/L, these values were 0.81, 0.91, 0.38 and 0.98, respectively.
Conclusions
At an appropriate negative predictive value (1.0) for a rule-out test, ferritin has too low a specificity to function as a stand-alone test and we cannot recommend it as an initial screen to be followed by spectrophotometry.
Introduction
It is of vital importance not to miss the diagnosis of subarachnoid haemorrhage (SAH). Where it is missed, and this is most likely to be in those who are neurologically intact at presentation, the consequences of re-bleeding and vasospasm can be very serious and leave the patient with significantly more neurological impairment than would have been the case had the original bleed been diagnosed and appropriately treated. 1 The reasons for misdiagnosis are: first, failure to appreciate the spectrum of clinical presentation; secondly, failure to understand the limitations of computed tomography (CT); and thirdly, failure to perform and correctly interpret the results of lumbar puncture. 1
Once the diagnosis has been entertained, the appropriate investigation is CT of the head without contrast for the presence of blood. 2 This will be positive in up to 98% of patients scanned within 12 h of a bleed, but positivity drops off with time such that it may be only 50% sensitive at one week after a bleed. 3,4 While new generation scanners may return an even higher sensitivity than quoted, 5 at present it is still well accepted that a negative scan does not exclude a SAH and a further investigation is necessary, especially where presentation is delayed, the bleed has been small or in an anatomically ill-defined area. 1
The further investigation when blood has been detected by CT is cerebral angiography. 2 This is not appropriate for CT-negative patients because of: the risk of stroke associated with conventional catheter angiography; the resource implications were magnetic resonance and CT angiography to be used; and the risk of finding unruptured aneurysms which may not require treatment. 2 Therefore before proceeding to angiography in CT-negative cases, an alternative investigation is required to delineate those few patients who are likely to have bled and who should proceed to angiography. Historically, laboratory investigations that predated CT by decades have been used, such as examination of cerebrospinal fluid (CSF) for red cells and visual inspection for xanthochromia. 6,7 More recently, detection of bilirubin in CSF by spectrophotometry at concentrations below those detectable by the naked eye has been adopted in Guidelines for best practice. 8 These have been published and are to date our recommended further investigation. However, in some parts of the world, particularly North America, visual inspection of the CSF supernatant for xanthochromia remains the favoured examination, despite its proven lack of sensitivity. 9–12 With spectrophotometry, there is a need for experience and expertise to perform such a manual analysis and to interpret the results; a necessary 12 h delay post ictus before lumbar puncture is appropriate; and the introduction of red cells from a bloody tap and subsequent lysis to release oxyhaemoglobin may cause uncertainty of interpretation. Very recently, spectrophotometry has been criticized as being non-specific and leading to unnecessary angiography. 12 An automated analysis with a single diagnostic cut-off could answer these criticisms.
Two approaches that might prove more suitable than spectrophotometry are the adaptation of a serum bilirubin method to measure CSF bilirubin, a procedure that has been validated and previously subject to a critique in this journal; 13,14 and the measurement of CSF ferritin. 15,16 Ferritin analysis is readily available in most blood science laboratories. Ferritin is synthesized in the CNS in response to the iron released from haem following an intracranial bleed. 15 While CSF ferritin has been shown to be elevated following an intracranial bleed in two studies and potentially of value in the group of patients we are considering, there has been no in-depth examination of its use in CT-negative patients as has been undertaken for CSF bilirubin. 16,17 In this study, we have examined the diagnostic efficiency of CSF ferritin in patients with suspected SAH who are negative on CT for an intracranial bleed.
Methods
The study involved four centres: two were District General Hospitals (DGH) providing an analytical service for their own admissions; one, a DGH/tertiary neurosurgery centre providing an analytical service for its own admissions and those of three neighbouring DGHs; and the fourth, a tertiary neurosurgical centre providing an analytical service via referrals for a population of about four million people. Research ethics approval was granted to all participating centres.
Recruitment was on an intention to consent basis following submission of a CSF for spectrophotometry because of the possibility of SAH in the presence of a negative or equivocal CT scan of the head for blood. After spectrophotometry according to published criteria, 8 CSF was stored at −20°C until analysis for ferritin in a single centre on a BN Prospec immunoassay platform (Dade-Behring, Milton Keynes, Buckinghamshire, UK–now part of Siemens Healthcare Diagnostics), without knowledge of spectrophotometry result or patient outcome. Spectrophotometry was interpreted with a positive cut-off for net bilirubin absorbance (NBA) of >0.007 AU. 8 A reference range for CSF ferritin had previously been established with an upper limit of 12 μg/L. 15,16 At concentrations of 5 and 13 μg/L, intra-assay SD (CV) were 0.31 (6.2%) and 0.62 (4.8%), respectively. At 5 μg/L, an increment (decrement) of 0.9 μg/L defines values that are analytically significantly different from 5 at the 95% confidence level. The lower limit of detection of the assay was 0.42 μg/L. Clinical outcome was deemed to be positive if angiography revealed an aneurysm or other cause of bleed that was treated or the discharge summary was one of non-anuerysmal SAH. Outcome was deemed negative for a bleed on the basis of discharge summary and, where available, subsequent review of the notes about one year after discharge, which indicated no subsequent admission for a proven intracranial bleed. Stata software, version 9.2, Statcorp, Texas, USA was used for statistical calculations.
Results
The four centres consented 263 patients, with ferritin results available for 252. Fourteen of the 252 were from centre 1, 37 from centre 2, 179 from centre 3 and 22 from centre 4. Notes from 233 patients were available for review at a median time of 14 months after discharge (range 8–24). Sixteen patients were given a final diagnosis of a SAH; six of these patients were treated for aneurysms, one for an arteriovenous malformation and nine were diagnosed with non-aneurysmal SAH. In four of the 16, the initial CT was initially reported as negative but reclassified as positive on neuroradiological review. There were 236 patients without a SAH. Figure 1 indicates the distribution of CSF ferritin between the two groups. Ferritin was greater than 12 μg/L in 13 of the SAH group (range 6.5–6790 μg/L) and in 20 of the non-SAH group (range 1.8–4410 μg/L); bilirubin was increased in all of the SAH group (range 0.008–0.335 AU). A receiver operating characteristic (ROC curve) indicates the relationship between the sensitivity and specificity for CSF ferritin as an indicator of SAH together with sensitivity, specificity, positive and negative predictive values (PPV, NPV) and positive and negative likelihood ratios at key CSF ferritin cut-offs (Figure 2, Table 1). Up to a cut-off of 6.4 μg/L, ferritin maintains a sensitivity of 1 and corresponding NPV of 1.0. At this point, the specificity is 0.48, PPV 0.12. At 8 μg/L, sensitivity is 0.81, specificity 0.70. At the upper limit of the reference range, sensitivity is 0.81, NPV is 0.98, specificity is 0.91 and PPV is 0.38. We have not determined corresponding values for bilirubin as an NBA >0.007 was a significant determinant in deciding whether the patient underwent angiography. However, at a cut-off of 6.4 μg/L, ferritin predicts with 100% sensitivity (NPV 1.0) all 29 patients with an increased NBA. To detect all patients with an increased NBA, 137 spectrophotometry scans would have been necessary.
Scattergram of cerebrospinal fluid (CSF) ferritin results for those patients with no subarachnoid haemorrhage (SAH) compared with those with SAH as represented in a Box and Whiskers plot. The upper line of the box represents the 75th percentile of data, the lower line the 25th and the midline the median and the bars encompass all points apart from those individually plotted. Two data points, of 373 and 4410 μg/L are omitted from the non-SAH group, and four, of 227, 457, 2510 and 6790 μg/L from the SAH group
ROC curve for ability of cerebrospinal fluid ferritin to predict subarachnoid haemorrhage. Area under the curve = 0.91 (95% CI 0.86–0.94)
Statistical parameters (±95% confidence limits) derived from ROC curve analysis (Figure 2) for ability of different cut-offs of cerebrospinal fluid ferritin to predict an intracranial bleed
PPV, positive predictive value; NPV, negative predictive value; LR+, positive likelihood ratio; LR−, negative likelihood ratio
Discussion
The requirement for a screening test that will determine which of those patients presenting with symptoms suggestive of SAH proceed to a more definitive study by angiography is that it should have an NPV of 1.0 such that no patients with the disease are missed. This is provided by a CSF ferritin of 6.4 μg/L (or lower, to allow for analytical uncertainty at this concentration). At this cut-off, the specificity is, at best only 0.48, indicating that were CSF ferritin to be used as a stand-alone test in this context, well over half the population would require angiography, an unacceptable number in terms of the resource required and the risk of finding unruptured aneurysms. Nevertheless, there remains the possibility of using ferritin as an initial test. Those with a ferritin below the cut-off could be rapidly discharged without the need for spectrophotometry, which would then be necessary only for those with a ferritin above the cut-off, at least 137 (54%) in this study. The fact that three patients had a CSF ferritin well within two previously determined reference ranges suggests that ferritin synthesis in response to a bleed is very variable. Finally, there would need to be an assessment of the balance between achieving a more simple rule-out test than at present and the need to perform the second test on more than half the population. One advantage of ferritin may be that it remains elevated for longer than bilirubin and may therefore be superior when there is delayed presentation. 16 There is no information in this study that would support or refute this hypothesis.
The findings in this study have to be compared with the few others that exist and reasons put forward for any discrepancies. Page and colleagues, in a study of patients with and without positive CT of the head for blood, identified four patients with a negative CT and a final diagnosis of SAH, three of whom had a CSF ferritin of >12 μg/L yielding a sensitivity of 75%. 17 Interestingly, one patient out of 12 with positive CTs had a ferritin of <12 μg/L. In a slightly more extensive study of 24 patients, all with negative CT, O'Connell and Watson demonstrated a sensitivity of 90% for a CSF ferritin cut-off of >12 μg/L in the 10 patients with an intracranial bleed. 16 These sensitivities are thus of the same order, given the confidence intervals likely to pertain to these studies where there are few true positive outcomes.
Three criticisms can be made of this study. The first is that of patient selection, which was determined by the ability of the authors to consent patients into the study. This manifested itself in two ways. First, there were certainly patients where there was an intention to recruit where patients declined the invitation to participate or where the patient could not even be contacted. Secondly, in at least one of the centres, a number of the patients who received a final diagnosis of an intracranial bleed were referred from other hospitals for treatment. These patients had CSF referred from the other hospitals for analysis and because they were then transferred for treatment could then be consented. The corresponding patients at these referring hospitals with negative CSF spectrophotometry were outside the scope of the ethics approval for the study. The second criticism is that of using a diagnosis of non-aneurysmal SAH as a criterion for positivity. Such a diagnosis can only be made with certainty in patients with a positive CT on admission. Where the CT is negative, the diagnosis of non-aneurysmal SAH has to rely on the finding of a positive CSF spectrophotometry that cannot be assigned to another cause. In this study such a consideration affected nine patients (56%). In patients presenting with blood on CT, up to 20% are given a diagnosis of non-aneurysmal SAH. 18 The third criticism relates to using follow-up rather than angiography as supportive of absence of an intracranial bleed. For reasons already given, it is not appropriate to carry out angiography where the risk of bleeding is very low. Lack of angiography to determine the outcome in all cases introduces a differential standard bias, is a breach of the Standards for Reporting Diagnostic Accuracy (STARD) criteria for a diagnostic study and will result in an overestimate of diagnostic power. 19,20 Where an initial incident could have been due to an occult bleed, the risk of re-bleeding remains about 3% per annum, considerably higher than that of the background population. 2 We cannot therefore exclude the fact that among the population negative for CSF findings supporting a bleed there were in fact patients who had experienced a bleed, but who yet had to suffer a re-bleed that would appropriately define the initial event.
Despite these reservations, we maintain that this study is of value in defining the role of ferritin in the investigation of CT-negative suspected SAH. While having a single cut-off would appear to give it the advantage of simplicity of interpretation, the use of a readily automated assay on platforms present in most analytical laboratories give it an advantage in ease of assay, the number of samples that would need to be referred for a second assay and uncertainty over the reliability of appearance of ferritin in the CSF lead us to conclude that at present we cannot recommend it for routine practice.
Footnotes
ACKNOWLEDGEMENT
We thank Roberto Melotti of the Research and Development Support Unit, North Bristol NHS Trust for statistical advice and provision of the figures.