Reporting the Degree of Certainty of CSF-Venous Fistulas in Patients with Spontaneous Intracranial Hypotension: The Duke CSF-Venous Fistula Confidence Score.

BACKGROUND AND PURPOSE: CSF-venous fistulas (CVFs) are a common cause of spontaneous intracranial hypotension (SIH). CVF identification and localization are critical for diagnosis and treatment, but inconsistent visualization of CVFs on myelography leads to diagnostic uncertainty. Diagnostic confidence impacts treatment decisions. However, there is currently no standardized method for reporting the degree of confidence about the presence or absence of a CVF on CT myelography (CTM). The purposes of this study are to present a novel instrument to provide structured communication of the degree of certainty about the presence of a CVF and to determine the interreader and intrareader agreement of this scoring system for determining the presence of a CVF at a given spinal level on CTMs.

MATERIALS AND

METHODS

This retrospective study assessed the interreader and intrareader reproducibility of a scoring system anchored in previously reported objective imaging findings, including the attenuation of paraspinal veins associated with CVFs. We included CTMs from patients with SIH performed between 10/2017-03/2024 at 1 institution. Exclusion criteria were CSF leak other than CVF, prior transvenous embolization, and nondiagnostic CTMs. Several potential iterations of the scoring system were developed. The study cohort consisted of a balanced set of cases representative of varying degrees of certainty: definite, high probability, low probability, and negative (25 each). Five radiologists (3-19 years of experience) provided their blinded subjective confidence assessment and then applied the scoring system. Interreader and intrareader agreements were calculated for the different scoring system models using kappa statistics.

RESULTS

The best-performing model produced substantial mean intrareader agreement, closely approximated the number of definite CVFs, and was adopted as the final model. Interreader agreement for the adopted model was moderate, replicating that for the subjective interpretations. Other versions of the model produced fair-to-moderate interreader agreements and were not adopted.

CONCLUSIONS

We developed a structured reporting system anchored in objective imaging findings that communicates the degree of certainty about the presence of CVF on CTM. This system replicates assessments by expert readers and meets a critical need for improved communication both in daily clinical practice and in research by providing a method for objectively quantifying the certainty of CVF diagnosis.