The PubMed/Medline, Cochrane Library, and EMBASE databases were searched using the terms 14-3-3 protein, CSF analysis, Creutzfeldt-Jakob disease, prion disease, dementia, and rapidly progressive dementia (exploded terms where appropriate) (appendix e-3). In addition, the reference lists of the articles identified were hand searched to identify articles that may have been missed in the initial search. Duplicates, reviews without original data, meeting abstracts, and case reports/series were excluded. The search included English-language articles and covered publications ranging from 1995 to January 1, 2011.
Studies in human subjects above 18 years of age were included. Nonsporadic cases from growth hormone use, genetic, iatrogenic (postsurgical), and new-variant (mad cow disease) prion diseases were excluded. Also excluded were non-CJD prion disorders and animal studies. In studies that looked at a mix of sCJD and other CJD subtypes, only the data on patients with sCJD were extracted for the analysis.
The articles were rated for their risk of bias according to the American Academy of Neurology classification of evidence criteria for diagnostic testing (appendix e-4), and recommendations were linked to the level of evidence (appendix e-5). In accordance with these criteria, studies with incorporation bias (the results of the 14-3-3 protein assay influenced the determination of the presence of sCJD) are rated Class IV, and studies with spectrum bias (which excluded a priori patients with uncertain diagnoses of sCJD) are rated Class III.
The number of patients, study design, data collection methods, patient population, diagnostic reference standard, and type of 14-3-3 assay used were collected. Other data extracted included the raw numbers of patients who tested positive vs negative for CSF 14-3-3 and their clinical or pathologic diagnosis, or both.
We constructed 2 × 2 tables with the presence or absence of 14-3-3 CSF protein as the independent variable and the presence or absence of sCJD as the dependent variable. We calculated sensitivities and specificities as the primary measures of diagnostic accuracy. Ninety-five percent confidence intervals (CIs) of the sensitivities and specificities were used as the measure of statistical precision.
CSF assays of 14-3-3 protein were dichotomized as either present or absent according to the authors' method of interpretation. For studies in which 14-3-3 analyses were performed with more than one technique,9,19
sensitivities and specificities were calculated for each technique used. To determine whether varying definitions of abnormal 14-3-3 (i.e., varying cutpoints for abnormal) introduced a threshold effect, we calculated the correlation between sensitivities and specificities (Spearman rank correlation coefficient).
We anticipated that the presence or absence of sCJD would be determined with varying levels of diagnostic certainty: definite sCJD—pathologic confirmation by biopsy or autopsy; probable sCJD—typical clinical presentation combined with confirmatory test results (e.g., periodic pattern on EEG, restricted diffusion on diffusion-weighted MRI); possible sCJD—typical clinical presentation without confirmatory ancillary tests; confirmed negative sCJD—pathologic confirmation or a prolonged course not compatible with sCJD.
For the primary calculation of sensitivity and specificity, we categorized patients meeting definite and probable criteria as sCJD present. Possible and confirmed negative sCJD were categorized as sCJD absent. To estimate a single value for the sensitivities and specificities we pooled the results of the identified studies with the lowest risk of bias in a meta-analysis. When studies appeared to acquire their patients from the same surveillance centers, signaling that the same population may have been used in more than one study,4,7,20
only the study enrolling the largest number of patients was considered in the pooled analysis.
To determine the dependence of the estimated sensitivities and specificities on the level of diagnostic certainty, we performed 2 sensitivity analyses. In these analyses, we categorized patients as either sCJD present or sCJD absent, using varying criteria as follows: primary analysis: sCJD present—definite and probable; sCJD absent—possible and confirmed negative; sensitivity analysis 1: sCJD present—definite; sCJD absent—probable, possible, and confirmed negative; sensitivity analysis 2: sCJD present—definite, probable, and possible; sCJD absent—confirmed negative.
The sensitivities and specificities resulting from these different analyses were compared through plotting of the data on a receiver operator characteristic (ROC) graph.