Improving Clinical Interpretation of the Anti-Platelet Factor 4/Heparin Enzyme-Linked Immunosorbent Assay for the Diagnosis of Heparin-Induced Thrombocytopenia Through the Use of Receiver Operating Characteristic Analysis, Stratum-Specific Likelihood Ratios, and Bayes Theorem
Journal Articles
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
BACKGROUND: Heparin-induced thrombocytopenia (HIT) is diagnosed using clinical criteria and detection of platelet-activating anti-platelet factor 4/heparin (anti-PF4/H) antibodies, usually through a surrogate enzyme-linked immunosorbent assay (ELISA). The high false-positive rate (FPR) of this ELISA prompted us to reexamine its interpretation. METHODS: We analyzed anti-PF4/H ELISA results from a previously published dataset of 1,958 patients, using clinical suspicion and serotonin-release assay (SRA) to diagnose HIT. We performed receiver operating characteristic (ROC) analysis using stratum-specific likelihood ratios (SSLRs) and used Bayes theorem to construct a clinical decision-support algorithm. RESULTS: The most discriminant single cutoff by anti-PF4/H ELISA for the diagnosis of HIT was found to be 0.8 optical density (OD) units, not 0.4 OD (currently accepted practice). This change reduced the FPR from 31% to 6% (95% CI, 5%-8%). ELISA results were grouped into five strata, which yielded SSLRs ranging from 0.02 (strongly ruling HIT out) to 104.4 (strongly ruling HIT in). Comparison of ROC curves demonstrated that this five-strata approach is statistically more accurate than current accepted practice at discriminating whether patients have HIT or not (area under the ROC curve, 0.97 [95% CI, 0.93-1.00] vs 0.83 [95% CI, 0.80-0.89]). Our decision-support algorithm incorporated clinical assessment into this stratified model and clarified HIT diagnosis with a high degree of certainty and without the need for SRA testing in approximately 90% of patients. CONCLUSIONS: Diagnostic accuracy of the anti-PF4/H ELISA can be optimized by using a higher cutoff and a stratified interpretation of the results. Our algorithm should significantly reduce overdiagnosis of HIT and the need for SRA testing.
