369. Using Hybrid Models and Blockchain Technology as a Means to Develop a Novel Propensity Score for Candidemia and Invasive Candidiasis

Background Early initiation of empiric antifungal therapy has been shown to decrease morbidity and mortality among patients with candidemia/invasive candidiasis (C/IC). However, the initiation of appropriate antifungal therapy is frequently delayed due to the severe limitations in early diagnosis. The goal of this study is to develop a high-risk scoring system to identify patients who may be eligible for preemptive antifungal therapy. The proposed new methodology combines hybrid modeling and blockchain technology. Methods Our approach is novel and using expert physicians’ perception of C/IC risk factors with those described in the hospitals through a set of models (hybrid model building from primary and secondary data). The goal is to improve the early detection of C/IC and initiate antifungal therapy. Once candidate hybrid models are derived, blockchain technology will be utilized. The methodology is based on vectors consisting of the ranking of candidiasis risk factors. These vectors will be constructed based on expert clinicians rank scores of known risk factors. Such methods are different than the usual statistical rank correlation computations, such as Spearman’s rank correlation, etc Results Preliminary analysis suggests threepotential models. Model 1: uses the following order of variables, by their relative importance: (1) major surgery within 0–3 days, (2)TPN-7–3 days, (3) steroids 0–3 days, (4) ECMO, (5) hemodialysis 0–3 days, (6) diabetes mellitus. Model 2 includes: (1)multifocal Candida colonization, 2.) central venous catheter 0–3 days, (3) LVAD, (4) medical ICU, (5) APACHE score > 20, (6) mechanical ventilation. Model 3 includes (1) pancreatitis –710 days, (2) diabetes mellitus, (3) hemodialysis 0–3 days, (4) central venous catheter 0–3 days, (5) TPN-7–3 days, 6.) APACHE score > 20. Conclusion Blockchain methods we propose are some of the first of their kind used in health research and are very suitable for the early detection of C/IC and other diseases where preemptive therapy is necessary. The following step will be to verify and use these models in the clinical realm and verify their effects on outcomes. Second we need to develop and evaluate our proposed methodology in building hybrid models, followed by algorithms for the early detection of diseases. These concepts still need to be fully evaluated on large population studies. Disclosures All authors: No reported disclosures.