International Workshop on Learning with Knowledge Graphs: Construction, Embedding, and Reasoning

A knowledge graph (KG) consists of numerous triples, in which each triple, i.e., (head entity, relation, tail entity), denotes a real-world assertion. Many large-scale KGs have been developed, e.g., general-purpose KGs Freebase and YAGO. Also, lots of domain-specific KGs are emerging, e.g., COVID-19 KGs, biomedical KGs, and agricultural KGs. By embedding KGs into low-dimensional vectors, i.e., representations of entities and relations, we could integrate KGs into machine learning models and enhance the performance of many prediction tasks, including search, recommendations, and question answering. During the construction, refinement, embedding, and application of KGs, a variety of KG learning algorithms have been developed to handle challenges in various real-world scenarios. Moreover, graph neural networks have also brought new opportunities to KG learning. This workshop aims to engage with active researchers from KG communities, recommendation communities, natural language processing communities, and other communities, and deliver state-of-the-art research insights into the core challenges in KG learning. © 2023 Owner/Author.

Original Article Combined CT score, blood mononuclear cell count, LDH, and plasma D-dimer for viral pneumonia diagnosis: a retrospective study

Objective: Due to a continued increase in viral pneumonia incidence and resulting high mortality, fast and accurate diagnosis is important for effective management. This investigation examined the significance of blood biomarkers and the CT score in the early diagnosis of viral pneumonia. Methods: Patients who were hospitalized due to radiologically-confirmed pneumonia and underwent virus antigen rapid test were enrolled. Their clinical information was compared. Blood mononuclear cell count, LDH, and plasma D-dimer were obtained. To evaluate the utility of biomarker levels in differentiating viral pneumonia from other pneumonia, ROC curves were developed to analyze the AUC. The optimal cut-off thresholds, specificity, sensitivity, and predictive values were assessed using the Youden index. The added value of the multi-marker approach was delineated using IDI and Reclassification analyses using NRI;IDI and NRI values were examined with 95% CI. Results: Overall, 1163 inpatients were recruited between January 2017 and January 2021. They were sub-divided into the viral pneumonia (n = 563) and non-viral pneumonia (n = 600) categories. We found that the CT score, blood mononuclear cell count, LDH, and plasma D-dimer were markedly elevated in viral pneumonia patients. At an LDH threshold of 693.595 U/L, an AUC of ROC was 0.805 in differentiating viral pneumonia. The combination of CT score and blood biomarkers had an ROC AUC value of 0.908. Conclusions: Combining elevated biomarkers with CT assessments outperformed the CT score alone in identifying viral pneumonia. It is crucial to better characterize the significance of biomarkers in combination with CT assessments in the diagnosis of viral pneumonia.

OAB-003: CARDAMON:Carfilzomib (K) maintenance following Autologous Stem Cell Transplant (ASCT) or carfilzomib-cyclophosphamide-dexamethasone (KCd) consolidation for newly diagnosed (NDTE) multiple myeloma (MM)

Background: The role of upfront ASCT for NDTE MM remains under evaluation with high MRD rates following novel induction and consolidation (cons) strategies. K maintenance represents an alternative strategy to lenalidomide maintenance. The CARDAMON trial investigated K maintenance following KCd induction plus either ASCT or KCd cons. Methods: NDTE pts received 4 x KCd induction (K 20/56 mg/m2 biweekly, C 500 mg D 1,8,15, d 40mg weekly) before 1:1 randomisation to ASCT or 4 x KCd cons followed by 18 cycles K maintenance (56mg/m2 D1,8,15). Flow cytometric MRD (10-5) was assessed post induction, pre-maintenance and at 6 months maintenance. Primary endpoints were ≥VGPR post induction and 2-year PFS from randomisation. Secondary endpoints included improvements in disease response and MRD conversion following ASCT/ cons and maintenance. Results: 281 patients were registered, with 218 randomised to either ASCT or cons. The median PFS for ASCT was not yet reached vs 3.4 years for cons, with cons failing to show non-inferiority (difference in 2-year PFS 6.5%, 70% CI 1.0% to 11.1%). 196 patients received K maintenance (99 ASCT, 97 cons), 17 remain on treatment. A median of 16 cycles (1-18) were given over a median of 15.9 months (0-21.5). COVID-19 led to maintenance treatment interruptions in 41 (8 ASCT, 6 Cons) and treatment discontinuation in 15 (9 ASCT, 6 Cons). The median K dose given was 50.6mg/m2 and was similar across both arms (51.2 vs 49.4mg/m2, p=0.03). K maintenance was discontinued for PD in 14.1% (ASCT) vs 22.7% (cons), and for adverse events (AEs) in 7.1% (ASCT) vs 4.1% (cons). Most common AEs were hypertension and infections and more ≥G3 AEs were noted in ASCT vs cons (p=0.01). Patient/ clinician withdrawals from maintenance were low but occurred more in the ASCT arm (9.1% vs 1%). MRD neg patients post ASCT/ Cons had a longer PFS than MRD pos (p=0.002);with a higher MRD neg rate in the ASCT arm (53.6% vs 35.1% in Cons, p=0.01). MRD neg patients at 6 months post maintenance also had longer PFS (p=0.004 cf MRD pos patients);again with higher MRD neg rates in the ASCT arm (58.1% ASCT vs 40.5% Cons, p=0.02). There was no difference in PFS for MRD neg patients according to treatment arm from PBSCH, post-ASCT/ Cons or 6 months maintenance timepoints. Overall, 27.8% of MRD pos patients converted to MRD neg post ASCT/ Cons with more converting with ASCT (39.1% ASCT vs 16.1%, p=0.004). 23.5% of MRD pos patients converted to neg during maintenance (30.6% ASCT, 17.8%: p=0.2). Maintenance of MRD negativity over the first 6 months was similar between ASCT and Cons arms (p=0.3). There was no evidence that the timing of achievement of MRD negativity impacted PFS. Conclusions: K maintenance at 56mg/m2 weekly was deliverable and tolerable, with continued higher MRD neg rates at 6 months post-ASCT compared to post-Cons. However more ≥G3 AEs and discontinuations for AEs/ patient choice were noted for K maintenance after ASCT.

EPIDEMIC MODELS WITH VARYING INFECTIVITY

We introduce an epidemic model with varying infectivity and general exposed and infectious periods, where the infectivity of each individual is a random function of the elapsed time since infection, those function being independent and identically distributed for the various individuals in the population. This approach models infection-age-dependent infectivity and extends the classical SIR and SEIR models. We focus on the infectivity process (total force of infection at each time) and prove a functional law of large number (FLLN). In the deterministic limit of this FLLN, the joint evolution of the mean infectivity and of the proportion of susceptible individuals is determined by a two-dimensional deterministic integral equation. From its solutions, we then obtain expressions for the evolution of the proportions of exposed, infectious, and recovered individuals. For the early phase, we study the stochastic model directly by using an approximate (non-Markovian) branching process and show that the epidemic grows at an exponential rate on the event of nonextinction, which matches the rate of growth derived from the deterministic linearized equations. We also use these equations to derive the expression for the basic reproduction number R-0 during the early stage of an epidemic, in terms of the average individual infectivity function and the exponential rate of growth of the epidemic, and apply our results to the Covid-19 epidemic.