Hao Liu

Selected Publications

1. AMIA

   Can Race-sensitive Biomedical Embeddings Improve Healthcare Predictive Models?

   Hao Liu, Nour Moustafa-Fahmy, Casey Ta, and Chunhua Weng

   AMIA Informatics Summits Mar 2023

   Abs HTML

   This reproducibility study presents an algorithm to weigh in race distribution data of clinical research study samples when training biomedical embeddings. We extracted 12,864 PubMed abstracts published between January 1st, 2000 and January 1st, 2022 and weighed them based on the race distribution data extracted from their corresponding clinical trials registered on ClinicalTrials.gov. We trained Word2vec and BERT embeddings and evaluated their performance on predicting length of hospital stay (LHS) and intensive care unit (ICU) readmission using MIMIC-IV electronic health record data. We observed that models trained using race-sensitive embeddings do not consistently outperform the neutral embeddings ones when used for LHS prediction (with similar Mean Absolute Error 1.975 vs. 2.008) or ICU readmission prediction (with similar accuracy 74.61% vs. 75.17% and the same AUC 0.775), respectively. We conclude that demographic sensitive embeddings do not necessarily significantly improve the accuracy of health predictive models as previously reported in the literature.

2. JBI

   Ontology-based categorization of clinical studies by their conditions

   Hao Liu, Simona Carini, Zhehuan Chen, Spencer Phillips Hey, Ida Sim, and Chunhua Weng

   J. Biomed. Inform. Nov 2022

   Abs HTML Code

   OBJECTIVE: The free-text Condition data field in the ClinicalTrials.gov is not amenable to computational processes for retrieving, aggregating and visualizing clinical studies by condition categories. This paper contributes a method for automated ontology-based categorization of clinical studies by their conditions. MATERIALS AND METHODS: Our method first maps text entries in ClinicalTrials.gov’s Condition field to standard condition concepts in the OMOP Common Data Model by using SNOMED CT as a reference ontology and using Usagi for concept normalization, followed by hierarchical traversal of the SNOMED ontology for concept expansion, ontology-driven condition categorization, and visualization. We compared the accuracy of this method to that of the MeSH-based method. RESULTS: We reviewed the 4,506 studies on Vivli.org categorized by our method. Condition terms of 4,501 (99.89%) studies were successfully mapped to SNOMED CT concepts, and with a minimum concept mapping score threshold, 4,428 (98.27%) studies were categorized into 31 predefined categories. When validating with manual categorization results on a random sample of 300 studies, our method achieved an estimated categorization accuracy of 95.7%, while the MeSH-based method had an accuracy of 85.0%. CONCLUSION: We showed that categorizing clinical studies using their Condition terms with referencing to SNOMED CT achieved a better accuracy and coverage than using MeSH terms. The proposed ontology-driven condition categorization was useful to create accurate clinical study categorization that enables clinical researchers to aggregate evidence from a large number of clinical studies.

3. JBI

   A knowledge base of clinical trial eligibility criteria

   Hao Liu, Chi Yuan, Alex Butler, Yingcheng Sun, and Chunhua Weng

   J. Biomed. Inform. May 2021

   Abs HTML

   OBJECTIVE: We present the Clinical Trial Knowledge Base, a regularly updated knowledge base of discrete clinical trial eligibility criteria equipped with a web-based user interface for querying and aggregate analysis of common eligibility criteria. MATERIALS AND METHODS: We used a natural language processing (NLP) tool named Criteria2Query (Yuan et al., 2019) to transform free text clinical trial eligibility criteria from ClinicalTrials.gov into discrete criteria concepts and attributes encoded using the widely adopted Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) and stored in a relational SQL database. A web application accessible via RESTful APIs was implemented to enable queries and visual aggregate analyses. We demonstrate CTKB’s potential role in EHR phenotype knowledge engineering using ten validated phenotyping algorithms. RESULTS: At the time of writing, CTKB contained 87,504 distinctive OMOP CDM standard concepts, including Condition (47.82%), Drug (23.01%), Procedure (13.73%), Measurement (24.70%) and Observation (5.28%), with 34.78% for inclusion criteria and 65.22% for exclusion criteria, extracted from 352,110 clinical trials. The average hit rate of criteria concepts in eMERGE phenotype algorithms is 77.56%. CONCLUSION: CTKB is a novel comprehensive knowledge base of discrete eligibility criteria concepts with the potential to enable knowledge engineering for clinical trial cohort definition, clinical trial population representativeness assessment, electronical phenotyping, and data gap analyses for using electronic health records to support clinical trial recruitment.

4. JBI

   Concept placement using BERT trained by transforming and summarizing biomedical ontology structure

   Hao Liu, Yehoshua Perl, and James Geller

   J. Biomed. Inform. Dec 2020

   Abs HTML

   The comprehensive modeling and hierarchical positioning of a new concept in an ontology heavily relies on its set of proper subsumption relationships (IS-As) to other concepts. Identifying a concept’s IS-A relationships is a laborious task requiring curators to have both domain knowledge and terminology skills. In this work, we propose a method to automatically predict the presence of IS-A relationships between a new concept and pre-existing concepts based on the language representation model BERT. This method converts the neighborhood network of a concept into “sentences” and harnesses BERT’s Next Sentence Prediction (NSP) capability of predicting the adjacency of two sentences. To augment our method’s performance, we refined the training data by employing an ontology summarization technique. We trained our model with the two largest hierarchies of the SNOMED CT 2017 July release and applied it to predicting the parents of new concepts added in the SNOMED CT 2018 January release. The results showed that our method achieved an average F1 score of 0.88, and the average Recall score improves slightly from 0.94 to 0.96 by using the ontology summarization technique.