Combining rules and machine learning for extraction of temporal expressions and events from clinical narratives

Kovačević, Aleksandar; Dehghan, Azad; Filannino, Michele; Keane, John; Nenadić, Goran

doi:10.1136/amiajnl-2013-001625

Cited by 73 publications

(48 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Traditionally, the sodium polystyrene sulfonate trigger is used to identify cases in which excessive potassium is administered, yet the poor PPV (27%) of sodium polystyrene sulfonate may in part be due to its recommended and appropriate use in TLS. 21 Hyperkalemia is a potentially life-threatening consequence of TLS, and frequent potassium measurements and oral administration of sodium polystyrene sulfonate are recommended. 21 Additionally, multiple nephrotoxic agents are often used in the treatment of oncology patients, and even careful monitoring may not prevent renal impairment with resultant hyperkalemia.…”

Section: Discussionmentioning

confidence: 99%

“…21 Hyperkalemia is a potentially life-threatening consequence of TLS, and frequent potassium measurements and oral administration of sodium polystyrene sulfonate are recommended. 21 Additionally, multiple nephrotoxic agents are often used in the treatment of oncology patients, and even careful monitoring may not prevent renal impairment with resultant hyperkalemia. Vitamin K was not ordered for patients on warfarin experiencing an ADE but instead was prescribed for patients experiencing chemotherapy or disease-induced thrombocytopenia, veno-occlusive disease, or malabsorption.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Adverse Drug Event Detection in Pediatric Oncology and Hematology Patients: Using Medication Triggers to Identify Patient Harm in a Specialized Pediatric Patient Population

Call

Burlison

Robertson

et al. 2014

The Journal of Pediatrics

View full text Add to dashboard Cite

Objective To investigate the use of a trigger tool for adverse drug event (ADE) detection in a pediatric hospital specializing in oncology, hematology, and other catastrophic diseases. Study design A medication-based trigger tool package analyzed electronic health records from February 2009 to February 2013. Chart review determined whether an ADE precipitated the trigger. Severity was assigned to ADEs, and preventability was assessed. Preventable ADEs were compared with the hospital’s electronic voluntary event reporting system to identify whether these ADEs had been previously identified. The positive predictive values (PPVs) of the entire trigger tool and individual triggers were calculated to assess their accuracy to detect ADEs. Results Trigger occurrences (n=706) were detected in 390 patients from six medication triggers, 33 of which were ADEs (overall PPV = 16%). Hyaluronidase had the highest PPV (60%). Most ADEs were category E harm (temporary harm) per the National Coordinating Council for Medication Error Reporting and Prevention (NCC MERP) index. One event was category H harm (intervention to sustain life). Naloxone was associated with the most grade 4 ADEs per the Common Terminology Criteria for Adverse Events (CTCAE) v4.03. Twenty-one (64%) ADEs were preventable; 3 of which were submitted via the voluntary reporting system. Conclusion Most of the medication-based triggers yielded low PPVs. Refining the triggers based on patients’ characteristics and medication usage patterns could increase the PPVs and make them more useful for quality improvement. To efficiently detect ADEs, triggers must be revised to reflect specialized pediatric patient populations such as hematology and oncology patients.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Adverse Drug Event Detection in Pediatric Oncology and Hematology Patients: Using Medication Triggers to Identify Patient Harm in a Specialized Pediatric Patient Population

Call

Burlison

Robertson

et al. 2014

The Journal of Pediatrics

View full text Add to dashboard Cite

show abstract

“…Our results in a similar clinical NER challenge in 2012 [24] showed that using supplementary training data (in addition to the one provided by the organizers) could have a positive effect on the performance of the ML models. Based on that, we experimented with enriching the training data with the data set provided in the 2014 de-identification challenge.…”

Section: Methodsmentioning

confidence: 88%

Learning to identify Protected Health Information by integrating knowledge- and data-driven algorithms: A case study on psychiatric evaluation notes

Dehghan

Kovačević

Karystianis

et al. 2017

Journal of Biomedical Informatics

Self Cite

View full text Add to dashboard Cite

De-identification of clinical narratives is one of the main obstacles to making healthcare free text available for research. In this paper we describe our experience in expanding and tailoring two existing tools as part of the 2016 CEGS N-GRID Shared Tasks Track 1, which evaluated de-identification methods on a set of psychiatric evaluation notes for up to 25 different types of Protected Health Information (PHI). The methods we used rely on machine learning on either a large or small feature space, with additional strategies, including two-pass tagging and multi-class models, which both proved to be beneficial. The results show that the integration of the proposed methods can identify Health Information Portability and Accountability Act (HIPAA) defined PHIs with overall F1-scores of ~90% and above. Yet, some classes (Profession, Organization) proved again to be challenging given the variability of expressions used to reference given information.

show abstract

“…Other top approaches used adapted regular expressions (Sohn et al, 2013) on top of the off the shelf Heideltime system (a general-domain NLP system for parsing time expressions) (Strötgen and Gertz, 2010). Another approach used a hybrid system where the output from a CRF-based system was combined with the output of a rule-based system (Kovačević et al, 2013).…”

Section: Introductionmentioning

confidence: 99%

Extracting Time Expressions from Clinical Text

Miller¹,

Bethard²,

Dligach³

et al. 2015

Proceedings of BioNLP 15

View full text Add to dashboard Cite

Temporal information extraction is important to understanding text in clinical documents. Temporal expression extraction provides explicit grounding of events in a narrative. In this work we provide a direct comparison of various ways of extracting temporal expressions, using similar features as much as possible to explore the advantages of the methods themselves. We evaluate these systems on both the THYME (Temporal History of Your Medical Events) and i2b2 Challenge corpora. Our main findings are that simple sequence taggers outperform conditional random fields on the new data, and higher-level syntactic features do not seem to improve performance.

show abstract

Combining rules and machine learning for extraction of temporal expressions and events from clinical narratives

Abstract: The system provides a state-of-the-art method that can be used to support automated identification of mentions of clinical events and temporal expressions in narratives either to support the manual review process or as a part of a large-scale processing of electronic health databases.

Cited by 73 publications

References 28 publications

Adverse Drug Event Detection in Pediatric Oncology and Hematology Patients: Using Medication Triggers to Identify Patient Harm in a Specialized Pediatric Patient Population

Adverse Drug Event Detection in Pediatric Oncology and Hematology Patients: Using Medication Triggers to Identify Patient Harm in a Specialized Pediatric Patient Population

Learning to identify Protected Health Information by integrating knowledge- and data-driven algorithms: A case study on psychiatric evaluation notes

Extracting Time Expressions from Clinical Text

Contact Info

Product

Resources

About