BEST_ISH_poster.pdf

1Office of Biostatics and Epidemiology/ Center for Biologics Evaluation and Research/FDA, Silver Spring, MD USA, 2IQVIA,Cambridge, MA USA; 3Observational Health Data Sciences and Informatics, New York, NY USA; 4Georgia Tech Research Institute, Atlanta, GA USA

Alan E Williams1, Azadeh Shoaibi1, Barbee I Whitaker1, Christian G Reich2,3, Sara E Dempster2,3, Jon Duke4, and Steven A Anderson1

Background

Acknowledgements

Conclusions

BEST Initiative Goals

Figure 1. Current Passive HV Reporting Paradigm

• Hemovigilance (HV) worldwide has typically been implemented as passive surveillance with observation and reporting by the responsible clinical care team (Figure 1).

• The effectiveness of passive HV depends upon local clinical signal detection, consistent analysis, and accurate reporting of transfusion-associated adverse events (AE). This is frequently challenging in an environment of complex patient histories and limited institutional resources.

• The US FDA has implemented systems for active post-market surveillance of most FDA-regulated products over the past decade1; however, specific hurdles to active HV have included the absence of interoperable healthcare data systems; historical reliance on administrative (billing and diagnosis) codes for identification of transfusion exposures and outcomes; and health data nomenclature and transfusion AE clinical case definitions that remain somewhat clinically subjective.

• For the BEST Initiative, FDA entered into a partnership with scientific collaborators IQVIA, the University of California Los Angeles, and the Georgia Tech Research Institute, and both scientific and data collaborators Columbia University, Regenstrief Institute / The University of Indiana, and Stanford University in 2017.

• Collaborating partners implemented the Observational Health Data Sciences and

Informatics (OHDSI, www.ohdsi.com) distributed Observational Medical Outcomes

Partnership (OMOP) common data model (CDM; Figure 2).

• With the assistance of local institutional blood bank staff, each partner site identified and downloaded ISBT-128 codes reflecting component transfusions from 2012-present at their respective institutions2.

• Procedures for NLP event “time-stamping” and text data mining from transfusion nursing notes were developed.

• CP were drafted based on definitions from the literature3and the CDC HV protocol4.

• BEST data queries are developed using the OHDSI distributed data model (Figure 2):

• OHDSI is a multi-stakeholder, interdisciplinary collaborative to use health data through large-scale analytics;

• The OMOP CDM allows interoperability and includes access to OHDSI Analytic Tools (design, analysis, and semi-automated chart review);

• All solutions are open-source and OHDSI tools are publicly available for interoperability and common analysis.

Results

• The first year of CBER BEST development has established an infrastructure and related procedures for enhanced active hemovigilance using BEST partner EHR databases.

• Enhanced interoperability of infrastructure is needed to fully realize the BEST Initiative goals.

• BEST will continue the development, refinement, and validation of transfusions and AE queries using coded and textual data.

• Distributed application of commonly established queries through resources such as these Common Data Models may be an ideal way to harmonize international hemovigilance efforts (Figure 2).

The authors would like to acknowledge the work of partners Columbia University, Stanford University, the Regenstrief Institute and the University of Indiana and the contributions of IQVIA Inc, Observational Health Data Sciences and Informatics (OHDSI), Georgia Tech Research Institute, and the University of California Los Angeles.

References

METHODS

Figure 3. Development and NLP Application of a Computable Phenotype (CP)

ISBT AE Definition

(e.g. TACO)

Develop CP for the AE

AE NLP Query that can be applied to text notes

Disclaimer: The views expressed in the poster are those of the authors and may not reflect the views of the Food and Drug Administration.

700,000 patient records

~1,000,000 Notes from those 24,000 patients

70,000 patients with transfusions

24,000 transfused patients with notes

AE

NLP

Query

• The Biologics Effectiveness and Safety (BEST) Sentinel Initiative was developed by the Office of Biostatics and Epidemiology (OBE), Center for Biologics Evaluation and Research (CBER), US FDA in late 2017 as a “big data” active HV approach with several specific goals:

• Improve identification and characterization of blood component exposures within electronic health records (EHR) through the importation of ISBT-128 blood component labeling codes.

• Add transfusion event “time-stamping” and text data mining from transfusion nursing notes through natural language processing (NLP).

• Improve identification and characterization of transfusion AEs through the development of computable phenotypes (CP) reflecting constellations of signs/symptoms/events that accurately reflect an AE as validated by clinical record review (e.g. septic reactions, Transfusion-Related Circulatory Overload (TACO)). CPs are developed by NLP data mining of EHR unstructured text.

• Assess transfusion AEs on a large scale using both existing structured outcome data and CP vs improved transfusion exposure data.

• Automate AE case identification and reporting to improve report quality and reduce reporter burden.

• ISBT-128 codes downloaded from EHRs of partner blood banks have been incorporated into the OMOP CDM for blood exposure analysis.

• Approximately 100 of the 14,543 available ISBT-128 labeling codes were sufficient to identify the common US transfusion products among the three BEST EHR partner databases.

• Structured AE queries of EHRs based on ISBT-128 and ICD-9/10 codes have begun, using TRALI as the test AE.

• Draft CP have been developed for TACO and post-transfusion (PT) sepsis and are undergoing iterative assessment against clinical chart review to define predictive value and accuracy metrics (Figure 3).

• ISBT-128 codes now available in the OMOP Common Data Model for partner EHRs are being used to indicate transfusion exposures to assess both structured outcome variables (e.g. ICD-9/ICD-10) as well as CPs developed for TACO and PT sepsis.

• NLP procedures for event “time-stamping” and text data mining have been successfully developed at one partner site and are being standardized for use at the other collaborating sites.

• AE assessments using ISBT-128 codes and CPs have begun for TACO and PT sepsis.

• Preliminary queries using the TACO CP in a test EHR data extraction of ~700,000 patient records yielded 70,000 transfused patients (10%), approximately 34% of whom had associated nursing notes in their record (Figure 3).

• Of these patients, nearly 1,000,000 notes are available for the TACO NLP queries.

• Sites can share the logic for how to run NLP-based analyses but must further develop interoperability to apply CPs across EHRs.

Future Innovative Active HV Reporting Paradigm:

Characteristics of the OHDSI OMOP CDM

Figure 2

Local Blood Bank or Transfusion Service

Patient receiving transfusion

Clinician observes and recognizes AE

– treats patient

National Hemovigilance System

Annual HV Report

If AE = yes or likely

Investigates

Investigates Aggregates and Communicates

HV

Report

Possible AE for chart review

1. www.sentinelinitiative.gov

2. Obidi, J, K Chada, J Gruber et al. 2018. Biologics Effectiveness and Safety (BEST) Initiative:

Incorporating ISBT-128 Codes into OHDSI’s OMOP Common Data Model to Build a National Hemovigilance System to Monitor Transfusion-Related Adverse Events. Poster presented at 2018 ISBT Conference, Toronto, CA.

3. Hong, H, W Xiao, HM Lazarus, et al. 2016. Detection of septic transfusion reactions to platelet transfusions by active and passive surveillance. Blood 127: 496-502.

4. CDC National Healthcare Safety Network Biovigilance Component Hemovigilance Module Surveillance Protocol v2.5.2. (https://www.cdc.gov/nhsn/pdfs/biovigilance/bv-hv-protocol-current.pdf) accessed 29 June 2018.

http://www.ohdsi.com/ http://www.sentinelinitiative.gov/ https://www.cdc.gov/nhsn/pdfs/biovigilance/bv-hv-protocol-current.pdf