Domain

Evidence (e.g. research results from CER, PCOR, or QI)

Type

Case Study or Comparative Case Study

Theme

effectiveness; quality

Start Date

7-6-2014 1:15 PM

End Date

7-6-2014 2:45 PM

Structured Abstract

Objectives: To demonstrate an application of Clinical Decision Support (CDS) data to detect prescribing errors that occurred in an acute care hospital setting. We present the development and application of a tool to measure physician quality of care.

Methods: We obtained electronic records for all medications ordered between March 2009 and December 2012 for adult inpatients at one academic medical center and data generated by a knowledge-based CDS alert system (Epic Systems, Verona WI). We used medication orders and CDS alerts to develop a prescribing errors detection algorithm (PEDA). A prescribing error was defined as a medication order that had high potential to cause harm to the patient. We developed the PEDA in multiple steps: 1) through a consensus process, we selected CDS alert categories for inclusion; 2) we established category-specific thresholds to define a PER event; 3) we evaluated thresholds using a stratified random sample; and 4) based on results from a chart review (conducted independently by at least one physician and one pharmacist), we refined error thresholds to improve the PEDA performance.

Findings: During the study period 6,079,783 medications were ordered, resulting in 147,420 CDS alerts. Only two CDS alert categories— dose and drug-drug interaction— were included in the final PEDA. With two exceptions, we defined a dose error as an order that exceeded the maximum daily or the maximum single dose by ≥20% (as defined by First Databank). Drug-drug interaction errors were derived from common recurring themes. For dose errors, the PEDA had a positive and negative predictive value of 96% and 32%, respectively. Prescribing error rates were 3.5 per 1000 medication orders and 14.4 per 100 CDS alerts. Error rates (per 100 CDS alerts) varied widely by physician type: 14.2 for interns (PGY1); 10.7 for residents (PGY2+); 33.2 for non-faculty fellows; and 16.6 for faculty. We found no statistical difference in the proportion of prescribing errors when we compared the day shift to the night shift. Error rates were highest on the ICU service (19.8 per 100 CDS alerts) but were similar among medicine, surgery and ED (range; 11.7-13.1 per 100 CDS alerts).

Discussion: We present an application of a novel tool to detect prescribing errors using repurposed EHR data. This tool may be used to measure physician quality of care.

Higher faculty and fellow error rates are driven by dose errors; these physician types may be more responsible for writing orders for clinically-appropriate high doses. The ICU’s higher patient acuity may contribute to the higher error rate detected by the PEDA in this setting.

Our method has limitations: only prescribing errors defined in the CDS alert database are considered; among these alerts, only prescribing errors due to inappropriate dose or certain drug-drug interactions can be detected.

Conclusion: The PEDA provides a systematic and reproducible method to detect some types of prescribing errors that occur in the hospital, which are relevant to quality of care measurement. Further work is required to expand the tool’s use and to determine its validity in other hospital settings.

Acknowledgements

The project described was supported by the National Center for Advancing Translational Sciences, National Institutes of Health, through grant number UL1 TR000002 and linked award TL1 TR000133.

Support was also provided by grant number 1T32HS022236-01 from the Agency for Healthcare Research and Quality (AHRQ) through the Quality Safety Comparative Effectiveness Research Training (QSCERT) Program.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Share

COinS
 
Jun 7th, 1:15 PM Jun 7th, 2:45 PM

Repurposing Decision Support Data to Detect Prescribing Errors—An Application for Quality Measurement

Objectives: To demonstrate an application of Clinical Decision Support (CDS) data to detect prescribing errors that occurred in an acute care hospital setting. We present the development and application of a tool to measure physician quality of care.

Methods: We obtained electronic records for all medications ordered between March 2009 and December 2012 for adult inpatients at one academic medical center and data generated by a knowledge-based CDS alert system (Epic Systems, Verona WI). We used medication orders and CDS alerts to develop a prescribing errors detection algorithm (PEDA). A prescribing error was defined as a medication order that had high potential to cause harm to the patient. We developed the PEDA in multiple steps: 1) through a consensus process, we selected CDS alert categories for inclusion; 2) we established category-specific thresholds to define a PER event; 3) we evaluated thresholds using a stratified random sample; and 4) based on results from a chart review (conducted independently by at least one physician and one pharmacist), we refined error thresholds to improve the PEDA performance.

Findings: During the study period 6,079,783 medications were ordered, resulting in 147,420 CDS alerts. Only two CDS alert categories— dose and drug-drug interaction— were included in the final PEDA. With two exceptions, we defined a dose error as an order that exceeded the maximum daily or the maximum single dose by ≥20% (as defined by First Databank). Drug-drug interaction errors were derived from common recurring themes. For dose errors, the PEDA had a positive and negative predictive value of 96% and 32%, respectively. Prescribing error rates were 3.5 per 1000 medication orders and 14.4 per 100 CDS alerts. Error rates (per 100 CDS alerts) varied widely by physician type: 14.2 for interns (PGY1); 10.7 for residents (PGY2+); 33.2 for non-faculty fellows; and 16.6 for faculty. We found no statistical difference in the proportion of prescribing errors when we compared the day shift to the night shift. Error rates were highest on the ICU service (19.8 per 100 CDS alerts) but were similar among medicine, surgery and ED (range; 11.7-13.1 per 100 CDS alerts).

Discussion: We present an application of a novel tool to detect prescribing errors using repurposed EHR data. This tool may be used to measure physician quality of care.

Higher faculty and fellow error rates are driven by dose errors; these physician types may be more responsible for writing orders for clinically-appropriate high doses. The ICU’s higher patient acuity may contribute to the higher error rate detected by the PEDA in this setting.

Our method has limitations: only prescribing errors defined in the CDS alert database are considered; among these alerts, only prescribing errors due to inappropriate dose or certain drug-drug interactions can be detected.

Conclusion: The PEDA provides a systematic and reproducible method to detect some types of prescribing errors that occur in the hospital, which are relevant to quality of care measurement. Further work is required to expand the tool’s use and to determine its validity in other hospital settings.