Background/Purpose

Chronic glucocorticoid (GC) use is a known risk factor for osteoporosis and fracture. Patients with chronic GC use often receive suboptimal osteoporosis prevention, diagnosis, and treatment. We sought to create a best practice alert (BPA) to identify chronic GC users in our electronic health records (EHR) to recommend bone density testing.  Daily dosage and duration of prescription data were not uniformly available for building the BPA. To improve identification of these patients, our objectives were to 1) describe the quality of medication data available for triggering a BPA to prompt bone density screening for patients on chronic GCs, and 2) assess alternative criteria using existing data.

Methods

Our target population was patients ≥50 years of age on chronic GCs defined as taking ≥7.5mg of prednisone daily or equivalent, for 30 days or more. We extracted medication orders from the University of Utah Healthcare clinical data warehouse for all GCs ordered between July 1 and December 31, 2013 for patients ≥ 50 years.  The extract included refill number, quantity dispensed, difference between order start and end date, frequency, signature, and dosage per episode.  We manually reviewed and classified each order as ‘yes’, ‘no’, or ‘unable to determine’ for chronic GC use.   We assessed the frequency of data available for each data field, and stratified by records created using the structured versus free-text order template.  We assumed that records without the paired dosage and frequency information were entered using the free text template.  Finally, we assessed sensitivity and positive predict value (PPV) of selected medication data elements to identify the target population.

Results

Among the 1,699 GC prescriptions identified, 17% (292) were determined to be chronic GC use; 52% (881) were entered using a structured template. The structured and free-text templates resulted in similar rates of data populating Refills (97% vs 98%), Quantity dispensed (97% vs 98%), Order start date (99.7% vs 99.9%), and ‘Sig’ (99.9% vs 99.8%), respectively.  In contrast, rate of data availability differed between structured verses free-text templates: Order end date (90.5% vs 71%), Frequency (99.9% vs 0%) and Dosage (100% vs 0%), respectively. 

Data Field	Structure of Data	% of Records with Computable Data	Criteria	Sensitivity	PPV
Refill	Structured	97.3%	1 or more refill	72%	41%
	Free text	97.7%		72%	42%
Quantity Dispensed	Structured	97.5%	dispensed >=30 tablets	97%	44%
	Free text	98.5%		93%	44%
	Structured	97.5%	dispensed >=45 tablets	79%	48%
	Free text	98.5%		84%	64%
	Structured	97.5%	dispensed >=60 tablets	75%	50%
	Free text	98.5%		79%	68%
Difference Between Order Start and End Date	Structured	90.2%	Date difference between order start and end date	67%	43%
	Free text	71.1%		50%	19%
Refill and Quantity Data	Structured	97.0%	Refill >0 and quantity data >=30	71%	46%
	Free text	97.6%		72%	64%

In the absence of daily dosage and duration information, quantity dispensed ≥ 30 tablets performed best with the highest sensitivity, and mid-range PPV (Table 1).

Conclusion

Medication data in the EHR are subject to variability and detecting chronic medication use requires adequate evaluation of medication data quality, available data fields, and clinician practice patterns.  This is particularly challenging with GCs given their widespread use both chronically and in short-term tapers.  Successful alert designers must evaluate both the accuracy of data used to generate an alert, and triggering criteria, to improve identification of the desired population.

---

« Back to 2014 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/best-practices-for-best-practice-alerts-evaluation-of-a-best-practice-alert-to-detect-chronic-glucocorticoid-use/