Factors Contributing to CPOE Opiate Allergy Alert Overrides

Abstract

Context

Increasing regulatory incentives to computerize provider order entry (CPOE) and connect stores of unvalidated allergy information with the electronic health record (EHR) has created a perfect storm to overwhelm clinicians with high volumes of low or no value drug allergy alerts. Data sources include the patient and family, non-clinical staff, nurses, physicians and medical record sources. There has been little written on how to collect hypersensitivity information suited for drug allergy alerting. Opiates in particular are a frequently ordered class of drugs that have one of the highest rates of allergy alert override and are often a component of pre-populated Computerized Provider Order Entry (CPOE) order sets. Targeted research is needed to reduce alert volume, increase clinician acceptance, and improve patient safety and comfort.

Design, Setting, and Patients

An FY10 retrospective, quantitative analysis of 30321 unique adults with opiate allergies triggering CPOE alerts at a large academic medical center.

Measurements

The prevalence of opiates ordered with opiate allergy alerts triggered and overridden is described. The effect of age, race, gender, visit type (medical, procedural), provider type (physician, advance practice nurse), and reaction/severity (e.g. nausea/mild) on the likelihood of provider override of the patient’s first opiate alert was analyzed using Generalized Estimating Equations (GEE).

Results

Analysis of a patient’s first opiate allergy alert (n=2767) showed that only prescriber role had a significant effect on alert override compared with all other variables in the model. Advanced practice nurses (APNs) were generally less likely to override the patient’s first opiate alert as compared to physicians (GEE, β=−.793, β=.001). However, override rates remained high, with 80% for APN’s and 90% for physicians.

Over half of all discharges had opiates ordered during their stay. Of those, 9.1% of the patients had recorded opiate allergies triggering 25461 CPOE opiate allergy alerts. The largest sub-group of alerts was triggered by gastrointestinal (GI) “allergies” such as nausea and constipation. Removing these types of non-allergic, low severity GI reactions from the alert pool reduced the first alert volume by 15% and the overall alert volume by 22%. Of note is that a history of codeine allergy triggered a significant volume of opiate alerts, yet was rarely ordered.

Conclusion

With an increasingly complex, information dependent healthcare culture, clinicians do not have unlimited time and cognitive capacity to interpret and effectively act on high volumes of low value alerts. Drug allergy alerting was one of the earliest and supposedly simplest forms of CPOE clinical decision support (CDS), yet still has unacceptably high override rates. Targeted strategies to exclude GI non-allergic type hypersensitivities, mild overdose, or adverse effects could yield large reductions in overall drug overrides rates. Explicit allergy and severity definitions, staff training, and improved clinical decision support at the point of allergy data input are needed to inform how we process new and re-process historical allergy data.

Introduction

In 2009 President Obama signed into law the Health Information Technology Economic and Clinical Healthcare Act (HITECH), which significantly ramped up the investment in this nation’s health care infrastructure by providing significant reimbursement incentives for “meaningful use” of electronic health record (EHR) technology. Computerized provider order entry (CPOE) adoption and drug-allergy checking, were two of the meaningful use objectives described for phase I. With increasing incentives to implement these objectives comes increasing need to mitigate adverse, unintended consequences of this evolving technology.

While there is evidence to support this safety measure, numerous alert fatigue studies cite the excessive alert volume and low clinical value as significant factors in intentional and unintentional alert overrides (van der Sijs, 2009). Two classes of drugs, antibiotics and opiates, represent the majority of CPOE alerts and alert overrides (Hsieh et al., 2004; Hunteman et al., 2009). This study focused on the latter. Opiates are not usually associated with the life-threatening allergic reactions (i.e. anaphylaxis, angioedema) as are antibiotics. Previous studies have identified that from 31–80% of the patients with opioid/narcotic allergies are labeled inappropriately (Gilbar 2004, Pilzer 1998). Commonly reported are known side effects or mild overdose due to potentiation with other drugs.

The European Academy of Allergy and Clinical Immunology proposed a definition, “Hypersensitivity causes objectively reproducible symptoms or signs, initiated by exposure to a defined stimulus at a dose tolerated by normal subjects” (Johansson I, 2005). It is doubtful that this definition is understood and consistently applied in the medication allergy history by those outside of the allergy field of medicine.

The increase in opiate allergies is likely the by-product of a longitudinal EMR that has moved unconnected, and often unsubstantiated, allergy records into a centralized data repository. It is facilitated by lack of staff education and reinforced by data entry screen design. Many EMRs have an associated data input screen named “Allergy” that was designed to store allergy data, but has expanded to collect any possible adverse drug reaction. The more appropriate term purposed for CPOE is “drug hypersensitivity” which is the umbrella term that includes both allergic (immune system mediated) and non-allergic types of reproducible adverse drug reactions.

Study objectives

The purpose of this study was to identify factors that contribute to high volume, low value alerts that are consistently overridden and which pose minimal patient safety concerns. These low value alerts represent known drug side effects of low/mild severity as well as low/mild hypersensitivities. Opiate alerts represent the most common over-alerting problem in the literature and comfort management is a significant component of clinical care in most settings. Comfort management refers not only to timely and effective pain relief, but management of undesirable opiate side effects such as nausea and constipation. Patient and prescribing provider attributes were also evaluated to determine if factors other than the allergy reaction influenced opiate allergy alerting.

Meaningful Use Stage I was about widespread adoption of the technology. However, the success of subsequent decision support depends heavily upon its ability to deliver trusted, clinically significant, actionable information at the point of care. The following exemplar highlights the need for this study:

A patient was urgently admitted and underwent open heart surgery. In recovery, an order was placed for needed pain medication for this intubated, agitated patient. A narcotic was ordered and the CPOE system alerted the provider that the patient was allergic to the ordered class of narcotics. Review of the documented allergy history showed an adverse reaction to codeine, but the reaction and severity were not documented. Family was unavailable. Nurse, physician and pharmacist conferred, during which time patient experienced increasing distress. The decision to give morphine was made, the patient was monitored closely, and no adverse response was observed. When questioned prior to discharge, the spouse had told the admission nurse that he “felt funny” after taking cough syrup with codeine a couple of years ago. The patient was discharged, and his electronic allergy history was not updated.

Allergy alerts differ from other types of alerts in that they often depend upon patient recall rather than those that depend on discrete lab values, diagnoses, or known interactions with other drugs. Allergy data is collected by clerical and/or clinical staff across multiple venues (clinic, inpatient, doctor’s office) and during care encounters of varied intensity (routine, urgent) from patients or their surrogates (spouse, aide, etc.). This suggests that allergy alerts, while simpler to program, may be more complex to successfully to design and implement than other types of drug alerts.

Setting

The study was conducted, with IRB permission, at a large urban, southeastern academic medical center with approximately 40,000 FY10 adult inpatient discharges. The center has a large biomedical informatics department, and extensive clinical systems development capacity. It has a long history of electronic medical record (StarPanel™) and inpatient CPOE applications (WizOrder™, McKesson/Horizon Expert Orders™). Two internally developed data input applications supplied allergy data which were used in both inpatient and outpatient settings. All allergy data, regardless of originating system is collected by a patient summary service (PSS) application that makes this data available to CPOE and other decision support decisions. The CPOE drug alert logic is commercially supplied by First DataBank, Inc.™

Methods

This was a retrospective, quantitative analysis of 25,461 CPOE opiate allergy alerts across 3,473 discharges (2,767 unique patients). Prevalence of opiate allergy alerts across all FY10 opiate orders was calculated. Descriptives of related attributes included: (1) patient (age, race, sex), (2) prescriber alert override (APN, physician), (3) medical or procedural visit, (4) reaction descriptions (ie. nausea) and (5) severity (mild, moderate, severe). Reaction and associated severity were combined to create a new allergy variable called NALS (Non-Allergic/Low Severity) with three mutually exclusive groups: NALS, Not NALS, and Unknown.

The next step was to extract all opiate orders (n=153,026) from the CPOE orders database in FY10 based on the American Hospital Formulary System (AHFS) opiate class code =28080800. These orders were matched to the MRN to calculate the percentage of discharges with opiates ordered. Of these, physician (82%) and Advance Practice Nurse (15.4%) prescriber roles were retained, excluding orders attributed to other staff (2.6%).

In the third step, the opiate allergy alert response log by prescriber ID was extracted from the CPOE allergy alert dataset. This log recorded whether the prescriber cancelled the order or overrode the alert. The two CPOE alert responses were (1) Override (place order) or (2) Cancel order (accept alert). The reason for override field was available, but was excluded since it was 99% missing.

To reduce the dataset, and minimize the influence of excessive repeating alerts for some patients with long lengths of stay, only the patient’s first opiate allergy alert was used in the regression analysis for each patient.

The fourth step was to assign allergy status: allergic vs. non-allergic. Each reaction within each group was evaluated (Table 1). Each patient was assigned to one of the study categories (NALS, Not_NALS, Unknown) using the coded severity and opiate reaction data in the patient allergy file. The following algorithm was used, followed by visual confirmation and assignment to one of the three following:

• – Unknown: Both reaction and severity are blank or stated unknown

• – NALS (GI): Reaction is nausea, vomiting, constipation, diarrhea, or GI upset and severity is low or unknown

• – Not_NALS: Assign everything else to Not_NALS

Table 1.

Characteristics of Opiate Reactions by severity and by group

Patients		Reaction	Severity
#	%		Mild	Moderate	Severe	Unknown
1323	28%	Other	1%	2%	7%	91%
1142	24%	Skin	4%	7%	15%	73%
1046	22%	Gastrointestinal	2%	9%	15%	74%
504	11%	Nervous/Mood	2%	9%	22%	68%
781	16%	Unknown/Blank	0%	0%	0%	99%
4796	100%	All	2.5%	6.5%	16.8%	74.3%

Note. Some patients had more than one reaction

The following is an explanation of how reactions were assigned to or excluded from the non-allergic/low severity (NALS) category.

Skin (24%)

The predominant reaction within the skin category was itching and rash. While there were mild reactions recorded, this group as a class was excluded from the NALS category, since it is difficult to tell if these skin reactions were histamine responses of a pseudo-allergy (NALS), or a true, immune mediated allergic reaction (Not_NALS).

Nervous/Mood (11%)

Those in this category ranged from nervousness, insomnia, confusion, hallucinations, to suicidal ideation. In this group, there were very few coded as mild severity. There were only 2 coded mild reaction types that could have been included in the NALS group (sleepy, headache) – but these were excluded due to low volume compared to the class.

Other Reactions (28%)

The information in this group was difficult to classify by body systems. It was highly variable, and had low volumes in any one category. It also included non-allergy type data such as:

• Patient received Narcan in the past (respiratory rescue)

• Patient has history of opiate abuse

• Patient has liver failure, dose appropriately

• Patient allergic to IV morphine, but can tolerate oral

• Patient has stomach ulcers, may bleed from NSAIDS

Gastrointestinal (22%)

The predominant reaction in the GI category was nausea and vomiting. Those coded or described within the free text as moderate or severe reactions were assigned to Not_NALS category. Those GI Reactions with mild, unknown, or blank severity were assigned to the NALS group if the descriptions were nausea, vomiting, upset stomach, constipation, diarrhea, or GI upset. The assumption being that if it were more than a mild reaction, it would likely been have recorded. More severe reactions such as bleeding or ulcer (exacerbation) were excluded from NALS.

Results

Four allergies triggered 82% of the opiate allergy alerts: Allergies to Codeine (32%), Morphine (28%), Hydrocodone (11%), and Oxycodone (11%). Table 2 displays the number of discharges with a specific opiate allergy and the number of alerts triggered.

Table 2.

Patient Allergies Triggering Alerts

# Allergic	%	Allergy	# Alerts	%
1600	36%	Codeine	8,163	32%
1079	24%	Morphine	7,152	28%
483	11%	Hydrocodone	2,783	11%
427	10%	Oxycodone	2,773	11%
211	5%	Hydromorphone	1,190	5%
147	3%	Tramadol	971	4%
61	1%	Nubain	595	2%
148	3%	NSAID/OTC	541	2%
70	2%	Butorphanol	536	2%
104	2%	Meperidine	306	1%
105	2%	Propoxyphene	239	1%
10	0%	Opioid	88	0%
19	0%	Fentanyl	60	0%
2	0%	Dihydrocodeine	23	0%
2	0%	Oxymorphone	21	0%
11	0%	Other	20	0%
4479	100%		25,461	100%

Note: Some patients may have more than one recorded opiate allergy

Four drugs triggered 96% of opiate alerts: Hydromorphone (34%), Oxycodone (25%), Morphine (19%), and Hydrocodone (18%).

Reaction and Severity

Severity (mild, moderate, severe) was missing for 74.3% of the reactions. Of those, 16% had no reaction or severity. Overall, alerts with severity coded or described are 2.5% mild, 6.5% moderate, and 16.8% severe.

Non-Allergic/Low Severity Reactions (NALS)

First alerts were stratified into 3 allergy reaction/severity classes as previously described. Non-allergic/low severity (NALS) GI reactions accounted for 15.4% of the first alerts. The override rate for the patient’s first alert for all opiate alerts was 89%. Removing the GI NALS alerts reduced the first alert volume for opiates by 15% (425/2767) but did not significantly change the 89% overall alert override rate.

Influence of patient, prescriber and reaction/severity factors on override

Patient characteristics, reason for admission, prescriber role and reaction/severity group on opiate alert overrides were analyzed. The Generalized Estimating Equations (GEE) procedure (SPSS v20) was used to extend the generalized linear model to allow for analysis of clustering and repeated measurements. Clustering can happen when a physician may treat primarily a specialty population like orthopedics or cancer that may unduly influence outcomes based on opiate use. Repeated measurements over time may reflect the fatigue that occurs when a provider has a lot of alerts over a short period of time. The dependent variable of interest was the override response (yes, no), so a binary logistic model was selected, with “no override” as the reference category. There were 697 providers (89% Physicians, 11% APN) who had from 1–28 first opiate allergy alerts (mode=1, median= 3).

For inpatients of all ages, 53% had opiates ordered (n=153026 orders). Inpatient opiate allergy prevalence was 9.1 % (2767/30321 patients). This number is understated as opiate allergic patients without opiates ordered were not captured. For those with opiate allergy alerts, the mean age was 54 years, 81.2% white and 68.6% female. For all admissions, medical discharges were higher (56%) than procedure related ones (44%). However, for the patient’s first visit, procedural admissions were higher (59.8%). This reversal was expected as chronic medical conditions are more likely to have readmissions. Physicians received 91.1% of the first alerts and overrode 90% of them. Advance Practice Nurses (APN) received 15% of the first alerts and overrode 80% of them. A summary of the study sample characteristics for the patients first opiate allergy alert are shown in Table 4 below.

Table 4.

First Opiate Allergy Alert Response

			Override = No		Override = Yes
	N	%	N	%	N	%
All	2767	(100%)	304	(11%)	2463	(89%)
Gender
Female	1900	69%	222	(12%)	1678	(88%)
Male	867	31%	82	(9%)	785	(91%)
Race
Black	302	11%	30	(10%)	272	(90%)
Other	80	3%	11	(14%)	69	(86%)
White	2385	86%	263	(11%)	2122	(89%)
Reason for Admission
Medical	1112	40%	111	(10%)	1001	(90%)
Procedural	1655	60%	193	(12%)	1462	(88%)
Allergy Reaction/Severity
Unknown	699	25%	78	(11%)	621	(89%)
NALS	425	15%	36	(8%)	389	(92%)
Not NALS	1643	59%	190	(12%)	1453	(88%)
Prescriber Role*
APN	246	9%	49	(20%)	197	(80%)
Physician	2521	91%	255	(10%)	2266	(90%)
			Mean	(SD)	Mean	(SD)
Patient Age in Years			54.7	16.7	54.5	16.4

^*Significant p=.001

NALS = Non-Allergic/Low Severity alert

Summary

Over half of all discharges had opiates ordered during their stay. Of those, patients with recorded opiate allergies (9.1%) triggered 25461 CPOE opiate allergy alerts. The largest sub-group of alerts was triggered by gastrointestinal (GI) “allergies” such as nausea and constipation. Removing non-allergic, low severity GI reactions from the opiate alert pool reduced the first alert volume by 15% and the overall alert volume by 22%.

Opiate allergy override rate was 93% for all admissions and re-admissions. It was 89% for the first admission’s alert. In the GEE analyses, provider role was the most significant variable in predicting alert overrides. Advanced practice nurses (APNs) were generally less likely to override the patient’s first opiate alert as compared to physicians (GEE, β=-.793, p=.001). However, override rates remained high, with 80% for APN’s and 90% for physicians. Other factors were not statistically significant in predicting override.

The study revealed two related phenomena in the evaluation of patient opiate allergy alerts. The first being that there was a high prevalence of opiate allergies recorded and alerted on for a reportedly rare occurrence. Particularly problematic were recalled allergies to codeine, which trigger alerts to any opiate order. Findings suggest that this is due, in part, from inappropriately broad allergy definitions and use of the allergy data collection field to alert the prescriber to other non-allergy clinical concerns.

The second was the high volume of common opiate side effects recorded as allergic reactions. Of particular importance was the high volume of overridden opiate allergy alerts triggered by non-allergic/low severity gastrointestinal reactions. Eliminating this group of triggers, would have resulted in a 9% reduction in opiate allergy alerts at the study site. It also opens for discussion that proactive management of drug side effects may be needed.

In this study, 16% of the opiate alerts had no reaction documented, and 74.3% of the alerts had no severity assigned. The absence of reaction detail has been identified as a significant hindrance in provider evaluation of alerts. Data intake screens must be designed to make it easier to capture accurate data, rather than “guess” or ignore these data fields.

This study has several limitations. The data is from a university medical center with advanced capacity to build and continuously tune its CPOE system. The alert trajectory is complex and involves many systems and disciplines which may differ across institutions. Those systems include the allergy data collection tools, allergy processing algorithms, and CPOE alert displays. Limiting the study to opiates may limit its generalizability to other drugs.

Discussion

Drug allergy alerting was one of the earliest and supposedly simplest forms of CPOE CDS, yet still has unacceptably high override rates. While alert logic may be simple, the complex environment in which it exists is not. This study highlights unintended consequences that can occur when computerized clinical decision support logic is not tightly aligned with the electronic medical record data upon which it depends. Much of our recorded healthcare data is influenced by human factors such as memory, personal experience, and varied interpretation by patients, clinicians and support staff.

The volume of side effects, recorded as allergies, should not be ignored or completely attributed to vague allergy definitions, human error or the influences of data entry screen design. Moreover, this reflects the need for better symptom management of opiates and all medications that are known to cause distressing side effects. Increased education and engagement of patients in understanding the use and effects of opiates, accurate reporting and partnering with their providers in comfort management is fundamental.

This research also highlights the need for continuous feedback and analysis of alerts in clinical practice, to identify opportunities to improve systems and reveal emerging healthcare patterns as more people use computers to record, deliver and improve care. The high use opiates over time may be introducing new hypersensitivities in susceptible individuals that have yet to be fully revealed. More than half of all inpatients had opiates ordered. Of these, 9% of the patients triggered over 25,000 opiate alerts based on how their allergy history was recorded and how often opiates were re-ordered for the visit. This count could be tripled when you consider that the same alert that may be presented to the pharmacist on dispensing, and the nurse on administering.

Table 3.

Drug Orders Triggering Alerts (all ages)

Drug Order Group	#	%
Hydromorphone	8,599	34%
Oxycodone	6,395	25%
Morphine	4,916	19%
Hydrocodone	4,590	18%
Tramadol	435	2%
Fentanyl	354	1%
Belladonna-opium suppository	82	0%
Acetaminophen w Codeine	34	0%
Meperidine	20	0%
Codeine	16	0%
Propoxyphene (Darvon)	10	0%
Methadone (Dolophine)	10	0%
Total	25,461	100%