drug-drug interaction alerts: time for a new paradigm
TRANSCRIPT
Drug-Drug Interaction Alerts: Time for a New Paradigm
Jon D. Duke, MD, MS, Regenstrief Institute
Reviewed 42,641 orders11% (4690) produced alerts
DDI alert override rate = 88%Allergy override rate = 69%
Refinement in order check logic could reduce override rates and may increase practitioner acceptance and effectiveness of order checks.
Ten Commandments for Effective Clinical Decision Support
1. Speed is everything
2. Anticipate needs and deliver in real time
3. Fit into user’s workflow
4. Little things can make a big difference (usability)
5. Physicians resist stopping
6. Changing direction is fine
7. Simple interventions work best
8. Asking for information is OK − but be sure you really need it
9. Monitor impact, get feedback, and respond
10. Knowledge-based systems must be managed/maintained
Bates et al. J Am Med Inform Assoc. 2003;10:523-30. (PMID: 12925543)
Reviewed 18,354 orders13% (2455) produced alerts
DDI alert override rate = 95%Allergy override rate = 91%
Conclusions: Despite intensive efforts to improve a commercial drug interaction alert system and to reduce alerting, override rates remain as high as reported over a decade ago. Alert fatigue does not seem to contribute. The results suggest the need To fundamentally question the premises of drug interaction alert systems.
Why can’t we move the needle?
• Typically interruptive pop-up alerts
– Computerized provider order entry (CPOE)
– Pharmacist verification/dispensing
• Required for Meaningful Use Stages 1 and 2 1
• Most organizations use commercially available drug-drug interaction (DDI) knowledgebases
– Impractical for most organizations to create/maintain
1) http://www.healthit.gov/providers-professionals/achieve-meaningful-use/core-measures/drug-interaction-check 2) http://www.healthit.gov/providers-professionals/achieve-meaningful-use/core-measures-2/clinical-decision-support-rule
Current Approach for Drug-Drug Interaction Decision Support (I)
• Alerts often perceived excessive or irrelevant
– Presentation is suboptimal 1
– Providers dissatisfied 2
– High override rates 3
• Customizing commercial knowledgebasesrequires substantial resources
– Organizations may turn off decision support
– Potential unintended consequences 4
1) Russ et al. Int J Med Inform. 2012;81:232-43. 2) Weingart et al.. Arch Intern Med 2009;169:1627-32. 3) van der Sijs et al. J Am Med Inform Assoc 2006;13:138-47. 4) van der Sijs et al. J Am Med Inform Assoc. 2008;15:439-48.
Current Approach for Drug-Drug Interaction Decision Support (II)
Drug-Drug Interactions and Harm (I)
• Exposure to DDIs is a source of preventabledrug-related harm1
• Estimated to harm 1.9-5 million inpatients3 and cause up to 220,000 ED visits per year4,5
Association Between Hospital Admission for Drug Toxicity and Recent Co-Prescription of Interaction Medications (Juurlink et al. 2003) 2
INTERACTING MEDICATIONS TOXICITY OR (95% CI)
Glyburide + co-trimoxazole Hypoglycemia 6.6 (4.5-9.7)
Digoxin + clarithromycin Digoxin toxicity 11.7 (7.5-18.2)
ACE inhibitor + potassium-sparing diuretic Hyperkalemia 20.3 (13.4-30.7)
1) IOM. Preventing medication errors. National Academies Press. 2007. 2) Juurlink et al. JAMA. 2003;289:1652-8. 3) Magro et al. Expert Opin Drug Saf. 2012;11:83-94. 4) CDC. FASTSTATS - Emergency Department Visits. 2012; http://www.cdc.gov/nchs/fastats/ervisits.htm; 5) CDC. FASTSTATS - Hospital Utilization. 2010; http://www.cdc.gov/nchs/fastats/hospital.htm.
Drug-Drug Interactions and Harm (II)
• Most potential DDIs are clinically inconsequential
• DDIs are responsible for a low proportion of adverse drug events overall (<5%)
• But DDIs trigger a high proportion of alerts
Low Satisfaction with DDI Alerts
• Physician survey (N=184)
• 53% not satisfied with DDI / allergy alerts
• Top complaints
– Alerts triggered by discontinued medications
– Failure to account for appropriate combinations
– Excessive volume of alerts
Weingart et al. Arch Intern Med. 2009;169:1627-32.
Low Adherence to DDI Alerts
• Varies study to study but continue to see 60%-95% override rates for interruptive DDI alerts
• Non-interruptive alerts generating 1-2% adherence
Van der Sijs et al. JAMIA 2006. 13(2):138-147.
Seidling et al. J Am Med Inform Assoc. 2011;18:479-84.
Hard Stops Work But…
• RCT including “hard stop” DDI alert
– 1981 prescribers, 2 academic medical centers
– Warfarin + trimethoprim/sulfamethoxazole
Strom et al. Arch Intern Med. 2010;170:1578-83.
Unintended Consequences
Strom et al. Arch Intern Med. 2010;170:1578-83.
Study stopped early due to unintended consequences in intervention group
UNINTENDED CONSEQUENCE RELATION TO
INTERVENTION
3-day delay in TMP/SMX therapy deemed necessary by infectious disease
Probable
Failure to prescribe TMP/SMX prophylaxis for critically ill patient
Probable
1-day delay in warfarin therapy Definite
3-day delay in warfarin therapy Definite
Alerts with Poor Specificity
Weingart et al. Arch Intern Med. 2009;169:1465-73.
Study of 279,476 alerts by 2,321 physicians over 6 months in the ambulatory care setting
10% of alerts accounted for 60% ADEs prevented and 78% of cost benefit
331 alerts toprevent 1 ADE
Lack of Consistency Across Systems
• 62 hospitals voluntarily participated for review of simulated DDI orders of varying severity
• Detected only 53% of medication orders that would result in fatality
• Detected 10-82% of orders that would have caused serious ADEs
• Did not correlate with specific vendors
Metzger et al. Health Aff. 2010;29:655-63.
Lack of Consistency Across Systems
Metzger et al. Health Aff. 2010;29:655-63.
Scores For Detection of Test Orders That Would Cause an Adverse Event - By Software Vendor
Similar Story in Pharmacies
• 64 inpatient and outpatient Arizona pharmacies
• Fictitious patient orders to evaluate 19 drug pairs
– 13 DDIs and 6 non-DDIs
• Median correct responses 89.5% (range 47-100%)
Saverno et al. J Am Med Inform Assoc. 2011;18:32-7.
89%
86%
88%
45%
81%
90%
75%
84%
87%
83%
80%
70%
75%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Carbamazepine + clarithromycin
Digoxin + amiodarone
Digoxin + clarithromycin
Digoxin + itraconazole
Nitroglycerin + sildenafil
Simvastain + itraconazole
Simvastatin + amiodarone
Simvastatin + gemfibrozil
Warfarin + amiodarone
Warfarin + fluconazole
Warfarin + gemfibrozil
Warfarin + naproxen
Warfarin + sulfamethoxazole/trimethoprim
Saverno et al. J Am Med Inform Assoc. 2011;18:32-7.
Lack of Consistency in Pharmacy Alerts
Any Good News?
Usability Does Help
• 50,788 DDI alerts analyzed
• Higher quality alert display increased adherence
Seidling et al. J Am Med Inform Assoc. 2011;18:479-84.
Factors Associated with Interruptive Alert AcceptancePARAMETER OR (95% CI) P-VALUE
Quality of alert display 4.75 (3.87-5.84) <0.001
Setting (inpatient vs. outpatient) 2.63 (2.32-2.97) <0.001
Level of the alert 1.74 (1.63-1.86) <0.001
Frequency of the alert 1.30 (1.23-1.38) <0.001
Dose-dependent toxicity 1.13 (1.07-1.21) <0.001
Usability Key Factors
• Consistent signal words, severity descriptions
• Consistent colors and icons
• Consistent placement of information
• Parsimonious use of text (details on demand)
• Directly actionable
• Present as early as possible
Providing Patient Context
Tamblyn et al. R. J Am Med Inform Assoc. 2012;19:635-43.
• Cluster RCT, 81 family physicians, 5,628 elderly patients
• Modified alerts with patient-specific estimates of fall risk with psychotropic medications
• Reduced risk of injury by 1.7 injuries per 1000 patients (95% CI 0.2 to 3.2; p=0.02)
Tisdale et al. R. J Am Col Cardiology. 2012;59(13):E1799.
• RCT of alert using predictive model to inform risk regarding QT prolongation
Providing Patient Context
• Reduced inappropriate prescribing by 21%
• Reduced odds of QT prolongation by 35%
Providing Patient Context
Getting to Providers Earlier
Getting to Providers Earlier
Getting to Providers Earlier
Improving the Knowledgebase
• Identifying high priority alerts
• Identifying suppressible alerts
• Emerging predictive models around certain adverse outcomes (e.g., DDIs associated with hyperkalemia)
• Ideas swirling around the learning healthcare system / feedback loops for improving alert delivery and appropriateness
Phansalkar S et al, JAMIA 2012. 19:735-743. Phansalkar S et al, JAMIA 2013 20:489-493.
Eschmann E et al, Eur J Clin Pharmacol 2014. 70(2):215-23. McCoy A et al, Ochsner 2014.14:195-202
So I’d Like to Conclude
• We just need to…
– Improve alert display and usability
–Optimize alert specificity and sensitivity
– Increase knowledgebase consistency
– Incorporate contextual factors
But How Much Will It Move the Needle?
But How Much Will It Move the Needle?
Our Real Problem Is
TRUST
Why Doctors Still Won’t Trust DDI Alerts
• Disregarding alerts has become part of the medical culture
• It is inculcated during training, just as medical slang and other aspects of the sub-culture
• It is of course reinforced by all the problems we’ve described above
• Fixing the problems with our alerts will not fix the trust problem (for a long, long time)
So How Do We Get Doctors to Listen to DDI Alerts?
• First, why do doctors listen to anyone?
You have received conflicting advice regarding the prescribing of an antibiotic for an inpatient with community acquired PNA. Whose advice are you more likely to trust? To follow?
Why Do Doctors Take Advice?
• “Positive” Drivers
– Authority / Hierarchy
– Specialty
– Perceived Experience / Knowledge
– Team-building
• “Negative” Drivers
– Fear (e.g., of mistakes, lawsuits)
– Embarrassment
Why Do We Adhere?
Note: You have ignored this DDI warning 27 times on 14 unique patients. Of
these, 2 patients have developed a bleeding-related condition.
Increase Visibility of Adverse Events
Note: This DDI has been associated with 17 serious adverse events at our
hospital in 2014.
Increase Visibility of Adverse Events
Note: 2,585 serious adverse event reports indicating concurrent use of
Amiodarone and Warfarin were submitted to FDA in 2014.
Increase Visibility of Adverse Events
Note: There were 12 lawsuits associated with concurrent use of Amiodarone
and Warfarin in Indiana between 2010 and 2014.
Increase Visibility of Adverse Events
Connect with Hospital Hierarchy
Steve Nissen, MDChair of Cardiology
ApprovedAlert
AllergyAllergy Warning
Persist and Propagate Override Status
Persist and Propagate Override Status
Persist and Propagate Override Status
Persist and Propagate Override Status
Addendum: AMOXICILLIN 500MG. Allergy Alert Override by Smith, JD. 11/14/2014 at 8:31am.
Embed in Chart
The New Paradigm?It’s People
• Recognize and leverage natural human emotions as part of system design
• Decisions should be visible to peers and authority figures
• DDI warnings should be ‘sponsored’ by specific local experts
• Drug safety decisions should be longitudinal rather than instantaneous events
Thanks!