The Search for Better Clinical Decision Support: Drug Interaction Alerts in Focus

AUTHOR: Susan Cheng, Product Specialist (Medication Management),  InterSystems

It comes as no surprise that the drug interaction alert screen is, by far, the most frequently ignored screen in any clinical software. Doctors, pharmacists and nurses know it: our finger is already anxiously hovering over the ‘Enter’ key before this seemingly obnoxious screen displays. We bypass the drug interaction details before we could even process the content that flashed by. Yet statistics indicate that clinical errors within major hospitals could occur as frequently as one error per patient and that prescribing errors account for 2.5% of all medicine orders.[1,2] If so, why are we still ignoring these very important screens?

Alert fatigue appears to be the overriding culprit based on research. [1-6] Users are constantly bombarded with too many alerts containing information that is either irrelevant, unhelpful or both – a trend seen in prescribing and dispensing software across both primary and acute healthcare settings. [3,4] A multitude of assessment tools have been used in studies to analyse differences between alert display in different software applications and to compare alerts in common secondary references to those displayed in software applications. [1-6]

Many of these studies, however, have one common shortcoming: they look at individual interactions in terms of content, format and language but do not take into consideration factors surrounding drug interactions as a whole. This oversimplifies the problem and results in suggested solutions that may not necessarily be helpful. Alert fatigue does not occur as the result of one poorly displayed drug interaction. It is often the consequence of many contributing factors.



Severity filters

A study states that ‘designers and vendors sharply limit the ability to modify system alerts because they fear being exposed to liability if they permit removal of warning that would prevent harmful prescribing errors’ [4] whereas, in actual fact, many knowledge providers and software designers already provide such a feature to enable adjustment of severity filters. Rather, it is the users and organisations they work for who decide to minimally adjust filters. This is not a design issue but a question of liability – ultimately, no party wants to bear the burden of responsibility for a medication-related error. This instinctive nature is something that even the most advanced computer system can not change.



Another study suggests that there is a ‘lack of information on clinical effects and management advice’. [3] A step back from this observation reveals two possible causes: (1) limited research into the drug interaction of interest, and (2) failure of knowledge providers to maintain an up-to-date database of drug interactions reflective of current clinical recommendations. It is noteworthy that only the latter is a software issue. If the level of evidence remains limited for a particular drug interaction, then this is not a software design issue but a clinical question that the scientific community, as a whole, needs to answer.


Route of administration

Research argues that drug interaction alerts can be made more relevant by being more specific to the route of administration (e.g. if the offending medication is a topical preparation with negligible systemic absorption). [3,4] While this is true in a generic sense, it is important to consider situations that are more complex, such as:

  • multiple interacting medications ordered at different prescribing levels (e.g. a doctor may wish to prescribe paracetamol and leave it to the nurse’s discretion to nominate a suitable route depending on the patient’s ability to swallow) or;
  • extemporaneous preparations containing ingredients normally administered via a different route (e.g. interactions for an oral agent may not necessarily apply when it is added into topical preparation)

Moreover, in a software application designed for the hospital setting at an international scale, it is important to consider the fact that different drug databases may be used in different countries, and that not all drug databases provide drug interaction data at the same level. For example, First Data Bank and MIMS provide interactions data specific to the route of administration, where as Thesorimed and Farmadati provide interactions data at the generic level. This poses a challenge for software developers because the system’s alert-triggering infrastructure needs to function seamlessly, despite differences within each data source. Route of administration is not as simple as it seems. We need to define what we want under all circumstances before jumping to a generic conclusion.


Discontinued medications

A study also recommends removal of drug interaction alerts for medications no longer taken by patient. [3] While this appears to be a logical suggestion for most situation, we need to consider outliers such as:

  • drugs with a long half-life (e.g. MAOIs and fluoxetine where drug interactions may occur 14 days following cessation of the drug)
  • antidepressant changeover periods (when it is important to acknowledge the risk of serotonin toxicity if the second antidepressant is initiated too closely to the first antidepressant which has been ceased)

The question of whether to exclude drug interactions for discontinued medications becomes a question of which drugs and for how long.



Situations with >1 drug interaction

Current research provides valuable suggestions on improvements to the layout, format and content of single drug interaction alerts but provides very few (if any) recommendations on the display of multiple drug interaction alerts, despite the fact that alert fatigue is more likely to occur with multiple alerts compared to a single alert. Aspects to avoid when designing a screen for multiple drug interaction alerts include:

  • unnecessary scrolling (because users never scroll)
  • additional clicking (which will impede workflow)
  • displaying everything on the same screen (information overload)

Yet, if we look at this list carefully, one might ask what other ways there are to display information when screen space is limited and we do not want to scroll, click or display all details at the same time? Screen design becomes an ultimate dilemma. It is a fine balance of many on-screen factors where nothing will be perfect regardless of what we decide on.


Other clinical alerts

Clinical decision support is not just about drug interactions. A myriad of other clinical alerts exist: duplicate therapy, drug allergy, drug-disease contraindications, pregnancy and lactation precautions and drug-dose thresholds. As healthcare professionals, we are trained to process and prioritise these clinical concerns within seconds of reviewing a patient’s medicines list. Our clinical experience and professional judgement empowers us with the ability to filter across different clinical alert types to pinpoint and act on the most important concerns for the patient. Replicating this skill in a computer system is much more difficult. Within a drug database, each clinical alert type is contained within a different dataset with a different set of severity ratings (e.g. moderate drug interaction, Category X for pregnancy). We are yet to generate an algorithm that can prioritise across different clinical alert types. How do we build a system that can outsmart years of clinical experience combined with professional judgement?


Alerts at different stages of patient care (e.g. admission/inpatient/discharge), episodes and visits

Around half of hospital medication errors occur on admission, transfer and discharge. [1] Of these, about 30% have the potential to cause patient harm. [1] Medication reconciliation plays a large role in minimising these errors, and clinical decision support (which includes drug interaction alerts) is an important component of the medication reconciliation process. However, if we have so many different aspects to consider during the medication reconciliation process (e.g. changes to the medication, dose, strength, timing of administration, associated vital signs and laboratory results, progressive changes to the patient’s condition, other relevant clinical aspects), how do we reduce the amount of information and number of alerts displayed when all the details are important, just for different patients and to different users at different times under different circumstances? How do we build a system that caters for everyone without making incorrect assumptions about what they know (and don’t know) by removing ‘insignificant’ and ‘irrelevant’ drug interaction alerts?



There is a clear need for guidelines to regulate the on-screen display of drug interaction alerts. [1-6] We are, however, far from reaching this target: questions still exist around the display of individual drug interaction alerts; research has not yet evaluated the effectiveness of screen display for multiple drug interaction alerts. It took years of speculation before the Australian Commission on Safety and Quality in Healthcare developed national guidelines for on-screen display of clinical medicine information in January 2016. Considering the complexity of issues around clinical decision support, it may take years before regulatory guidelines are developed for the on-screen display of drug interaction alerts.

This makes it even more important for us to take a proactive approach. Organisations such as MSIA are in the perfect position to start a conversation among knowledge bases, software developers and users. Our knowledge and understanding needs to be shared vertically amongst different parties, as well as horizontally between competing parties.

Only through collaboration can we match the needs of users with what knowledge bases and software developers can realistically provide. Only through collaboration can we excel clinical decision support to the next level.


About the Author

Susan Cheng is a registered pharmacist, currently working as an eMedication Management Specialist at Intersystems. Her areas of specialty include drug databases and clinical decision support.



  1. Australian Commission on Safety and Quality in Health Care (2013), Literature Review: Medication Safety in Australia. ACSQHC, Sydney
  2. Duguid M.The importance of medication reconciliation for patients and practitioners. Aust Prescr 2012; 35: 15-9.
  3. Yu K, Sweidan M, Williamson M, Fraser A. Drug interaction alerts in software — what do general practitioners and pharmacists want? Med J Aust 2011; 195: 676-680.
  4. Kesselheim AS, Cresswell K, Phansalkar S, Bates DW, Sheikh A. Clinical decision support systems could be modified to reduce ‘alert fatigue’ while still minimizing the risk of litigation. Health Affairs 30, no.12 (2011): 2310-2317.
  5. Sweidan M, Reeve JF, Brien JE, Jayasuriya P, Martin JH, Vernon GM. Quality of drug interaction alerts in prescribing and dispensing software. Med J Aust 2009; 190: 251-254.
  6. Phansalkar S, Desai A, Choksi A, Yoshida E, Doole J, Czochanski M, Tucker AD, Middleton B, Bell D, Bates DW. Criteria for assessing high-priority drug-drug interactions for clinical decision support in electronic health records. BMC Medical Informatics and Decision Making 2013, 13: 6

Leave a Comment