Discussion

We observed a high frequency of discrepancies between medication and intravenous fluid orders within a CPOE system and programmed settings on ‘smart’ infusion pumps at the beside. The observed frequency of discrepancy (24%) between medication order and the medication infusions in our study is similar to the frequency observed by Barker et al (19%) in 2002.26 There are no published frequencies of fluid discrepancies to compare with our observed frequency of intravenous fluid discrepancies (42%). Among the medication classes of anti-infectives, concentrated electrolytes and anticoagulants, the observed frequency in order-programming discrepancies ranged from 40 to 100%. This is of particular concern because concentrated electrolytes and anticoagulants have been identified as high-risk medications.27

Why were these discrepancies present? The exact cause for any specific observed discrepancy is unclear; however, a number of potential origins have been identified. Problems with order entry include omission of entry as well as incorrect order entry. Alternatively, pump programming may be the source of the discrepancy and could be related to a lack of recognition of active orders, failure to execute programming, and incorrect programming. Regardless of the origin, it should be noted that there are inherent risks and potential for harm within the process of reconciling these discrepancies. Specifically, there is a very real risk that a nurse may discover the discrepancy between order and infusion pump, and decide to reconcile it by changing one to match the other. As long as the nurse correctly reconciles to the accurate item (order or pump), the patient does well. However, reconciliation of the incorrect item may create additional risk of patient harm.

Due to the limitations of this study, several questions remain. First, which is correct: the medication order in the CPOE system or the actual pump settings? Our study cannot answer this, as it was impossible to tell which technology reflects the intent of the prescribing physician and which technology was incorrect. Anecdotally, asking the bedside providers or physicians may not help clarify whether the order or the pump is correct because these providers typically go to the pump or order to aid their recall. A second question that remains unanswered by this study is, why might this phenomenon be occurring in the first place? This question is perhaps easier to answer by considering the flow of actions and information associated with titration of infusions of medications.

For medications that are titrated for clinical effect, such as epinephrine, the assumed process by the engineers designing CPOE and infusion technologies is as follows: Provider conceives of order → Provider enters order → Provider communicates order to nurse → Nurse reprogrammes infusion pump.

While this may happen, there are other variations of this process that appear to occur related to the patient being the point of initiating care and NOT the computer. For instance, the sequence may be: Provider conceives of order → Provider enters order → Provider fails to communicate order to nurse → Nurse does not act or acts only after later discovering order.

More commonly, an evaluation of the patient at the bedside leads to one of the following occurrences: Provider exams patient → Provider conceives of order → Provider communicates order to nurse → Nurse reprogrammes infusion pump → Provider may or may not later enter order → Nurse may or may not enter order as verbal order in the event they realise the provider failed to enter the order. Or: Provider exams patient → Provider conceives of order → Provider reprogrammes infusion pump → Provider may or may not later enter order.

In this last example, the nurse may not recognise that a pump change was made for some time.

An additional factor that likely contributed to the increased frequency of discrepancies between CPOE orders and pump programming relates to vascular access and the number and relative timing of infusions, and compatibility of medications. For example, orders within a CPOE system may indicate that multiple medications are to be given within a narrow time frame. Depending on the duration of infusion and compatibility of the various medications, it may be necessary for the bedside nurse to delay administration of one medication while another is infusing. This situation could have led to observation of either omitted or unauthorised medications under our study design, and it may help explain the high frequency of discrepancies within the anti-infective and concentrated electrolyte medication groups.

This study was based on a several concepts from the science of human factors engineering including a model of understanding systems in healthcare as well as the multidimensional nature of nursing work. The issue of nursing work is critical to this study, as nurses are the providers who program infusion pumps in the studied PICU. This task-level activity (programming a pump) is performed in a system that involves multiple prescribing physicians and clinicians who are charged with entering orders into the CPOE system. The nurses in the PICU act as one of several ‘people’ in the care delivery system who must perform the task-level action of order discovery and pump programming while also performing their job- and unit-level activities.

An additional concept that is important to this study is the understanding that errors and harm are both manifestations of flawed systems of care delivery.19 These system flaws can be identified as risks, which can then be analysed for potential impact and reduced, when appropriate. Using this concept, this project did not attempt to identify harm from medication errors. How harm might manifest in any given patient would vary by the medication or fluid involved in a discrepancy and by the cause of the error. An unauthorised medication might create harm that would be different from an omitted medication or wrong dose. The heterogeneity of diagnoses, severity of illness, comorbidities, ages and other therapies that exist in a PICU setting confound the detection of harm. Thus, the goal was to identify any discrepancy between a medication order and the actual programmed infusion that created the potential for harm to patients. Only through identifying and analysing risks can the process of risk elimination and mitigation begin.

This leads to the final question that remains unanswered by this study: how best to eliminate or mitigate this risk? The question is most difficult to answer. Because the expectation of information and work flow as envisioned by designers of technologies used in medication delivery is very different from the reality at the bedside, any solutions may serve as ‘workarounds’ that increase risk. Thus, the introduction of technologies intended to reduce risk and harm may, by virtue of inadequate understanding of the pre-existing and resultant systems of medication delivery, create increased risk. This is particularly true in ICU environments where the changes are often made at the patient's bedside. Ironically, these patients happen to be among the most ill and most complex, receiving large numbers of potentially dangerous medications.