Abstract
Paramount to patient safety is the ability for nurses to make clinical decisions free from human error. Yet, the dynamic clinical environment in which nurses work is characterized by uncertainty, urgency, and high consequence, necessitating that nurses make quick and critical decisions. The aim of this study was to examine the influence of human and environmental factors on the decision to administer among new graduate nurses in response to alert generation during bar code-assisted medication administration. The design for this study was a descriptive, longitudinal, observational cohort design using EHR audit log and administrative data. The study was set at a large, urban medical center in the United States and included 132 new graduate nurses who worked on adult, in-patient units. Research variables included human and environmental factors. Data analysis included descriptive and inferential analyses. This study found that participants continued with administration of a medication in 90.75% of alert encounters. When considering the response to an alert, residency cohort, alert category, and previous exposure variables were associated with the decision to proceed with administration. It is important to continue to study factors that influence nurses’ decision-making, particularly during the process of medication administration, to improve patient safety and outcomes.
Keywords: nurse decision-making, medication administration, alerts, patient safety, human factors, clinical decision support systems, alarm fatigue
Introduction
The first reported sentinel event resulting in patient hard due to the failure of a clinician to respond to an alarm was published in 1974 by the ECRI Institute (formerly the Emergency Care Research Institute).1 Though the case predated the concept of alarm fatigue, it marked the first reported sentinel event resulting from the failure of a clinician to respond to an alarm. Between 2009 and 2012, The Joint Commission (TJC) received reports of 98 alarm-related sentinel events, 80 of which resulted in patient deaths.2 The sudden increase in alarm-related reports prompted TJC to recognize the significance of alarm fatigue in the clinical setting and institute a National Patient Safety Goal (NPSG) directed at alarm safety.3 Since then, the concept of and research related to alarm and alert fatigue is increasingly described in the healthcare literature. Additionally, researchers are starting to explore alert optimization methods and implementation strategies.4–6 The majority of research to date, however, examines alerts intended for prescribers and pharmacists, and few studies examine the impact of alerts on nurses, specifically. This study contributes to the existing body of literature about alarm and alert fatigue, and further, addresses an existing gap in the literature specifically related to understanding decision-making in response to alert generation during bar code-assisted medication administration (BCMA).
Background
The use of technology in the clinical setting has rapidly expanded. A recent study reported that nurses spend an average of one third of their shifts interacting with various technologies.7 Indeed, more than 80% of hospitals use electronic health records (EHRs) for clinical documentation.8 and the use of clinical decision support systems (CDSSs) has also increased. While CDSSs are designed to enhance decision-making and improve the safety and quality of care, these systems may lead to error when not used as intended.9,10
One specific type of CDSS is BCMA which is used by nurses when administering medications to patients. Studies show that BCMA is effective in reducing rates of medical error related to medication administration.11,12 Research also shows, however, that nurses have developed workarounds in some instances during BCMA use, such as when an alert is triggered by the BCMA system.13 These workarounds introduce additional risks for error.14,15 Human factors, such as physical characteristics (e.g., anthropometric measurements), cognitive characteristics, and interactions between humans and the overall work system or environment16, and environmental factors may impact or influence nurse cognition and decision-making. What remains unclear is which factors (human, environmental, and other) influence alert generation and impact the nurse’s decision to accept or override an alert.
Further, research suggests that repeated exposure to alarms or alerts, particularly those of low clinical meaningfulness, may influence clinician response and contribute to the development of alarm or alert fatigue.3,17 Regarding BCMA, a recent review found that the number of alerts nurses are exposed to during medication administration ranges from 1% to upwards of 40% across all medication administration encounters.18 Additionally, human factors research suggests that repeated exposure to alarms or alerts influences clinician response.19 However, despite advances toward optimal deployment of alerts, more research is needed to explore how nurses, in particular, respond to alert generation. To our knowledge, no studies to date examine the influence that both human and environmental factors may have on nurse response to alerts that are generated during BCMA.
To guide methodological decisions about study design, we used a research model to illustrate study variables adapted from a conceptual model published by Sloss and Jones.20 The conceptual model synthesizes theories on human cognition and proposes a framework showing the process of nurse cognition, along with factors that may influence nurse decision-making during BCMA. The primary objectives of this study were: 1) to examine patterns of the decision to administer medication after alert generation among new graduate nurses and 2) to examine the influence of human and environmental factors on the decision to administer after alert generation during among new graduate nurses. Our goal is that this work can help to better understand the impact of alert generation on nurse decision-making and to improve nurse workflow while at the same time reducing medication administration errors.
Methods
Study Design and Research Setting
Study objectives were addressed through a descriptive, longitudinal observational cohort design using EHR audit log data. The setting for this study was a large, urban academic medical center in the United States. BCMA was integrated into the existing EHR system at the facility in the mid 2010s. Facility procedures for BCMA stipulate that after selecting the correct patient record in the EHR, the nurse verifies patient identity by visually checking and electronically scanning the patient’s identification bracelet.
Sample and Recruitment
The sample included new graduate nurses who were hired into a new graduate nurse residency program between January 1, 2018 and December 31, 2018. The rationale for the selection of new graduate nurses as eligible participants was twofold. First, new graduate nurses are new to the role of registered nurse as well as the independent, unsupervised use of BCMA during the medication administration process. Both the process of BCMA and alerts that are generated are relatively novel to this population. As an alert-naïve population, they are optimal for the evaluation of the effects of alert exposure. Second, medication administration errors are notably high among new graduate nurses,21 and the study of factors that influence nurse cognition during medication administration by new graduate nurses is highly relevant and significant to reduce errors and improve patient safety.
Both existing health system employees who transitioned into a new graduate nurse role (e.g., from patient care assistant), as well as new graduate hires to the facility were eligible for inclusion. Additionally, participants must have been hired to work on an in-patient care unit that uses BCMA to document medication administration. New graduate nurses were excluded if their employment tenure at the health system was less than one year and thus ended before the end of the data collection period, if they transferred units within the study period, or if they had more than two weeks off between any two shifts indicating prolonged absence or leave. No exclusions were made based upon age, gender, or race.
Data Sources
Data were collected from EHR audit log and administrative data sources. Medication administration and alert encounter information including alert type, prescribed drug, dose, and route were collected from the EHR audit log. Administration data pertaining to whether or not a scanned medication was administered, and if administered the time of administration, were also contained within the EHR audit log datafile. Demographic information for each nurse participant was obtained from a nurse residency program administrative datafile.
Data Collection
An honest broker, defined as a neutral intermediary between the health system and research team,22 was given a list of new graduate registered nurses and associated demographic data (age, gender, residency start date, education, and assigned unit) from administrative records. The honest broker replaced the participant’s user ID with the unique identifier in both the administrative and EHR audit log datafiles before being shared with the researchers.
Protection of Human Subjects
Personally identifiable information (PII) was collected about each nurse participant and patients who received care from the nurses in the sample. Nurse participant and patient PII was removed by an honest broker prior to data analysis using standard de-identification procedures. Institutional Review Board (IRB) approval was obtained prior to study implementation.
Study Variables
The primary outcome variable of this study was BCMA decision output (whether or not the nurse decided to proceed with administration). Decision output is a binary (Yes/No) outcome variable derived from data fields in the EHR audit log. If there was a documented administration time for an alert encounter, the decision output variable (Administered) was recorded as “Yes” and if there was no medication administration for an alert encounter the variable was recorded as “No.” During inferential analyses, Administered Yes/No was recoded as “No” equals 0 and “Yes” equals 1.
Independent variables include alert encounter characteristics, individual factors, human factors, and environmental factors. Alerts were categorized as action required or display only, where actionable alerts required a decision to be made by the nurse to proceed to accept the alert (thereby stopping the process of medication administration) or override the alert (continuing with administration). For example, an actionable alert is one that prompts the nurse to verify lab results before moving forward with administration. Alerts categorized as display only were informational and only required acknowledgment before proceeding, such as drug-drug interaction.
Individual factors refer to demographic characteristics of the nurse participant, including age, gender, education (highest nursing degree achieved), residency cohort (one of four cohorts to which new graduates were hired into, denoted by start date), and experience (ranging from 0 days on date of hire to 364 days at the end of the study period). Residency cohort was determined by date of hire, with four distinct residency program start dates over the year-long recruitment period. Unit types were used to describe work environment in this study and included general, intermediate, and intensive care unit (ICU).
Human factors variables included patient workload, medication workload, and prior exposure to alerts. Patient workload represented the number of unique patients that a nurse is assigned to care for over a 12-hour shift and was calculated using the number of unique patient IDs that a nurse experienced an alert or medication administration encounter for over a 12-hour shift. Medication workload was the total number of alert or medication administration encounters that a nurse experienced on a given shift. Exposure variables described previous experience with, or exposure to, an alert during BCMA. For this study, exposure was measured four different ways: 1) previous exposure to an alert of any type, 2) previous exposure to an alert of the same type, 3) previous exposure to an alert of the same type for the same medication, and 4) previous exposure to an alert of the same type for the same medication in the same patient.
Data Cleaning, Preparation, and Analysis Methods
Datafiles containing (1) participant demographic information and (2) medication administration and alert encounters were received from the honest broker. The datafiles contained information for new graduate nurses hired between January 1, 2018 and December 31, 2018 and included medication administration and alert encounters from January 1, 2018 to December 21, 2019, to capture all encounters through one year after the participant’s hire date. After examining the data for missing fields, each medication administration and alert encounter were linked to nurse participant demographic variables across unique nurse identifier using Python.
Descriptive analyses were conducting using JMP Pro 15 and inferential analyses were applied using SASⓇ software (SAS Institute Inc., Cary, NC, USA). Descriptive statistics included calculation of means, standard deviations, ranges, and frequencies of alert encounter characteristics. For inferential analyses, repeated-measures analysis of covariance (RMANCOVA) was used for the binary outcome variable, Administered (Yes/No). Interactions and covariates were removed through a backward stepwise elimination approach. In the following stages of the model, predictors were removed sequentially starting with the variable having the highest p-value until the final model included all variables with a p-value less than or equal to 0.05.
During data analysis, we discovered that due to a reconfiguration, the “Acetaminophen Max Dose” alert was incomplete. We conducted descriptive and inferential analyses on all alerts, including “Acetaminophen Max Dose.” However, due to incomplete data related to this alert type and possible impact on the validity of our models, we present analyses that do not include “Acetaminophen Max Dose” alert data.
Further, we recognized the “Patch Removal” alert comprised a large percent of alerts in relation to other types of alerts. Given the aims of this study were to explore nurse cognition and decision outputs in response to alert types, it was thought that inclusion of the Patch Removal alert type could introduce noise into alert analyses and that better insights could be gleaned with exclusion of the alert type. Though we did not intend to conduct inferential analyses including or excluding selected alert types, after reviewing the data we thought that this approach would help us to better identify and understand factors that may influence nurse cognition and decision making in response to alert generation. Consequently, we present inferential analyses for two scenarios: 1) Scenario 1: Including Patch Removal Alerts and 2) Scenario 2: Excluding Patch Removal Alerts.
Results
The original datafile contained information for 249 new graduate nurses and 784,237 alert and administration medication encounters. After excluding nurses who could not be linked to medication administration and/or alert encounters (n=41), the sample of nurses was then screened against inclusion and exclusion criteria. Seventy-six nurses were removed for the following reasons: missing or ineligible unit of hire (n=41); employment tenure less than one year (n=30); and long gaps (greater than 22 days) in between alert encounter or medication administrations (n=5). The final sample for this study consisted of 132 nurse participants, associated with 587,879 medication administration of which there were 20,018 alert encounters. Findings that describe medication administration and alert encounter patterns are reported elsewhere.
Overall Decision Output Patterns
Table 1 displays overall decision output frequencies by nurse participants across residency cohort and unit type by alert category. Among new graduate nurses during the first year of practice, participants continued with administration of a medication in 90.75% of alert encounters and stopped the medication administration process in 9.25% of encounters, regardless of alert category. However, after excluding the Patch Removal alert, participants decided not to administer medications at slightly higher rates, choosing to administer a medication in only 73.09% of alert encounters.
Table 1:
Overall Decision Output Frequencies in Response to Alerts During BCMA by Alert Category
| Admin Status by Alert Cat. | Residency Cohort | Unit Type | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1. All Alerts | 1 | 2 | 3 | 4 | ICU | Intermediate | General | All | |||||||||
| N | % | N | % | N | % | N | % | N | % | N | % | N | % | N | % | ||
| Yes | Display Only | 1 | 0.22 | 1 | 0.14 | 0 | 0.00 | 0 | 0.00 | 0 | 0.00 | 1 | 0.21 | 1 | 0.09 | 2 | 0.10 |
| Action Req | 408 | 91.69 | 642 | 90.68 | 634 | 90.06 | 168 | 90.32 | 400 | 91.53 | 432 | 89.81 | 1020 | 90.67 | 1852 | 90.65 | |
| Total | 409 | 91.91 | 643 | 90.82 | 634 | 90.06 | 168 | 90.32 | 400 | 91.53 | 433 | 90.02 | 1021 | 90.76 | 1854 | 90.75 | |
| No | Display Only | 1 | 0.22 | 3 | 0.42 | 1 | 0.14 | 1 | 0.54 | 2 | 0.46 | 2 | 0.42 | 2 | 0.18 | 6 | 0.29 |
| Action Req | 35 | 7.87 | 62 | 8.76 | 69 | 9.80 | 17 | 9.14 | 35 | 8.01 | 46 | 9.56 | 102 | 9.07 | 183 | 8.96 | |
| Total | 36 | 8.09 | 65 | 9.18 | 70 | 9.94 | 18 | 9.68 | 37 | 8.47 | 48 | 9.98 | 104 | 9.25 | 189 | 9.25 | |
| 2. Excluding Patch Removal Alerts | 1 | 2 | 3 | 4 | ICU | Intermediate | General | All | |||||||||
| N | % | N | % | N | % | N | % | N | % | N | % | N | % | N | % | ||
| Yes | Display Only | 1 | 0.92 | 1 | 0.95 | 0 | 0.00 | 0 | 0.00 | 0 | 0.00 | 1 | 1.49 | 1 | 0.68 | 2 | 0.57 |
| Action Req | 89 | 81.65 | 75 | 71.43 | 81 | 66.94 | 11 | 61.11 | 112 | 81.16 | 42 | 62.69 | 102 | 68.92 | 256 | 72.52 | |
| Total | 90 | 82.57 | 76 | 72.38 | 81 | 66.94 | 11 | 61.11 | 112 | 81.16 | 43 | 64.18 | 103 | 69.60 | 258 | 73.09 | |
| No | Display Only | 1 | 0.92 | 3 | 2.86 | 1 | 0.83 | 1 | 5.56 | 2 | 1.45 | 2 | 2.99 | 2 | 1.35 | 6 | 1.70 |
| Action Req | 18 | 16.51 | 26 | 24.76 | 39 | 32.23 | 6 | 33.33 | 24 | 17.39 | 22 | 32.84 | 43 | 29.05 | 89 | 25.21 | |
| Total | 19 | 17.43 | 29 | 27.62 | 40 | 33.06 | 7 | 38.89 | 26 | 18.84 | 24 | 35.83 | 45 | 30.40 | 95 | 26.91 | |
There were notable differences in administration rates between cohorts when Patch Removal alerts were excluded. Nurses in the first residency cohort proceeded with administration 82.57% of the time whereas nurses in the fourth residency cohort only proceeded with administration 61.11% of the time. Similar differences were detected among nurses by unit type. Nurses working in the ICU settings experienced administration rates of 81.16%, while nurses working on general units experienced administration rates of 69.60%.
Decision Output Patterns in Response to Alert Type
Table 2 shows decision outputs by unit type and alert type. Decision outputs varied somewhat by alert type. In general, Patch Removal and Neuroblock Vent, that confirms a patient must be on a ventilator to receive the medication, alerts were all associated with high rates of decision output by the nurse to proceed with administration, ranging from 92.41% to 94.44%. Rates of administration decreased slightly for Vancomycin Trough Task Check alerts (84.76%). The Diabetes Task alert was associated with the lowest rate of decision to proceed with (43.66%) administration following an alert (43.66%). The Dilution Required alert was rarely generated (n=10) but was associated with the decision to continue with administration 100% of the time.
Table 2:
Decision Outputs by Unit Type and Alert Type
| Alert Type | Admin Status | Unit Type | |||||||
|---|---|---|---|---|---|---|---|---|---|
| ICU | Intermediate | General | All | ||||||
| N | % | N | % | N | % | N | % | ||
| Patch Removal | Yes | 288 | 96.32% | 390 | 94.20% | 918 | 93.96% | 1596 | 94.44% |
| No | 11 | 3.68% | 24 | 5.80% | 59 | 6.04% | 94 | 5.56% | |
| Vancomycin Trough Task Check | Yes | 32 | 84.21% | 12 | 75.00% | 45 | 88.24% | 89 | 84.76% |
| No | 6 | 15.79% | 4 | 25.00% | 6 | 11.76% | 16 | 15.24% | |
| Neuroblock Vent | Yes | 72 | 92.31% | 0 | 0.00% | 1 | 100.00% | 73 | 92.41% |
| No | 6 | 7.69% | 0 | 0.00% | 0 | 0.00% | 6 | 7.59% | |
| Diabetes Task | Yes | 1 | 33.33% | 15 | 51.72% | 15 | 38.46% | 31 | 43.66% |
| No | 2 | 66.67% | 14 | 48.28% | 24 | 61.54% | 40 | 56.34% | |
| BCMA_Factors (triggered when dose drawn from multiple containers) | Yes | 4 | 30.77% | 1 | 100.00% | 18 | 78.26% | 23 | 62.16% |
| No | 9 | 69.23% | 0 | 0.00% | 5 | 21.74% | 14 | 37.84% | |
| BCMA_Routes | Yes | 1 | 50.00% | 1 | 100.00% | 15 | 75.00% | 17 | 73.91% |
| No | 1 | 50.00% | 0 | 0.00% | 5 | 25.00% | 6 | 26.09% | |
| Dilution Required | Yes | 0 | 0.00% | 3 | 100.00% | 7 | 100.00% | 10 | 100.00% |
| No | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | |
| Reassess Pain | Yes | 1 | 100.00% | 5 | 83.33% | 1 | 33.33% | 7 | 70.00% |
| No | 0 | 0.00% | 1 | 16.67% | 2 | 66.67% | 3 | 30.00% | |
| Med Drug Allergy | Yes | 0 | 0.00% | 0 | 0.00% | 1 | 50.00% | 1 | 16.67% |
| No | 2 | 100.00% | 2 | 100.00% | 1 | 50.00% | 5 | 83.33% | |
| Buprenorphine Total Dose | Yes | 0 | 0.00% | 4 | 100.00% | 0 | 0.00% | 4 | 100.00% |
| No | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | |
| Drug-Drug Interaction | Yes | 1 | 100.00% | 0 | 0.00% | 0 | 0.00% | 1 | 25.00% |
| No | 0 | 0.00% | 3 | 100.00% | 0 | 0.00% | 3 | 75.00% | |
| Kinetic Consult | Yes | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% |
| No | 0 | 0.00% | 0 | 0.00% | 2 | 100.00% | 2 | 100.00% | |
| Dose Range Check | Yes | 0 | 0.00% | 1 | 100.00% | 0 | 0.00% | 1 | 100.00% |
| No | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | |
| Pharm Height Weight | Yes | 0 | 0.00% | 1 | 100.00% | 0 | 0.00% | 1 | 100.00% |
| No | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | 0 | 0.00% | |
There was some variation of decision output across alerts by unit type. For instance, 75% of nurses working on intermediate units proceeded with administration in response to a Vancomycin Trough Task Check alert whereas 88.24% of general unit nurses and 84.21% of ICU nurses decided to proceed with administration. Additionally, nurses working on intermediate units were more likely to proceed with administration (51.72%) than nurses working on general units (38.46%) in response to the Diabetes Task alert. Many of the remaining alert types only occurred in a few instances across units, so it is difficult to glean additional insight from decision output patterns.
Factors Influencing Decision Outputs
Table 3 shows results of the decision output model for the two scenarios. In both scenarios, residency cohort was inversely related to the decision output to continue with administration. Additionally, actionable alerts were positively associated with the decision to proceed with administration.
Table 3:
Models for Decision Output
| Scenario | Solution for Fixed Effects | Type III Tests of Fixed Effects | Tests of Covariance Parameters (Based on the Restricted Likelihood) | ||
|---|---|---|---|---|---|
| Estimate | Pr > |t| | Pr > F | Pr > ChiSq | ||
| 1. All Alerts | Intercept | −0.5117 | 0.5575 | ||
| Residency Cohort | −0.1797 | 0.0657 | 0.0657 | ||
| Alert Category | 0.0006 | ||||
| Action Required | 2.9307 | 0.0006 | |||
| Display Only | . | . | |||
| Previous Exposure to Any Alert Type | −0.0616 | <.0001 | <.0001 | ||
| Previous Exposure to the Same Alert Type | 0.1435 | <.0001 | <.0001 | ||
| Nurse Effect | 0.0174 | ||||
| 2. Excluding Patch Removal Alerts | Intercept | −0.4157 | 0.6464 | ||
| Residency Cohort | −0.4345 | 0.0049 | 0.0049 | ||
| Alert Category | 0.0169 | ||||
| Action Required | 2.0694 | 0.0169 | |||
| Display Only | . | . | |||
| Previous Exposure to the Same Alert Type | 0.4997 | 0.0003 | 0.0003 | ||
| Nurse Effect | 0.1243 | ||||
Previous exposure variables were associated with decision output in both scenarios; however, the specific exposure variables and effects were inconsistent. Previous exposure to any alert was associated with decision output only in Scenario 1. In both models, previous exposure to the same alert type was associated with the decision to proceed with administration. Previous exposure to the same alert type and drug, as well as previous exposure to the same alert type, drug, and patient, was not associated with decision output in either model.
Several variables did not have significant association with decision outputs in either scenario, including unit type, education, experience, patient workload, and medication workload. Nurse effect was found to have significance only when Patch Removal alerts were included (Scenario 1). Because the decision output models used binary outcomes, the exact percentage of variation attributable to unmeasured individual nurse characteristics could not be calculated.
Discussion
This is believed to be the first study that explores the influence of human and environmental factors on nurse cognition and decision-making in response to alert generation during BCMA. Building on previous studies on alert fatigue, we found that factors such as previous alert exposure over time may influence how nurses think about and respond to alerts during the process of medication administration.
Patterns and Predictors of Decision Outputs
Including all alert types, nurses continued with administration of medication in 90.75% of all alert encounters. This rate of administration is high when compared to prior studies, that reported override rates from 10.30% to 78.00%.23,24 It is important to note, however, that none of the alert types in this study included the rights of medication administration (i.e., wrong patient, wrong medication, etc.). While there were likely instances when alerts appropriately notified nurses of potential error, it was also the case that alerts were generated during medication encounters where continuing with administration was the appropriate response. Therefore, it is necessary to consider the 90.75% administration rate in the context of numerous potential and appropriate medication administration encounters. Further, it is notable that nurses decided not to proceed with administration in response to 9.25% of alert encounters. While this number represents a small percentage of total medication encounters, it is meaningful when considering the number of potential errors averted.
This study found that decision outputs differed by alert type. Alerts that were more likely to result in stopping administration included Diabetes Task, Med Allergy, Drug-Drug Interaction, and Kinetic Consult. While not accounting for a large percentage of alerts in general, the Diabetes Task alert was most commonly associated with insulins lending to the significance that decision outputs when this alert occurred were more likely to result in stopping the administration of insulin, a high-risk medication. This may indicate that this alert was effective in preventing inappropriate administration of insulin. Still, additional research is needed to determine rationale for proceeding with administration in cases where nurses documented insulin administration.
When examining descriptive statistics related to decision output, nurses working in ICUs had slightly higher decision output rates to continue with administration. However, unit type was not associated with decision output in inferential analyses. Several of the exposure variables were significant when modeling decision output. In Scenario 1, when “Patch Removal” alerts were included, there were some differences in directionality of the previous exposure variables. Exposure to any alert type was inversely associated with the decision to proceed with administration while previous exposure of the same alert type was positively associated with a decision to proceed with administration. Again, this could be attributable to exposure or experience. Because the previous exposure to any alert was associated with the decision to not proceed with administration, it may be the case the exposures to the same alert type do increase alert fatigue related to that specific alert, leading to nurses more frequently proceed with administration. In keeping with the alert fatigue literature, it is important to consider that high exposure to alerts of low clinical meaningfulness contribute to the development of alert fatigue.25
Indeed, in the second decision output model excluding Patch Removal alerts, nurses were more likely to proceed with administration when exposed to a previous alert of the same type. This is notable, as this model only included 353 alert encounters. It was hypothesized that previous exposure, particularly previous exposures of the same alert type and in the same context, may contribute to the development of alert fatigue. However, the inconsistencies among previous exposure associations calls into question the utility of these measures and additional research is needed.
Another significant variable in the decision output models was residency cohort. Specifically, nurses in earlier residency cohorts were more likely to proceed with administration. Though it cannot be determined what specifically resulted in this association, the association between residency cohort and decision output could be due to work environment changes, changes to the residency training program related to BCMA processes or alert response, or other event that drew attention to alert response during BCMA.
Limitations
Several limitations to this study arose from the use of retrospective, EHR audit log data for analysis. Due to the large size of the data set, Python was used to merge datafiles and calculate several of the variables. Though merged datafiles and calculated variables were validated, it is possible that errors occurred during data cleaning, linkages, and calculations. Further, the reconfiguration of the Acetaminophen Max Dose alert during the study period resulted in incomplete alert encounter data.
While this study examined factors that influenced decision making in response to alert generation, we do not know whether or not the alert was generated in error, or if continuing with administration was the appropriate response by the nurse after an alert was generated during BCMA. Also, the study sample included new graduate nurses from one acute care setting in the US. Therefore, findings from this study may not be generalizable nationally or internationally.
Implications for Clinical Practice
Key implications for practice include the importance of thoughtful selection and deployment of alerts, given the influence of human and environmental factors of nurse decision making. It is important to recognize that individual and work environment factors, such as clinical unit or alert category, may influence how nurses think about and respond to alerts. Special attention should be paid to alert optimization. For example, rather than have blanket alerts such as “Patch Removal” that is triggered any time a nurse scans a medication administered via patch, the rule could be written to only generate an alert when the nurse has not already documented removal of the existing patch. Because alerts are interruptive in nature, we know that clinicians may develop workarounds or become desensitized to the alarm or alert stimulus. It is therefore essential that hospitals and health systems continually evaluate the meaningfulness and effectiveness of alerts used in BCMA and other CDSSs.
Though this study explored the decision by the new graduate nurse to proceed with administration in response to varying alert types during BCMA, future research is needed to describe nurse cognition during BCMA or similar CDSSs. While this study explored human and environmental factors that influence nurse decision making in response to alert generation during BCMA, it is important to examine the impact of high alert exposure and subsequent development of alert fatigue over time. Future studies could explore how often new graduate nurses seek guidance from a fellow nurse, preceptor, manager, or pharmacist or the effect of orientation process, particularly between units, could influence responses to alerts. Significantly, we must continue to optimize EHR workflow in a way that continues to promote thoughtful clinical practice in a technology-rich environment.
As work environments, workloads, and previous exposures change over time, it is necessary to consider the impact that these factors can have on nurse decision making and ensure that BCMA or other CDSSs can work to optimize nurse decision-making. By identifying factors associated with response to an alert, particularly among new graduate nurses, interventions or changes to alert rules can be implemented to improve the safety of medication administration. In the future, possible strategies can include not only establishing customizable alert parameters by hospital or unit type, but also for individual clinicians. Building on initial findings and utilizing methods from this study, subsequent research should explore decision-making among nurses while using CDSSs generally, as well as the continued focus on human factor approaches in CDSS design.
Conclusion
This study explored the influence of human and environmental factors on decision making among nurses during BCMA. Gaining a better understanding of the influence of human factors on nurse decision making during care can lead to better CDSS design that reduces the occurrence of medical errors and improves patient safety outcomes. Further, optimizing EHR and clinical workflows can uphold thoughtful and deliberate practice by nurses in technology-rich, clinical environments.
Acknowledgements:
Work by ES was supported by the National Institute of Nursing Research of the National Institutes of Health under Award Number T32NR013456 and the University of Utah Senior Vice-President for Health Sciences Research Unit and College of Nursing. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the University of Utah.
Footnotes
Conflict of Interest Statement: The authors declare no conflicts of interest.
Contributor Information
Elizabeth A. Sloss, School of Nursing, Virginia Commonwealth University, Richmond, VA.
Terry L. Jones, Department of Adult Health and Nursing Systems, School of Nursing, Virginia Commonwealth University.
Kathy Baker, UVA Health.
Jo Lynne W. Robins, Department of Adult Health and Nursing Systems, School of Nursing, Virginia Commonwealth University.
Leroy R. Thacker, Department of Biostatistics, School of Medicine, Virginia Commonwealth University
References
- 1.Sendelbach S, Funk M. Alarm fatigue: A patient safety concern. AACN Adv Crit Care. 2013;24(4):378–386. doi: 10.1097/NCI.0b013e3182a903f9 [DOI] [PubMed] [Google Scholar]
- 2.Johnson KR, Hagadorn JI, Sink DW. Alarm safety and alarm fatigue. Clin Perinatol. 2017;44(3):713–728. doi: 10.1016/j.clp.2017.05.005 [DOI] [PubMed] [Google Scholar]
- 3.Ruskin KJ, Hueske-Kraus D. Alarm fatigue: Impacts on patient safety. Curr Opin Anaesthesiol. 2015;28(6):685–690. doi: 10.1097/ACO.0000000000000260 [DOI] [PubMed] [Google Scholar]
- 4.Bhakta SB, Colavecchia AC, Haines L, Varkey D, Garey KW. A systematic approach to optimize electronic health record medication alerts in a health system. Am J Health-Syst Pharm AJHP Off J Am Soc Health-Syst Pharm. 2019;76(8):530–536. doi: 10.1093/ajhp/zxz012 [DOI] [PubMed] [Google Scholar]
- 5.Chaparro JD, Hussain C, Lee JA, Hehmeyer J, Nguyen M, Hoffman J. Reducing Interruptive Alert Burden Using Quality Improvement Methodology. Appl Clin Inform. 2020;11(1):46–58. doi: 10.1055/s-0039-3402757 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Hussain MI, Reynolds TL, Zheng K. Medication safety alert fatigue may be reduced via interaction design and clinical role tailoring: a systematic review. J Am Med Inform Assoc. 2019;26(10):1141–1149. doi: 10.1093/jamia/ocz095 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Higgins LW, Shovel JA, Bilderback AL, et al. Hospital nurses’ work activity in a technology-rich environment: A triangulated quality improvement assessment. J Nurs Care Qual. 2017;32(3):208–217. doi: 10.1097/NCQ.0000000000000237 [DOI] [PubMed] [Google Scholar]
- 8.Adler-Milstein J, Holmgren AJ, Kralovec P, Worzala C, Searcy T, Patel V. Electronic health record adoption in US hospitals: the emergence of a digital “advanced use” divide. J Am Med Inform Assoc JAMIA. 2017;24(6):1142–1148. doi: 10.1093/jamia/ocx080 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Greenes RA, Bates DW, Kawamoto K, Middleton B, Osheroff J, Shahar Y. Clinical decision support models and frameworks: Seeking to address research issues underlying implementation successes and failures. J Biomed Inform. 2018;78:134–143. doi: 10.1016/J.JBI.2017.12.005 [DOI] [PubMed] [Google Scholar]
- 10.Lopez KD, Gephart SM, Raszewski R, Sousa V, Shehorn LE, Abraham J. Integrative review of clinical decision support for registered nurses in acute care settings. J Am Med Inform Assoc. 2016;24(2):ocw084. doi: 10.1093/jamia/ocw084 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Hutton K, Ding Q, Wellman G. The effects of bar-coding technology on medication errors. J Patient Saf. Published online February 2017:1. doi: 10.1097/PTS.0000000000000366 [DOI] [PubMed] [Google Scholar]
- 12.Poon EG, Keohane CA, Yoon CS, et al. Effect of bar-code technology on the safety of medication administration. N Engl J Med. 2010;362(18):1698–1707. doi: 10.1056/NEJMsa0907115 [DOI] [PubMed] [Google Scholar]
- 13.van der Veen W, Taxis K, Wouters H, et al. Factors associated with workarounds in barcode-assisted medication administration in hospitals. J Clin Nurs. 2020;29(13–14):2239–2250. doi: 10.1111/jocn.15217 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Holden RJ, Rivera-Rodriguez AJ, Faye H, Scanlon MC, Karsh BT. Automation and adaptation: Nurses’ problem-solving behavior following the implementation of bar coded medication administration technology. Cogn Technol Work Online. 2013;15(3):283–296. doi: 10.1007/s10111-012-0229-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Strudwick G, Reisdorfer E, Warnock C, et al. Factors Associated With Barcode Medication Administration Technology That Contribute to Patient Safety: An Integrative Review. J Nurs Care Qual. Published online June 2017:1. doi: 10.1097/NCQ.0000000000000270 [DOI] [PubMed] [Google Scholar]
- 16.Russ AL, Fairbanks RJ, Karsh BT, Militello LG, Saleem JJ, Wears RL. The science of human factors: Separating fact from fiction. BMJ Qual Saf. 2013;22(10):802–808. doi: 10.1136/bmjqs-2012-001450 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Johansen ML, O’Brien JL. Decision Making in Nursing Practice: A Concept Analysis. Nurs Forum (Auckl). 2016;51(1):40–48. doi: 10.1111/nuf.12119 [DOI] [PubMed] [Google Scholar]
- 18.Sloss EA, Jones TL. Alert types and frequencies during bar code-assisted medication administration: A systematic review. J Nurs Care Qual. 2020;35(3):265–269. doi: 10.1097/NCQ.0000000000000446 [DOI] [PubMed] [Google Scholar]
- 19.Phansalkar S, Edworthy J, Hellier E, et al. A review of human factors principles for the design and implementation of medication safety alerts in clinical information systems. J Am Med Inform Assoc. 2010;17(5):493–501. doi: 10.1136/jamia.2010.005264 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Sloss EA, Jones TL. Nurse cognition, decision support, and barcode medication administration: A conceptual framework for research, practice, and education. CIN Comput Inform Nurs. Published online 2021. doi: 10.1097/cin.0000000000000724 [DOI] [PubMed] [Google Scholar]
- 21.Saintsing D, Gibson LM, Pennington AW. The novice nurse and clinical decision-making: How to avoid errors. J Nurs Manag. 2011;19(3):354–359. doi: 10.1111/j.1365-2834.2011.01248.x [DOI] [PubMed] [Google Scholar]
- 22.Choi HJ, Lee MJ, Choi CM, et al. Establishing the role of honest broker: bridging the gap between protecting personal health data and clinical research efficiency. PeerJ. 2015;3:e1506. doi: 10.7717/peerj.1506 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Koppel R, Wetterneck T, Telles JL, Karsh BT. Workarounds to barcode medication administration systems: Their occurrences, causes, and threats to patient safety. J Am Med Inform Assoc JAMIA. 2008;15(4):408. doi: 10.1197/JAMIA.M2616 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Sakowski J, Leonard T, Colburn S, et al. Using a bar-coded medication administration system to prevent medication errors in a community hospital network. Am J Health Syst Pharm. 2005;62(24):2619–2625. doi: 10.2146/ajhp050138 [DOI] [PubMed] [Google Scholar]
- 25.Kane-Gill SL, O’Connor MF, Rothschild JM, et al. Technologic distractions (Part 1): Summary of approaches to manage alert quantity with intent to reduce alert fatigue and suggestions for alert fatigue metrics. Crit Care Med. 2017;45(9):1481–1488. doi: 10.1097/CCM.0000000000002580 [DOI] [PubMed] [Google Scholar]