The anesthesia information management system (AIMS) is now a mature technology, as evidenced by evolutionary rather than revolutionary changes in function between version upgrades. It is estimated as adopted in the United States by 80% of academic anesthesiology departments (ie, those with residency programs) as of 2016.1 As often occurs for widely adopted innovations, substitution of previous generations of AIMS technology is occurring,2 most often when hospitals replace their current electronic health record software with enterprise-wide systems3 (eg, Epic's OpTime or Cerner's SurgiNet). At each of the authors’ institutions, such conversions are currently underway. The process of migrating to another vendor’s AIMS represents an opportunity to revisit choices made during the original configuration, informed by use of the local system over many years.
Accurate documentation of drug administration is a core requirement of anesthesia care.a Thus, it is optimal, from a performance perspective, that default drug doses, selectable by quick entry buttons, reflect the most common practices of the local anesthesia care team (Figure). Automation is helpful because errors are common when drugs are manually entered into an AIMS.4,5 Efficient use of an AIMS is enhanced by quick buttons because fewer mouse and key clicks are required than when typing in doses.6,7 In addition, the range of the defaults might prevent inadvertent dosing errors for unfamiliar drugs by anchoring the low and high range of the most typical quantities administered. However, such benefit would require that the providers check the defaults before administration. This is essentially a passive implementation of Guardrails technology.8
In this study, we examined multiple years of drug administration data from the AIMS at 3 institutions. These were compared with the existing default doses, configured at the 2 sites with the same AIMS, to determine the extent to which reconfiguration would be necessary to match existing practices. Our first hypothesis was that most (ie, >50%) of the drugs would need at least one change to the current default doses. If true, the implication would be that those responsible for maintaining the AIMS or involved in configuring a new AIMS should analyze the historical dose data at their facility and adjust the default doses, as necessary.
We also examined the potential anchoring effect of having default doses for the 20 most frequently administered drugs. Our second hypothesis was that for most (>50%) of these drugs, the 4 most common doses at the site where default doses were not implemented would be included among the 4 most common doses at the other 2 sites. If true, this would suggest that having default doses does not affect the typical drug administration behavior of anesthesia providers.
The study was determined by the institutional review boards of Thomas Jefferson University and the University of Miami as “not human subjects research.”
Data were extracted from the AIMS at Thomas Jefferson University Hospital (TJUH) (Innovian, Draeger, Telford, PA), the University of Miami Hospital (UMH) (Innovian), and Jackson Memorial Hospital (JMH) (Picis Anesthesia; Picis Clinical Solutions, Inc, Wakefield, MA). These included the medication name, dosage, and units of administration for all anesthesia records for which the patient’s age was >16 years. The case date range was January 2006 through January 2016 at TJUH and JMH, and September 2013 through January 2016 at UMH.b Medications available for selection as of January 2016 were analyzed. The default doses configured for each medication at TJUH and UMH were extracted from the AIMS database (limited by Innovian to a maximum of 4 choices); default doses had not been configured in Picis. During the study period, there were no alterations to the default doses.
At all sites, there were separate listings for medications given either by bolus or by infusion; therefore, medication-unit combinations (eg, propofol-mg, propofol-µg/kg/min) were analyzed separately. For convenience, we hereafter refer to each medication-unit combination as a drug. For each drug, the number of administrations was determined, and the fraction of total administrations among all drugs was calculated. These frequencies were sorted in descending order, and the N drugs corresponding to the cumulative 99.9% of all drugs were then analyzed. For each drug, the frequency of use of each dose was determined, and such doses were sorted in descending order. Up to the 5 most common doses of each drug occurring with a frequency of ≥5%c were determined; the top 4 became the new recommended default doses. For example, if one dose was selected >95% of the time, only one default would be provided, whereas if the fourth most common drug was administered 5% of the time, 4 defaults would be provided. These new recommended default drug doses were compared with the existing defaults. A change was noted if an old default dose was either removed or replaced with a new default dose or if a new default dose was added. The minimum number of changes required for the transformation was determined. Thus, the default doses for each drug in the current AIMS could have between 0 and 4 changes. The percentages of drugs among all drugs analyzed with 0, 1, 2, 3, or 4 changes were then calculated. The fewer the changes, the greater the similarity between the originally configured defaults and those informed by actual use, analogous to computation of the Levenshtein distance between 2 string sequences.9
The 20 most commonly administered drugsd during cases that were common to all 3 sites were determined for analysis of the anchoring effect of having specified default doses. (Each drug was only counted once per case, even if multiple doses were entered.) Antibiotics (because doses are fixed by protocol) and drugs not administered intravenously (because doses vary according to the site of injection) were excluded. From the Picis data, the 4 most frequently documented drug doses representing ≥5% use among all doses for that drug were determined. For each of these drugs in Picis, we determined whether all these typical doses were represented among the 4 most common doses (used ≥5% of the time for the drug) at the 2 sites with Innovian. If all Picis doses were represented, the drug was scored as a 1, and if any were missing, the drug was scored as a 0.
The one-sided binomial test in R version 3.2.3 (the R Foundation for Statistical Computing, Vienna, Austria) was used to calculate the probability that the proportion of Innovian drugs requiring ≥1 change was >0.5 (first hypothesis), and that the proportion of Picis drugs with all typical doses included in the Innovian typical doses was >0.5 (second hypothesis). P < .01 was required to claim statistical significance.
At TJUH, 84.7% of the drugs representing 99.9% of administered doses would need at least 1 change in the configured default doses (P < 10–6 compared with 0.5). At UMH, 77.5% of such drugs would require at least 1 change (P < 10–6 compared with 0.5). Only 15.3% and 22.5% of drugs at TJUH and UMH, respectively, would not require any changes (Table 1). The first hypothesis, that most drugs with default doses configured would need at least one change to the default values, was accepted.
Of the 20 most commonly administered drugs at the site lacking default doses (JMH), 100% of the doses were included within the most commonly administered doses at TJUH or UMH (Table 2; P < 10–6 vs 0.5). The second hypothesis, that there would be lack of evidence of alteration of typical drug administration behavior created by the presence of default doses, was accepted.
At the 2 facilities where default drug doses had been configured, the majority of drugs required at least 1 change in the default values to match actual dosing, with <25% of the drugs needing no changes. The original default doses had been selected on an ad hoc basis, without formal analysis. The implication is that when changing to a new AIMS, the database should be analyzed to determine the most appropriate default drug doses. Unless actual usage had been evaluated and default doses were adjusted accordingly, simply copying over the default settings from the previous system is not recommended because this would simply perpetuate the suboptimal status quo. Such analysis would still be useful on a periodic basis to capture changes in practice over time. Indicated changes to the default doses have been implemented at the 2 Innovian facilities as a result of this study, and the recommended defaults have been provided to the information systems personnel responsible for configuring the new enterprise-wide systems. For AIMS allowing >4 default doses, additional defaults might be desirable, especially for drugs for which the loading doses are much higher than subsequent bolus doses (eg, propofol and rocuronium). Thus, we recommend manual examination of the recommended defaults for drugs with a bimodal pattern of administration, and adjustment, as necessary.
There were modest differences in the drug doses administered, with a frequency of ≥5% among the facilities. This suggests that, ideally, one would use local data to inform the default configuration process. However, in the absence of such data (eg, converting from a paper anesthesia record to an AIMS), use of the Supplemental Digital Content (which includes suggested defaults for all drugs administered ≥100 times during the study interval at each of the study facilities; Supplemental Digital Content 1, http://links.lww.com/AA/B512) can help guide the selection of a reasonable set of initial default doses. Subsequently, the dose information in the AIMS database could be analyzed to refine the default doses to those most often administered by the anesthesia providers in a given practice.
There did not appear to be an anchoring effect on typical doses administered by having a range of defaults available for selection via quick buttons. That is, providers at the facility where defaults had not been configured (JMH) did not differ in their pattern of administration of typical doses from the facilities where defaults were provided (TJUH and UMH). However, the generalizability of this observation will require additional study. For several reasons, we were not able to analyze outlier doses to see whether there might have been a beneficial (ie, patient safety) effect of anchoring. First, there was sometimes ambiguity attributable to likely selection of the wrong units, but an appropriate actual dose. For example, in some instances, 50 mg of fentanyl was entered, undoubtedly a documentation error, with 50 μg the likely amount given. Second, we could not determine whether large doses still within the realm of possible usage (eg, 500 μg of fentanyl) were data entry errors or if they were actually administered. Third, because the site without default doses used a different AIMS than the other 2 sites, we could not separate the potential benefit of anchoring from effects because of the user interface. Finally, there was no way to determine whether a dose was entered into Innovian using the quick button or typed in manually using the keyboard or virtual keypad (Figure). We also lacked data to assess whether drug documentation was more efficient when the quick buttons were used versus alternative methods of entering doses (ie, virtual keypad or keyboard). Such an assessment would likely need to be addressed via simulation, as was described for the arrangement of drugs in an AIMS, alphabetically versus categorically.10
There are several limitations of this study. First, we studied 2 sites at which the same AIMS (Innovian) had been installed, so results might be different with other systems. However, having the same software allowed us to eliminate potential confounding effects from the user interface itself because the layout of the drug dialog boxes in Innovian (Figure) is not modifiable. Another limitation is the shorter duration of available data at UMH versus the other 2 facilities. However, given that the percentages of drugs requiring 0, 1, 2, 3, or 4 changes were similar between the 2 facilities, we think that this is not an important issue. A third limitation is that we excluded patients ≤16 years of age, so the most common doses we found do not apply to pediatric patients. In such patients, dosing is often specified more rigorously with weight-based dosing, as opposed to adults, who often received doses less tightly coupled to their body mass. However, the process we describe can still be applied to facilities caring for pediatric patients, modified to adjust dosing for all drugs on a per body mass basis. At facilities where both children and adults receive anesthesia care, consideration should be given to creating separate drug default profiles for pediatric patients.11 Some AIMS (eg, Epic) allow specifying default doses on both a fixed and a per mass basis (eg, 2 mg/kg of propofol and 100 mg of propofol), with calculation of the actual dose based on the entered patient body weight (Allan F. Simpao, MD, The Children’s Hospital of Philadelphia, personal communication). Finally, although the nurse anesthetists who worked regularly at UMH and JMH were different, there was some overlap in the anesthesiologists and anesthesia residents at the 2 facilities because anesthesia services are provided at both locations by the University of Miami Department of Anesthesiology. Thus, the 2 facilities were not completely independent. This may have resulted in a greater convergence of common drug doses between the 2 facilities than would otherwise have occurred if they had been completely separate. However, 94.8% of cases at UMH during the study period involved nurse anesthetists, not anesthesia residents, mitigating the impact of this commonality.
In conclusion, we showed that for the majority of drugs, the originally configured quick button doses in the AIMS at 2 separate institutions differed by at least 1 default from the most commonly administered doses. Periodic analysis of the AIMS database is recommended to determine changes that need to be made to the default drug doses to reflect current clinical practice, and especially when migrating to a new AIMS.
Name: Luis I. Rodriquez, MD.
Contribution: This author helped design the study, conduct the study, and write the manuscript.
Name: Todd J. Smaka, MD.
Contribution: This author helped write the manuscript.
Name: Michael Mahla, MD.
Contribution: This author helped conduct the study and write the manuscript.
Name: Richard H. Epstein, MD, CPHIMS.
Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.
This manuscript was handled by: Nancy Borkowski, DBA, CPA, FACHE, FHFMA.
1. Stol IS, Ehrenfeld JM, Epstein RH. Technology diffusion of anesthesia information management systems into academic anesthesia departments in the United States. Anesth Analg. 2014;118:644–650.
2. Norton JA, Bass FM. A Diffusion theory model of adoption and substitution for successive generations of high-technology products. Manag Sci. 1987;33:1069–1086.
3. Sanders DS, Read-Brown S, Tu DC, et al. Impact of an electronic health record operating room management system in ophthalmology on documentation time, surgical volume, and staffing. JAMA Ophthalmol. 2014;132:586–592.
4. Avidan A, Dotan K, Weissman C, Cohen MJ, Levin PD. Accuracy of manual entry of drug administration data into an anesthesia information management system. Can J Anaesth. 2014;61:979–985.
5. Wax DB, Feit JB. Accuracy of vasopressor documentation in anesthesia records. J Cardiothorac Vasc Anesth. 2016;30:656–658.
6. Saitwal H, Feng X, Walji M, Patel V, Zhang J. Assessing performance of an electronic health record (EHR) using cognitive task analysis. Int J Med Inform. 2010;79:501–506.
7. Rose AF, Schnipper JL, Park ER, Poon EG, Li Q, Middleton B. Using qualitative studies to improve the usability of an EMR. J Biomed Inform. 2005;38:51–60.
8. Manrique-Rodríguez S, Sánchez-Galindo A, Fernández-Llamazares CM, et al. Developing a drug library for smart pumps in a pediatric intensive care unit. Artif Intell Med. 2012;54:155–161.
9. Levenshtein VI. Binary codes capable of correcting deletion, insertions, and reversals. Soviet Physics Doklady. 1966;10:707–710.
10. Marian AA, Dexter F, Tucker P, Todd MM. Comparison of alphabetical versus categorical display format for medication order entry in a simulated touch screen anesthesia information management system: an experiment in clinician-computer interaction in anesthesia. BMC Med Inform Decis Mak. 2012;12:46.
11. Patterson ES, Zhang J, Abbott P, et al. Enhancing electronic health record usability in pediatric patient care: a scenario-based approach. Jt Comm J Qual Patient Saf. 2013;39:129–135.
Supplemental Digital Content
Copyright © 2016 International Anesthesia Research Society