Oral quinolones in hospitalized patients: an evaluation of a computerized decision support intervention*

Authors


  • *

    The concept and design of the intervention was presented as a poster at the Annual Symposium of the American Medical Informatics Association, November 2002, San Antonio, TX, USA.

  • Present address: Preeti Bansal, Department of Ophthalmology, University of Kentucky, Lexington, KY, USA.

    Douglas A. Talbert, Department of Computer Science, Tennessee Technological University, Cookeville, TN, USA.

Todd Hulgan MD, MPH, Division of Infectious Diseases, Department of Medicine, Vanderbilt University, 345 24th Ave N; Suite 105, Nashville, TN 37203, USA.
(fax: 615 467-0158; e-mail: todd.hulgan@vanderbilt.edu).

Abstract.

Objective.  To determine whether a computerized decision support system could increase the proportion of oral quinolone antibiotic orders placed for hospitalized patients.

Design.  Prospective, interrupted time-series analysis.

Setting.  University hospital in the south-eastern United States.

Subjects.  Inpatient quinolone orders placed from 1 February 2001 to 31 January 2003.

Intervention.  A web-based intervention was deployed as part of an existing order entry system at a university hospital on 5 February 2002. Based on an automated query of active medication and diet orders, some users ordering intravenous quinolones were presented with a suggestion to consider choosing an oral formulation.

Main outcome measure.  The proportion of inpatient quinolone orders placed for oral formulations before and after deployment of the intervention.

Results.  There were a total of 15 194 quinolone orders during the study period, of which 8962 (59%) were for oral forms. Orders for oral quinolones increased from 4202 (56%) before the intervention to 4760 (62%) after, without a change in total orders. In the time-series analysis, there was an overall 5.6% increase (95% CI 2.8–8.4%; P < 0.001) in weekly oral quinolone orders due to the intervention, with the greatest effect on nonintensive care medical units.

Conclusions.  A web-based intervention was able to increase oral quinolone orders in hospitalized patients. This is one of the first studies to demonstrate a significant effect of a computerized intervention on dosing route within an antibiotic class. This model could be applied to other antibiotics or other drug classes with good oral bioavailability.

Introduction

Antibiotic selection is an area of particular complexity within medical decision-making, especially in the hospital setting. Antibiotic decisions not only have immediate effects on therapeutic outcomes and toxicities, but also on health care expenditures, with antibiotics accounting for 10–20% of hospital pharmacy acquisition costs [1–3]. These decisions also have long-term effects on selection of resistant organisms, outcomes of treatment failures and societal costs. Appropriate antibiotic use involves more than choosing the agent(s) with the most appropriate spectrum of activity. The prescriber must also consider the optimal dose, route of administration, duration of therapy and cost-effectiveness of their selection. With an ever-increasing number of choices and frequent changes in indications and formularies, it will probably become even more difficult for healthcare providers to maintain an up-to-date knowledge base of appropriate antibiotic use.

Healthcare providers routinely use intravenous (i.v.) antibiotics in hospitalized patients preferentially to oral forms for several reasons. Hospitalized patients generally have greater acuity and severity of illness than outpatients, often have conditions that compromise gastrointestinal absorption of oral medications, and may require a broader spectrum of antimicrobial activity than is provided by most oral antibiotic agents. In many cases, i.v. administration of antibiotics provides greater plasma and tissue concentrations of the agents. Doctor decisions regarding i.v. antibiotic use, however, may be based on beliefs that are not substantiated by the available literature. One in five doctors surveyed about treatment of community-acquired pneumonia felt that hospitalized patients should receive a ‘standard duration of intravenous antibiotics,’ and more than half believed that patients should remain on i.v. antibiotics until they are afebrile for more than 24 h [4]. Although the most appropriate time to transition to oral therapy can vary according to characteristics of the patient and of the selected antibiotic(s), early transition from i.v. to oral antibiotics has been studied in controlled trials of patients with community-acquired pneumonia, with favourable clinical and economic results [5–8].

Intravenous quinolone use is an area of antibiotic utilization that has attracted some attention, primarily because gastrointestinal absorption and bioavailability of this class of drugs are excellent [9, 10]. Additionally, i.v. quinolones cost substantially more; three to four times more per dose at our institution. Although the situations in which oral therapy may be safely used as initial therapy are not well described, there is no known clinical benefit of i.v. dosing of these agents in patients who can tolerate oral medications and are clinically stable. Clinical trials have shown that oral quinolones are effective as initial therapy in hospitalized patients with urinary tract infections [11] and community-acquired pneumonia [5], and in rapid transition from i.v. to oral forms in many settings, including severe infections [12–16]. Evidence-based and cost-effective clinical practice would minimize i.v. dosing of quinolones, although the appropriate minimum has not yet been defined and might vary depending on the patient population. A single-institution study found that a pharmacist-based intervention to facilitate transition from i.v. to oral levofloxacin after computerized identification of candidate patients was cost-effective [17]. To our knowledge, no published study has evaluated the use of computerized decision support systems (CDSS) at the point-of-care to promote initial oral antibiotic use in patients requiring hospitalization.

Computers have long been an attractive method of assisting doctors with complex medical decisions. Clinical trials have shown that computer reminders can improve drug dosing [18, 19], primary care screening [20–23] and clinical guideline adherence [24]. A systematic review of almost 70 clinical trials has shown that computerized interventions can improve clinician performance [25]. Computer-based provider order entry systems incorporating CDSS can also impact antibiotic use [26, 27]. Investigators at one institution have reported improved perioperative antibiotic use and decreased postoperative infection rates as a result of computer-based reminders [28]. The same group has used CDSS to improve empiric antibiotic selection [29], decrease excessive doses and adverse drug events [30], and implement antibiotic practice guidelines with improved clinical and economic outcomes [3]. These interventions have focused primarily on the decision to use antibiotics and selection of the most appropriate drug. The impact of CDSS on more subtle but perhaps equally important aspects of antibiotic use, such as dosing route, has not been well characterized.

Our primary hypothesis was that a targeted CDSS intervention integrated into an existing provider order entry system would increase the use of oral quinolones by decreasing unnecessary i.v. orders. We also sought to identify reasons why providers would choose to order i.v. forms of quinolones when the patient was prescribed other oral medications or an oral diet. This information should provide insight into the limitations of the intervention in accurately identifying patients capable of tolerating oral medications, and inform the design of future CDSS interventions. The Department of Biomedical Informatics (DBMI) at Vanderbilt University Medical Center (VUMC) has developed and implemented a computerized provider order entry system with CDSS capabilities [31–35]. This system, called WizOrder, can incorporate patient-specific data into decision support, give immediate point-of-care feedback and recommendations, and provided the optimal environment in which to test our hypothesis.

Methods

Study design

The primary outcome of our study was the change in weekly proportion of oral quinolone orders after deployment of the intervention. Based on pilot data from February 2001 to July 2001, we estimated a detectable difference based on average weekly quinolone orders. There were an average of 71 orders per week for oral quinolones at VUMC during this time period, which were 56% (SD = 4%) of weekly quinolone orders. Using a sample size of 52 weeks postintervention and α = 0.05, we were able to detect a 2.2% effect of the intervention on weekly quinolone orders with 80% power and a 2.9% effect with 95% power. PS software program v2.1.30 was used for power calculations [36].

We used an interrupted time-series analysis to account for secular trends and determine the effect of the intervention comparing the 52 weeks before and the 52 weeks after the intervention. For parameter estimation, a first-order autocorrelation was assumed. The value of this method in the evaluation of interventions of this type has been described previously [37, 38]. We examined the effects of the intervention across various hospital units, in intensive care unit (ICU) and non-ICU patients, and compared orders placed by medical doctor (MD) users to those placed by non-MD users. The secondary outcome measure, rates of prescriber justification for use of i.v. quinolones in cases identified as ‘able to take oral medications’, are reported and compared using Fisher's exact test. Statistical analyses were performed using Stata SE version 8.0 (Stata Corporation, College Station, TX).

To limit confounding, one of the authors (DSK), who served as chair of the VUMC Antibiotic subcommittee of the Pharmacy and Therapeutics committee throughout the study period, ensured that there were no pharmacy-initiated i.v.-to-oral conversion interventions and no quinolone formulary changes during the study. The VUMC Institutional Review Board reviewed and approved this study, and waiver of patient and provider consent was granted.

Description of the intervention

The intervention was initiated by any order for levofloxacin or ciprofloxacin (Fig. 1). Upon recognizing a relevant order, the CDSS searched the patient's current active orders for the presence of an oral medication or a solid diet. An order for either of these combined with the absence of a ‘nothing by mouth’ (NPO) order identified the patient as being ‘able to take oral medications’. Prescribers entering an order for an oral quinolone in a patient ‘able to take oral medications’, or an order for an i.v. quinolone in patients not ‘able to take oral medications’ were allowed to complete their order through a menu of doses and suggested indications based on quinolone choice and renal function (Fig. 2a). If an order for an i.v. quinolone was initiated in a patient identified as being ‘able to take oral medications’, the intervention presented the prescriber with a statement suggesting that the patient could potentially tolerate an oral quinolone (Fig. 2b). To place an order for an i.v. quinolone despite a recommendation to use an oral form, the prescriber selected from a list of predefined reasons for the use of i.v., or entered a free-text indication (Fig. 2c).

Figure 1.

Flow chart representation of the quinolone CDSS. NPO, nothing per os; i.v., intravenous; CrCl, creatinine clearance.

Figure 2.

Figure 2.

(a) Representative provider order entry screen for an oral quinolone order in a patient with an active order for another oral medication or diet. The user enters text in the bottom right box and can select from single mouse-click menu options in the middle box of the right column; CDSS-generated messages are displayed in the top right box; active orders are displayed in the left column. (b) Representative provider order entry screen displayed to user when an i.v. quinolone is ordered in a patient with an active order for another oral medication or diet. A suggestion that an oral form be considered is in the upper right-hand message portion of the screen. (c) Representative provider order entry screen displayed after selection of an i.v. quinolone in a patient with an active order for another oral medication or diet (option 2 in b). The user may complete the order for an i.v. form without restriction, but must select from the listed indications or enter a free-text indication.

Figure 2.

Figure 2.

(a) Representative provider order entry screen for an oral quinolone order in a patient with an active order for another oral medication or diet. The user enters text in the bottom right box and can select from single mouse-click menu options in the middle box of the right column; CDSS-generated messages are displayed in the top right box; active orders are displayed in the left column. (b) Representative provider order entry screen displayed to user when an i.v. quinolone is ordered in a patient with an active order for another oral medication or diet. A suggestion that an oral form be considered is in the upper right-hand message portion of the screen. (c) Representative provider order entry screen displayed after selection of an i.v. quinolone in a patient with an active order for another oral medication or diet (option 2 in b). The user may complete the order for an i.v. form without restriction, but must select from the listed indications or enter a free-text indication.

Figure 2.

Figure 2.

(a) Representative provider order entry screen for an oral quinolone order in a patient with an active order for another oral medication or diet. The user enters text in the bottom right box and can select from single mouse-click menu options in the middle box of the right column; CDSS-generated messages are displayed in the top right box; active orders are displayed in the left column. (b) Representative provider order entry screen displayed to user when an i.v. quinolone is ordered in a patient with an active order for another oral medication or diet. A suggestion that an oral form be considered is in the upper right-hand message portion of the screen. (c) Representative provider order entry screen displayed after selection of an i.v. quinolone in a patient with an active order for another oral medication or diet (option 2 in b). The user may complete the order for an i.v. form without restriction, but must select from the listed indications or enter a free-text indication.

Data collection

The VUMC order entry system recorded completed orders in a database maintained by the DBMI. Query of this database using the roots of the drug names (e.g. ‘cipro’ for ciprofloxacin and ‘levoflox’ for levofloxacin) identified initial orders for quinolones placed in the system from 1 February 2001 to 31 January 2003. All forms of quinolones other than oral or i.v. (e.g. ophthalmologic preparations) were excluded. Orders for quinolones other than levofloxacin and ciprofloxacin were not analysed, as they made up <5% of overall quinolone use at our institution at the time of the study and were not included in the intervention. Newer quinolones (e.g. gatifloxacin and moxifloxacin) were nonformulary during the study and orders for these drugs were directed to ciprofloxacin or levofloxacin via the order entry system. Indications for overriding the CDSS and prescribing i.v. quinolones were collected in a separate database maintained by the VUMC Department of Pharmacy.

Results

Effects on quinolone orders

A total of 15 194 orders for quinolones were placed during the 105-week study period (Table 1). Of these, 7623 orders (50.2%) were placed after the intervention was deployed. Of these, 4760 (62.4%) orders were for oral quinolones, compared with 4202 (55.5%) before the intervention. This was an absolute increase of 558 oral orders (13.3%). In the time-series analysis, the intervention increased the proportion of oral quinolone orders per week by 5.6% (95% CI 2.8–8.4%; P < 0.001; Fig. 3). There were no significant prescribing trends noted before or after the effect of the intervention.

Table 1.  Preintervention and postintervention quinolone orders, by unit and provider type
Hospital unitPreintervention ordersPostintervention orders
TotalOralTotalOral
  1. Values are expressed as n (%). aOther units included paediatric, obstetric and gynaecological, rehabilitation and neurological units. bIntervention effect P ≤ 0.001 by interrupted time-series analysis. Orders from subunits within medical and surgical units were not analysed.

All hospital units75714202 (55.5)76234760 (62.4)b
All surgical units2316 (30.6)1130 (48.8)2277 (29.9)1274 (56.0)
Trauma unit402150 (37.3)345152 (44.1)
All medical units4534 (59.9)2700 (59.6)4272 (56.0)2872 (67.2)b
Cardiac units638409 (64.1)687530 (77.1)
Myelosuppression units684460 (67.3)707492 (69.6)
Renal unit383250 (65.3)295196 (66.4)
Other unitsa721 (9.5)372 (51.5)1074 (14.1)614 (57.2)
Non-ICUs6028 (79.6)3601 (59.7)6256 (82.1)4105 (65.6)b
MD users only5913 (78.1)3317 (56.1)5708 (74.9)3721 (65.2)b
Figure 3.

Interrupted time-series analysis of overall quinolone orders during the study period. The intervention increased overall oral quinolone orders by 5.6% (95% CI 2.8–8.4%; P < 0.001).

MDs placed 77% of all quinolone orders during the study period. The remaining orders were entered by other health care providers under the supervision of MDs. The majority of these were entered by Registered Nurses (9%), pharmacists (8%) or nurse practitioners (3%). To assess the true effect of the intervention on decision-making, we performed a sub-analysis using only those orders placed by MD providers, presumably the primary decision-makers with the potential for greatest influence by the CDSS. Amongst these MD providers, there was an increase in the proportion of oral quinolone orders per week of 6.0% (95% CI 2.8–9.3%; P < 0.001; Fig. 4).

Figure 4.

Interrupted time-series analysis of quinolone orders placed by doctor providers only. The intervention increased doctor-placed oral quinolone orders by 6.0% (95% CI 2.8–9.3%; P < 0.001).

Orders for quinolones were placed on 39 different hospital units. These units are categorized by specialty as shown in Table 1. The majority of quinolone orders (58.0%) were from medical units, where there was a 7.9% increase (95% CI 3.9–11.9%; P < 0.001; Fig. 5) in the proportion of weekly oral quinolone orders as a result of the intervention. Orders for oral quinolones on surgical units increased slightly, but this was not statistically significant (1.1% increase per week; 95% CI −6.7–9.0%; P = 0.78). Overall non-ICU oral quinolone orders increased by 4.6% per week (95% CI 2.0–7.3%; P = 0.001), but there was no significant effect of the intervention on quinolone orders placed in the ICUs (7.3% increase per week; 95% CI −2.1–16.6%; P = 0.13).

Figure 5.

Interrupted time-series analysis of quinolone orders on general and sub-specialty medical units. The intervention increased oral quinolone orders on these units by 7.9% (95% CI 3.9–11.9%; P < 0.001).

Reported reasons for overriding the intervention

We collected user-selected reasons for ordering i.v. quinolones despite the presence of other active oral medications or diet orders and the absence of ‘NPO’ orders. Of 1361 i.v. quinolone orders placed during the first 26 weeks after the intervention was deployed, 1017 (75%) met the CDSS criteria for being able to take oral medications and thus required either a predefined or free-text rationale. The most common reason for overriding the CDSS recommendations was ‘patient unable to take oral medications’, which was entered in 49% of the cases and was significantly more likely to be entered by a user on a surgical unit than a nonsurgical unit (P = 0.002). More than one quarter of the time, users entered a free-text rationale for the use of i.v. quinolones. The greatest proportion of these orders (23%) were for ‘postictal’ patients and were all entered on the neurosurgical ICU.

Estimated cost savings

Although we did not perform a formal cost analysis, our data allow for estimates of cost savings as a result of the intervention. There was an absolute increase of eight oral quinolone orders per week as a result of the intervention (95% CI four to 12 orders per week). Using the most conservative assumptions that only one dose of i.v. quinolone per order would be saved as a result of the intervention, and an acquisition cost difference between an i.v. and oral quinolone dose of $10, the intervention would save $2100–6400 year−1. Any increase in the average number of i.v. doses saved per oral order would increase the potential cost savings.

Discussion

The ability of CDSS to influence doctor behaviour has been well established, but not specifically in selection of dosing route, an important aspect of antibiotic prescribing. These data demonstrate that a CDSS integrated into an existing provider order entry system was able to influence initial dosing of quinolones in hospitalized patients by increasing the use of oral quinolones. The intervention had its greatest effect on quinolone prescribing on nonsurgical and non-ICU units. Prescribing on surgical and ICU units was not significantly affected by the intervention. This may have been due to fewer quinolone orders on these units and less power to detect a difference, or, perhaps more likely, patients on these units were less able to tolerate oral medications and an i.v. quinolone was more often the most appropriate choice.

To confirm the ability of our intervention to influence behaviour of those users most likely to make dosing decisions, we identified orders as those placed by MD users or non-MD users. Just over 75% of quinolone orders during the study were placed by MDs, whilst ‘surrogate’ prescribers (e.g. nurses or pharmacists) entered the remainder. We observed an effect of the intervention on MD prescribers that was slightly more than that seen amongst all prescribers (6.0% vs. 5.6% increase in oral orders per week).

We also observed that a majority of i.v. quinolones in the hospital setting are prescribed for patients with characteristics that suggest they could tolerate an oral antibiotic with good bioavailability. Despite the significant impact of the intervention, the majority of i.v. quinolone orders placed in the 6 months after the intervention was deployed required user override. There are several possible conclusions from this observation. First, providers frequently order oral diets and/or oral medications (other than antibiotics) in patients with questionable ability to tolerate them. Secondly, our intervention was unlikely to influence a prescriber to order an oral quinolone when they felt it was not clinically appropriate. Thirdly, future interventions should incorporate more specific clinical criteria to better identify the most appropriate patients for oral antibiotic dosing.

Our study has several limitations that merit discussion. We chose not to perform a randomized, concurrently controlled trial for several reasons. Primarily, the most appropriate unit of randomization was unclear. At our institution, like most, patient care ‘teams’ are made up of persons at various levels of training and experience who care for the same group of patients. The decision-making process is often, and appropriately, diluted. It is unclear whether teams, individual users or individual orders are the most appropriate unit of randomization. The first two are subject to hierarchy bias, in which the decision is not guided by the intervention, but by the direction of a supervisor who may or may not be part of the user's (intervention or control) group. Randomizing orders would be subject to learning effect in which an order placed without the intervention may be guided by a previous exposure rather than an independent decision. Additionally, the randomization of users to be presented with a CDSS in the academic setting is logistically difficult. There are frequent changes in team members, cross-coverage of patients by various providers outside the team and staggered clinical rotation dates. These factors could threaten the even distribution of known and unknown confounders that is the primary benefit of randomization.

We have not confirmed the accuracy of the user-entered reasons for overriding the CDSS. It would be useful to know the ‘predictive value’ of these reasons (i.e. was a patient truly unable to tolerate oral medications at the time they were identified as such by the user) for the design of future interventions and a better understanding of provider behaviour. Due to the fact that there is no definitive diagnostic test for ‘sepsis’ or gastrointestinal absorption, this would be difficult to assess retrospectively in cases other than those in which patients were identified as being intubated.

A conservative estimate of the potential cost savings as a result of the intervention was more than $6000 year−1. A more realistic estimate using a mean difference of four i.v. doses saved per oral order resulting from the intervention [12] and a cost difference of $15 per dose would lead to a savings of $13 000–38 000 year−1. These estimates do not include indirect costs such as i.v. tubing apparatus, pharmacy preparation and nursing time. Additionally, studies comparing early switch therapy interventions have shown a decrease in length of stay [7, 39], cost [12] and episodes of infusion phlebitis [5], all of which would also be expected in a setting where appropriate patients are treated initially with oral antibiotics.

As with any CDSS implemented at the local level, there are limitations in generalizability. The intervention described here is one of several CDSS interventions implemented into our order entry system over recent years to address quality of care and cost-containment at our institution. Although we have not surveyed users regarding the acceptance of this specific intervention, we have found our system to be well integrated into the workflow of the hospital and generally well accepted by users [40]. The primary objectives of this study were twofold: to evaluate the effects of a specific intervention at our institution and to prove the broader principle of CDSS effectiveness on dosing within a class of antibiotics. Despite the uniqueness of the WizOrder system, the logic of the intervention could be generalized to other provider order entry systems and to other medications with good oral bioavailability.

Antibiotics make up a large part of hospital pharmacy expenditures, their use varies widely between individuals and institutions, and they are prescribed across diverse medical specialties. In contrast to most clinical care algorithms or practice guidelines, antibiotic education must be directed towards virtually all health care providers to achieve appropriate and consistent antibiotic use at an institutional level. Because this is logistically impossible at all but the smallest hospitals, computerized interventions are an appealing, if not requisite mode of antibiotic education. Computerized order entry and decision support can be an effective tool to educate a diverse group of providers, and, as our results demonstrate, influence subtle but important aspects of high-volume hospital prescribing in a way that is not feasible with traditional methods of formulary control. More studies at a greater number of institutions are needed to better quantify and generalize the impact of such interventions.

Conflict of interest statement

DAT and RAM receive authorship royalties through Vanderbilt University from the commercial distribution of WizOrder.

STR has received consulting fees from McKesson Information Solutions, which has licensed WizOrder for commercial distribution.

None of the other authors have related disclosures or potential conflicts of interest.

Acknowledgements

The authors thank Dominik Aronsky MD, PhD, for helpful discussions during the early conception of the study, and William Schaffner MD, for critical review of the manuscript.

Financial support: NIH Training Grant T32 AI 07474-08 and Vanderbilt Clinical Research Scholar Award K12 RR17697 (TH).

Ancillary