Article Text

Download PDFPDF

Long-term effects of a multifaceted intervention to encourage the choice of the oral route for proton pump inhibitors: an interrupted time-series analysis
  1. I Colombet1,2,3,
  2. B Sabatier4,
  3. F Gillaizeau1,2,3,
  4. P Prognon4,
  5. D Begué4,5,
  6. P Durieux1,2,3
  1. 1
    INSERM, UMR_S 872, Paris, France
  2. 2
    School of medicine, Université Paris Descartes, Paris, France
  3. 3
    APHP, Georges Pompidou European Hospital, Paris, France
  4. 4
    Pharmacy Department, APHP, Georges Pompidou European Hospital, Paris, France
  5. 5
    School of Pharmacy, Université Paris Descartes, Paris, France
  1. Dr I Colombet, Département d’Informatique Hospitalière et Santé Publique, Hôpital Européen Georges Pompidou, 20–40 rue Leblanc, 75908 Paris cedex 15, France; isabelle.colombet{at}


Problem: Drugs are often given intravenously even when the patient is able to swallow and when an oral form would be more cost-effective.

Design: Evaluation of the impact of a multifaceted intervention on the early switch from intravenous to oral administration of proton pump inhibitors (PPI) in a hospital setting. The interrupted time series of intravenous PPI consumption was analysed.

Background and setting: At a French University Hospital, the Drug Committee, composed of multidisciplinary pharmacy and medical staff, addressed the issue of increasing consumption of intravenous PPI drugs (May 2003).

Strategy for change: Letters to department heads, academic analyses from members of the Drug Committee, paper reminders at the point of care and audit-feedbacks by pharmacists. Monitoring of consumption and repeated reminder letters were planned.

Effect of change: The consumption of PPI was stable before the first intervention (mean level: 954 units/month). An immediate decrease occurred after the first Drug Committee letter (30% relative reduction, 95% CI −16% to −46%; p<0.001) with a significant trend change during the first multifaceted intervention (−24 units/month, 95% CI −42 to −7; p = 0.007). After the end of the outreach visits (July 2004), the consumptions increased (+32 units/month, 95% CI: 14 to 50, p<0.001). The second intervention had no significant impact.

Lessons learnt: A complex intervention (audit, feedbacks, outreach visits) had an effect on practice. It was not sustained even after a less resource-intensive intervention. Other types of interventions are needed that could be continuously implemented to improve ordering practices long term.

Statistics from

In the current context of decreasing healthcare resources and increasing demand for quality control, hospitals are mandated to enforce the evidence-based use of drugs. One of the various forms of inappropriate use of drugs is the choice of intermittent intravenous infusion instead of oral administration. As part of operative and intensive care, the suitability of intravenous administration can be questioned if the oral route is possible or if a nasogastric tube is available: in such cases, a suspension or encapsulated granules may be preferable if they provide the same benefit with a better cost-effectiveness than the intravenous form. For example, Ripouteau et al1 studied an early switch from intravenous to oral acetaminophen to treat postoperative pain; they found that clinicians’ were unaware of the associated costs and that nurses believed that the intravenous drug was more effective for managing pain despite clear evidence of oral and intravenous forms being of similar efficacy. An educational intervention has been effective in correcting an inappropriate use of intravenous form of acetaminophen. Another example is the use of proton pump inhibitors (PPIs) for the prevention of peptic ulcer in intensive care patients, as recommended by guidelines.2 Three different forms of PPI are available: the infusion form, the suspension and encapsulated granules. All three have the same bioavailability. However, the hospital acquisition cost of one unit of intravenous PPI and the medical device for the infusion is 1000 (omeprazole) or 5800 (pentaprazole) times higher than that for one unit of equivalent oral treatment.


Clinical guidelines are considered as tools to improve medical practice. They may lead to changes in prescribing behaviour if their implementation is very straightforward. Their impact can be increased if supported by interventions demonstrated to be effective for quality improvement, like academic detailing and papers or computerised reminders at the point of care.3 Good results have been obtained with some multifaceted interventions aimed at improving the management of postoperative pain in a French hospital.1

However, the effectiveness of various interventions for quality improvement has been found to be moderate, and few data are available on the persistence of these effects.4 It is currently difficult to maintain a long-term effect with a single or repeated cost-effective intervention.

In France, a Drug Committee is mandatory in all hospitals, and its mission is to enforce the evidence-based use of drugs: (1) agreement for the introduction of new drugs, (2) update of the hospital’s drug reference list and (3) promotion of guidelines for the evidence-based use of drugs. The Committee members include hospital pharmacists, administrative staff, chief-nurses and referent physicians from all specialty departments of the hospital.

At Georges Pompidou European Hospital, a teaching hospital with 950 acute care beds in various medical and surgical departments, the president of the Drug Committee is the head of a medical department. In early 2003, the hospital pharmacist brought to the attention of the Committee the large increase in the hospital’s consumption of intravenous forms of PPI. The Committee decided to evaluate this problem and to design an intervention to promote the early switch from intravenous to oral administration of these drugs.


Details of approach taken

The Georges Pompidou European Hospital opened in July 2000. Consequently, the consumption of the intravenous form of PPI could be retrospectively analysed from January 2001 to May 2003.

Our general approach was to develop, implement and evaluate simple advice on the choice of the administration route of any PPI molecule (rabeprazole or omeprazole) used for the prevention of peptic ulcer. We used a multifaceted intervention to distribute this advice, directed by the Drug Committee. The intervention consisted of general information by letters to department heads, academic detailing, paper reminders at the point of care and audit-feedbacks.

We also aimed to ensure a continuing effect of this intervention, by regular monitoring and, if necessary, repetition of a less resource-consuming intervention.

We evaluated the effects of the intervention by analysing the interrupted time series of the monthly hospital consumption of drugs until December 2006. The analysis consisted of a segmented regression analysis.56 This allowed assessment of both changes in level (break) and changes in the trend of PPI consumption following the intervention.

Standard used

In France, the inhibition of gastric acid secretion is recommended in postoperative and acute care. The official guidelines associated with the marketing authorisation of the intravenous form of PPI recommend that it only be used in cases where oral aministration is not possible.

Measurement of problem

The hospital pharmacist records the number of intravenous PPI units delivered monthly to each department. We computed, for each department, the number of units of drug delivered per 10 000 patient-days, and we considered this an estimate of the true consumption of PPI by patients in the department.

The problem was assessed as the time series of the adjusted monthly consumption of PPI as recorded by the pharmacist for the whole hospital, from January 2001 to May 2003.


Graphic inspection of the times series before May 2003 (29 time points available) showed that the monthly consumption oscillated around a mean value of 954 units/10,000 patient-days. AutoRegressive, Integrated, Moving Average (ARIMA) models78 were used to analyse associations between observations of the preintervention series. The outcome of this analysis satisfied all assumptions of the general linear regression model (independence,9 constant variance10 and normality of the residuals11). Consequently, a segmented regression model including time and intervention terms could be applied to the original time series without any correction.



The first multifaceted intervention lasted 9 months, from May 2003 to July 2004. It consisted of the following:

  • The consumption of intravenous PPI was analysed for the 6 months up to May 2003 and communicated to the physicians of the Drug Committee. The committee decided to monitor the practice prospectively over the following months.

  • In early November 2003, the head of the Drug Committee sent a letter to every department head, promoting the use of encapsulated granules rather than the intravenous form when possible.

  • In January 2004, the pharmacist and the head of the Drug Committee visited each department head in the hospital (outreach visits). The visits involved explaining the problem and obtaining a commitment to act. The department heads were expected to understand the message and pass it on to their whole teams. Clinicians are not necessarily aware of the cost of drugs which they are prescribing, so the relevant information was particularly emphasised. The department heads were directly informed of their department’s monthly consumption of PPI and their consumption relative to the hospital’s total. A monthly feedback on consumption of PPI was planned until July 2004.

Between July 2004 and January 2006, the consumption of PPI was still monitored but without any feedback or other intervention.

In January 2006, a new increase in PPI consumption motivated a second intervention. This consisted of a letter sent by the head of the Drug Committee to all department heads without any other intervention. The members of the Drug Committee consensually assumed that a less resource-consuming intervention could be sufficient. The monitoring of PPI consumption was continued until December 2006.

Method of analysis of time series

We consider four delimited periods: before the first campaign (January 2001–May 2003), during the first multifaceted campaign (June 2003–June 2004), between the two campaigns (July 2004–January 2006) and after the second campaign (February 2006–December 2006). The first and second interventions started in mid-May 2003 and in mid-January 2006, respectively, but they were considered to start at the beginning of the following month (June 2003 and February 2006) because stocks were initialised at the beginning of each month. Fig. 1 presents the monthly time series of PPI consumption per 10 000 patient-days. A segmented regression model, inspired by Wagner et al,6 was used to estimate the effects of the two interventions (see table 1).

Figure 1

Time series of monthly proton pump inhibitor (PPI) unit used (per 10 000 patient-days) and predicted values from the segmented regression model.

Table 1 Segmented regression model for the monthly time series of the PPI consumption per 10 000 patient-days between 2001 and 2006

A stepwise method with two-tailed Student t tests was used to select significant variables in the final model. All analyses were conducted using SAS software (8.02 version), and a result was considered significant if p<0.05.


Assessment of change

Change was evaluated by the linear segmented model of effects of the interventions on mean monthly adjusted PPI consumption. The parameter estimates of the full model containing all explanatory variables, and the most parsimonious model obtained after elimination of non-significant terms, are described in table 2. This final model met all criteria of a reasonable linear segmented model (independence, constant variance and normality of the residuals).

Table 2 Parameter estimates from the full and the most parsimonious (final) segmented regression models predicting the mean monthly number of proton pump inhibitor units used per 10 000 patient-days

The mean monthly PPI consumption before the first intervention was estimated to be 954 units/10 000 patient-days (95% CI 910 to 998), and there was no significant month-to-month change.

There was an immediate change after the first Drug Committee letter and a trend change was observed during the first multifaceted intervention: the monthly consumption dropped by 295 units/10 000 patient-days (30% relative reduction, 95% CI −16% to −46%, p<0.001) and decreased significantly by 24 units/10 000 patient-days per month (95% CI −42 to −7, p = 0.007). After the end of the outreach visits, in July 2004, consumption started to increase with a significant monthly trend change of 32 units/10 000 patient-days per month (95% CI 14 to 50, p<0.001). The second intervention had no significant effect.

After the stepwise selection, the significant variables included in the most parsimonious model were: the mean number of PPI units used before the first intervention, the level and trend changes (decrease) during the first multifaceted intervention, and the level and trend changes (increase) between the two interventions.

Lessons learnt

This retrospective analysis of data covered the period from January 2001 to December 2006; it shows a significant discontinuity in the consumption of PPI drugs at the beginning of the first multifaceted intervention in May 2003. The various actions (audit and feedbacks, outreach visits) were associated with a persistent decrease in consumption which was significant throughout the intervention period. Consumption increased progressively again after July 2004, during the period without any intervention. This motivated implementation of the second intervention which took the form of a reminder of the first and was deliberately less resource-consuming. Analysis at six time points after the second intervention showed that there was no significant change. This study demonstrates that the significant effect of a multifaceted intervention is not maintained after its interruption. Furthermore, a subsequent reminder involving a simple letter was not sufficient to obtain the same effect.

The positive effect of the intervention may be explained both by the type of intervention and by the content of the guideline which was simple and acceptable in the current context of routine medical practice. Audit-and-feedback and outreach visits have been demonstrated to be among the most effective interventions to modify practices.1213 Although postintervention follow-up is included as an item of the checklist used for the Cochrane EPOC (Effective Practice an Organisation of Care) group’s systematic reviews, its length is rarely reported in publications.3 Moreover, some evaluation of an intervention on postoperative anticoagulant prescribing with an alternate time series design showed a significant effect, without any learning effect during control periods.14 Even when an intervention successfully improves prescription practices, its effect remains difficult to sustain.

This study has several limitations. The criterion defined for evaluating the effect of the interventions was the monthly consumption of PPI in the hospital, adjusted to the activity of the hospital. However, we analysed the number of intravenous PPI units delivered monthly to each department as recorded in real time by the hospital pharmacist as a surrogate for the true consumption. For each department, the difference between the monthly quantity delivered and the true monthly consumption was most probably compensated over a period of one to several months. The use of the quantity delivered as an approximation of the true consumption has been validated in another context.1 Moreover, the components of the first intervention were defined a priori by the Drug Committee, and the collection of data was prospective. However, the statistical analysis of data was not planned a priori, and the observational follow-up of the PPI consumption led to the design and implementation of the second intervention.

The main lesson of this study is that a complex intervention to modify a routine prescribing practice may have an effect on practice, but this effect may not be sustainable. The repetition of such interventions is time- and resource-consuming. A computer physician order entry (CPOE) system would allow real-time follow-up of ordering practice and in particular the choice of administration route. It also would make it possible to design interventions based on computerised reminders and real-time collection of indicators.1516 The durability of such interventions could be ensured with fewer resources. Moreover, the nature of such interventions could be generalised to all drugs for which intravenous and oral administration provide equivalent bioavailability.



  • Competing interests: None.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.