Reducing Inappropriate Proton Pump Inhibitors Use for Stress Ulcer Prophylaxis in Hospitalized Patients: Systematic Review of De-Implementation Studies

EMBASE (Ovid) and Medline (Ovid) electronic databases were searched by an information specialist (RS) on the 8th of January 2020 from inception, without restrictions on publication date or language. Searching included indexing terms, free text terms, and synonyms for proton pump inhibitor combined with terms for low-value care and hospitalized patients (detailed information in Supplementary Table 2a/b). Expert in the field (BvM) retrieved one study reference independently from the systematic literature search.

We included studies of adult, hospitalized patients in non-ICU settings, in which an intervention to reduce the use of inappropriate PPI was evaluated. (Quasi-) randomized controlled trials and comparative observational studies reported in English, Dutch, or German were eligible for inclusion. Studies that addressed both PPI and H2RA medication as SUP were included if data on PPI use could be extracted separately. Studies combining inpatient and outpatient data were included when inpatient data could be extracted separately.

Pairs of authors (CO, PH, JJKvD) independently screened all titles and abstracts that were retrieved from the literature search, using Rayyan Software³⁰. Subsequently, they assessed final eligibility based on full-text assessment. Disagreements between the authors were resolved by discussion.

Data were extracted by one author (CO/JJKvD) and checked by another author (PH). Disagreements were resolved by discussion or involvement of a third author (LH). A predefined, piloted digital form was used for data extraction (including details of study design, participants, setting, de-implementation strategies (components and targets), outcomes). The interventions used for de-implementation were classified based on the taxonomy provided by the Cochrane Effective Practice and Organisation of Care (EPOC) Group (Suppl. Table 3)³¹. Four categories of target audiences were distinguished: healthcare providers, patients, organization, and system. Besides our primary outcome (inappropriate PPI prescription or use), secondary outcomes of interest included pharmaceutical effects (symptoms of acid reflux; ulcer and upper gastrointestinal bleeding), adverse pharmaceutical effects (diarrhea or obstipation, abdominal pain, Clostridium difficile infections, hospital-acquired pneumonia, electrolyte disturbances), and healthcare use (e.g., length of stay (LOS), ICU or hospital admission, emergency department visit, alternative medication use). Two authors (PH, CO or JJKvD) independently assessed the risk of bias (RoB) using suggested criteria for EPOC reviews³¹.

Descriptive characteristics of studies were summarized narratively. To quantify the effectiveness of de-implementation strategies, we calculated the proportion of inappropriate PPI prescriptions or inappropriate PPI use for each study arm. PPI inappropriateness was defined as no indication for SUP in patients without risk factors, but definitions of indication for SUP differed between included studies.

We distinguished two groups of patients: one consisted of patients who were using PPI medication prior to hospitalization that was continued during hospitalization and the second group were patients who started PPI medication during hospitalization. To translate this distinction in patient groups to different categories of PPI medication application, we defined PPI use as all PPI prescriptions during hospitalization (i.e., continued or new PPI prescriptions). We defined PPI prescriptions as PPI prescriptions that started during hospitalization.

Results for the primary outcome are presented in forest plots that were generated with Review Manager (RevMan5.3) software. Meta-analysis revealed considerable heterogeneity between studies (based on visual inspection of forest plot and I² > 50%). Therefore, a pooled effect estimate was not presented. The results for the secondary outcomes of interest were described narratively. Funnel plot analysis, to address the issue of publication bias, was not performed, because too few studies for a similar outcome were retrieved.

We retrieved 2264 studies from database searches and one study through experts in the field. After duplicate removal, 1863 studies remained, of which 75 were selected for full-text review (Fig. 1). Main reasons for excluding studies were non-relevant population or setting and when authors indicated that de-implementation was not directed at reduction of PPI for SUP. Finally, ten studies were included in the analyses.

The studies were published from 1998 to 2018 and all, but one, had a before-after study design without a parallel control group. They were mainly conducted in high- and middle-income countries and as single-center studies (Table 1).

The summary of the risk of bias (RoB) assessment is presented in Fig. 2, and substantiation for RoB judgement is presented in Supplementary Table 6.

The study of Del Giorno was the only trial which included a control group and therefore four additional RoB items were assessed³⁷. As this study was a non-randomized trial, we scored RoB for random sequence generation and allocation concealment at “high risk.” The interventions were conducted at the internal medicine departments of five hospitals, with the surgery departments in the same hospitals serving as control. As a consequence, the participant population at baseline were dissimilar, but this was not further addressed by the authors. Also, the interventions conducted at one department could have potentially contaminated the results at the control department within the same hospital. Therefore, both these items were scored “unclear risk” of bias.

Most studies scored “unclear risk” of bias for blinding of outcome assessment as they did not report how outcome assessment was performed^{32‐35,37‐40}. Only two studies scored “low risk” of bias as outcome assessment was automated or standardized^13,36. Other studies were scored “unclear” risk of bias for blinding of outcome assessment. Six studies scored “low risk” for similar baseline characteristics as participant population baseline characteristics were comparable^{13,33,35,36,38,39}. Two studies scored “unclear risk” as no information about the baseline characteristics of participants was reported^32,40. Two studies score “high risk” as large differences between the two participant groups were observed^34,37.

In agreement with other systematic reviews addressing effectiveness of interventions to change healthcare, we observed large heterogeneity between studies and outcomes, in combination with low study design quality^42‐44. Also, the number of interventions in the included studies was limited mostly to educational interventions directed at providers. None of the studies targeted interventions at patients, even though patient participation in the reduction of inappropriate PPI prescriptions has been shown effective^25,45‐47. Combinations of educational interventions, a reflective practice and supportive environment, are required for high-value healthcare^27,48. None of the studies included e-health solutions, while computer-assisted decision aids and web-based information for both patients and medical professionals could be instrumental in de-implementing PPI⁴⁹.

Additionally, none of the studies provided sufficient intervention details to allow knowledge transfer of effective intervention strategies. De-implementation, giving up a clinical behavior, has repeatedly been shown to be psychologically more challenging than adopting a new behavior^22,50. Therefore, it should be acknowledged that de-implementation strategies to change the prescription behavior of healthcare providers require thorough analysis of the local clinical setting and identification of barriers and facilitators^22,51. This prior analysis was minimally addressed and reported in the included studies.

Future de-implementation studies should take all the contextual factors into account when designing a strategy, and standardization in data collection and reported outcomes would further improve knowledge transferability. Subsequently, for impact evaluation of the de-implementation strategies, authors need to collect and report all essential information needed to interpret and apply their results into practice. This includes knowledge on barriers and facilitators, de-implementation strategy details, sustainability of observed effects, and insight into unintended consequences of the de-implementation strategy. There are several relevant reporting guidelines that can assist authors^52‐54. Finally, the field of de-implementation science in healthcare would benefit from high-quality studies with more rigorous study designs (i.e., cluster RCTs, interrupted time series studies, etc.) that adhere to international reporting guidelines and recommendations on implementation strategy classification. Altogether, we recommend that future de-implementation studies of clinical pathways involve a multidisciplinary approach in which clinicians collaborate with patient representatives and facilitatory services in their organization (i.e., communication, finance, education departments) to ensure that all aspects of an effective implementation are addressed and are supported throughout the organization.

The strength of this systematic review is the focus on the reduction of inappropriate PPI prescriptions/use in hospitalized patients in non-ICU settings. This is not often specifically addressed, and is justified by the increased popularity of PPI use in non-ICU-hospitalized patients^11‐14 and several reports on adverse effects attributed to long-term PPI use^3,9,17‐19.

A limitation of this review is that terminology to describe de-implementation strategies varies widely. In an effort to retrieving all relevant studies, we applied an extensive search strategy. Nevertheless, we cannot exclude the possibility that we missed relevant studies.