Deprescribing versus continuation of chronic proton pump inhibitor use in adults

Gastroesophageal reflux disease

GERD affects millions of adults globally. Approximately five million Canadians experience heartburn or regurgitation (or both) at least once weekly, where heartburn is the most common symptom reported (17%) followed by regurgitation (11%) (CDHF 2015; Tougas 1999). Similar figures also extend globally. One systematic review evaluating the epidemiology of GERD found that 10% to 20% of adults in Western countries are also affected by GERD (Dent 2005).

Many GERD publications and guidelines exist though a uniform definition of GERD among them is lacking. The Canadian Consensus on the Management of Gastroesophageal Reflux Disease in Adults (Canadian Consensus) defines GERD as the "reflux of gastric contents into the esophagus causing (a) symptoms sufficient to reduce quality of life, and/or (b) esophageal injury". The hallmark symptoms are heartburn and acid regurgitation. The Canadian Consensus also classifies GERD as either 'mild' or 'moderate and severe' disease (Armstrong 2005). The American College of Gastroenterology's (ACG) Guidelines for the Diagnosis and Management of Gastroesophageal Reflux Disease recognizes that GERD is "defined by consensus and as such is a disease comprising symptoms, end‐organ effects and complications related to the reflux of gastric contents into the esophagus, oral cavity and/or the lung" (Katz 2013). The ACG categorizes GERD based on endoscopic findings: nonerosive GERD (NERD) is symptomatic GERD with negative endoscopy, while GERD with erosive findings on endoscopy is referred to as erosive reflux disease or erosive esophagitis (EE) (Katz 2013). In 2006, the Global Consensus Group published the Montreal Definition and Classification of GERD to determine a globally acceptable definition that would be useful for patients, practitioners, and health agencies. This group defined GERD as, "a condition which develops when the reflux of stomach contents cause troublesome symptoms and/or complications", such as heartburn, regurgitation, and esophageal injury (i.e. reflux esophagitis) (Vakil 2006).

Reflux esophagitis

PPIs are also commonly used on a long‐term basis for esophagitis due to chronic GERD. The severity of EE detected on endoscopy is classified by the Los Angeles (LA) Grading criteria as grades A to D, defined as follows: grade A ‐ at least one mucosal break of 5 mm or less, not extending between the tops of two mucosal folds; grade B ‐ at least one mucosal break with a length greater than 5 mm not extending between the tops of two mucosal folds; grade C ‐ at least one mucosal break involving less than 75% of the esophageal circumference, but extending between the tops of at least two mucosal folds; grade D ‐ at least one mucosal break involving 75% or more of the esophageal circumference (Lundell 1999). LA Grade C and D esophagitis are considered severe stages of esophagitis, have lower rates of healing with PPIs, and are more prone to relapse in the absence of maintenance therapy (Katz 2013). The Canadian Dyspepsia Working Group suggest that healing and symptom relief of EE is superior with PPIs as compared to histamine H2 receptor antagonists (H₂RAs), and that standard doses of PPIs are superior to half or low doses of PPIs for this indication (Veldhuyzen van Zanten 2005). Barrett's esophagus is another long‐term sequelae of GERD and esophagitis that is believed to be developed from the change of normal esophageal epithelial cells to specialized intestinal glandular epithelial cells. This metaplasia, detectable via endoscopy, can progress to dysplasia, and finally adenocarcinoma (Armstrong 2005). The ACG guidelines recommend ongoing maintenance PPI therapy for people with EE and Barrett's esophagus (Katz 2013). Since Barrett's esophagus is believed to result from chronic acid reflux, the Canadian guidelines suggest that regardless of symptoms, it should be treated with lifelong PPIs of at least standard dose, and managed by specialists (Veldhuyzen van Zanten 2005).

Peptic ulcer disease

Gastric and duodenal ulcers (bleeding or nonbleeding) are often referred to under the rubric of PUD (ACPG 2000; Katz 2013; Talley 2005). There are many risk factors and possible causes for PUD, the most common being infection of the gastric mucosa with H. pylori; however, other causes include NSAID‐ or drug‐induced PUD, and hypersecretory states. The standard for H. pylori‐positive PUD is treatment with triple therapy (two antibiotics and one PPI) to eradicate the bacteria. This is typically followed by four to eight weeks of PPI therapy for gastric ulcers, but does not require subsequent PPI therapy in the setting of uncomplicated duodenal ulcers. If eradication is achieved, the rate of rebleeding is very low (1.3% over 11 to 53 months) (Gisbert 2004; Laine 2012; Malfertheiner 2009).

NSAIDs, and to a lesser extent newer selective cyclooxygenase‐2 (COX2) inhibitors, are known to cause dyspeptic symptoms and increase the risk of PUD through the inhibition of cyclooxygenase‐1 (COX1) enzyme in the gastric mucosa. There is support for the concomitant use of gastroprotective agents (PPIs, misoprostol, or H₂RAs) as primary or secondary prophylaxis of peptic ulcers, especially in the presence of risk factors. These include age 65 years or over; history of bleeding PUD; and use of corticosteroids, anticoagulants, or antiplatelet agents (Veldhuyzen van Zanten 2005). If people are to be maintained on long‐term NSAID therapy, it is advised to test and treat for H. pylori due to the additive risk (Veldhuyzen van Zanten 2005).

Types of studies

We included RCTs and quasi‐randomized trials. We applied no limitation on language or abstracts.

Types of participants

Participants were adults (aged 18 years or greater) using a PPI chronically for one month (28 days) or greater. Participants also required an indication for PPI use: GERD (nonerosive reflux disease or GERD LA Grade A to D), functional dyspepsia, or undiagnosed dyspepsia. Participants were also considered eligible if they had a history PUD (bleeding or nonbleeding, or both), H. pylori‐positive or ‐negative diagnoses, recurrent peptic esophageal stricture or Barrett's esophagitis. Participants using a PPI as SUP or for gastroprotection in the setting of low‐dose acetylsalicylic acid (ASA) (less than 325 mg/day), anti‐inflammatory doses of NSAIDs, and steroids were eligible. We excluded trials if participants were taking high‐dose NSAIDs who require chronic PPI use for gastroprotection; however, those with occasional NSAID use were not excluded.

Types of interventions

Included studies compared at least one of the deprescribing modalities with a control consisting of no change in continuous daily PPI use. Deprescribing was defined as one or more of the following interventions:

1. Stopping PPI therapy: either via abrupt discontinuation or a tapering regimen.

2. Step down: following abrupt discontinuation or tapering of PPI, an H₂RA was prescribed (any H₂RA at any approved dose and dosing interval per drug monograph).

3. Reduction in PPI, which included the following subcategories:

   1. intermittent PPI use as previously defined by the Canadian Consensus (Armstrong 2005);

   2. on‐demand PPI use as previously defined by the Canadian Consensus (Armstrong 2005);

   3. lower dose: continuous daily PPI therapy at a lower dose compared to a control arm within the trial.

4. Treatment of H. pylori (using any approved triple or quadruple therapy) followed by four to eight weeks of PPI maintenance therapy for bleeding or nonbleeding PUD and then PPI discontinuation.

We excluded studies if treatment arms consisted of concomitant interventions such as rescue therapy (e.g. with antacids), lifestyle, diet modifications, or a combination of these. This did not affect inclusion if these additional interventions were handled similarly in all treatment arms.

Types of outcome measures

Electronic searches

We searched the following databases:

1. Cochrane Central Register of Controlled Trials (CENTRAL) (2016, Issue 10) (Appendix 3);

2. MEDLINE (1946 to 15 November 2016) (Appendix 4);

3. Embase (1980 to 15 November 2016) (Appendix 5);

4. ClinicalTrials.gov;

5. World Health Organization International Clinical Trials Registry Platform (WHO ICTRP).

As deprescribing is a new concept in the literature, we used a broad range of key words. Search terms for PPIs included both the generic and common brand names (Donnellan 2004; Van Herwaarden 2014). A librarian from Cochrane reviewed the search strategy and a research librarian completed the search strategy (see Appendix 3; Appendix 4; Appendix 5). Review authors will conduct future literature searches.

Searching other resources

We handsearched reference lists of all primary studies and review articles.

Selection of studies

Three review authors (FJR, TB, and WT) independently reviewed search results, assessed study eligibility and trial quality, and extracted data. We resolved disagreements by discussion with another review author (CRF). The review authors retrieved and independently screened full‐text study reports and publications and identified studies for inclusion and identified/recorded reasons for exclusion of the ineligible studies. We made every attempt to exclude duplicates. We constructed a PRISMA flow diagram (Figure 1) and a Characteristics of excluded studies table based on the information collected during this process. We used a standardized eligibility checklist.

Data extraction and management

We used a standard data collection form for study characteristics and outcome data. Prior to data collection, two review authors independently completed a pilot eligibility and data extraction exercise to ensure criteria were consistently applied (Higgins 2011). Three review authors (FJR, TB, and WT) extracted the following study characteristics from included studies:

1. methods: study design, total duration study and run in, number of study centers and location, study setting, withdrawals, date of study;

2. participants: number, mean age, age range, gender, severity of condition, types of condition, inclusion criteria, and exclusion criteria;

3. interventions: intervention, comparison, concomitant medications;

4. outcomes: primary and secondary outcomes, time points reported;

5. other: author contact information, funding for trial, notable conflicts of interest of trial authors.

We resolved disagreements by discussion with a third review author (CRF). One review author (FJR) transferred the data into Review Manager 5 (RevMan 2014). A second author (TB) reviewed data entry for accuracy.

Assessment of risk of bias in included studies

We assessed the quality and strength of the evidence using GRADE and study bias using Cochrane's tool for assessing risk of bias (Higgins 2011).

Three review authors (FJR, TB, and WT) independently assessed risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We resolved disagreements by discussion with a third review author (CRF). We ranked each potential source of bias as high, low, or unclear and provided a quote from the study to justify the bias grade (refer to Characteristics of included studies table for risk of bias assessments) using the following domains:

1. random sequence generation;

2. allocation concealment;

3. blinding of participants and personnel;

4. blinding of outcome assessment;

5. incomplete outcome data;

6. selective outcome reporting;

7. other bias.

Measures of treatment effect

We reported results as risk ratios (RR) where possible with 95% confidence intervals (CI). We pooled dichotomous data using a random‐effects model (Dersimonian 1986), and continuous outcomes using mean differences (MD) when outcomes were measured on the same scale.

Unit of analysis issues

The unit of analysis was the individual participant undergoing PPI treatment. We conducted a scoping exercise prior to commencing the review and did not identify any cluster‐randomized trials for this comparison. As expected, we found no cluster‐randomized trials for this comparison after conducting comprehensive literature searching during the review.

Dealing with missing data

During the data extraction process, we attempted to contact the primary author for clarification regarding missing or unclear data. No data were imputed.

Assessment of heterogeneity

We used the Chi² statistic and I² statistic to measure heterogeneity among the trials in each analysis. If we identified substantial heterogeneity (Chi² statistic with a P value less than 0.15 or I² statistic greater than 25%) via either statistic, we planned to explore this through prespecified subgroup analysis. Due to the paucity of included trials, we did not perform predefined subgroup analyses to investigate causes of heterogeneity.

Assessment of reporting biases

We used the methods outlined in the Cochrane Handbook for Systematic Reviews of Interventions for assessment of bias (Higgins 2011). We attempted to contact study authors to request missing outcome data. Refer to Characteristics of included studies table for risk of bias assessment.

Data synthesis

We used Review Manager 5 for analysis (RevMan 2014). We calculated 95% CIs for the treatment effect and used the Mantel‐Haenszel method for dichotomous data and the inverse variance method for continuous data. We used a random‐effects models for the analysis (DerSimonian 1986).

Subgroup analysis and investigation of heterogeneity

We planned to carry out the following subgroup analyses.

1. Deprescribing interventions (stopping PPI, lower‐dose PPI compared to control arm, on‐demand PPI, intermittent PPI, stopping PPI and changing to H₂RA, treatment of H. pylori with four to eight weeks of PPI therapy).

2. Diagnosis (GERD, esophagitis, functional dyspepsia, PUD, H. pylori infection).

3. Trial setting (primary versus secondary care).

4. Age (young adults versus older adults).

We planned to assess the robustness of the conclusions based on the risk of bias assessments for blinding and allocation. We were unable to conduct subgroup analysis given the limited data.

Sensitivity analysis

We planned to perform sensitivity analysis defined a priori to assess the robustness of our conclusions by: outcome reporting and participant populations, and to assess any effect of risk of bias on our estimates. Data were too limited to perform sensitivity analysis.

Results of the search

We identified 2357 references; 2317 through electronic searches of CENTRAL (Wiley), MEDLINE (OvidSP), and Embase (OvidSP), and 40 through other resources. We excluded 2268 clearly irrelevant references by reading the abstracts and retrieved 89 references for further detailed assessment. We excluded 83 references for the reasons listed in the Characteristics of excluded studies table. Six references fulfilled the inclusion criteria (Characteristics of included studies table). The results of the search are presented in Figure 1.

Included studies

We included six RCTs with 1753 participants (Bayerdörffer 2016; Bour 2005; Janssen 2005; Morgan 2007; Pilotto 2003; Van der Velden 2010). The Characteristics of included studies table outlines details (e.g. sample size, participant characteristics, interventions, etc.) of all studies that met inclusion criteria.

Excluded studies

We excluded 83 studies. Three studies had the wrong study design (i.e. not RCTs) (Bautista 2006; Inadomi 2001; Ponce 2004). Six studies included chronic NSAID users (Annibale 1998; Bate 1997; Dent 1994; Gough 1996; Hatlebakk 1997a; Hatlebakk 1997b). In one study, participants did not have an indication for PPI use (i.e. healthy volunteers) (Niklasson 2010). Seventy‐three studies did not have a control arm (i.e. no continuous/no change in chronic PPI use) (Archambault 1988; Bardhan 1990; Bardhan 1995; Bate 1989; Bate 1990; Bate 1995; Bayerdörffer 1993; Bigard 2005; Björnsson 2006; Bytzer 2004; Carlsson 1998; Chen 1993; Dabholkar 2011; Deboever 2003; Dent 1989; DeVault 2006; Earnest 1998; Eggleston 2009; Escourrou 1999; Farley 2000; Fass 2011; Fass 2012; Fennerty 2009; Goh 1995; Goh 2007; Havelund 1988; Hoshino 1995; Howden 2001; Howden 2009; Hsu 2015; Inadomi 2003; James 1994; Johnson 2001; Johnson 2010; Jones 1997; Juul‐Hansen 2008; Kaplan‐Machlis 2000; Kaspari 2005; Klinkenberg‐Knol 1987; Kovacs 2010; Labenz 2005; Lauristen 2003; Lind 1999; Lundell 1991; Metz 2003; Nagahara 2014; Norman Hansen 2005; Peura 2009; Peura 2016; Reimer 2010; Richter 2000; Richter 2004; Robinson 1995; Sandmark 1988; Scholten 2005; Sjöstedt 2005; Sontag 1997; Sun 2015; Talley 2001; Talley 2002a; Talley 2002b; Tan 2011; Tsai 2004; Vakil 2001; Valenzuela 1991; Van Rensburg 1996; Van Rensburg 2008; Vantrappen 1988; Van Zyl 2000; Venables 1997; Vigneri 1995; Wulff 1989; Zwisler 2015). Refer to Characteristics of excluded studies table for deprescribing trials that did not meet all inclusion criteria.

Allocation

Contribution of allocation bias was difficult to determine, as four trials did not describe how they performed allocation (Bayerdörffer 2016; Bour 2005; Morgan 2007; Pilotto 2003). The remaining two trials used sound methods: Janssen 2005 used sealed boxes containing study medication, and participants were allocated according to the numbering in ascending order, while Van der Velden 2010 used a computer‐generated list, enlisted an external party to generate sealed identical containers which were sent to the general practitioners, and participants were allocated the next available number of the set on entry into the trial.

Blinding

Of the six trials included, two trials were double blinded (Pilotto 2003; Van der Velden 2010), although the authors of van der Velden and colleagues noted that the data were analyzed unblinded. Van der Velden 2010 blinded participants by providing identical containers and by providing the control arm with daily dose of PPI plus placebo rescue medication to use as needed and providing the inverse to the intervention group (placebo daily plus PPI when needed). Pilotto 2003 did not comment how blinding was maintained and whether pills looked identical in both arms. The remaining four trials had an open‐label design (Bayerdörffer 2016; Bour 2005; Morgan 2007; Pilotto 2003). Many of the outcomes related to symptom relief, despite use of scales, relied on subjective reporting, and thus were subject to bias. For example, Bour 2005 evaluated participant‐rated symptom relief (primary outcome) using a Likert scale and it was unclear if investigators saw participant satisfaction measures before they assessed satisfaction. Janssen 2005 also measured subjective outcomes (symptom control and satisfaction/unwillingness to continue), creating high risk of bias. However, the objective measure of pill count was unbiased (measurement was not likely to be influenced by lack of blinding). Participant‐rated effect of treatment and satisfaction in the trial by Morgan 2007 were also inherently subject to bias as neither participants nor assessors were blinded.

Incomplete outcome data

Reporting of outcome data revealed a relatively high risk of bias as the majority of trials had participants lost to follow‐up or did not report on all predefined outcomes, or both.

Although the study completion rate was high for Bour 2005 (greater than 80%), the reason for dropout was unclear in all instances and there was a statistically significant higher number of participants who withdrew from the trial for reasons unknown, defined as 'other' in the on‐demand treatment group. Only 67/81 participants were analyzed for pill usage in the continuous treatment group, and only 55/71 for the on‐demand treatment group, with no explanation provided.

Janssen 2005 had a relatively higher dropout rate (22%) but maintained sufficient participation to achieve their power calculation. Although they did not report if characteristics differed between those who stayed in study, versus those who were excluded, the unwillingness to continue as cause of dropout was similar between groups (6% versus 8%) for the long‐term phase.

Morgan 2007 and Pilotto 2003 reported overall dropout rates making it difficult to assess if reasons for withdrawal from the study were similar between treatment and control groups. Endpoints in the Pilotto 2003 trial were not clearly defined, also making it difficult to assess completion of outcome data reporting. Van der Velden 2010 clearly reported dropouts and reasons, however did not comment on whether differences were significant between groups (33% dropout in the on‐demand arm, which was greater than the 20% anticipated). The authors of this study stated that 40 participants were required in each arm (Van der Velden 2010 page 44), though a power calculation was not described. Bour 2005 arbitrarily set a fixed sample size of 200, with no discussion of power or adequate sample size. In total, three of the five trials did not report a power calculation, therefore it is unclear how significant these dropout rates were (Bour 2005; Pilotto 2003; Van der Velden 2010). Bayerdörffer 2016 presented a comprehensive review in their figure two, regarding dropout rates including reasons within in each study arm.

Selective reporting

Van der Velden 2010 appeared to have significant reporting bias for several outcomes because of incomplete reporting of results. Number of rescue pills used weekly was not reported as absolute numbers, but rather participants were divided into groups based on usage that were not defined a‐priori (less than two pills, two to five pills, and six or seven pills weekly). The authors did not predefine how they would assess participant satisfaction. Predefined secondary outcomes of symptom control and quality of life were not assessed separately and rather inferred from results of two surveys (Quality of Life in Reflux and Dyspepsia (QolRad), and SF‐36 Health Survey). The use of a modified per protocol population lead to results that were more difficult to interpret, and additionally, standard deviations (SD) were not reported for all their predefined outcomes.

There was significant bias in the Pilotto 2003 trial, as outcomes were not predefined in their methods and event rates were not reported. Despite methods suggesting that investigators classified symptom severity as mild, moderate, or severe, symptom assessment was only reported as percentage of participants with symptoms. Baseline participant demographics/characteristics were presented as overall numbers rather than for each group, so it is unknown if groups were balanced at randomization and if risk factors in one arm outweighed the other.

Three trials reported all predefined outcomes (Bour 2005; Janssen 2005; Morgan 2007). However, Morgan 2007 combined satisfaction ratings in their results, which was not predefined (e.g. reported results as a combination of 'good' or 'very good' ratings and 'satisfied' or 'very satisfied' ratings). This trial did not report SDs for pill usage. We were unable to use participant satisfaction scale in the meta‐analysis as SDs were not reported for this scale.

Bour 2005 reported additional analysis, but acknowledged the change to their protocol, to include statistical analysis rather than just descriptive data (e.g. subanalysis of relapse rates in different levels of on‐demand use). Janssen 2005 presented comprehensive tables and figures and included means, SDs, event numbers, absolute differences calculated, P values, and CIs. Janssen 2005 reported additional data in the results section although they were not predefined in their study design (e.g. tolerability, number of pills used). Bayerdörffer 2016 addressed all outcomes and reported data for both primary and secondary outcomes.

Other potential sources of bias

The use of perceived mean daily symptom load (MDSL) score by Janssen 2005 could contribute to bias, as this scale was defined by the authors, was not validated, and psychometric properties were unknown. Evaluating bias for the trial published by Pilotto 2003 was challenging as insufficient data were provided to make an accurate assessment.

Five of the six trials listed drug manufacturers as external funding sources: Bour 2005 by Janssen‐Cilag; Morgan 2007 by Janssen‐Ortho; Pilotto 2003 by Pharmacia, Milano, Italy; Van der Velden 2010 by Nycomed BV; and Bayerdörffer 2016 by AstraZeneca. Their role was unclear/unreported in four trials (Bayerdörffer 2016; Bour 2005; Morgan 2007; Pilotto 2003), while Van der Velden 2010 explicitly stated drug manufacturer was not involved in any aspect of participant selection, data collection, analysis, interpretation, writing and submission of manuscript. Janssen 2005 did not list sources of funding.

Primary outcomes

Secondary outcomes

On‐demand proton pump inhibitor deprescribing

Five trials evaluated deprescribing via an on‐demand approach (Bayerdörffer 2016; Bour 2005; Janssen 2005; Morgan 2007; Van der Velden 2010). Participants were aged 48 to 57 years and either had NERD or a milder form of esophagitis (LA grade A and B or Savary‐Miller grade 1 and 2).

For the primary outcome of change in upper GI symptoms, deprescribing demonstrated a return of symptoms (e.g. heartburn, regurgitation) for a proportion of people. These results were expected as participants in the on‐demand group were instructed to initiate on‐demand PPI use upon sensation of symptoms. In the combined cohort of 1653 participants randomized to on‐demand therapy, approximately 16.3% had either treatment failure or inadequate symptom relief (compared to around 9.2% on continuous PPI therapy, P less than 0.0001). This indicates that while the risk of recurrent symptoms with on‐demand therapy is greater than continuous daily PPI use, the majority of people will still tolerate the intervention. Further studies are required to identify which subpopulations comprise this group.

We pooled data for drug burden/PPI use from three on‐demand trials (n = 1152) (Bayerdörffer 2016; Bour 2005; Janssen 2005). The pooled data favored deprescribing (MD 3.79 pills weekly) and was statistically significant with a P value of less than 0.00001, although heterogeneity was high. The pooled results of this review were consistent with what is currently published in primary literature. Trials evaluating on‐demand PPI use have reported a mean pill consumption of 0.3 pills per day (Bigard 2005; Kaspari 2005; Pace 2008; Ponce 2004; Tsai 2004).

Participant satisfaction was determined by evaluating unwillingness to continue treatment and inadequate symptom relief. Data for this outcome favored chronic PPI use. Interestingly, when assessed individually, not all trials showed a significant difference in participant satisfaction ratings between groups. Janssen 2005, Bour 2005, and Bayerdörffer 2016 had CI value crossing one; however, the overall treatment effect showed significance in favor of deprescribing (RR 1.82, 95% CI 1.26 to 2.65, P < 0.002). One on‐demand PPI trial provided data for unwillingness to continue secondary to inadequate relief, which was interpreted for this review as a marker for participant dissatisfaction (and thus the inverse, as participant satisfaction) (Janssen 2005). The open‐label nature of Janssen 2005 likely influenced the results as participants were aware that they were in control of PPI use and likely favored this method rather than daily PPI use. In past literature, participant satisfaction has been reported to be in favor of chronic PPI use, though the difference from on‐demand therapy is small (Pace 2005). Our results parallel data that have been previously published. We believe that on‐demand therapy could be a viable option for people with NERD or mild‐to‐moderate GERD, without compromising satisfaction of symptom control. Further studies using more targeted measurements of participant satisfaction are needed to confirm or refute these results.

One on‐demand trial had data for ADWE (Bayerdörffer 2016). Participants in this trial were assessed endoscopically before and after randomization. Those who did not present with esophagitis were included and randomized. Investigators performed an endoscopy at the end of the trial to assess the development of esophagitis. It found that 5% (n = 15) of participants using on‐demand PPI developed mild esophagitis (14 with LA grade A, 1 with grade B).

Abrupt discontinuation deprescribing

We identified one trial of abrupt discontinuation (Pilotto 2003). Participants in this trial experienced mild‐to‐moderate esophagitis (Grade 1 or 2), though they were much older than participants taking on‐demand therapy (mean age 73 years). This trial demonstrated a statistically significant increase in the primary outcome of lack of symptom control. Participants experienced return of symptoms, which is expected, as transient and reversible rebound effects have been previously described for both healthy people and people with GI disorders upon abrupt PPI discontinuation (Reimer 2009b). Once again, symptom relapse was not observed in the entire cohort (67.9% in abrupt discontinuation versus 22.4% in the continuous arm), suggesting almost a third of the studied population tolerated abrupt discontinuation.

The Pilotto 2003 trial also addressed the secondary outcome of ADWEs. The trial evaluated changes in endoscopic findings (i.e. healing rates), which were interpreted within this review as relapse rates by assuming the inverse. The results from this trial indicated an increased risk of ADWE with deprescribing (i.e. abrupt discontinuation) that was statistically significant in people with esophagitis grades 1 or 2. Despite these findings, it is difficult to make any conclusions as the results are based on a single study of older adults with mild‐to‐moderate disease. Furthermore, the trial had a small sample size with no reported power or predefined primary endpoints. Current recommendations also suggest avoiding abrupt discontinuation, as relapse acid hypersecretion has been demonstrated even in healthy participants (Reimer 2009b). Other potential ADWEs such as upper GI bleeding or progression of esophagitis to Barrett's esophagus, among others, were not addressed by any of the trials included.