When PPIs Fail

In this issue of the Journal, Sharma et al report on a large clinical series evaluating persistent symptoms in acid suppressed patients undergoing impedance pH monitoring [1]. In doing so, they highlight two key challenges that have come to dominate the post-PPI world of gastroenterology: 1) how to manage patients when PPIs “fail” and 2) how to conduct and interpret impedance-pH monitoring studies. They conclude that: 1) patients on PPIs still experience symptoms, 2) 24-hour ambulatory impedance-pH monitoring identifies similar reflux-symptom associations regardless of refluxate acidity, and 3) impedance-pH monitoring clarifies symptoms unrelated to reflux. Each of these is an important observation that warrants further exploration.

Gastroenterologists were spoiled by PPIs. Heralded as breakthrough drugs, they were viewed as omnipotent. At first PPIs were used to treat ulcers and esophagitis, domains in which their efficacy was easily proven. And then the PPIs were brought to bear on gastroesophageal reflux disease (GERD) and all that that entailed: non-erosive reflux disease, laryngitis, throat clearing, cough, sinusitis, otitis, asthma, chest pain, you name it. In the rare instance that taking one a day didn’t work, taking two a day did. It was just a matter of dose and duration of therapy. So much for science. Eventually, however, this line of thinking ran its course with the realization that more was not always better. The concept of a ceiling effect emerged. Maybe there was an upper limit to what could be achieved by more and more intense acid inhibition. Maybe a symptom was unresponsive to therapy because it was unrelated to reflux. Maybe we need to devise ways to test whether or not symptoms are related to reflux.

It is within this framework that we should consider the utility of ambulatory multichannel intraluminal impedance-pH (MII-pH) monitoring. MII-pH monitoring is the most recent adaptation of an ingenious technology described by Silny in 1991 wherein impedance monitoring could be used to detect the flow of liquids or gas through hollow viscera [2]. The added value with MII-pH is that by adding a pH electrode to the multichannel impedance catheter, refluxate can be characterized not only as liquid or gaseous but also by its acidity. Another advantage of MII-pH recording compared to conventional pH monitoring is of an increased sampling rate: 40 per second compared to as low as once every 6 seconds. Finally, by incorporating multiple impedance rings along the catheter the direction of flow can be determined, minimizing the confounding effect of swallowed liquids and, in the case of reflux, determining the length of esophagus exposed to the refluxate. Thus, from a technological viewpoint, MII-pH recorders are the best instruments available for the detection of all reflux events and, secondarily, to characterize individual reflux events by acidity, gas/liquid composition, and proximal extent reached within the esophagus.

A reflux event, be it of pH <4 or pH ≥4, can trigger two sensory signals from the esophagus; one related to its acidity, so called chemo-stimulation and one related to its volume, often referred to as mechano-stimulation. Heartburn caused by acid reflux and mediated by chemo-stimulation is very responsive to a PPI; heartburn caused by esophageal distention and mediated by mechano-stimulation is not. Both exist. Fortunately, evident by the billions of dollars, euros, and pounds spent on PPIs, most heartburn is caused by acid reflux. However, can the same statement be made about regurgitation, indigestion, chest pain, cough, throat clearing, nausea and “other”? Only a rhetorical question; the answer is an increasingly emphatic, No! Why then would one want to lump these diverse symptoms together into a single analysis? It gets back to the point that gastroenterologists were spoiled by the perceived omnipotence of PPIs coupled with a snowballing enthusiasm for attributing all that ails the human race to reflux disease. Under such a scenerio, of course there will be “PPI failures”. However, if we regain rationality for a moment and consider each of these symptoms in unique pathophysiological terms, it becomes apparent that heartburn is unique. Heartburn exhibits a much stronger relationship to acid reflux (Figures 3 & 4 in Sharma et al [1]), particularly acid reflux that extends into the proximal esophagus and is cleared slowly [3, 4]. Regurgitation, which becomes the dominant reflux symptom in the setting of acid inhibition [4, 5], is much less dependent on acidity and consequently much more likely to be “refractory” to PPI therapy. Chest pain lies somewhere in the middle in terms of mechano-stimulation vs chemo-stimulation with the added complexity that it is not always of esophageal origin. As for cough, throat clearing, nausea and “other”, is it even reasonable to look for event-by-event reflux correlation? I wouldn’t have thought so, but yet that is what Sharma et al [1] suggest based on their retrospective review of symptom index (SI) values in 200 clinical MII-pH studies. What might account for this?

Although it is tempting to gloss over the details, I believe that there is a serious methodological limitation to the SI as used by Sharma et al in this paper. The authors have defined the SI to be positive if the number of symptoms occurring within 5 minutes of a reflux event divided by the total number of symptom events is ≥50%. The original definition of a positive SI stipulated 2-minute windows [6]. Let’s consider the impact of this small numerical change with some sample numbers. In this paper there were 415 reported symptoms in 200 patients so an average of two symptom occurrences per patient. Although not reported in this manuscript, other analogous studies determine that in a patient population such as this, MII-pH recording will detect an average of about 60 reflux events per 24 hours, mainly weakly acidic if the patient is taking a PPI [3, 4]. Finally, patients do sleep a little during these studies, so assume that they were awake, not eating, and able to report symptoms for 20 of the 24 hours of the study. So, given these parameters, what is the probability of finding a positive SI purely on the basis of chance? Before your eyes glaze over, I’ve made the numbers really easy. With 60 reflux events dispersed over 20 hours of evaluable recording, there would be 5 hours during which a randomly activated event marker would be scored as reflux associated (5 min per event). That means that if a patient reported only 1 symptom during the study they would have a positive SI (SI=100%) 25% of the time (5 hrs/20 hrs) and a negative SI (SI=0%) 75% of the time purely by chance. Now in this study, patients reported an average of 2 symptoms meaning that they could potentially have an SI of 0%, 50%, or 100%. Of the three possibilities, only the 0% is a negative SI, the others are both positive. Thus, the probability of getting a negative SI (0%) was 0.75 × 0.75 or 0.56. Hence, the probability of getting a positive SI purely by chance is 44%. The authors report that 42% of patients had a positive SI. These numbers are suspiciously similar.

Admittedly, I cooked the numbers a bit for the sake of computational simplicity, but the point is that a positive SI does not prove a significant temporal association between symptoms and reflux events because it does not consider the possibility of chance association. This was true in the analysis of pH monitoring studies using a 2-minute symptom association window and it is even more true with combined MII-pH studies using a 5-minute window for symptom association. After all, with MII-pH studies there are going to be substantially more reflux events detected and now each one now has 2.5 times the vulnerability to chance-driven symptom association. Before applying an SI analysis, one has to first test the hypothesis that reflux events and symptoms are temporally associated by chance. Only after that hypothesis has been rejected is it valid to apply an SI analysis. The likelihood of chance association between reflux events and symptoms is most commonly tested using the symptom association probability (SAP) [7]. The SAP calculation amounts to doing Fisher’s exact test to determine the probability that the observed association between reflux events and symptoms is attributable to chance. If the SAP is 95% the probability of the association being driven by chance is 5% (p=0.05); if the SAP is 99% the probability of the association being driven by chance is 1% (p=0.01). It is customary to accept a p<0.05 as adequate so only after establishing that the SAP is greater than 95% is it valid to then calculate an SI as an indication of the strength of association between symptoms and reflux. Given the sparse number of symptoms reported in the Sharma et al study, it is unlikely that many of these patients would have an SAP greater than 95%.

The third conclusion of Sharma et al is that doing combined impedance-pH monitoring clarifies symptoms unrelated to reflux. This is an increasingly important concept as the diagnostic value of endoscopy dwindles in the post-PPI world. In the common setting of a healed mucosa, reflux-monitoring studies become the most objective indication of presence or absence of GERD. And given the diverse symptom complexes being attributable to GERD, it is increasingly the role of the consulting gastroenterologist to establish that reflux disease is not present. In this role, I would argue that the most useful MII-pH study would be a negative study (normal acid exposure, negative SAP) in a patient studied while acid suppression therapy was withheld. Only then is one maximizing MII-pH as both a tool for identifying patients with abnormal esophageal acid exposure and as a reflux event detector. I would also be quite restrictive as to which specific symptoms were included in computing an SAP; heartburn, regurgitation and chest pain until someone convinces me otherwise. True, equivocal studies will still be encountered, but nobody said this was always going to be easy. Better to err on the side of caution, especially in the context of considering a potentially morbid therapeutic intervention such as antireflux surgery.

In conclusion, for a simple disease, GERD is remarkably complex. It was a mistake to think that PPIs were omnipotent or for that matter that response to PPIs equated to a diagnosis of GERD [8]. However, let’s not extend the streak of oversimplification into the domain of diagnostic testing. There is no doubt that MII-pH monitoring is a more sophisticated and advanced technology than the pH monitoring studies of the past. There is no doubt that MII-pH monitoring is the best tool that exists for the identification and characterization of gastroesophageal reflux events. But the clinical significance of these reflux events needs to be cautiously assessed within the context of other pathophysiological concepts. Otherwise, in our zeal to embrace a new technology and redefine an established disease entity, we risk branding everyone as having reflux disease based on flawed analysis paradigms.