Summary
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
Background : In subjects with gastrooesophageal reflux disease treated with a gastric antisecretory agent, the extent to which gastric acidity needs to be reduced to prevent pathological oesophageal acid exposure is not known.
Methods : Gastric and oesophageal pH were measured in 26 healthy subjects and in 59 subjects with gastrooesophageal reflux disease. In 27 of the subjects with gastrooesophageal reflux disease, pH was also recorded on days 1, 2 and 8 of treatment with 20 mg omeprazole and 20 mg rabeprazole in a randomized, twoway, crossover fashion.
Results : Receiver operating characteristic analysis was used to determine values for the integrated oesophageal acidity and time oesophageal pH≤4 that gave optimal cutoff points for distinguishing between normal and pathological oesophageal reflux. Using these cutoff points, we found that the probability of no pathological oesophageal reflux (Y) could be best fitted by an exponential equation, Y = a(e^{–bX}) + c, where a, b and c are constants and X is the value of the integrated gastric acidity. There was close agreement between the predicted and observed percentages of subjects with pathological oesophageal reflux during different days of treatment.
Conclusions : In subjects with gastrooesophageal reflux disease treated with a proton pump inhibitor, the value of the integrated gastric acidity can predict the likelihood of pathological oesophageal reflux.
Results
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
Figure 1 illustrates the results used to determine the value of the integrated oesophageal acidity that best distinguishes subjects with GERD from healthy controls. Figure 1a shows the percentage of controls and subjects with GERD with an integrated oesophageal acidity value above that indicated on the x axis of the graph. As expected, values in subjects with GERD were higher than those in controls. Figure 1a, however, does not clearly indicate the extent to which the integrated oesophageal acidity values differ between controls and subjects with GERD. Figure 1b addresses this issue and is a receiver operating characteristic curve of the data in Figure 1a.^{10} This curve plots the sensitivity (i.e. true positives or percentage of subjects with GERD with values above the cutoff point) vs. 1 − specificity (i.e. false positives or percentage of control subjects with values above the cutoff point) for each value of the integrated oesophageal acidity. In Figure 1b, the greater the displacement of the points from the diagonal identity line, the greater the extent to which the integrated oesophageal acidity distinguishes between controls and subjects with GERD. If, in Figure 1a, there was no overlap between the two groups with respect to the distributions of the integrated acidity values, the points in Figure 1b would increase upwards on the y axis to 100% and then spread horizontally across the line labelled 100%. On the other hand, if, in Figure 1a, the distributions were the same in subjects with GERD and controls, the points in Figure 1b would all fall on the diagonal identity line. The area under the receiver operating characteristic curve, which is conceptually equivalent to the Wilcoxon statistic,^{11–13} was 0.812 ± 0.047. (The standard error of the area was calculated using the procedure described in Hanley and McNeil.^{13}) The area under the receiver operating characteristic curve is a quantitative measure of the extent to which the integrated oesophageal acidity differs between controls and subjects with GERD, and represents the probability that a subject selected randomly from the GERD group would have a higher integrated acidity value than a subject selected randomly from the control group.^{10–13} If there was no overlap in the integrated oesophageal acidity between the two groups, the area would be 1.00 and the probability would be 100%. If the distributions were identical, the area would be 0.50 and the probability would be 50%, i.e. the same as chance alone.
The time oesophageal pH≤4 is commonly used to estimate the oesophageal acid exposure in normal and disease states.^{1}Figure 2 illustrates these values, which can be used to determine the value that best distinguishes subjects with GERD from healthy controls.
From the results in Figures 1 and 2, the optimal values that distinguished between subjects with GERD and controls were: integrated oesophageal acidity, 8.1 mmol.h/L; time oesophageal pH≤4, 4.5%(Table 1). ‘Optimal’ was chosen as the value that gave the maximum combined values for sensitivity and specificity.
Table 1. Comparison of the integrated oesophageal acidity and time oesophageal pH≤4 in distinguishing subjects with gastrooesophageal reflux disease (GERD) from controls Measure  Integrated oesophageal acidity 8.1 mmol.h/L  Time oesophageal pH≤4 4.5% 


Sensitivity (%)  72  89 
Specificity (%)  88  81 
Positive predictive value (%)  93  82 
Negative predictive value (%)  59  73 
The sensitivity and specificity assess how well a particular value of the integrated acidity or time pH≤4 distinguishes between subjects with GERD and controls. As pointed out by others,^{11}^{,} ^{14} the sensitivity and specificity are properties of the measure used to distinguish between subjects with GERD and controls and do not depend on the prevalence of GERD in the sample. The positive predictive value addresses the question: ‘If a subject has a value above the cutoff point, what is the probability that the subject has GERD?’. The negative predictive value addresses the question: ‘If a subject has a value below the cutoff point, what is the probability that the subject does not have GERD?’. An important feature of the predictive values is that they depend on the prevalence of GERD in the sample, and the lower the prevalence, the lower the ratio of true positives to false positives (a particularly clear discussion is given in Motulsky^{14}). This issue will become more important later in this paper when we address the value of the integrated gastric acidity that is necessary to prevent pathological oesophageal reflux in subjects with GERD.
Initially, it seemed like a straightforward process to determine the value of the integrated gastric acidity that would prevent pathological oesophageal reflux in subjects with GERD. That is, we thought that we could use all values from subjects with GERD at baseline and during treatment with a proton pump inhibitor, and plot 24h values for the integrated oesophageal acidity or for the time oesophageal pH≤4 vs. the corresponding value of the integrated gastric acidity. We could then determine graphically the highest value of the integrated gastric acidity that was associated with an integrated oesophageal acidity below 8.1 mmol.h/L and a time oesophageal pH≤4 below 4.5%, and use these values of gastric acidity as cutoff points.
Figure 3 illustrates the results of such plots. For both the integrated oesophageal acidity and time oesophageal pH≤4, normal oesophageal reflux values (i.e. values below the full horizontal lines in Figure 3) were associated with the entire range of integrated gastric acidity values, and there was no obvious relationship between either measure of oesophageal acidity and integrated gastric acidity.
We next stratified all values of the integrated gastric acidity for the accompanying value of the integrated oesophageal acidity using 8.1 mmol.h/L as a cutoff point. Figure 4a illustrates that the percentage of integrated gastric acidity values with pathological oesophageal reflux (integrated oesophageal acidity above 8.1 mmol.h/L) was significantly higher than the percentage of values with no pathological oesophageal reflux (integrated oesophageal acidity below 8.1 mmol.h/L) at all values of the integrated gastric acidity (P < 0.0001 by Wilcoxon matched pairs test). Figure 4b shows a receiver operating characteristic curve of the percentage of recordings with pathological oesophageal reflux vs. the percentage of recordings with no pathological oesophageal reflux at all values of the integrated gastric acidity. The area under the curve was 0.783 ± 0.033 which, as pointed out previously, gives the probability that a randomly selected subject with pathological oesophageal reflux would have a higher value of the integrated gastric acidity than a randomly selected subject with no pathological oesophageal reflux.
Figure 5 illustrates the results from the same type of analysis as illustrated in Figure 4, but using a time oesophageal pH≤4 of 4.5% as the cutoff point. The results in Figure 5 are essentially the same as those in Figure 4. That is, the percentage of integrated gastric acidity values with pathological oesophageal reflux (time oesophageal pH≤4 above 4.5%) was significantly higher than the percentage of values with no pathological oesophageal reflux (time oesophageal pH≤4 below 4.5%) at all values of the integrated gastric acidity (P < 0.0001 by Wilcoxon matched pairs test). The area under the receiver operating characteristic curve was 0.712 ± 0.036.
Table 2 gives the cutoff points for the integrated gastric acidity that best distinguish between recordings with pathological and normal oesophageal reflux using two different measures to define pathological reflux. The optimal cutoff point for the integrated gastric acidity was 943 mmol.h/L using the integrated oesophageal acidity to define pathological reflux, and 800 mmol.h/L using the time oesophageal pH≤4 to define pathological oesophageal reflux. With either cutoff point, the discrimination between pathological and normal oesophageal reflux was higher when the integrated oesophageal acidity was used rather than the time oesophageal pH≤4. We also conducted many of the analyses illustrated in the present paper using the time gastric pH≤4 instead of the integrated gastric acidity, but were unable to distinguish between pathological and normal oesophageal reflux (results not shown).
Table 2. Cutoff points for integrated gastric acidity that optimally distinguish between pathological and normal oesophageal acid exposure Cutoff point for integrated gastric acidity (mmol.h/L)  Integrated oesophageal acidity  Time oesophageal pH≤4 

Pathological oesophageal reflux (%)  Normal oesophageal reflux (%)  Pathological oesophageal reflux (%)  Normal oesophageal reflux (%) 


943  77  27  55  21 
800  80  34  61  24 
The data used for the analyses illustrated in Figures 4 and 5 can also be employed to address the question of the probability that a subject with a particular value of the integrated gastric acidity has no pathological oesophageal reflux. We calculated the probability of no pathological reflux by dividing the number of recordings with an integrated oesophageal acidity below 8.1 mmol.h/L, or a time oesophageal pH≤4 below 4.5%, by the total number of recordings at each value of the integrated gastric acidity. Figure 6a illustrates the results obtained using the integrated oesophageal acidity to define pathological oesophageal reflux. In this figure, the probability of no pathological reflux is 100% with a value of the integrated gastric acidity of near zero, and with increasing values of gastric acidity, the probability falls progressively to 60%. As mentioned above, the predictive values given previously in Table 1 are dependent on the prevalence of what is being detected. In Figure 6, values for the probability of no pathological reflux depend on the percentage of recordings without pathological oesophageal reflux. In the case of no pathological reflux defined as an integrated oesophageal acidity below 8.1 mmol.h/L, 60% of the recordings used for analysis had no pathological reflux; therefore, the probability of no pathological reflux can never be less than 60%.
The results in Figure 6a were best fitted by an equation with a single exponential plus a plateau, i.e.
where Y is the probability of no pathological reflux, k is a constant with a value of 7.383 × 10^{−4} and X is the value of the integrated gastric acidity. This equation fitted the data in Figure 6a with R^{2} = 0.965, indicating that virtually all of the variation in the probability of no pathological reflux could be accounted for by values of the integrated gastric acidity.
Figure 6b illustrates the results obtained using a time oesophageal pH≤4 below 4.5% to define pathological oesophageal reflux. In this figure, the probability of no pathological reflux is 53% with a value of the integrated gastric acidity of near zero, and with increasing values of gastric acidity, the probability falls progressively to 26%. The results in Figure 6b could also be fitted by an equation with a single exponential plus a plateau, i.e.
where X and Y are the same as defined above and k is a constant with a value of 2.662 × 10^{−5}. This equation fitted the data in Figure 6b with R^{2} = 0.934.
As a previous study has demonstrated that the results for predictive values, sensitivity and specificity are often misinterpreted,^{15} it is worth reviewing how the present findings should be interpreted, using the data in Figures 4 and 6 as examples. Figure 4a illustrates that, with a value of the integrated gastric acidity of 527 mmol.h/L as an example, 10% of subjects with pathological oesophageal reflux have values of the integrated gastric acidity below 527 mmol.h/L and 51% of subjects without pathological oesophageal reflux have values of the integrated gastric acidity above 527 mmol.h/L. It is not correct to conclude that the probability of no pathological oesophageal reflux in a subject with an integrated gastric acidity of 527 mmol.h/L can be calculated from these percentages, i.e. 51/(51 + 10) = 84%. The reason that this approach is not correct is that the probability depends on the relative abundance of recordings with no pathological oesophageal reflux. For example, if there were nine times more recordings with pathological oesophageal reflux than with no pathological reflux, the correct calculation would be 51/[(51 + (9 × 10)] = 36%. The results in Figure 6a display the probability of no pathological oesophageal reflux at each value of the integrated gastric acidity, calculated in a manner that takes into account the relative abundance of recordings with no pathological oesophageal reflux. Using the bestfit equation for the data in Figure 6a, the calculated probability of no pathological oesophageal reflux with a value of the integrated gastric acidity of 527 mmol.h/L is 87%.
Figure 7 compares the calculated probability of no pathological oesophageal reflux using the equations above with the observed percentage of subjects with no pathological oesophageal reflux on days 1, 2 and 8 of treatment with a proton pump inhibitor. There was close agreement between the calculated and observed values on days 2 and 8 using either criterion for no pathological oesophageal reflux. On day 1, however, the predicted probability was approximately twice the observed value. These differences on day 1 probably result from the fact that some subjects have oesophageal reflux before the onset of action of the proton pump inhibitor on gastric and oesophageal acidity. That is, even though on day 1, subjects have decreased integrated gastric acidity over the 24h period, which reduces the calculated probability of no pathological oesophageal reflux, some subjects experience oesophageal reflux before the onset of action of the proton pump inhibitor on gastric acidity.
Discussion
 Top of page
 Summary
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
The main purpose of the present study was to examine the extent to which gastric acidity in subjects with GERD treated with a proton pump inhibitor needs to be decreased to prevent pathological oesophageal reflux. Although the reduction of gastric acidity with a gastric antisecretory agent is the mainstay of treatment of GERD, the relationship between gastric and oesophageal acidity has not been examined. Others have published values for the time gastric pH > 4 with different antisecretory agents, and have related these values to the time oesophageal pH < 4 and to the percentages of subjects with healed erosive oesophagitis;^{5}^{,} ^{6} however, the values of gastric acidity were obtained from subjects with peptic ulcers, whereas the values of oesophageal acidity and oesophagitis were obtained from subjects with GERD.
We defined pathological oesophageal reflux in the same way as others,^{16}^{,} ^{17} using the value of the oesophageal acid exposure, calculated from 24h oesophageal pH recordings, that best distinguished between subjects with GERD and normal subjects. The optimal cutoff point in terms of the time oesophageal pH≤4 was 4.5%, and this value agrees with the cutoff point determined in other studies.^{16}^{,} ^{17} We are unaware of values of the integrated oesophageal acidity being used to distinguish between normal subjects and those with GERD, but the calculation is straightforward, and it should be possible for others to perform such calculations using existing oesophageal pH recordings from normal subjects and those with GERD. The cutoff point that we used to define pathological oesophageal reflux was that which gave the maximal values for sensitivity and specificity in distinguishing between subjects with GERD and normal subjects. Under other circumstances, other cutoff points might be used. For example, a different value might optimally distinguish between subjects with GERD with and without healed oesophagitis, whilst another value might optimally distinguish between subjects with GERD with and without resolved heartburn.
There was no apparent relationship between the integrated gastric acidity values and the corresponding values of the integrated oesophageal acidity or time oesophageal pH≤4 in subjects with GERD treated for different times with a proton pump inhibitor. We were initially surprised by this finding, because proton pump inhibitors decrease both the gastric and oesophageal acidity, and the magnitude of the decrease tends to be greater after several days of treatment than after 1 day of treatment.^{18} In contrast with the results in the present study, a study by others has reported a linear relationship between the time oesophageal pH < 4 and the time gastric pH > 4;^{5} however, as mentioned above, oesophageal pH values were obtained from subjects with GERD, whereas gastric pH values were obtained from subjects with peptic ulcer.
Our primary focus in the present study was not on the value of the oesophageal acidity per se, but on whether the value was pathological using a cutoff point based on findings in normal subjects and those with GERD. We then determined the extent to which corresponding values of the integrated gastric acidity could distinguish between pathological and nonpathological oesophageal reflux. Integrated gastric acidity values from subjects with pathological oesophageal reflux were significantly higher than values from subjects without pathological reflux. Moreover, the area under the receiver operating characteristic curve indicated that the probability that a subject chosen randomly from the group with pathological oesophageal reflux would have a higher value of gastric acidity than a subject chosen randomly from the group without pathological reflux was 0.712 or 0.783, depending on how pathological reflux was defined.
The present finding that the 24h integrated gastric acidity in subjects with pathological oesophageal reflux was significantly higher than that in subjects without pathological reflux might appear to disagree with the finding in a preceding paper that the 24h integrated gastric acidity in subjects with GERD was not significantly different from that in controls.^{9} The basis for the difference is probably that the values for no pathological oesophageal reflux in the present report were caused by treatment with a proton pump inhibitor, which decreased the gastric acidity over the entire 24h period, not just during the postprandial period. In contrast, in the preceding study, gastric acidity was significantly increased only during the postprandial period, not during the entire 24h period.
We found that the probability of pathological oesophageal reflux in subjects treated with a proton pump inhibitor could be predicted with high precision using an equation with a single exponential plus a plateau that reflected the percentage of the total recordings with no pathological oesophageal reflux. There is no mechanistic or pathophysiological significance to this equation that we are aware of. It was simply the equation that best fitted the data. We also found that there was close agreement between the observed and predicted percentages of subjects with no pathological oesophageal reflux on days 2 and 8 of treatment. As mentioned above, the difference between the observed and predicted percentages on day 1 probably occurred because some subjects experienced pathological oesophageal reflux before the onset of action of the proton pump inhibitor on gastric acidity.
There are, however, several caveats related to the present analyses. First, the results are probably only applicable to subjects with GERD treated with a proton pump inhibitor. Similar subjects treated with a histamine2 receptor antagonist would probably give different results, because these agents have a different duration of action on gastric acid secretion, have different effects on gastric pH during fasting, and inhibit fasting gastric acid secretion to a greater extent than mealstimulated secretion.^{19–21} Second, in the present study, subjects with GERD treated with a proton pump inhibitor were those with an oesophageal pH≤4 for at least 10% of a 24h baseline recording period. The parameter values for the exponential equation that give the probability of no pathological reflux might differ in important ways in subjects with GERD who have lower values of oesophageal acidity. Third, we and others have reported that subjects with GERD with erosive oesophagitis have higher values of oesophageal acid exposure than those without erosive oesophagitis.^{7}^{,} ^{22} It may be that the relationship between gastric and oesophageal acidity during treatment with a proton pump inhibitor may differ between these two groups. Fourth, we have found that subjects with GERD can be classified in terms of whether most of their oesophageal acid exposure occurs before or after 23.00 h.^{7}^{,} ^{8} The relationship between gastric and oesophageal acidity during treatment with a proton pump inhibitor might differ between these two groups.
Previously, we have commented on some of the shortcomings of time pH≤4 compared with integrated acidity in the evaluation of gastric and oesophageal pH recordings.^{7}^{,} ^{8} The present study also illustrates that the time oesophageal pH≤4 is less sensitive than the integrated oesophageal acidity in assessing the relationship between the integrated gastric acidity and the probability of no pathological oesophageal reflux. Our observation that the use of the time gastric pH≤4 was unsatisfactory in evaluating the relationships between gastric acidity and the presence or absence of pathological oesophageal reflux agrees with our previous findings that this measure is unsatisfactory in assessing the extent to which rabeprazole reduces gastric acidity.^{8}
To the extent that the subjects in the present study represent the overall population of subjects with GERD, we can use the parameter values for the exponential equation to predict the probability of normalizing oesophageal acid exposure during treatment with a proton pump inhibitor from the corresponding value of the integrated gastric acidity. The present study also illustrates an analytical approach that can be used to explore the relationship between the effect of a gastric antisecretory agent on gastric or oesophageal pH and clinical endpoints in GERD, such as heartburn or oesophagitis.