pHoenix score discriminative ability in suspected gastroesophageal reflux disease: a high-volume single-center external validation

SUMMARY

The pHoenix score was recently developed to reduce the proportion of inconclusive diagnoses associated with using total acid exposure time (AET) alone. The aim of this study was to compare the discriminative performance of the pHoenix score to total AET and DeMeester score (DMS) in patients undergoing 24-hour transnasal pH-monitoring (24-h pH). This cross-sectional study included consecutive patients (2017–2024) undergoing 24-h pH for suspected gastroesophageal reflux disease. Exclusions criteria were prior foregut/bariatric procedures, outflow obstruction disorders and inadequate pH-studies (<18 hours). The pHoenix score calculation was (%upright AET × 0.991) + (% supine AET × 1.286), with thresholds: <7.06 (normal), 7.06–8.45 (borderline), >8.45 (pathological). Total AET thresholds were: <4% (normal), 4–6% (borderline), and >6% (pathological). The DMS (pathological if >14.72) was the reference standard. Of 500 patients (50% females, median age 51 years, median BMI 24.65 kg/m²), 213 (43%) had pathological DMS. The pHoenix score and total AET identified a similar proportion of normal cases (54% vs. 56.2%, P > 0.99), but different pathological (40.4% vs. 30%, P < 0.01) and borderline diagnoses (5.6% vs. 13.8%, P < 0.01; with a 59% reduction with the pHoenix score). The pHoenix model showed strong performance (pseudo R²:0.877; Akaike information criterion = 83.57). Sensitivity/specificity were high at both 7.06 (99.1%/93.4%) and 8.45 (93.9%/99.3%) cutoffs. The AUC was 0.995 (95% CI: 0.987–1) for the pHoenix score, and 0.992 (95% CI: 0.987–0.997) for total AET. The pHoenix score, while maintaining a high diagnostic accuracy, offers a refined classification of acid exposure weighting supine/upright AET, thus reducing borderline diagnoses and potential need for further testing.

INTRODUCTION

Gastroesophageal reflux disease (GERD) affects approximately one in four individuals in Western countries, with a rising global prevalence and a significant economic burden.^1,2 According to the Lyon Consensus, a conclusive diagnosis of GERD can be made either endoscopically, through findings such as Los Angeles grade B to D esophagitis, biopsy-proven Barrett’s esophagus, or peptic stricture, or by ambulatory reflux monitoring, using a threshold of >6% total acid exposure time (AET).³ However, an AET between 4% and 6% is considered a borderline area, requiring further diagnostic evaluation. Despite additional testing, GERD may remain neither definitively confirmed nor excluded in these cases. Furthermore, relying on AET alone overlooks the distribution of reflux episodes between the upright and supine positions, an important component of the DeMeester score (DMS) and a relevant clinical factor.⁴

Historically, the DMS has been the standard metric for evaluating gastroesophageal reflux on pH monitoring.^5,6 Despite its strong validation, it does not define a borderline range, that would better capture the diagnostic gray area often encountered in clinical practice. Additionally, the DMS requires the exclusion of meal periods, the recording of which relies on patient cooperation and is often inaccurate.

Recently, Latorre-Rodríguez et al. introduced the pHoenix score, a novel metric designed to overcome these limitations.⁷ The pHoenix score is easy to calculate, does not require meal exclusion, and incorporates both upright and supine AET. Moreover, in the development and validation cohorts, it generated a narrower borderline zone compared to total AET, potentially offering greater diagnostic precision. Reducing indeterminate diagnoses, the pHoenix score may minimize the need for further testing, thereby lowering healthcare costs and improving patient satisfaction.

The aim of the present study was to evaluate the performance and conduct an external validation of the pHoenix score in a population setting different from the one in which it was originally developed. Additionally, the study sought to compare its diagnostic classifications with those based on total AET and DMS.

METHODS

Population

This cross-sectional study included consecutive adult (>18 years) patients undergoing 24-hour pH-monitoring between 2018 and 2024 for suspected GERD at the Department of Surgery, Oncology and Gastroenterology, University of Padova (Italy), an academical referral center for esophageal diseases. Exclusion criteria included prior foregut or bariatric procedures, outflow obstruction disorders, and inadequate pH-studies (<18 hours or poor meal records). No restrictions were applied to AET for inclusion, unlike in the development and internal validation cohort, which included only patients with an overall 48-hour total AET between 2% and 6%.⁷ This study was conducted in accordance with the Declaration of Helsinki and received approval from the institutional review board of our department. Written informed consent was waived due to the nature of the study.

We adhered to the guidelines outlined in the TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) statement and checklist (Supplementary Material) to guarantee rigorous and standardized reporting of the validation phase of this study. Demographic and clinical characteristics of all patients were prospectively recorded. GERD symptoms were recorded using the GerdQ questionnaire.⁸

Ambulatory 24-hour pH-monitoring

Esophageal acid exposure was assessed using 24-hour pH-monitoring, rather than 48-hour pH monitoring (Bravo capsule) as used in the pHoenix score development cohort. The procedure was performed in all patients at least 15 days after suspending any proton pump inhibitors (PPI), H2-blockers or promotility agents. The test was used to identify any abnormal acid exposure by positioning an electrode 5 cm above the upper border of the lower esophageal sphincter (LES according to the standard procedure used at our laboratory.

The probe was connected to a portable solid-state monitoring device (Digitrapper, Medtronic), and the acid reflux parameters assessed were as follows: number of episodes with pH <4, number of episodes lasting more than 5 min, duration of the longest episode, total percentage of time when the pH was <4 (total AET), percentages of time when the pH was <4 while supine, and while upright.

A DMS was considered pathological if >14.72. Total AET was classified as follows: <4% was considered normal, 4%–6% was considered borderline, and >6% was defined as pathological.³ The pHoenix score was calculated as (%upright AET × 0.991) + (% supine AET × 1.286). Diagnostic categories of the pHoenix score were determined through receiver operating characteristic (ROC) curve analysis in a previous publication from Latorre-Rodrìguez et al. Score categories were defined as follows: normal (<7.06), borderline (7.06–8.45), and pathological (>8.45)⁷ In contrast to the original pHoenix score development, where meal periods were included in the pH study interpretation, our analysis excluded meal periods, in line with the standard practice at our institution. Data on DMS and AET were extracted independently of the pHoenix score calculations.

Sample size

During the study period, approximately 2500 24-hour pH studies were performed at out Institution, and based on our experience, about 50% of patients are evaluated for suspected GERD. Therefore, considering the estimated prevalence (50%) and the potential eligible individuals, using a 95% confidence interval (CI) with a 5% marginal error, a minimum sample of 334 patients was calculated. A convenience sampling method was employed, enrolling the first 500 consecutive patients who met the inclusion criteria, achieving a power greater than 0.99.

Statistical analysis

A descriptive statistical analysis was employed using median and interquartile range (IQR) for continuous variables and counts with percentages for categorical variables. Chi-squared tests were used to compare categorical variables and assess differences in distribution between groups. The Kolmogorov–Smirnov test was used to assess normality of continuous variables. We used the Mann–Whitney U test to compare continuous variables between independent groups. Correlations between continuous variables were assessed using Pearson’s correlation coefficient or the Spearman rank correlation, as appropriate.

The ability of the pHoenix score to predict pathological DMS was assessed using pseudo R² for model fit, Akaike Information Criterion (AIC) for model quality, as well as discrimination metrics such as the area under the curve (AUC), sensitivity, and specificity. These outcomes were compared against those from a model based on total AET. The calibration of the predictive model was performed using brier score, calibration slope, and observed-to-expected event ratio. A significance threshold of α < 0.05 was applied throughout. All results are presented with corresponding 95% CIs and P-values where relevant. Missing data were handled using imputation methods. Specifically, missing values were identified only for body mass index (BMI) and symptom scores, and were addressed using median imputation.

Analyses were performed using R version 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

The study population comprised 500 patients, 248 (49.6%) females, and 252 (50.4%) males, with a median age of 51 years (IQR: 41–61) and a median BMI of 24.65 kg/m² (IQR: 22.10–27.46). Demographic and clinical characteristics of the study population compared to the development and interval validation cohorts are shown in Table 1.

Table 1.

Demographic and clinical characteristics of the present study population compared to the development and interval validation population from Latorre-Rodríguez et al.⁷publication

	External validation cohort (present study) n = 500	Entire development and internal validation cohort (from Latorre et al.) n = 114	Development cohort (from Latorre et al) n = 39
Age, years	51 (41–61)	55 (44–66)	51 (41–64)
Gender, female	248 (49.6)	80 (70.2)	24 (61.5)
BMI, kg/m2	24.65 (22.10–27.46)	28.9 (24.8–32.4)	29.5 (25–32.8)
GerdQ score	10 (4–15)	–	–
	24-hour pH-study	Day 1 of 48-hour pH study
DeMeester score	18.3 ± 25.7	12.41 ± 7.2	17.01 ± 5.1
Total AET, %	5.1 ± 7.4	3.5 ± 1.9	4.5 ± 1.3
Upright AET, %	5 ± 6.1	4.5 ± 3	5.4 ± 3.5
Supine AET, %	4.6 ± 9.4	2 ± 3	2 ± 3
Number of reflux episodes	44.3 ± 42.7	22.3 ± 13.5	26 ± 13.1
Duration of longest episode, min	13.4 ± 18.7	14 ± 12.8	19.6 ± 12

Data are expressed as no. (%), median (IQR), or mean ± SD. AET, acid exposure time; BMI, body mass index; IQR, interquartile range.

Based on the DMS, 213 (42.6%) patients had pathological reflux, and were diagnosed with GERD, while 287 (57.4%) had a normal pH monitoring. Using the total AET classification, 150 (30%) studies were pathological, 69 (13.8%) borderline, and 281 (56.2%) normal. According to the pHoenix score, 202 (40.4%) pH tests were pathological, 28 (5.6%) borderline, and 270 (54%) normal. Figure 1 shows the proportion of diagnoses according to the DMS, total AET and pHoenix score classifications.

Fig. 1.

Proportion of diagnoses according to the DMS, total AET, and pHoenix score classifications. AET, acid exposure time; DMS, DeMeester score.

According to pHoenix score and total AET, the proportion of normal cases was similar (54% vs. 56.2%, P > 0.99), while the pathological (40.4% vs. 30%, P < 0.01) and borderline diagnoses (5.6% vs. 13.8%, P < 0.01) were significantly different. Using the pHoenix score, a 59.4% relative decrease in borderline diagnoses was achieved compared to total AET. Patients classified as borderline by the pHoenix score had a narrower range in DMS (11.7 to 18.1) compared to borderline studies according to total AET (10.6 to 28.3). The median DMS between patients classified as borderline by the pHoenix score and those by total AET was significantly different (median DMS:14.13 [IQR: 13.36–15.75] vs. 17 [IQR: 14.81–18.4]; P < 0.001). Table 2 shows the number of patients and the corresponding DMS ranges for each diagnostic category of total AET and pHoenix score, stratified by whether the DMS was pathological or normal. The visual representation is shown in Fig. 2.

Table 2.

Number of patients and the corresponding DMS ranges for each diagnostic category of total AET and pHoenix score, stratified by DMS classification

	No. of patients (%)	DMS range
Pathological DMS	213 (42.6)	14.75–321.7
Total AET
Normal	9 (1.8)	14.76–18.9
Borderline	54 (10.8)	14.75–28.3
Pathological	150 (30)	17.43–321.7
pHoenix score
Normal	2 (0.4)	14.75–19.7
Borderline	11 (2.2)	14.8–18.1
Pathological	200 (40)	14.76–321.7
Normal DMS	287 (57.4)	0–14.59
Total AET
Normal	272 (54.4)	0–14.5
Borderline	15 (3)	10.6–14.59
Pathological	0	-
pHoenix score
Normal	268 (53.6)	0–14.59
Borderline	17 (3.4)	11.7–14.5
Pathological	2 (0.4)	13.8–13.9

Note: All percentages refer to the total study cohort (n = 500). AET, acid exposure time; DMS, DeMeester score.

Fig. 2.

Pie charts showing the reclassifications of patients with normal and pathological DMS using total AET or pHoenix score. AET, acid exposure time; DMS, DeMeester score.

A strong positive correlation was found between the pHoenix score and DMS (ρ = 0.992, 95% CI: 0.982–0.996, P < 0.001), total AET (ρ = 0.978, 95% CI: 0.966–0.985, P < 0.001), and distal reflux episodes (ρ = 0.836, 95% CI: 0.795–0.868, P < 0.001) (Fig. 3).

Fig. 3.

Spearman rank correlation revealing a strong positive correlation between pHoenix score and DMS, total AET, and distal reflux episodes, all of them showing a statistical significance (P < 0.001). AET, acid exposure time; DMS, DeMeester score.

Using the lower threshold of 7.06, the pHoenix score demonstrated a sensitivity of 99.1% (95% CI: 96.6–99.9%) and a specificity of 93.4% (95% CI: 90–96%). At the upper threshold of 8.45, sensitivity was 93.9% (95% CI: 89.8–96.7%) and specificity reached 99.3% (95% CI: 97.5–99.9%). In comparison, total AET at the 4% threshold showed a sensitivity of 95.8% (95% CI: 92.1–98%) and specificity of 94.8% (95% CI: 91.5–97%), while at the 6% threshold, sensitivity was 70.4% (95% CI: 63.8–76.5%) and specificity 100% (95% CI: 98.7–100%). Sensitivity and specificity of both metrics are summarized in Table 3. At their upper threshold (8.45 pHoenix score vs. 6% AET), the pHoenix score was more sensitive than AET (93.9% vs. 70.4%) while maintaining a high specificity (99.3% v. 100%).

Table 3.

SE and SP of pHoenix score and AET at their lower and upper thresholds, having the DMS as reference standard

pHoenix score			AET
Threshold	SE (95% CI)	SP (95% CI)	Threshold	SE (95% CI)	SP (95% CI)
7.06	99.1(96.6–99.9)	93.4(90.0–96.0)	4%	95.8(92.1–98)	94.8(91.5–97)
8.45	93.9(89.8–96.7)	99.3(97.5–99.9)	6%	70.4(63.8–76.5)	100(98.7–100)

AET, acid exposure time; CI, confidence interval; DMS, DeMeester score; SE, sensitivity; SP, specificity.

The pHoenix score model demonstrated adequate performance in predicting GERD as defined by the DMS, with strong indicators of model fit and quality (pseudo R² = 0.866, null deviance = 682.16, residual deviance = 83.57, AIC = 89.57). In comparison, the model based on total AET alone showed a lower performance (pseudo R² = 0.764, residual deviance = 151.9, AIC = 157.9), indicating that the pHoenix score may be a better predictor of GERD. The pHoenix score showed an excellent discrimination (AUC: 0.995 [95% CI: 0.987–1]), slightly higher than total AET (AUC: 0.992 [95% CI: 0.987–0.997]) (Fig. 4). Further, the pHoenix score demonstrated good calibration performance, as reflected by a Brier score of 0.041 (95% CI: 0.028–0.054), a calibration slope ranging approximately from 0.577 to 1.246, and an observed-to-expected event ratio of 0.99 (95% CI: 0.97–1.03).

Fig. 4.

Receiver operating characteristics curves representing the discriminative ability of the pHoenix score and the total AET as compared to the DMS. AET, acid exposure time; AUC, area under the curve; CI, confidence interval; DMS, DeMeester score.

DISCUSSION

As surgical and endoscopic treatment options for GERD continue to expand, the need for accurate and reliable diagnostic tools has become increasingly important to support effective clinical decision-making in an era of informed and engaged patients.^8,9 The primary objective of this study was to assess the discriminative performance of the pHoenix score in comparison to total AET and DMS, as well as validating its ability to reduce the rate of indeterminate diagnoses compared to total AET, as observed in the original development and validation cohorts.⁷ Our main finding is that the pHoenix score, while maintaining diagnostic accuracy, offers a refined classification by appropriately weighting upright and supine acid exposure and narrowing the borderline diagnostic area. Its clinical relevance lies in the potential to decrease the number of indeterminate diagnoses, thereby reducing the need for additional diagnostic procedures. This, in turn, could lower healthcare costs and enhance patient satisfaction. Furthermore, by preserving a small gray area, the pHoenix score may better reflect the inherently non-dichotomous, continuum-like nature of the disease, which is often inadequately represented by the binary classification provided by the DMS.

The DMS was introduced in the 70s by Lawrence Johnson and Tom DeMeester in the pioneering era of esophageal physiology.⁵ It has been the cornerstone of GERD diagnosis, guiding medical and surgical management for decades. However, despite its widespread use and strong validation, the DMS has some limitations: it relies on multiple parameters, requires meal recording and exclusion, and involves a complex formula, though this has largely been automated in modern practice. Perhaps most importantly, the DMS converts a complex condition like GERD to a binary outcome, a simplification that clinicians have come to understand does not fully reflect the reality of the disease.

To simplify the reporting of ambulatory reflux monitoring, the Lyon Consensus, recommends the use of total AET alone to establish a conclusive diagnosis of GERD based on pH studies.³ In its classification of acid exposure, it appoints a range of 4–6% total AET as a borderline area to accounts for diagnostic uncertainty in these intermediate values. The major drawback of this classification is that it leads to an indeterminate diagnosis in 9–19% of patients, as reported in previous studies.^10–13 Our findings are consistent with these data, as 14% of patients in our cohort had a borderline total AET. This results in the need for adjunctive metrics, among which mean nocturnal baseline impedance (MNBI) has been proposed as a promising tool. ^14–16 However, impedance-pH catheters are usually more expensive and less widely adopted than the standard ones, limiting the broader applicability of MNBI. In settings where adjunctive metrics are unavailable, particularly in patients with non-erosive reflux disease, a definitive diagnosis may not be reached, potentially resulting in undertreatment or mismanagement. Wang et al. analyzed a cohort of 225 patients who underwent laparoscopic fundoplication and had a preoperative pathological DMS but either borderline or pathological total AET according to the Lyon Consensus prior to surgery.¹⁷ Their findings showed no significant differences between patients with borderline and pathological AET in long-term outcomes, including improvements in GERD-related quality of life, PPI use, patient satisfaction, willingness to undergo the procedure again, and need for reoperation. These results suggest that patients with borderline AET may, in fact, have clinically significant reflux that can be underrecognized, and could benefit from surgical treatment. This observation is consistent with the findings of Jiang et al., who reported no significant difference in MNBI, symptom index, symptom association probability, and total number of reflux episodes between patients with borderline and pathological total AET.¹¹

To address these limitations, Latorre-Rodríguez et al. recently introduced the pHoenix score, a straightforward metric calculated as the sum of supine AET and upright AET, each multiplied by their respective coefficients (0.991 for upright AET and 1.286 for supine AET).⁷ This approach is grounded in the observation that supine AET is typically higher in patients with more severe disease phenotypes, such as those with erosive esophagitis or Barrett’s esophagus, compared to individuals with non-erosive reflux disease.^18–20 Using the DMS as the dependent variable, the authors aimed to define a narrower borderline diagnostic zone than total AET while simplifying the calculation process relative to the DMS. Through ROC curve analysis, they identified a borderline range for the pHoenix score of 7.06–8.45.

It is worth noting that a key difference between the development and validation cohorts of the original study and our cohort lies in the use of catheter-based 24-hour pH monitoring, rather than the 48-hour wireless system employed by Latorre-Rodríguez et al.⁷ This difference may influence reflux detection due to the shorter monitoring period and potential behavioral effects related to catheter discomfort. In particular, the presence of a transnasal catheter may cause transient discomfort and potentially modify eating habits or physical activity, leading to slightly lower acid exposure compared with the more physiologic conditions achieved during wireless monitoring. However, in our cohort, pH-metric parameters were comparable to those reported for day 1 in the development and validation cohorts of the Latorre-Rodríguez study, as shown in Table 1, and even tended to be slightly higher, contrary to what would be expected if the presence of the catheter had induced habits changes. This finding suggests that, in our population, the behavioral impact of catheter placement was likely minimal and did not substantially influence reflux activity or the recorded pH outcomes.

Despite this technical difference, the pHoenix score maintained consistent, or even higher, performance in identifying normal and pathological reflux categories, supporting its applicability across both catheter-based and wireless modalities, as well as across different populations. In our study, the pHoenix score achieved a sensitivity of 93.9%, compared with 87% in the original development cohort. This difference may reflect the inclusion criteria of the development cohort, which was limited to patients with an overall 48-hour AET between 2% and 6%. Importantly, the pHoenix score demonstrated the ability to effectively classify patients, with only 5.6% designated as borderline. This finding highlights the strength of the pHoenix score in significantly reducing the number of indeterminate diagnoses (59% less compared to total AET), while maintaining high sensitivity for detecting pathological reflux (93.9% with phoenix score vs. 70.4% with total AET).

Interestingly, in our cohort, patients classified within the borderline range of the pHoenix score had DMS values ranging from 11.7 to 18.1 (vs. a range 10.6–28.3 in borderline total AET). The DMS range identified by the borderline pHoenix score closely mirrors the clinical ‘gray zone’ in which surgeons and gastroenterologists often hesitate to recommend surgical or endoscopic intervention due to diagnostic uncertainty. These intermediate DMS values may not provide a strong enough indication for antireflux procedures, yet they also do not definitively rule out pathologic reflux. The pHoenix score offers a valuable tool to support clinical decision-making in these cases. Its ability to clearly identify a small subset of patients as borderline, with corresponding intermediate DMS values, provides an evidence-based rationale for conservative management. This may help avoid unnecessary procedures in patients unlikely to benefit, while improving confidence in the decision to defer intervention.

This study has several limitations. First, the analysis relied on previously recorded clinical data, which may be subject to inaccuracies, inconsistencies, or unavailable information. Although selection bias is possible, critical variables were complete, with no missing data. Second, detailed symptoms assessment and endoscopic data were not obtainable, precluding correlation of each score with symptom severity or endoscopic findings. Third, the study is conducted at a single high-volume center in Italy, and this represents a limitation in terms of generalizability of the results. Multicenter, international validations are warranted to confirm these findings across diverse populations and clinical settings. Lastly, the absence of post-treatment outcome data limits our ability to assess the clinical impact of each diagnostic classification. Despite these limitations, the study provides methodological insights into the discriminative power of the pHoenix score and lays a foundation for future prospective studies. Future validations should consider potential variability related to different pH monitoring devices (e.g. Diversatek, Laborie, Medtronic) as well as regional differences in pH metrics across populations. Such considerations are important to ensure the reproducibility and generalizability of the score in diverse clinical contexts, as highlighted in the consensus analysis by Sifrim et al.²¹

In conclusion, our findings highlight the effectiveness of the pHoenix score as a diagnostic tool for patients with suspected GERD, even in a population different from the one in which it was originally developed. This score is generated by a simple calculation and features a narrow borderline zone, which align closely with the clinical uncertainty often encountered in cases having intermediate DMSs. By offering diagnostic precision, the pHoenix score could streamline clinical decision-making, potentially reducing the need for additional testing in patients previously classified as borderline based on total AET. Ultimately, this may lead to improved diagnostic accuracy, more targeted management, and greater patient satisfaction. Further validation in larger, multicenter cohorts is needed to confirm the generalizability and clinical utility of these findings, incorporating adjunctive clinical tools and treatment outcomes to strengthen the assessment of the pHoenix score performance.

Renato Salvador and Michele Valmasoni contributed equally to this work and are acknowledged as joint final authors.

Specific author contributions: Conceptualization: Arianna Vittori, Renato Salvador, Michele Valmasoni; Data curation: Arianna Vittori, Luca Provenzano, Matteo Pittacolo, Loredana Nicoletti, Matteo Santangelo; Formal analysis: Arianna Vittori, Giovanni Capovilla; Investigation: Arianna Vittori, Loredana Nicoletti; Methodology: Arianna Vittori, Lucia Moletta, Renato Salvador; Validation: Arianna Vittori, Matteo Pittacolo, Michele Valmasoni; Visualization: Arianna Vittori, Luca Provenzano, Matteo Pittacolo, Francesca Forattini; Writing—original draft: Arianna Vittori, Giovanni Capovilla, Luca Provenzano, Francesca Forattini, Matteo Santangelo; Writing—review & editing: Arianna Vittori, Giovanni Capovilla, Luca Provenzano, Matteo Pittacolo, Loredana Nicoletti, Francesca Forattini, Matteo Santangelo, Lucia Moletta, Renato Salvador, Michele Valmasoni; Software, Matteo Pittacolo; Resources: Lucia Moletta, Renato Salvador; Supervision: Lucia Moletta; Project administration: Renato Salvador, Michele Valmasoni.