Do we have an alternative for the Rome III gastroduodenal symptom-based subgroups in functional gastroduodenal disorders? A cluster analysis approach


  • L. Van Oudenhove,

    1. Department of Pathophysiology, Gastroenterology Section, University of Leuven, Leuven, Belgium
    2. Liaison Psychiatry, University Psychiatric Centre, University Hospital Gasthuisberg, University of Leuven, Leuven, Belgium
    Search for more papers by this author
  • L. Holvoet,

    1. Department of Pathophysiology, Gastroenterology Section, University of Leuven, Leuven, Belgium
    Search for more papers by this author
  • J. Vandenberghe,

    1. Liaison Psychiatry, University Psychiatric Centre, University Hospital Gasthuisberg, University of Leuven, Leuven, Belgium
    2. Department of Neurosciences, Psychiatry Division, University of Leuven, Leuven, Belgium
    Search for more papers by this author
  • R. Vos,

    1. Department of Pathophysiology, Gastroenterology Section, University of Leuven, Leuven, Belgium
    Search for more papers by this author
  • J. Tack

    1. Department of Pathophysiology, Gastroenterology Section, University of Leuven, Leuven, Belgium
    Search for more papers by this author

Address for Correspondence
Dr. Lukas Van Oudenhove, Translational Research Centre for Gastrointestinal Diseases (TARGID), University Hospital Gasthuisberg, O&N1, 7th floor, bus 701, Herestraat 49, B-3000 Leuven, Belgium.
Tel: +3216330858; fax: +3216345939;


Background  Functional dyspepsia (FD) is a heterogeneous biopsychosocial disorder. The Rome III consensus proposed a subdivision into epigastric pain syndrome and postprandial distress syndrome, based on gastroduodenal symptom pattern only; nausea/vomiting- and belching disorders were classified as separate functional gastroduodenal disorders (FGD). We aimed to investigate an alternative subdivision of FGD, taking into account gastric sensorimotor function, anxiety & depression and ‘somatization’, besides gastroduodenal symptoms.

Methods  Gastroduodenal symptom data were available for 857 consecutive FGD patients (Rome II criteria). In a subsample (n = 259), additional data were obtained on gastric sensitivity, anxiety, depression and ‘somatization’. Two separate cluster analyses were performed. In analysis 1, clustering was based on individual gastroduodenal symptom scores. In analysis 2, gastric sensitivity, anxiety & depression and ‘somatization’, besides total gastroduodenal symptoms score, were used for clustering.

Key Results Analysis 1 identified four clusters, largely supporting the Rome III classification, with early satiation, pain and nausea/vomiting clusters, besides a limited severity cluster (R2 = 0.32). Analysis 2 suggested a five-cluster solution (R2 = 0.48). Anxiety, depression and ‘somatization’ were the most important variables separating the clusters. ‘Primary somatization’ (with low psychiatric symptom levels) as well as ‘secondary somatization’ (with high anxiety & depression scores) subgroups were identified, besides three other subgroups characterized by psychiatric/gastroduodenal symptoms, mild anxiety symptoms and limited overall severity, respectively.

Conclusions & Inferences  We propose an alternative to the current subgrouping in FGD that is exclusively based on gastroduodenal symptoms. This may have consequences for future classification of FGD, as well as broader relevance towards the debate on subgrouping ‘functional somatic syndromes’.


Based on the consensus opinion of an international panel of clinical investigators, the Rome III classification divides the functional gastroduodenal disorders (FGD) into four categories: functional dyspepsia (FD), belching disorders, nausea & vomiting disorders and rumination syndrome.1 Functional dyspepsia is defined as ‘the presence of symptoms thought to originate in the gastroduodenal region, in the absence of any organic, systemic, or metabolic disease that is likely to explain the symptoms’.1 According to Rome III consensus, the cardinal dyspepsia symptoms are epigastric pain, epigastric burning, postprandial fullness and early satiation.1 Heartburn or any other symptoms suggestive of esophageal disease may co-exist but are not considered typical FD symptoms by the Rome committee. Functional dyspepsia is then further subdivided in two symptom-based subgroups:1 meal-induced dyspeptic symptoms or postprandial distress syndrome (PDS, with postprandial fullness and/or early satiation as predominant symptom), and2 epigastric pain (EPS, with epigastric pain/burning as predominant symptom).1

To date, although some efforts have been made to link symptoms to gastric sensorimotor (dys)function,2 the current FGD classification and diagnostic criteria are almost entirely based on gastroduodenal symptom patterns. The Rome III process was the result of expert consensus, but is also supported by some empirical evidence, mostly derived from factor analytic studies on gastroduodenal symptom data obtained in groups of FGD patients from various settings [overview in(1)]. However, most of these studies were not limited to cardinal FGD symptoms (most of them, for example, included esophageal symptoms), the amount of variance explained was often rather small3 and only a few of these analyses have been performed after the introduction of Rome III criteria.4,5

Furthermore, FGD, like other ‘functional somatic syndromes’ (FSS) with which they frequently overlap, are heterogeneous disorders in which symptoms result from a complex and reciprocal interaction between biological, psychological and social factors.6,7 Therefore, in the psychiatric/psychosomatic literature, better integration of psychosocial factors into diagnostic criteria and classification systems for FSS/somatoform disorders has been repeatedly proposed.8–11 This may equally apply to functional gastrointestinal disorders (FGID). However, it remains unknown whether taking into account gastric sensorimotor function, anxiety & depression and ‘somatization’, besides gastroduodenal symptoms, would lead to a better subdivision of F(G)D. ‘Somatization’ will be descriptively defined here as a tendency to experience and report multiple (i.e. not limited to the GI tract) somatic symptoms that cannot be adequately explained by organic findings.12

The aim of the present study was to twofold. Firstly, we wanted to determine whether we could identify symptom-based FGD clusters which align with the current Rome III subgroups. Secondly, we aimed to investigate a subdivision of FGD based on gastric sensorimotor function, total gastroduodenal symptoms level, anxiety, depression and ‘somatization’, as an alternative to the Rome III subdivision based on individual FGD symptom levels. The variables used in the second analysis were chosen a priori based on existing evidence on their important role in FGD2,6,13,14 as well as on a study in fibromyalgia, another FSS.15

Materials and methods

Patient samples

Consecutive Dutch-speaking patients who recently received a clinical diagnosis of FGD [either at their visit to the general GI or GI motility clinic at the University Hospital Leuven (tertiary care) or at a recent secondary care gastroenterologist visit that lead to the referral] were asked to complete an FGD symptom questionnaire based on Rome II criteria,16 between 1999 (when Rome II criteria became available) and 2009. Additional data on anxiety, depression and ‘somatization’ as well as barostat data were available for a subsample obtained between 2002 and 2009. The patient samples reported on in the present paper do partially overlap with samples from other studies by our group.13,17–20 However, the hypotheses tested and analyses performed in the present article are novel and have not been reported on elsewhere. Further details about patient selection have been published in detail elsewhere.17 The protocol had been approved by the University Hospitals Leuven ethical committee prior to the start of patient recruitment.

Gastric sensorimotor function testing

Details about gastric sensorimotor function testing have been published elsewhere.17 Briefly, we used our standard barostat protocol. During isobaric stepwise distension, patients scored their perception of upper abdominal sensation at the end of every distending step using a graphic rating scale (0–6) with verbal descriptors (0 corresponding with no sensation, 1 with first sensation, 5 with discomfort and 6 with pain). Perception, discomfort & pain thresholds were defined as the lowest pressure above minimal distending pressure evoking a perception score of 1 or more, 5 or more & 6 or more, respectively.

Psychological factors & psychiatric disorder measurements

Trait anxiety  The 20-item trait scale of the State-Trait Anxiety Inventory (STAI) measures stable individual differences in ‘anxiety proneness’, that is ‘differences between people in the tendency to perceive situations as threatening and to respond to them with elevations in state anxiety’21,22 and yields a continuous total score.

Current anxiety & depression  The 14-item Hospital Anxiety and Depression Scale (HADS) was used to screen for current anxiety or depressive disorders.23,24 The subscale scores for anxiety and depression were used as separate continuous variables.

Somatic symptoms measurements

‘Somatization’  The PHQ somatoform disorder module (PHQ-15) is a well-validated self-report questionnaire composed of 15 somatic symptoms, including 14 of the 15 most prevalent DSM-IV somatization disorder criteria.25,26 All items are rated on a Likert scale (0–2). Current ‘somatization’ (past month) was measured. The PHQ-15 includes two FGD symptoms (‘stomach pain’ and ‘nausea, gas or indigestion’), which were not used in the present analysis because of their overlap with the gastroduodenal symptoms score. Thus, ‘somatization’ scores were calculated as the sum of the remaining 13 non-FGD items.

A limitation of the PHQ-15 as a self-report measure (without interview) is that it cannot distinguish between ‘medically explained’ and ‘unexplained’ symptoms,25 which is an important feature of the ‘somatization’ concept.12,27 In this study, adequate clinical and technical investigations were performed to rule out a medical explanation of GI symptoms, but not systematically of other somatic symptoms as included in the PHQ-15, although major non-GI medical comorbidity that may account for these somatic symptoms was ruled out on an ‘as needed’ basis (i.e. when considered clinically relevant). Nevertheless, it should be emphasized that the PHQ symptom count in this study can only be characterized as indicative of ‘somatization’. However, total self-reported PHQ somatic symptom counts are highly associated with physician-rated somatoform disorder symptom counts.25,28 Furthermore, the distinction between medically explained and unexplained symptoms may be problematic and less relevant than previously thought.10,29–31

Gastroduodenal symptoms score  The intensity of eight gastroduodenal symptoms was scored on a Likert scale [0–3 (absent, mild, moderate, severe)], as previously reported.16 These symptoms include the six typical dyspepsia symptoms as defined by Rome II criteria (epigastric pain, epigastric discomfort, early satiation, upper abdominal fullness, bloating and nausea), as well as the cardinal symptom of the other two Rome II FGD: belching (as a symptom of aerophagia) and vomiting.32 Gastroduodenal symptoms score is defined as the sum of all eight items.16

Statistical analysis

SAS 9.1.3 and Enterprise Miner 4.3 (SAS Institute, Cary, NC, USA) were used. Data are presented as mean ± standard deviation (SD). Significance level was set at P = 0.05. The aim of cluster analysis is to identify groups of subjects that are internally homogeneous but clearly different in some respect(s) from other clusters (i.e. with maximal homogeneity within the groups but maximal heterogeneity between the groups).

Firstly, hierarchical cluster analysis was performed using Ward’s clustering method. The result of this analysis was used to determine the most appropriate number of clusters for subsequent non-hierarchical cluster analysis, based on the R2 statistic.

Secondly, non-hierarchical cluster analysis with full seed replacement was performed to determine the final cluster solution based on the number of clusters derived from the hierarchical analysis. anova’s with post hoc comparison of all pairwise differences (with Tukey correction for multiple comparisons) were performed to determine significant differences between cluster means for each variable.

The clinical interpretation of each cluster was aided by describing a cluster profile that compromised the mean score per variable per cluster, as well as an overall mean score per variable for comparison purposes.


Descriptive data

Gastroduodenal symptoms data were available for 857 FD patients, additional data on ‘somatization’, current anxiety & depression and gastric sensorimotor function were available for a subgroup of 259 patients. The mean age was 41.5 ± 15.1 and 39.5 ± 12.9 in the whole sample and the subsample, respectively. The gender distribution was also similar (69.4% women in the whole sample, 75.3% in the subsample). The mean gastroduodenal symptoms score was higher in the subsample than in the whole sample (13.4 ± 4.8 and 10.7 ± 4.3, respectively). Overall means for the eight individual FGD symptoms fell in the mild to moderate range (Table 1).

Table 1.   Results of non-hierarchical cluster analysis on eight gastroduodenal symptoms in 857 functional gastroduodenal disorders patients
 PainFullnessBloatingEarly satiationNauseaVomitingBelchingEpigastric burningTotal symptom score
  1. Overall R2 = 0.32.

  2. anova’s with Tukey post hoc comparison for all pairwise differences were performed within each variable; cluster means with different superscripts are significantly different from each other (all P < 0.025 and most P < 0.001), cluster means with the same superscript do not differ significantly.

  3. n/a, not applicable.

Cluster 1 (n = 252)0.37 ± 0.59a1.46 ± 1.00a1.27 ± 1.00a0.95 ± 0.94a0.57 ± 0.82a0.17 ± 0.52a0.98 ± 0.97a0.52 ± 0.79a6.3 ± 2.4a
Cluster 2 (n = 228)1.69 ± 0.99b2.29 ± 0.62b2.25 ± 0.61b2.36 ± 0.53b1.41 ± 0.99b0.19 ± 0.40a1.33 ± 1.05b1.11 ± 1.02b12.6 ± 2.3b
Cluster 3 (n = 166)1.70 ± 1.10b2.34 ± 0.74b2.12 ± 0.88b2.12 ± 0.91c2.43 ± 0.62c2.47 ± 0.59b1.24 ± 1.02a,b1.46 ± 1.09c15.7 ± 3.5c
Cluster 4 (n = 211)2.22 ± 0.64c1.63 ± 0.89c1.75 ± 0.91c0.43 ± 0.64d1.32 ± 1.00b0.36 ± 0.72c1.06 ± 1.06a1.23 ± 1.05b,c10.0 ± 2.4d
Overall mean (n = 857)1.43 ± 1.101.89 ± 0.921.81 ± 0.951.42 ± 1.111.34 ± 1.090.66 ± 1.051.14 ± 1.031.04 ± 1.0410.7 ± 4.3

Analysis 1: subgroups based on individual gastroduodenal symptom scores

Hierarchical cluster analysis suggested a 4 cluster solution, based on the R2 statistic, as within-cluster homogeneity appeared to have reached a plateau at 4 clusters.

The results of the non-hierarchical cluster analysis with 4 clusters are shown in Table 1. Vomiting, early satiation and pain (R2 = 0.71, 0.52 and 0.43, respectively) were found to be the most important variables for forming clusters. Epigastric burning and belching had the lowest R2 (0.12 and 0.02, respectively), indicating that they are not important for clustering. The overall R2 was rather low (0.32), which indicates a relatively weak clustering solution (i.e. homogeneity within clusters and heterogeneity between clusters being rather low).

Cluster 1 was characterized by low scores for all variables, with significant differences from all other clusters for most variables (‘limited severity cluster’). Cluster 2 was characterized by a particularly high score on early satiation (significantly higher than in all other clusters), high scores on bloating and fullness, intermediate scores on pain and nausea and a low score on vomiting (‘early satiation cluster’). The striking feature of cluster 3 was a particularly high score on nausea and vomiting (significantly higher than in all other clusters), besides high scores on fullness and bloating (comparable with cluster 2), a relatively high score on early satiation (though significantly lower than in cluster 2) and an intermediate pain score (‘nausea & vomiting cluster’). Cluster 4, finally, was characterized by a particularly high score on pain (significantly higher than all other clusters), a particularly low score on early satiation (significantly lower than all other clusters) and intermediate to low scores on all other variables (‘pain cluster’). The mean cumulative gastroduodenal symptoms score per cluster, as well as the relative importance of each individual symptom, are graphically represented in Fig. 1. Total gastroduodenal symptom scores were significantly different between all four clusters (all pairwise differences P < 0.0001 after Tukey correction for multiple comparisons), with the highest total score in cluster 3 and the lowest score in cluster 1 (Table 1).

Figure 1.

Analysis 1: mean cumulative gastroduodenal symptoms score per cluster and the relative importance of each individual symptom.

Analysis 2: subgroups based on total gastroduodenal symptoms score, ‘somatization’, trait anxiety, current anxiety & depression and gastric sensitivity

Because the variables used in this analysis were measured/scored on different scales, all variables were standardized with a mean of 0 and SD of 1 prior to performing the cluster analysis. Due to the resulting unit normal distribution of variables, numeric values can be interpreted as SD units. For example, a cluster mean of −1.0 would indicate the cluster was depressed on that variable by an average of 1 SD.

Hierarchical cluster analysis suggested a 5 cluster solution, based on the R2 statistic, as within-cluster homogeneity appeared to have reached a plateau at 5 clusters. Fig. 2 shows the non-standardized total gastroduodenal symptoms scores per cluster, for comparison purposes with the scores from analysis 1.

Figure 2.

Analysis 2: mean total gastroduodenal symptoms score per cluster, error bars represent standard deviations.

The results of the non-hierarchical cluster analysis with 5 clusters are shown in Table 2. None of the variables had a low R2, indicating that they all contribute to clustering to a certain extent. Trait anxiety had the highest R2, with current anxiety and depression as well as somatization having a high R2 as well (0.63, 0.55, 0.57 and 0.49, respectively). Gastroduodenal symptoms score and gastric discomfort threshold had the lowest R2 (0.38 & 0.26, respectively). The overall R2 was 0.47.

Table 2.   Results of non-hierarchical cluster analysis on total gastroduodenal symptoms score, ‘somatization’, gastric sensitivity, trait anxiety and current anxiety and depression in 259 functional gastroduodenal disorders patients: numerical values
 Discomfort thresholdGastro-duodenal symptomsTrait anxietyCurrent anxietyCurrent depressionSomatization
  1. Overall R2=0.48; *variables are standardized with mean 0 and standard deviation 1; non-standardized mean and standard deviation for the whole sample.

  2. anova’s with Tukey post hoc comparison for all pairwise differences were performed within each variable; significant differences (all P < 0.0284 and most P < 0.001) are indicated by superscripts as follows: 1different from cluster 1, 2different from cluster 2, 3different from cluster 3, 4different from cluster 4, 5 different from cluster 5.

Cluster 1* (n = 43)−0.65 ± 0.603,4,50.63 ± 0.703,4,50.61 ± 0.772,4,50.76 ± 0.922,3,4,51.05 ± 0.612,3,4−0.23 ± 0.642,4,5
Cluster 2* (n = 62)−0.53 ± 0.763,4,50.74 ± 0.753,4,5−0.44 ± 0.621,3,4,5−0.37 ± 0.651,3,4,5−0.36 ± 0.601,3,4,50.95 ± 0.731,3,4
Cluster 3* (n = 65)0.42 ± 0.911,2−0.80 ± 0.731,2,4,50.33 ± 0.612,4,50.23 ± 0.691,2,4,50.00 ± 0.701,2,4,5−0.46 ± 0.692,5
Cluster 4* (n = 62)0.21 ± 0.931,2,5−0.29 ± 0.931,2,3−0.95 ± 0.431,2,3,5−0.95 ± 0.391,2,3,5−0.90 ± 0.351,2,3,5−0.72 ± 0.671,2,5
Cluster 5* (n = 27)0.81 ± 1.161,2,40.00 ± 0.841,2,31.55 ± 0.741,2,3,41.36 ± 0.801,2,3,41.28 ± 1.132,3,40.92 ± 0.981,3,4
Overall mean (n = 259)9.9 ± 3.613.4 ± 4.841.1 ± 11.56.2 ± 4.14.5 ± 3.79.8 ± 4.6
R2variable 0.26 0.38 0.63 0.55 0.57 0.49

A summary of cluster characteristics is provided in Table 3. Cluster 1 was characterized by high FGD symptom levels, high depression and, to a lesser extent, anxiety scores, intermediate to low ‘somatization’ scores and a low gastric discomfort threshold (‘psychiatric – FGD cluster’). Cluster 2 was characterized by a high ‘somatization’ score, high gastroduodenal symptoms, low discomfort threshold as well as low scores for trait anxiety, current anxiety and depression (‘primary somatization cluster’). Cluster 3 was characterized by intermediate to high anxiety levels as well as low somatic symptoms (FGD as well as ‘somatization’) (‘mild anxiety cluster’). Cluster 4 was characterized by low scores for all symptom ratings and intermediate to high gastric discomfort threshold (‘limited severity cluster’). Cluster 5 was characterized by very high scores for ‘somatization’ and psychiatric symptoms, as well as intermediate FGD symptom levels and a high gastric discomfort threshold (‘psychiatric – secondary somatization cluster’).

Table 3.   Results of non-hierarchical cluster analysis on total gastroduodenal symptoms score, ‘somatization’, gastric sensitivity, trait anxiety and current anxiety and depression in 259 functional gastroduodenal disorders patients: summary of cluster characteristics
 Name clusterDiscomfort thresholdGastro-duodenal symptomsTrait anxietyCurrent anxietyCurrent depressionSomatization
Cluster 1 (n = 43)Psychiatric – FGDLowHighHighHighVery highIntermediate low
Cluster 2 (n = 62)Primary somatizationLowHighLowLowLowVery high
Cluster 3 (n = 65)Mild anxietyIntermediate highVery lowIntermediate highIntermediate highIntermediateLow
Cluster 4 (n = 62)Limited severityIntermediate highIntermediate lowVery lowVery lowVery lowLow
Cluster 5 (n = 27)Psychiatric – secondary somatizationHighIntermediateVery highVery highVery highVery high


This is, to the best of our knowledge, the first study identifying subgroups within the FGD in an alternative way compared with the Rome III classification, which is exclusively based on individual FGD symptoms. This was done empirically, using the technique of cluster analysis.

Analysis 1

In a first analysis, clustering was attempted based on individual gastroduodenal symptom profiles, as in the Rome III classification of FGD. The majority of the results are in line with this classification: vomiting (and nausea, although to a lesser extent), early satiation and epigastric pain contributed most to clustering, and these are cardinal symptoms of Rome III nausea and vomiting disorders, FD/PDS and FD/epigastric pain, respectively. However, postprandial fullness (another cardinal symptom of FD/PDS according to Rome III criteria) and belching (the defining symptom of belching disorders) were not found to be important for separating clusters.

These results should be interpreted against previous studies that attempted to identify FGD subgroups based on gastroduodenal symptoms, even though most of these studies used a different statistical approach (factor analysis rather than cluster analysis). It should be noted that the use of different definitions for symptoms as well as FGD syndromes, reflecting the change in classification and diagnostic criteria over time, as well as the different nature of the study populations used, render the comparison between studies even more difficult.

Some studies based their analyses on upper GI symptoms only. An Australian population-based study identified four symptom factors based on a list of 13 upper GI symptoms (seven of the present FGD symptoms and six esophageal symptoms). Three of the factors were related to FGD: nausea/vomiting/retching, early satiation/postprandial fullness and epigastric pain/bloating/belching – the other one being esophageal.3 The three FGD factors together accounted for 37.7% of the variance. A more recent and bigger population-based study reported on a factor- and subsequent cluster analysis, which led to similar results. However, no information on the amount of variance explained or the importance of each individual factor in the clustering is provided in this study.33 Fischler et al. found four factors in a tertiary care FD sample, based on the same eight FGD symptoms as used in the present study: nausea/vomiting/early satiation, bloating/fullness, epigastric pain/burning and belching.18 Two recent studies, finally, aimed to investigate the association between symptom-based FD subgroups and weight loss using factor- and subsequent cluster analysis. In a population-based study, four clusters were found that were roughly comparable to the results from the present study, except for the fact that bloating clustered with pain. The early satiation/postprandial fullness cluster was most associated with weight loss.5 In a tertiary care study, three factors (nausea/vomiting/early satiation, fullness/bloating/early satiation and pain/belching) resulted in six clusters. The clusters scoring high on the nausea/vomiting/early satiation factor were most associated with weight loss.4

A number of other studies used symptom scores from the entire GI tract. In one of the earliest studies on the subject, Agreus et al. found no evidence for any significant clustering of GI symptoms using principal component analysis in a large population-based sample. However, their list of 24 GI symptoms did not include all cardinal FGD symptoms as used in the later Rome II and III definitions as well as in the present study.34 Kwan et al. performed a factor analysis on 40 GI symptoms in 1012 Rome II FGID patients attending 14 Asian clinics. Two FD factors were identified (postprandial fullness/early satiation/bloating/belching and upper abdominal pain/discomfort), besides a functional vomiting factor (nausea/vomiting/retching).35 These three factors collectively explained 11% of the total variance in GI symptoms. In a similar analysis on US and Italian samples based on a Rome I questionnaire, four dyspepsia factors were identified: two epigastric pain factors, an early satiation/nausea/retching/vomiting factor and an upper abdominal bloating factor.36

In conclusion, the results of the present study are relatively well in line with previous studies on subgrouping FGD, although some differences exist in methodology as well as results. This provides further support for the current Rome III classification of FGD. However, the amount of variance explained by the first analysis is low (overall R2 = 0.32), indicating that clustering based on gastroduodenal symptom pattern only remains far from optimal. This is in line with clinical evidence showing much overlap between FD subgroups and with other FGD, as well as lack of stability over time.34,37 This justifies attempting alternative strategies for subgrouping, as in analysis 2.

Analysis 2

We performed a second analysis in which clustering was based on a combination of total gastroduodenal symptoms score, ‘somatization’, anxiety (both trait and current), depression and gastric sensitivity. Incorporating psychosocial factors into diagnostic criteria and classification systems for FSS/somatoform disorders has been repeatedly suggested before,8–11 mostly in conceptual papers, but this type of alternative subgrouping strategy for FGD based on empirical data has, to the best of our knowledge, not been reported in the literature before. Although this analysis is exploratory in nature, the variables used for clustering were chosen a priori based on existing evidence on their important role in FGD2,6,13,14 as well as a study in fibromyalgia, another FSS.15 The results indicate that all variables contributed at least moderately to clustering, as reflected by their individual R2 values, with anxiety (both current and trait), depression and ‘somatization’ being the most important ones. Although it is difficult to compare both analysis directly given the different nature of the variables included, the overall R2 value was higher than in the first analysis (0.48), indicating a somewhat better clustering solution (i.e. more homogeneity within and heterogeneity between clusters), although still suboptimal.

The results of this second cluster analysis may provide interesting new evidence on the nature of ‘somatization’ in general and its relationship with anxiety & depression in particular. It remains a matter of debate whether ‘somatization’ needs to be conceptualized as a primary phenomenon (‘functional somatization’) or rather as the result of a complex psychobiological process in which psychological processes or psychiatric comorbidity such as hypervigilance, (symptom-specific) anxiety and/or depression lead to amplification of bodily signals through central sensitization of interoceptive/pain-processing neural systems (secondary phenomenon, ‘presenting somatization’).12,38–40 The present results suggest it may be both, in the following way. Cluster 2 (‘primary somatization’) is characterized by the highest ‘somatization’ scores, with anxiety and depression scores well below average. This cluster may therefore represent a group of patients characterized by primary (i.e. not driven by psychiatric comorbidity) ‘somatization’. This cluster also has a high level of FGD symptoms and a low gastric discomfort threshold. This suggests a role for gastric hypersensitivity in this subgroup of patients, besides ‘somatization’. Cluster 5 (‘psychiatric – secondary somatization’) has equally high ‘somatization’ scores, combined with the highest scores for anxiety & depression. This may correspond to a group of patients characterized by ‘somatization’, which is driven mainly by psychological processes/psychiatric comorbidity. This is in agreement with a recent finding that psychiatric comorbidity influences, but not completely explains, multiple somatic symptoms comorbidity (i.e. ‘somatization’) in irritable bowel syndrome (IBS).41 The intermediate level of FGD symptoms and high gastric discomfort threshold (hyposensitivity) suggest that this subgroup may be primarily characterized by a psychologically driven tendency towards general rather than GI-specific symptom reporting, with a more limited role for GI-specific mechanisms.

Cluster 1 (‘psychiatricFGD’) has high FGD & psychiatric (especially depression) symptom levels, combined with a low gastric discomfort threshold and low ‘somatization’ scores. This suggests a subgroup characterized by FGD but not extraintestinal symptoms (i.e. without a generalized tendency towards symptom reporting), which may be influenced by psychiatric comorbidity and/or gastric hypersensitivity.

A somewhat similar study investigated subgroups in fibromyalgia based on depression, trait anxiety, pain-related cognition and somatic pain sensitivity. ‘Somatization’ was lacking here, but the three clusters found in this study are largely in line with the present study in FGD: a mild anxiety – hyposensitivity cluster (corresponding largely to cluster 3 in the present study), a psychiatric – hypersensitivity cluster (corresponding largely to cluster 1 in the present study) and a primary (i.e. without psychiatric comorbidity) hypersensitivity cluster (corresponding largely to cluster 2 in the present study).15


Although the variables used for clustering in the second analysis were carefully chosen based on previous evidence, this should be regarded as an exploratory analysis which needs confirmation in other samples and/or with confirmatory statistical methods. Furthermore, the different nature of the variables included in both analyses as well as the fact that the second analysis was performed in a subsample only renders direct comparison difficult. The amount of variance explained in the second analysis, although higher than in the first analysis, remains rather low. This indicates that we are still missing variables that may be important for subgrouping. Although speculative, candidate include gastric motor function (accommodation, emptying) as well as (symptom-specific) psychological mechanisms including GI-specific anxiety and pain catastrophizing, among others. Finally, the study population was recruited in a single center with some second but mostly tertiary care functions, which may affect generalizability of the results towards other populations. However, we do believe these results have relevance towards a broader clinical audience, including general practitioners, in that screening for psychiatric and non-gastrointestinal somatic symptoms may be relevant for identifying subgroups who may benefit from different forms of treatment. It has indeed been shown in IBS that particularly patients with psychiatric co-morbidity or ‘somatization’ benefit from psychopharmacological or psychotherapeutic treatment, and that this subgroup responds poorly to ‘standard treatment’.42,43 To our opinion, though, the results clearly show that FGD as a group cannot be reduced to or explained by psychiatric disorders such as anxiety & depression, as psychiatric comorbidity plays a role in a subgroup only.

General conclusion

Finally, the relatively low R2 values in both analyses may lead us to a more fundamental question regarding the very sense of the search for subgroups within FG(I)D only. There is a broader ongoing debate about whether all FSS should be primarily ‘lumped’ together or ‘splitted’ into different entities, based on the evidence that they overlap frequently and have a lot of common features (such as high comorbidity with anxiety and depression) without, however, being completely identical and/or overlapping.41,44–46 Therefore, rather than only looking for an ever increasing number of FGID and subgroups within these FGID, we should also attempt to adopt a complementary broader perspective and start looking for subgroups within the category of FSS as a whole, over the boundaries of medical specialist (sub)disciplines. This may bring us closer to a pathophysiology-based classification of FSS and may especially provide important new information on central biopsychosocial processes that may be common to all FSS.47 It would, for example, be interesting to test whether the existence of ‘primary and secondary somatization’ clusters can be confirmed in a mixed FSS sample. The findings in fibromyalgia discussed above, which are consistent with the present study, provide a further indication that this may be a worthwhile approach. This does not mean, however, that research on potentially relevant peripheral mechanisms in each group of FSS separately should be neglected; we believe both approaches could be complementary.

In summary, we have presented an empirical alternative for gastroduodenal symptom-based subgroups within the FGD as proposed by the Rome III criteria, based on a combination of gastric sensitivity, anxiety & depression, ‘somatization’ and total gastroduodenal symptoms score. The results may be relevant towards future revisions of the FGD classification system, but also have broader consequences towards the debate on the nature of ‘somatization’ (primary phenomenon or secondary to psychiatric comorbidity). Finally, adopting a similar strategy in which a combination of biological and psychosocial factors as well as a broad range of somatic symptoms is used as the basis for clustering in the FSS as a whole may provide important new information on different central processes that are common to all FSS. For example, it may indicate that the relevant boundaries for subgrouping FSS do not coincide with the boundaries of current medical specialist disciplines, which may have an important impact on the difficult debate on how to conceptualize, classify and diagnose FSS.


The authors wish to thank Prof. Dr. An Carbonez, Leuven Statistics Research Centre (LStat), for excellent statistical advice. Dr. Van Oudenhove is a Postdoctoral Research Fellow of the Research Foundation-Flanders (FWO-Vlaanderen). This study was funded with a grant from the Research Foundation-Flanders and a Methusalem grant from the University of Leuven to Prof. Tack.

Author Contribution

LVO, JVBD and JT contributed to study design; LVO contributed to the analysis of the data; LVO and JT contributed to writing of the paper; JVDB and LH contributed to the data collection; RV, LH and JT contributed to patient recruitment.