Symptom clustering in subjects with and without diabetes mellitus

THE AMERICAN JOURNAL OF GASTROENTEROLOGY © 2003 by Am. Coll. of Gastroenterology Published by Elsevier Science Inc.

Vol. 98, No. 2, 2003 ISSN 0002-9270/03/$30.00 doi:10.1016/S0002-9270(02)05904-X

Symptom Clustering in Subjects With and Without Diabetes Mellitus: A Population-Based Study of 15,000 Australian Adults Johann Hammer, M.D., Stuart Howell, B.A. (Hons) M.D., Peter Bytzer, B.A. (Hons) M.D., Michael Horowitz, M.B.B.S., Ph.D., F.R.A.C.P., and Nicholas J. Talley, M.D., Ph.D., F.A.C.G. Abteilung fu¨r Gastroenterologie und Hepatologie, Universita¨tsklinik fu¨r Innere Medizin IV, Vienna, Austria; Department of Medicine, University of Sydney, Nepean Hospital, Sydney; and Department of Medicine, University of Adelaide, Royal Adelaide Hospital, Adelaide, Australia

OBJECTIVES: GI symptoms form distinct symptom clusters in community samples when factor and cluster analysis is applied. However, this has not been studied in diabetic populations, despite clear evidence that GI complaints are common in patients with diabetes mellitus (DM). This study aimed to describe clustering of GI symptoms among individuals with and without diabetes mellitus, and to describe associations of symptom clustering in diabetes mellitus, with self-reported glucose control and treatment. METHODS: A large population survey (n ⫽ 15,000) was used to identify a cohort with diabetes mellitus. Items assessing therapy and quality of glycemic control were included, as were those assessing 16 common GI symptoms. Latent GI symptom factors were extracted by factor analysis and used in a k-means cluster analysis. The latter serves to group individuals according to commonalities in symptom profiles. The association of cluster group membership to glycemic control and diabetic treatment was described by logistic regression. RESULTS: Factor analysis identified four latent symptom factors, which accounted for 69.3% of the total variance. These were labeled Upper GI/Dysmotility, Diarrhea, Constipation, and Vomiting/Nausea. The k-means analysis produced a five-cluster solution, which included a “health” group and four “diseased” groups, each identified by a predominant symptom: Upper GI/Dysmotility symptoms, Nausea/Vomiting, Diarrhea, and Constipation. After adjustment for age and gender, poor glycemic control predicted membership in all disease clusters, when compared separately with the health group. Oral hypoglycemic drugs predicted membership in the Nausea/Vomiting cluster (OR ⫽ 5.13) when used alone, and membership in the Nausea/ Vomiting (OR ⫽ 10.12) and Upper GI/Dysmotility cluster (OR ⫽ 10.12) when used in combination with insulin. CONCLUSION: Diabetes can be grouped according to common GI symptoms. Glycemic control and treatment for DM predict membership of symptom clusters. (Am J Gastroen-

terol 2003;98:391–398. © 2003 by Am. Coll. of Gastroenterology)

INTRODUCTION It is now generally accepted (albeit only recently) that complications involving the GI tract represent an important cause of morbidity in patients with diabetes mellitus (1, 2). Although patients with diabetes mellitus often complain of GI symptoms that have no identifiable structural cause (3– 13) and affect quality of life adversely (14), the prevalence and determinants of GI symptoms in patients with diabetes remain controversial. Studies in some outpatient samples have reported prevalence estimates of 70% or higher in patients with type 1 or type 2 diabetes (6, 15, 16); other outpatient studies have suggested much lower prevalence rates of GI symptoms in diabetic patients (17–21), and prevalence estimates have often also been lower in population-based studies (21–27). These differences probably relate in part to the methodology applied and the types of populations studied. Although GI symptoms were usually assessed by interview or by standard questionnaire, the criteria applied to identify relevant symptoms differed between the studies, and few studies compared symptoms in diabetic patients with adequately matched controls. There is also some controversy as to whether GI symptoms are more common in diabetic samples than they are in the general population. Whereas some population-based studies have reported that symptoms such as anorexia, vomiting, abdominal pain, and constipation are more common in type I and/or type II diabetics (22–24), others have failed to detect any differences when comparing individuals with diabetes mellitus with their respective control samples (25– 27). Several mechanisms have been implicated in the pathogenesis of GI symptoms, including Helicobacter pylori infection (28), psychological factors (29), and impaired sensory function (30, 31). However, disordered motor function resulting from autonomic (vagal) neuropathy (32, 33) and

392

Hammer et al.

abnormal blood glucose levels seem to be most important in diabetes (22). Acute changes in blood glucose concentrations (independent of neuropathy) can have a profound effect on motor function throughout the GI tract (34, 35) and modulate the perception of sensations arising from the GI tract (36 –39). An association between GI symptoms and glycemic control, as assessed by HbA1c levels, is also supported by cross-sectional epidemiologic studies (22, 40). However, overall, the association between symptoms and putative etiologic factors in diabetes have been relatively weak. Recent studies have shown that GI symptoms form distinct clusters in community samples (41– 47). Because a common etiology is implied in symptoms that consistently form clusters, this approach may prove useful in developing theoretical frameworks for studying the pathologic mechanisms that underlie GI complaints. Furthermore, a comparative, population-based study conducted in four nations (United States, Germany, Sweden, and Australia) has suggested that key clusters are formed from similar symptoms in Western cultures (47). It remains uncertain as to whether GI symptoms form similar clusters in subjects with diabetes mellitus, as in community-based samples. Further, the relationship between symptom clustering in subjects with diabetes mellitus and established risk factors for diabetic complications remains to be quantified. Accordingly, this study aimed to describe clustering of GI symptoms among individuals with and without diabetes mellitus. We also aimed to describe associations of symptom clustering in diabetes mellitus, with self-reported glucose control and treatment.

MATERIALS AND METHODS Subjects These data were collected as part of a larger study of GI symptoms in diabetes mellitus (24). The study was conducted in the catchment area of the Wentworth Area Health Service. This is a state government health authority, which serves the Penrith and Blue Mountains regions of western Sydney. This area has a population of 155,258 and is demographically similar to the Australian population as a whole, according to 1996 census data. The Federal Electoral Commission provided names and addresses for a random sample of 15,000 individuals. These were selected from the electoral rolls of all local government authorities that fall within the boundaries of the Wentworth Area Health Service. The sample was gender-stratified, and equal numbers of males and females were selected. In Australia, all adults aged 18 yr or older are required by law to be registered on the electoral rolls. A two-page symptom questionnaire was sent to all subjects. A $1.00 lottery ticket was included as a method of improving survey response rates (48). Two reminder letters were sent at 3-wk intervals. The study was approved by the Research and Ethics Committee of the Wentworth Area Health Service.

AJG – Vol. 98, No. 2, 2003

Assessment of Symptoms The questionnaire contained 16 items concerning the frequency of troublesome GI symptoms over a 3-month presurvey period. The frequency of symptoms was recorded on a five-point Likert scale, with the response options being “not at all,” “rarely,” “sometimes,” “often,” and “very often.” Symptoms that were not completely self-explanatory were anchored to a standard description, consistent with the Rome II Criteria (49, 50). The response options of “often” and “very often” were used to identify a symptom as present. The subjects were asked to rate control of their blood glucose levels in general on a five-point Likert scale that included the options “excellent control,” “good control,” “fair control,” “poor control,” and “very poor control.” This scale has been shown to correlate significantly with HbA1c (40). The questionnaire also contained questions regarding treatment of diabetes; specifically, questions asking whether insulin and/or oral medication was used for treatment and which medication was used. Subjects were classified as having type I diabetes if their age at diagnosis was less than 30 yr and they had used insulin ever since they were diagnosed with diabetes (16). All other diabetic subjects were classified as having type 2 diabetes. The prevalence of GI symptoms in diabetes mellitus and the relationship between symptoms and socioeconomic status has been reported elsewhere (21, 24). Statistical Analysis Cluster analysis is a generic term, which describes a subset of statistical procedures that can be used to create a classification. These procedures start with a data set, which contains information (e.g., symptoms) about a sample of entities (e.g., individuals) and attempts to sort these entities into relatively homogenous and mutually exclusive groups (or clusters). In this article, cluster analysis is used to sort individuals into “disease” clusters or groups, based on similarities in the symptoms that they have reported. The terms cluster and cluster membership are synonymous with the more common epidemiologic terms group and group membership. Indeed, these terms can be used interchangeably in the current setting. We have elected to use the term cluster in lieu of group to maintain consistency with earlier literature that has reported these techniques. The complexity of a cluster solution is determined, at least in part, by the number of items used to form a classification. For example, a cluster solution that is based on 16 symptom items is likely to be more complex than one based on fewer items. Indeed, when cluster solutions are derived from a large number of items, two or more clusters are often indistinguishable in terms of their underlying content or structure. Thus, we have adopted the following two-stage approach as a method of identifying the most parsimonious cluster solution:

AJG – February, 2003

1. Data reduction: Factor analysis was applied to the 16 symptom items in an effort to identify the underlying latent factors that these symptoms represent. The factors were extracted using principal components analysis, with the criterion of an eigenvalue greater than or equal to one. Varimax rotation was applied to extract a solution containing statistically independent (i.e., orthogonal) factors. 2. Cluster analysis: A k-means cluster analysis was applied, using factors extracted from the principal components analysis as the basis for forming the cluster solution. The analysis commenced with a three-cluster solution and proceeded by generating increasingly complex cluster solutions (i.e., four to five clusters). The choice of three clusters as the starting point was based on the expectation of a health cluster and at least two “diseased” clusters, reflecting upper and lower GI symptom groups. Three criteria were used to select the appropriate cluster solution. First, comparisons were made of cluster membership across increasingly complex cluster solutions: if the more complex solution seemed to systematically break a large cluster into substantive subclusters, the complex solution was adopted; however, if the more complex solution seemed to randomly allocate members of several clusters to a new cluster or clusters, the simpler solution was adopted. Second, the distance metric (Euclidean distance) method was used to judge whether a more complex solution improved within-cluster homogeneity. If the average distance metric was reduced with a more complex solution, the more complex solution was favored. Third, to preserve the reliability of within-cluster estimates, no cluster could be made up of less than 5% of the entire sample. The interpretation of each cluster was aided by describing a cluster profile that comprised the mean score per factor per cluster. For each cluster, there is a series of mean scores centered about zero. A mean of zero indicates that the cluster is average (i.e., undistinguished) on that particular factor. The unit of measurement is the SD, because of the unit normal distribution of factor scores: a score of ⫾2.0 indicates that the cluster is within the top or bottom 5% in terms of that factor. Scores of less than ⫺1.0 or greater than ⫹1.0 were interpreted as indicating clear differentiation; scores between 0.5 and 1.0 (positive or negative) were interpreted as indicating possible differentiation. Subjects with no more than three missing values across all variables were included in the factor and cluster analysis. Separate solutions were developed for subjects with and without diabetes mellitus. Assessment of Risk Factors for Cluster Membership in Subjects With Diabetes Mellitus The associations between cluster membership and diabetic risk factors (self-reported glycemic control, type of diabetic control) were summarized using age- and gender-adjusted

GI Symptom Correlates in Diabetes Mellitus

393

Table 1. Factor Structure for the Sample Without Diabetes Mellitus Symptom Factor 1 Nausea Food staying in stomach Early satiety Vomiting Pain Bloating Heartburn Dysphagia Factor 2 Loose/watery stools Urgency ⬎3 bowel movement/day Fecal incontinence Factor 3 Hard/lumpy stools Blockage in the anus ⬍3 bowel movement/wk Constipation/diarrhea

Upper GI/ Dysmotility Diarrhea Constipation 0.78 0.70 0.69 0.69 0.62 0.58 0.56 0.49

0.15 0.17 0.15 0.08 0.33 0.21 0.19 0.17

0.09 0.33 0.26 ⫺0.04 0.33 0.44 0.13 0.15

0.24 0.23 0.17 0.10

0.79 0.77 0.74 0.60

0.12 0.13 0.02 0.03

0.19 0.20 0.12 0.31

0.09 0.18 ⫺0.08 0.51

0.82 0.78 0.70 0.59

n ⫽ 7955.

ORs. The ORs were obtained by comparing the cluster with the most favorable symptom profile to each remaining cluster in separate logistic regression models.

RESULTS Response Rate Of the 15,000 questionnaires mailed, 429 were returned undelivered, and 99 were sent to people who had recently died or were away from home throughout the duration of the survey. A further 102 individuals returned completed questionnaires but reported postal codes that were outside of the catchment area; these individuals were removed from the sample. Overall, 8555 eligible subjects returned a completed questionnaire, yielding a response rate of 60%. Comparisons between respondents and the population of the surveyed area indicated very similar age strata (p ⫽ 1.00, ␹2 test), but a slightly higher proportion of females among the respondents (53.5% vs 50.9%, p ⬍ 0.001). In total, 423 subjects (4.9%) reported diabetes mellitus. Diabetic subjects were older than control subjects (mean ⫾ SD: 59.5 ⫾ 14.1 yr vs 44.6 ⫾ 15.6 yr, p ⬍ 0.001); 18% of the diabetics were insulin users, and 94.5% had type 2 diabetes. Subjects Without Diabetes Mellitus Details of the factor structure for the nondiabetic sample are shown in Table 1. Three latent symptom factors were identified: these were labeled Upper GI/Dysmotility, Diarrhea, and Constipation. The minimum eigenvalue was 1.33, and the factors accounted for 37.7%, 10.4%, and 8.3% of the variance, respectively. Table 2 shows the results of the cluster analysis, which was applied to subjects from the subsample without diabetes

394

Hammer et al.

AJG – Vol. 98, No. 2, 2003

Table 2. Cluster Centers for the Six-Factor Solution Identified in the Subsample Without Diabetes Mellitus

Cluster Health Upper GI Diarrhea Constipation Upper GI with diarrhea Upper GI with constipation

Table 4. Cluster Centers for the Six-Factor Solution Identified in the Subsample With Diabetes Mellitus

Latent Symptom Factor

Latent Symptom Factor

Upper GI/ Dysmotility Diarrhea Constipation

Upper GI/ Consti- Nausea/ Dysmotility Diarrhea pation Vomiting

⫺0.38 1.68 ⫺0.56 ⫺0.55 1.41 1.32

⫺0.40 ⫺0.49 1.44 ⫺0.26 1.97 ⫺0.24

⫺0.40 ⫺0.59 ⫺0.01 1.87 ⫺0.30 1.43

Cluster Health Upper GI/dysmotility Diarrhea Constipation Nausea/vomiting

⫺0.49 1.59 0.41 0.29 ⫺0.49

⫺0.41 0.11 1.87 ⫺0.22 ⫺0.41

⫺0.37 ⫺0.39 0.04 1.85 ⫺0.37

⫺0.16 ⫺0.27 ⫺0.28 ⫺0.55 ⫺0.16

n ⫽ 396.

n ⫽ 7955.

mellitus. This describes the manner in which individuals group according to commonalities in symptom profiles, and the cluster parameters represent mean factor scores for each latent symptom variable within each cluster of individuals. The nondiabetic subsample produced a six-cluster solution, which consisted of five diseased clusters and an undifferentiated group. The undifferentiated group was defined by lower-than-average scores on all symptom factors and was accordingly labeled Health. Three of the five disease clusters were defined by a single, predominant symptom and were labeled according to that symptom: Upper GI/Dysmotility, Diarrhea, and Constipation. The remaining two clusters included a group with Upper GI/Dysmotility symptoms in conjunction with diarrhea and Upper GI/Dysmotility symptoms in conjunction with constipation. Subjects With Diabetes Mellitus Details of the factor structure for the diabetic sample are shown in Table 3. Four latent symptom factors were idenTable 3. Factor Structure for the Subsample With Diabetes Mellitus Symptom Factor 1 Bloating Food staying in stomach Pain Heartburn Early satiety Dysphagia Factor 2 Urgency Loose/watery stools ⬎3 bowel movement/ day Fecal incontinence Factor 3 Hard/lumpy stools Blockage in the anus ⬍3 bowel movement/ week Constipation/diarrhea Factor 4 Vomiting Nausea n ⫽ 396.

Upper GI/ Consti- Vomiting/ Dysmotility Diarrhea pation Nausea 0.78 0.76

0.17 0.22

0.26 0.19

0.16 0.30

0.76 0.72 0.63 0.44

0.30 0.21 0.19 0.18

0.16 0.17 0.14 0.20

0.20 ⫺0.03 0.29 0.39

0.24 0.37 0.29

0.83 0.80 0.77

0.13 0.09 ⫺0.06

0.14 0.09 0.08

0.04

0.72

0.25

0.18

0.14 0.27 0.17

0.12 0.18 ⫺0.01

0.88 0.79 0.68

0.09 0.13 0.16

0.44

0.48

0.53

0.18

0.14 0.38

0.12 0.24

0.16 0.19

0.90 0.74

tified: these were labeled Upper GI/Dysmotility, Diarrhea, Constipation, and Nausea/Vomiting. The minimum eigenvalue was 1.01, and the factors accounted for 44.8%, 10.4%, 7.8%, and 6.3% of the variance, respectively. The cluster analysis of the four latent symptom factors (Table 4) produced a five-cluster solution. This was characterized by an undifferentiated Health group and four diseased clusters. The disease clusters were each defined by higher-than-average scores on a single symptom and were labeled according to that symptom: Upper GI/Dysmotility, Nausea/Vomiting, Diarrhea, and Constipation. Risk Factors for Cluster Membership Among Subjects With Diabetes Mellitus Table 5 shows cluster characteristics among subjects with diabetes according to age, gender, and duration of diabetes. Cluster group membership was significantly related to subject age (Kruskal-Wallis ␹2 ⫽ 13.56; p ⫽ 0.01) and gender (␹2 ⫽ 12.87; p ⫽ 0.01) but was not related to duration of diabetes (Kruskal-Wallis ␹2 ⫽ 2.79; p ⫽ 0.59). In general, subjects in the Upper GI and Nausea/Vomiting clusters were younger than those in the remaining groups; “diseased” females were more likely to fall into the Constipation cluster, whereas “diseased” males were more likely to fall into the Upper GI or Diarrhea clusters. Self-Reported Glycemic Control After adjustment for age and gender, self-reported glycemic control was significantly associated with cluster membership when each symptom-based cluster was independently compared with the Health group (Table 6). Poor control was associated with a threefold increase in the odds of membership of the Upper GI/Dysmotility cluster (OR ⫽ 3.61; 95% CI ⫽ 1.14 –11.42), a sevenfold increase in the odds of belonging to the Diarrhea cluster (OR ⫽ 7.69; 95% CI ⫽ 2.44 –24.24), a fivefold increase in the odds of belonging to the Constipation cluster (OR ⫽ 5.38; 95% CI ⫽ 1.65– 17.56), and a sixfold increase in the odds of belonging to the Nausea/Vomiting cluster (OR ⫽ 6.09; 95% CI ⫽ 1.78 – 20.84). These effects remained significant after further adjustment for the type of treatment used to control diabetes (Table 6), although the magnitude of the ORs was attenuated in all comparisons.

AJG – February, 2003

GI Symptom Correlates in Diabetes Mellitus

395

Table 5. Characteristics of Clusters According to Age, Gender, and Duration of Diabetes Cluster

n Age (yr) Mean SD Gender (%) Male Female Duration (yr) Median Range

Health

Upper GI/ Dysmotility

Diarrhea

Constipation

Nausea/ Vomiting

5205

162

246

348

435

60.8 14.1

55.8 12.7

61.3 10.9

59.7 16.9

53.8 12.3

59.5 40.5

54.8 45.2

58.7 41.3

31.3 68.8

54.3 45.7

5.0 0.08–54.00

5.0 0.25–37.00

4.4 0.17–30.00

5.0 0.04–30.00

5.0 0.25–37.33

Type of Treatment Used to Control Diabetes After adjustment for age and gender, the type of treatment used, whether insulin or oral medication, was not related to cluster membership in comparisons of the Health group to the Diarrhea cluster or to the Constipation cluster (Table 7). Membership in the Upper GI/Dysmotility cluster was significantly related to the use of insulin in conjunction with hypoglycemic medication (OR ⫽ 3.24; 95% CI ⫽ 1.08 – 9.75), whereas membership in the Nausea/Vomiting cluster was significantly related to insulin use in combination with oral hypoglycemic medication (OR ⫽ 10.12; 95% CI ⫽ 1.88 –54.34) and to the use of oral hypoglycemic medication alone (OR ⫽ 5.13; 95% CI ⫽ 1.43–18.36). These effects remained significant after further adjustment for self-reported glycemic control; in addition, insulin use alone was significantly related to membership in the Nausea/Vomiting cluster after this adjustment (OR ⫽ 6.71; 95% CI ⫽ 1.16 – 38.74).

DISCUSSION Numerous studies have shown that GI symptom groups exist in the general population when factor and cluster

analysis is applied (41– 47). However, the findings have often been inconsistent, although this may reflect differences in the questionnaires used and in the populations studied. Until now, it has not been clear as to whether GI symptoms form clusters in populations with diabetes mellitus. We have found that GI symptoms form clusters in subjects with (largely type II) diabetes mellitus and that these differ substantially from the clusters that form in nondiabetic subjects. Three symptom clusters were common to both diabetics and nondiabetics, including an Upper GI/ Dysmotility cluster, a Diarrhea cluster, and a Constipation cluster. However, a Nausea/Vomiting cluster was found among diabetic subjects but not among nondiabetics, whereas two Upper GI/Dysmotility subgroups (one with concomitant diarrhea and the other with constipation) were found among nondiabetics but not among diabetics. These results, at least in part, reflect differences in the underlying latent symptom factors that formed the basis for cluster analysis: nausea and vomiting loaded on the Upper GI/ Dysmotility factor among nondiabetics but formed a separate factor among those with diabetes mellitus. There were

Table 6. Logistic Regression Models for Cluster Membership and Self-Rated Glycemic Control Model 1* Upper GI/dysmotility‡ Good control Poor control Diarrhea‡ Good control Poor control Constipation‡ Good control Poor control Nausea/vomiting‡ Good control Poor control

Model 2†

OR

95% CI

OR

95% CI

1.00 3.61§

1.14–11.42

1.00 3.28§

1.01–10.71

1.00 7.69ⱍⱍ

2.44–24.24

1.00 7.55¶

2.35–24.18

1.00 5.38¶

1.65–17.56

1.00 4.71§

1.40–15.81

1.00 6.09¶

1.78–20.84

1.00 4.89§

1.31–18.32

* Adjusted for age and gender. † Adjusted for age, gender, and type of treatment for diabetes mellitus. ‡ Compared with the Health group. § p ⬍ 0.05. ⱍⱍ p ⬍ 0.001. ¶ p ⬍ 0.01.

396

Hammer et al.

AJG – Vol. 98, No. 2, 2003

Table 7. Logistic Regression Models for Cluster Membership by Type of Treatment for Diabetes Mellitus Model 1* Upper GI/dysmotility‡ Diet alone Insulin Insulin/oral hypoglycemics Oral hypoglycemics Diarrhea‡ Diet alone Insulin Insulin/oral hypoglycemics Oral hypoglycemics Constipation‡ Diet alone Insulin Insulin/oral hypoglycemics Oral hypoglycemics Nausea/vomiting‡ Diet alone Insulin Insulin/oral hypoglycemics Oral hypoglycemics

Model 2†

OR

95% CI

OR

95% CI

1.00 0.31 3.24§

0.08–1.12 1.08–9.75

1.00 0.31 3.14§

0.09–1.14 1.04–9.45

1.06

0.55–2.08

1.00

0.51–1.97

1.00 0.69 1.00

0.17–2.78 0.19–5.12

1.00 0.74 1.25

0.16–3.02 0.24–6.52

1.33

0.63–2.79

1.41

0.65–3.09

1.00 1.21 0.72

0.44–3.31 0.14–3.71

1.00 1.05 0.86

0.36–3.04 0.17–4.46

0.66

0.31–1.41

0.46

0.34–1.62

1.00 3.87 10.12储

0.82–18.13 1.88–54.34

1.00 6.71§ 12.39§

1.16–38.74 1.85–83.21

5.13§

1.43–18.36

7.12§

1.56–32.46

* Adjusted for age and gender. † Adjusted for age, gender, and self-rated glycemic control. ‡ Compared with the Health group. § p ⬍ 0.05. 储 p ⬍ 0.01.

too few cases of type I diabetes to separately analyze this group. The lack of a common cluster solution suggests that the etiology of GI complaints is different in diabetes mellitus, when compared with the general population. In this study, we found that poor self-reported glycemic control increased the odds of belonging to a GI symptom cluster. This was observed for all symptom clusters, when each was independently compared with an asymptomatic health group, with the ORs ranging from 3.61 to 7.69. We also found that diabetic treatment was associated with membership in the Nausea/Vomiting cluster and, to a lesser extent, in the Upper GI/Dysmotility cluster. The use of oral hypoglycemic medications, in conjunction with insulin, was associated with membership in the Nausea/Vomiting and Upper GI/Dysmotility clusters. Oral medication use alone was associated with membership in the Nausea/Vomiting cluster, as was insulin use alone, after adjustment for self-reported glucose control. In the present analysis, we have grouped oral medications into one group. This might be a possible limitation of the analysis, because drugs such as metformin might be associated with GI disorders more often than other oral drugs. However, no association has been previously found between metformin use and upper GI symptoms (18). The etiology of GI symptoms in diabetes mellitus has been a matter of research since the middle of the 20th century. The early literature has emphasized the high prev-

alence of GI symptoms in diabetic patients with autonomic neuropathy and gastric dysmotility (51), although some have suggested that this concept is overly simplistic (1, 2); specific attempts to identify a direct link between upper GI symptoms and autonomic neuropathy have met with little success (15), and the link between gastric motor disorder and symptoms has generally been modest (52). For example, most studies have failed to demonstrate strong associations between symptoms and either delayed (53–55) or accelerated (56) gastric emptying. However, the contribution of motility abnormalities to symptoms may still be important in some patients with diabetes mellitus. The quality of glycemic control has been suggested previously as an important factor in symptom onset in studies of pathophysiology as well as in epidemiologic studies. A number of studies have shown that acute changes in blood glucose concentration can have a profound effect on motor function throughout the GI tract in both normal subjects and patients with diabetes mellitus (34, 35, 57). Recent studies have demonstrated that blood glucose concentration may also modulate the perception of sensations arising from the GI tract (36 –39). In a population-based study, glycemic control, measured by HbA1c was predictive of upper but not lower GI symptoms in 892 randomly selected patients from a diabetes support group (40). A HbA1c level of 10 mg% or more was associated with an increased prevalence of nausea and dysphagia in patients with diabetes (40). The present

AJG – February, 2003

study confirms that poor glycemic control is related with GI symptoms in diabetes mellitus. Although we applied a selfreported measure, we have previously shown that the results correlate significantly with objective assessments of glycemic control, including HbA1c (40). Poor glycemic control was significantly related to all GI symptom clusters that were identified compared with the asymptomatic cluster. GI symptoms are frequently reported as side effects of oral hypoglycemic drugs. In clinical practice especially, biguanides are often seen as a cause of GI symptoms in diabetics. Dandona et al. (18) observed that patients with diabetes mellitus type 2 who received biguanides had a higher prevalence of diarrhea (20%), whereas the prevalence of diarrhea in patients who were taking insulin or other oral hypoglycemics was low (6%). Similarly, in a cohort study including over 950 patients with type 2 diabetes, metformin was associated with chronic diarrhea and with fecal incontinence (58). No other oral hypoglycemic could be identified to be associated with GI symptoms in this study (58). The present results confirm oral hypoglycemic drugs as risk factors for GI symptoms; however, use of oral therapy was only associated with membership in the Nausea/Vomiting cluster, not with membership in Diarrhea cluster and Constipation cluster. A combination of oral therapy and insulin increased the odds of membership in the Nausea/ Vomiting cluster, even though this should improve glycemic control. Our observations most likely reflect the presence of more advanced disease in type 2 diabetes and hence higher rates of underlying autonomic neuropathy and gastroparesis, rather than a causal link between treatment and symptoms. We have not assessed diabetes complications in this study, which also might be etiologic factors for GI symptoms in diabetes mellitus. Future studies will be required to determine if other proposed risk factors of GI symptoms in patients with diabetes mellitus, including autonomic neuropathy, GI dysmotility, visceral hypersensitivity, and psychological factors are related to cluster membership in the diabetes population, and whether different clusters exist in type 1 diabetics. In conclusion, this is, to our knowledge, the first study to demonstrate GI symptom clusters in diabetic patients that are associated with several risk factors, namely glycemic control and type of therapy. Cluster analysis can be used as a tool to provide a theoretical framework for the underlying mechanisms in the pathogenesis of GI symptoms in patients with diabetes mellitus.

ACKNOWLEDGMENTS Supported by research grants from The National Health and Medical Research Council (Australia) and Diabetes Australia. Reprint requests and correspondence: Nicholas J. Talley, M.D., Ph.D., F.A.C.G., The Nepean Hospital–University of Sydney, Department of Medicine, Level 5 South Block, P.O. Box 63, 2751 Penrith, NSW, Australia. Received Mar. 14, 2002; accepted July 16, 2002.

GI Symptom Correlates in Diabetes Mellitus

397

REFERENCES 1. Horowitz M, Fraser R. Disordered gastric motor function in diabetes mellitus. Diabetologia 1994;37:543–51. 2. Horowitz M, Wishart JM, Jones KL, Hebbard GS. Gastric emptying in diabetes: An overview. Diabetic Med 1996;13: S16 –22. 3. Katz LA, Spiro HM. Gastrointestinal manifestations of diabetes. N Engl J Med 1966;275:1350 –61. 4. Goyal RK, Spiro HM. Gastrointestinal manifestations of diabetes mellitus. Med Clin N Am 1971;55:1031–44. 5. Taub S, Mariani A, Barkin JS. Gastrointestinal manifestations of diabetes mellitus. Diabetes Care 1979;2:437–47. 6. Feldman M, Schiller LR. Disorders of gastrointestinal motility associated with diabetes mellitus. Ann Intern Med 1983;98: 378 –84. 7. Atkinson M, Hosking DJ. Gastrointestinal complications of diabetes mellitus. Clin Gastroenterol 1983;2:633–50. 8. Yang R, Arem R, Chan L. Gastrointestinal tract complications of diabetes mellitus. Arch Intern Med 1984;144:1251–6. 9. Niakan E, Harati Y, Comstock JP. Diabetic autonomic neuropathy. Metabolism 1986;35:224 –34. 10. Nompleggi D, Bell SJ, Blackburn GL, Bistrian BR. Overview of gastrointestinal disorders due to diabetes mellitus: Emphasis on nutritional support. J Parent Ent Nutr 1989;13:84 –91. 11. Rothstein RD. Gastrointestinal motility disorders in diabetes mellitus. Am J Gastroenterol 1990;85:782–5. 12. Locke GR, III. Epidemiology of gastrointestinal complications of diabetes mellitus. Eur J Gastroenterol Hepatol 1995;7: 711–6. 13. Camilleri M, Bharucha AE. Gastrointestinal dysfunction in neurologic disease. Semin Neurol 1996;16:203–16. 14. Talley NJ, Young L, Hammer J, et al. Impact of chronic gastrointestinal symptoms in diabetes mellitus on health-related quality of life. Am J Gastroenterol 2001;96:71–6. 15. Clouse RE, Lustman PJ. Gastrointestinal symptoms in diabetic patients: Lack of association with neuropathy. Am J Gastroenterol 1989;84:868 –72. 16. Ko GT, Chan WB, Chan JC, et al. Gastrointestinal symptoms in Chinese patients with type 2 diabetes mellitus. Diabet Med 1999;16:670 –4. 17. Maxton DG, Whorwell PJ. Functional bowel symptoms in diabetes—the role of autonomic neuropathy. Postgrad Med J 1991;67:991–3. 18. Dandona P, Fonseca V, Mier A, Beckett AG. Diarrhea and metformin in a diabetic clinic. Diabetes Care 1983;6:472–4. 19. Keshavarzian A, Iber FL. Gastrointestinal involvement in insulin-requiring diabetes mellitus. J Clin Gastroenterol 1987; 9:685–92. 20. Maser RE, Pfeifer MA, Dorman JS, et al. Diabetic autonomic neuropathy and cardiovascular risk. Pittsburgh epidemiology of diabetes complications study III. Arch Intern Med 1990; 150:1218 –22. 21. Bytzer PM, Howell S, Leemon M, et al. Low socio-economic class is a risk factor for upper and lower gastrointestinal symptoms: A population based study in 15,000 adults. Gut 2001;49:66 –72. 22. Schvarcz E, Palme´ r M, Ingberg CM, et al. Increased prevalence of upper gastrointestinal symptoms in long-term type 1 diabetes mellitus. Diab Med 1996;13:478 –81. 23. Spånge´ us A, El-Salhy M, Suhr O, et al. Prevalence of gastrointestinal symptoms in young and middle-aged diabetic patients. Scand J Gastroenterol 1999;34:1196 –202. 24. Bytzer PM, Talley NJ, Leemon M, et al. Diabetes mellitus is associated with an increased prevalence of gastrointestinal symptoms: A population-based survey of 15,000 adults. Arch Intern Med 2001;161:1989 –96.

398

Hammer et al.

25. Janatuinen E, Pikkarainen P, Laakso M, Pyorala K. Gastrointestinal symptoms in middle-aged diabetic patients. Scand J Gastroenterol 1993;28:427–32. 26. Ricci JA, Siddique R, Stewart WF, et al. Upper gastrointestinal symptoms in a U.S. national sample of adults with diabetes. Scand J Gastroenterol 2000;35:152–9. 27. Maleki D, Locke GR, III, Camilleri M, et al. Gastrointestinal tract symptoms among persons with diabetes mellitus in the community. Arch Intern Med 2000;160:2808 –16. 28. Oldenburg B, Diepersloot RJA, Hoekstra JBL. High seroprevalence of Helicobacter pylori in diabetes mellitus patients. Dig Dis Sci 1996;41:458 –61. 29. Clouse RE, Lustman PJ, Reidel WL. Correlation of oesophageal motility abnormalities with neuropsychiatric status in diabetics. Gastroenterology 1986;90:1146 –54. 30. Holtmann G, Goebell H, Talley NJ. Gastrointestinal sensory function in functional dyspepsia. Gastroenterology 1995;109: 331–2. 31. Samson M, Salet GAM, Roelofs JMM, et al. Compliance of the proximal stomach and dyspeptic symptoms in patients with type I diabetes mellitus. Dig Dis Sci 1995;40:2037–42. 32. Mearin F, Malagelada J-R. Gastroparesis and dyspepsia in patients with diabetes mellitus. Eur J Gastroenterol Hepatol 1995;7:717–23. 33. Horowitz M, Edelbroek M, Fraser R, et al. Disordered gastric motor function in diabetes mellitus. Recent insights into prevalence, pathophysiology, clinical relevance, and treatment. Scand J Gastroenterol 1991;26:673–84. 34. Hebbard GS, Sun WM, Dent J, Horowitz M. Acute hyperglycaemia increases proximal gastric compliance. Gastroenterology 1994;106:A509. 35. Schvarcz E, Palme´ r M, Åman J, et al. Hypoglycaemia increases the gastric emptying rate in patients with type 1 diabetes mellitus. Diabet Med 1993;10:660 –3. 36. Hebbard GS, Sun WM, Dent J, Horowitz M. Hyperglycaemia affects gastric motor and sensory function in normal subjects. Eur J Gastroenterol Hepatol 1996;8:211–7. 37. Rayner CK, Samsom M, Jones KL, Horowitz M. Relationships of upper gastrointestinal motor and sensory function with glycemic control. Diabetes Care 2001;24:371–81. 38. Hebbard GS, Samsom M, Sun WM, et al. Hyperglycemia affects proximal gastric motor and sensory function during small intestinal nutrient infusion. Am J Physiol 1996;271: G814 –9. 39. Chey WD, Kim M, Hasler W, Owyang C. Hyperglycaemia alters perception of rectal distension and blunts the recto-anal inhibitory reflex in healthy volunteers. Gastroenterology 1995; 108:1700 –8. 40. Bytzer PM, Talley NJ, Young LJ, et al. Gastrointestinal symptoms in diabetes mellitus are associated with diabetic complications and poor glycemic control. Australian Gastroenterology Week. J Gastroenterol Hepatol 2000;15(suppl):J41. 41. Whitehead WE, Crowell MD, Bosmajian L, et al. Existence of irritable bowel syndrome supported by factor analysis of

AJG – Vol. 98, No. 2, 2003

42. 43. 44.

45. 46. 47.

48. 49. 50. 51. 52. 53. 54. 55. 56. 57.

58.

symptoms in two community samples. Gastroenterology 1990; 98:336 –40. Taub E, Cuevas JL, Cook EW, III, et al. Irritable bowel syndrome defined by factor analysis: Gender and race comparisons. Dig Dis Sci 1995;40:2647–55. Kay L, Jorgensen T. Redefining abdominal syndromes. Results of a population-based study. Scand J Gastroenterol 1996; 31:469 –75. Schlemper RJ, Van der Werf SDJ, Vandenbroucke JP, et al. Peptic ulcer, non-ulcer dyspepsia and irritable bowel syndrome in The Netherlands and Japan. Scand J Gastroenterol 1993;28(suppl 200):33–41. Robbins JM, Kirmayer LJ, Hemami S. Latent variable models of functional somatic distress. J Nerv Ment Dis 1997;185: 606 –15. Talley NJ, Boyce P, Jones M. Identification of distinct upper and lower gastrointestinal symptom groupings in an urban population. Gut 1998;42:690 –5. Talley NJ, Holtmann G, Agre´ us L, Jones M. Gastrointestinal symptoms and subjects cluster into distinct upper and lower groupings in the community: A four nations study. Am J Gastroenterol 2000;95:1439 –47. Kalantar JS, Talley NJ. The effects of lottery incentive and length of questionnaire on health survey response rates: A randomized study. J Clin Epidemiol 1999;52:1117–22. Thompson WG, Longstreth GF, Drossman DA, et al. Functional bowel disorders and functional abdominal pain. Gut 1999;45(suppl II):43–7. Talley NJ, Stanghellini V, Heading RC, et al. Functional gastroduodenal disorders. Gut 1999;45(suppl II):37–42. Rundles RW. Diabetic neuropathy: General review with report of 125 cases. Medicine 1945;24:111–60. Wegener M, Bo¨ rsch G, Schaffstein J, et al. Gastrointestinal transit disorders in patients with insulin-treated diabetes mellitus. Dig Dis 1990;8:23–6. Horowitz M, Harding PE, Maddox AF, et al. Gastric and oesophageal emptying in patients with type 2 (non-insluindependent) diabetes mellitus. Diabetologia 1989;32:151–9. Horowitz M, Maddox AF, Wishart JM, et al. Relationships between oesophageal transit and solid and liquid gastric emptying in diabetes mellitus. Eur J Nucl Med 1991;18:229 –34. Keshavarzian A, Iber FL, Vaeth J. Gastric emptying in patients with insulin-requiring diabetes mellitus. Am J Gastroenterol 1987;82:29 –35. Nowak TV, Johnson CP, Kalbfleisch JH, et al. Highly variable gastric emptying in patients with insulin dependent diabetes mellitus. Gut 1995;37:23–9. Schvarcz E, Palme´ r M, Åman J, et al. Physiological hyperglycemia slows gastric emptying in normal subjects and patients with insulin-dependent diabetes mellitus. Gastroenterology 1997;113:60 –6. Bytzer P, Talley NJ, Jones MP, Horowitz M. Oral hypoglycemic drugs and gastrointestinal symptoms in diabetes mellitus. Aliment Pharmacol Ther 2001;15:137–42.