SEARCH

SEARCH BY CITATION

Keywords:

  • urination disorders;
  • nocturia;
  • urinary incontinence;
  • symptoms;
  • cluster analysis;
  • epidemiology

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix

OBJECTIVES

To classify lower urinary tract symptoms (LUTS) in a large, representative sample of men in the USA by means of cluster analysis and to investigate risk factors and comorbidities associated with the resulting cluster patterns.

SUBJECTS AND METHODS

A combination of hierarchical and non-hierarchical cluster methods was used to assign men with LUTS in the Boston Area Community Health (BACH) study to symptom-based categories or clusters. Of the 2301 men in the BACH study, those reporting one or more of 14 common LUTS (1592 men) were included in the analysis. The prevalence and frequency of symptoms in each cluster was assessed, in addition to the demographic, lifestyle risk factors, comorbidities, quality of life, and interference with activities of daily living associated with each cluster. We used anova methods for assessing cluster effects on continuous variables, and cross-classification and chi-square tests for categorical measures. Internal validity of the cluster solution was tested by means of a split-half replication, and external validity by comparison with previously published data.

RESULTS

Five clusters were identified among symptomatic men. About half of the symptomatic men were assigned to Cluster 1, which included individuals with a low prevalence and frequency of urological symptoms and a correspondingly low level of interference with activities of daily living. There were intermediate levels of symptom frequency and prevalence in Clusters 2–4, which were characterized by mixed patterns of voiding, storage and postvoiding symptoms. Cluster 5 consisted of predominantly older men (mean age 58.9 years), with a high prevalence and frequency of urological symptoms with a mean (sd) number of symptoms of 9.9 (2.1), and with elevated levels of comorbid cardiovascular disease (P < 0.001). These men also had higher rates of kidney and bladder infections and previous urological surgery. Men with increased waist circumference and more sedentary lifestyles were over-represented in the more symptomatic clusters.

CONCLUSION

Cluster analysis provides an empirically based method for categorizing men with LUTS. These findings provide a new framework for examining aetiological pathways and mechanisms, the potential impact of and consequences for comorbid conditions, and for assessing prognosis and outcomes associated with common urological disorders.


Abbreviations
UI

urinary incontinence

BACH

Boston Area Community Health (survey)

ICSI

Interstitial Cystitis Symptom Index

BMI

body mass index

PASE

Physical Activity Scale for the Elderly

SES

socioeconomic status

CVD

cardiovascular disease.

INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix

LUTS include a wide range of storage, voiding and postvoiding symptoms in men and women, which may be related to anatomical, physiological or unknown causes [1–3]. Benign prostate enlargement is a common cause of LUTS in ageing men, although symptoms of voiding dysfunction are often reported in the absence of histological or anatomical changes in the prostate [4,5]. Similarly, detrusor overactivity and BOO may be present in varying degrees in men with and without symptoms [6,7]. Male LUTS are currently divided into storage, voiding and postvoiding symptoms, although older men frequently report a combination of urological symptoms [6,7]. These symptoms are often bothersome and have significant impact on daily living in affected individuals [8].

Clinical research and observational studies have been hampered by diagnostic imprecision and the need for a standardized terminology and classification of LUTS. Current diagnostic terminology includes a combination of symptom- and mechanism-based concepts, such as overactive bladder syndrome, benign prostatic obstruction, and stress urinary incontinence (UI). These diagnostic categories lack precise definitions, and show substantial overlap in both presenting symptoms and objective findings [9,10]. Management approaches, including surgical and nonsurgical treatments, provide varying degrees of symptom relief for patients with multiple, heterogeneous symptoms across diagnostic categories [11–13]. As symptoms and their associated interference with daily activities are the basis for help-seeking and healthcare utilization, we propose to develop a new symptom-based classification using an empirically grounded, statistical approach, known as cluster analysis [14,15]. A symptom-based classification of this type can be used for identifying common mechanisms, risk factors, or associated comorbid conditions, and for evaluating disease progression or outcomes associated with treatment.

Cluster analysis is a statistical method for categorizing groups of individuals with similar properties or characteristics [14]. The method has been used in previous studies to classify patients with complex medical or psychiatric disorders and to delineate diagnostic categories and common symptom profiles [15–18]. Only one study to date has applied this method to the classification of urological symptoms in men [19]. The present study extended the use of cluster analysis to men with symptoms of LUTS enrolled in the Boston Area Community Health (BACH) Survey. The BACH study is a community-based, random population sample, which includes approximately equal numbers of White, Black and Hispanic participants, and is unique in the breadth of sociodemographic, lifestyle and health history information obtained for each participant [20–22].

Specific objectives of the current study were: (i) to assess cluster-symptom distributions among symptomatic men in our sample; (ii) to replicate a previous cluster analysis of urological symptoms in men; and (iii) to investigate the relationship of symptom clusters to major risk factors and covariates in the BACH study. We report results for men in this paper; while cluster analysis results for the female participants in BACH are reported in a separate paper [23].

SUBJECTS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix

STUDY POPULATION

The BACH study is supported by the USA National Institutes of Health (National Institute of Diabetes and Digestive and Kidney Diseases), and is a population-based study of urological symptoms among 5506 men and women aged 30–79 years who are residents of the city of Boston, MA, USA. A multistage stratified cluster sampling design was used for the purposes of recruiting approximately equal numbers of persons to pre-specified age groups (30–39, 40–49, 50–59, 60–79 years), race and ethnic groups (Black, White and Hispanic) and gender. Interviews for 63.3% of eligible subjects were completed, with a resulting study population of 2301 men and 3205 women comprised of 1770 Blacks, 1877 Hispanics and 1859 Whites. After written informed consent, data were collected between April 2002 and June 2005 during a 2-h, in-person interview conducted by a trained, bilingual interviewer, usually in the home. Information collected included urological symptoms, medical conditions, use of over-the-counter and prescription drugs, anthropometric measures, lifestyle factors, education, income and psychosocial factors. All protocols and procedures were approved by New England Research Institutes’ Institutional Review Board. Further details of study design and procedures are available [22].

UROLOGICAL SYMPTOM DATA AND CLASSIFICATION

Fourteen non-pain-related urological symptoms in the study were selected for inclusion. These items were selected as representative of items assessed in the earlier Canadian study [19] and in the EPIC study cluster analysis [24]. The latter is reported in an accompanying article [25]. Participants who were considered symptomatic for these 14 symptoms (as detailed in Table 1) were eligible for the cluster analysis. UI was defined as reporting any leakage over the past year of at least once per month. Three items were assessed with a time-frame of the past 7 days: nocturia (the wording of the BACH nocturia item is based on the IPSS, but the time-frame for this question differs from the six other IPSS based items: ‘last seven days’ vs ‘the previous month’) and two subtypes of UI (urge UI, leaking when unable to get to the toilet soon enough; and stress UI, leaking with physical activity including coughing and sneezing). The remaining urological symptoms were assessed on a 6-point ordinal scale in reference to the past month using the IPSS, and included incomplete emptying, intermittency of urination, weak urinary stream, straining to begin urination, frequency (repeat urination within 2 h), and urgency (difficulty postponing urination). Two postvoiding symptoms were assessed over the past month: dribbling after urination and wet clothes due to dribbling. Perceived frequency of urination was assessed by asking if ‘frequent urination during the day’ occurred. Finally, an additional question on urgency of urination using the Interstitial Cystitis Symptom Index (ICSI) scale [26] was included, urgency (ICSI). This question asks if participants experienced a strong urge or pressure to urinate with little or no warning.

Table 1.  Question descriptions and thresholds for symptoms used in the cluster analysis
VariableBACH question descriptionScaleRange-standardized scale
  • *

    Subjects with the lowest scores on all items were considered asymptomatic (except UI, where those with scores of <2 were considered asymptomatic). Scores in bold indicate positive responses for prevalence graphs (Fig. 2).

Incomplete emptyingDuring the last month, how often have you had: A sensation of not emptying your bladder completely after you have finished urinating?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
IntermittencyDuring the last month, how often have you had: To stop and start again several times while you urinate?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
Weak streamDuring the last month, how often have you had: A weak urinary stream?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
StrainingDuring the last month, how often have you had: To push or strain to begin urination?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
FrequencyDuring the last month, how often have you had: To urinate again less than 2 h after you finished urinating?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
Perceived frequencyDuring the last month, how often have you had: Frequent urination during the day?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
UIMany people complain that they leak urine (wet themselves) or have accidents. In the last 12 months, have you leaked even a small amount of urine?0 None*0
1 Less than once per month*0.25
2 One or more times per month0.5
3 One or more times per week0.75
4 Everyday1
UrgencyDuring the last month, how often have you had: Difficulty postponing urination?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
Urge UIDuring the last 7 days, how many times did you accidentally leak urine when you had the strong feeling that you needed to empty your bladder but you couldn’t get to the toilet fast enough?0*0
10.14
20.29
30.43
40.57
50.71
60.86
7(Free response – Capped at 7 in cluster analysis)1
Urgency (ICSI)During the last month, how often have you had: A strong urge or pressure to urinate immediately, with no, or little warning?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
NocturiaIn the last 7 days, on average, how many times have you had to go to the bathroom to empty your bladder during the night after falling asleep?0*0
10.2
20.4
30.6
40.8
5 (Free response – Capped at 5 in cluster analysis)1
Stress UIDuring the last 7 days, how many times did you accidentally leak urine when you were performing some physical activity such as coughing, sneezing, lifting, or exercise?0*0
10.14
20.29
30.43
40.57
50.71
60.86
7 (Free response – Capped at 7 in cluster analysis)1
Wet clothesDuring the last month, how often have you had: Wet clothes because of dribbling after urination?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1
DribblingDuring the last month, how often have you had: Dribbling after urination?1 I do not have the symptom*0
2 Rarely0.2
3 A few times0.4
4 Fairly often0.6
5 Usually0.8
6 Almost always1

For the displays of prevalence within clusters, conditions were defined as follows. Participants who reported two or more nightly voids (on average) in the past 7 days were considered positive for nocturia. Subjects were deemed to have UI if they reported any problems with leakage over the past year, while urge UI and stress UI were considered positive if subjects reported at least one episode in the past 7 days. A report of ‘a few times or more’ over the past month was considered a positive response for the remaining variables. Further details of the specific symptom queries and response scales used in the cluster analysis are shown in Table 1.

RISK FACTORS AND COVARIATES

A range of demographic, lifestyle and comorbidity variables were assessed as potential risk factors for urological outcomes. These analyses were conducted in SUDAAN version 9.0.0 [27] to account for the complex sampling design (multistage cluster sampling). Consequently to allow for generalizability to the Boston, MA population, population estimates in Table 2 were weighted inversely proportional to their probability of selection. Differences across groups were tested using Wald’s F-test for equality of means across groups for continuously scaled variables, and chi-square tests of independence or homogeneity for categorically scaled variables. These included age, race/ethnicity (Black, White, Hispanic), body mass index (BMI), waist circumference, alcohol consumption (0, <1, 1–2, ≥3 drinks per day), current cigarette smoking status (yes vs no), and physical activity as measured by the Physical Activity Scale for the Elderly (PASE) [28] divided into three categories (<100, 100–249, ≥250, with the ≥250 category indicating the highest level of activity). BMI was calculated from interviewer measurements and was examined on the continuous scale and in categories of ≥30 kg/m2 and <30 kg/m2. Socioeconomic status (SES) was defined as a combination of education and income [29] and was categorized as lower, middle and upper with half of the BACH population in the middle category and a quarter in each of the other two categories.

Table 2.  Risk factors and comorbidities in symptomatic and asymptomatic men*, N = 2276
VariableAsymptomatic menSymptomatic men (n = 1592)P
Cluster 1Cluster 2Cluster 3Cluster 4Cluster 5
  • *

    Weighted by the inverse of the probability of being sampled;

  • P for test of equality of means across strata (Wald F-test), all the others are P values for test of homogeneity of distributions across strata (chi-square).

N684801282184195130 
Sociodemographic
 Mean (sem) age, years 44.2 (0.7) 47.1 (0.7) 47.0 (1.1) 50.4 (1.5) 51.8 (1.6) 58.9 (1.7)<0.001
 SES, %
  Lower 22.6 22.8 26.0 22.4 14.2 40.0 
  Middle 49.3 50.9 43.4 47.5 58.2 45.20.086
  Upper 28.1 26.3 30.6 30.0 27.5 14.8 
 Race/ethnicity, %
  White 53.3 62.6 62.2 66.7 76.2 66.4 
  Black 28.8 25.6 25.6 21.7 14.0 26.00.005
  Hispanic 17.9  11.8 12.2  11.5  9.8  7.6 
Cardiovascular risk
 Mean (sem) BMI, kg/m2 28.1 (0.3) 28.7 (0.3) 28.5 (0.4) 28.4 (0.8) 29.5 (1.6) 29.3 (1.0)0.609
 BMI ≥30 kg/m2, % 30.0 34.0 30.2 25.5 37.6 43.20.414
 Mean (sem) waist circumference, cm 95.5 (0.7) 97.9 (0.6) 97.3 (0.9) 98.0 (2.5)102.0 (4.4)102.8 (2.3)0.009
 High blood pressure, % 18.2 25.8 24.6 38.8 28.7 46.5<0.001
 Diabetes, %  4.9  7.2 10.7 17.2  11.9 24.30.001
 CVD, % 14.3 18.4 18.6 30.6 29.3 39.3<0.001
Urological risk
 Previous bladder or prostate surgery, %  1.1  2.6  2.4  7.2  8.8 16.0<0.001
 Previous UTI and/or kidney infection, %  6.8  8.4 10.4 19.4 15.8 32.00.004
Interference with activities of daily living
 Mean (sem) Epstein score (range 0–28)  0.1 (0.0)  0.6 (0.1)  1.1 (0.2)  4.3 (1.1)  1.9 (0.3)  9.1 (1.0)<0.001
Lifestyle factors, %
 Current smoking 23.6 30.9 22.0 20.4 22.2 29.70.199
 Alcohol use:
  None 26.2 26.8 24.8 30.2 27.0 41.4 
  <1/day 39.6 37.3 44.5 48.3 35.7 27.80.328
  1–2/day 26.4 25.1 20.9 14.6 28.6 18.2 
  ≥3/day  7.8 10.8  9.8  6.9  8.7 12.5 
Physical activity (PASE score):
  Low (<100) 19.9 26.4 30.3 27.0 30.0 53.7 
  Moderate (100–249) 53.3 43.6 45.4 56.7 48.3 30.7<0.001
  High (≥250) 26.7 30.1 24.3 16.3 21.7 15.6 

Comorbid illnesses were assessed by the question: ‘Have you ever been told by a healthcare provider that you have . . . ?’ Specific diseases included high blood pressure, a history of UTI or kidney infection, type I or type II diabetes, and a composite cardiovascular disease (CVD) variable that included any of: a history of coronary artery bypass surgery or angioplasty, heart attack, angina, irregular heartbeat and/or having a pacemaker, congestive heart failure, transient ischaemic attack, stroke, carotid artery surgery, intermittent claudication, surgery or angioplasty for arterial disease of the leg, pulmonary embolism, aortic aneurysm, heart-rhythm disturbance, deep vein thrombosis, Raynaud’s disease and peripheral vascular disease. Interference with activities of daily living was assessed by means of a validated measure [30]. This scale comprises seven questions on a 5-point scale (0–4) of interference for each.

CLUSTER ANALYSIS METHODS

Cluster analysis was performed using SAS, version 9.1.3 [31]. Participants were included in the cluster analysis if they reported at least one symptom on the scales in Table 1, and data were not missing for any of the symptoms. Of the 2301 men in the BACH study, 25 were excluded from the analysis due to missing data and an additional 684 were excluded due to reporting none of the symptoms of interest. The remaining 1592 subjects were used for the cluster analysis using items from the BACH questionnaire, which were selected as representative of 14 common LUTS.

The distribution of responses to each question was range standardized on a 0–1 scale, as symptom measures included both ordinal and continuous scales. Thus, each item has a minimum score of 0 and a maximum of 1, with the remaining responses evenly distributed across the range (Table 1). This standardization method is recommended whenever distances between clusters are assessed using Euclidean distance measures (which were used here because they are applicable to both ordinal and continuous data). Exploratory models led to the determination of an appropriate number of clusters and split-half validation was used to examine internal validity of the resulting cluster solution. Further details of the cluster analysis procedures can be found in the Appendix[31,32].

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix

Among the 1592 symptomatic men, five symptom clusters were identified (Fig. 1). The prevalence of symptoms in each cluster is shown in Fig. 2. Radar plots are used to show the simultaneous distribution of symptom scores in each cluster (Fig. 3). There were no statistical differences between the two split-half samples, indicating split-half reliability and adequate internal validity.

image

Figure 1. Classification of BACH men into asymptomatic and five symptomatic clusters*. Each subject is classified as either asymptomatic (as detailed in Table 1) or into one of the five clusters based on their symptom profiles. *Unweighted for sampling design.

Download figure to PowerPoint

image

Figure 2. Symptom distribution within clusters. A yes on each variable is defined by occurring ‘a few times’ or more frequently considering the past month for all variables except UI, stress UI, urge UI and nocturia. A yes refers to once or more in the past year for UI, once or more over the past 7 days for stress UI and urge UI, and twice or more nightly on average over the past 7 days for nocturia.

Download figure to PowerPoint

image

Figure 3. Radar plots of symptom distribution among clusters. Points represent cluster averages of the range-standardized scores (as detailed in Table 1), so that 0 is the lowest possible value (at the centre of the figure) and 1 is the most extreme value. Points farther out from the centre indicate higher average scores for the symptom.

Download figure to PowerPoint

CLUSTER 1

The largest group of symptomatic men, which included 35% of the sample (half of the symptomatic population), was assigned to Cluster 1 (Fig. 1). These men had a mean (sd) number of symptoms of 1.0 (1.0), with nocturia (30.1%) and frequency of urination (25.5%) being the most prevalent symptoms in the cluster (Fig. 2), while other urological symptoms were relatively rare in this cluster.

CLUSTER 2

This was the second largest cluster, representing 12% of the total sample and nearly 18% of the symptomatic men. Perceived frequency was the distinguishing symptom in Cluster 2, affecting all of the men in this cluster, with frequency (45.7%) and nocturia (33.7%) as other prevalent symptoms. Men in this cluster had a mean (sd) number of symptoms of 2.4 (1.1). The preponderance of these symptoms is shown in the accompanying radar plots (Fig. 3).

CLUSTER 3

Representing 8% of the total sample (≈12% of the symptomatic men), men in this cluster reported a mean (sd) number of symptoms of 4.9 (1.8). A mixed profile of voiding and storage symptoms were reported by men in Cluster 3, with high rates of frequency (99.5%), perceived frequency (88.6%), nocturia (67.9%), incomplete emptying (44.0%), urgency based on the ICSI scale (38.6%) and intermittency (33.2%).

CLUSTER 4

This cluster had a similar prevalence (9% of the total sample, 12% of the symptomatic men) and a mean (sd) number of symptoms of 4.8 (2.0) compared with Cluster 3. However, the symptom profile was characterized by a higher rate of postvoiding symptoms (dribbling after urination, wet clothes due to dribbling after urination) and UI (55.4%). Nocturia (35.9%), incomplete emptying (39.0%), perceived frequency (36.4%) and frequency (61.5%) were other common symptoms in this cluster.

CLUSTER 5

The least prevalent cluster (6% of the total sample, 8% of the symptomatic men) consisted of the most highly symptomatic men with a mean (sd) number of symptoms of 9.9 (2.1). In most participants there was a mixed profile of voiding, storage and postvoiding symptoms. The prevalence of each symptom was >50%, with the exception of stress UI, which was reported by only 32% of the men in this cluster.

DEMOGRAPHICS, RISK FACTORS, AND COMORBIDITIES

The relationship between symptom clusters and common risk factors, including sociodemographic, lifestyle factors, and comorbidities are shown in Table 2. There was a significant increase in age across the groups, with the most symptomatic men being older than men in the other groups (mean age of 58.9 years in Cluster 5, compared with 44.2 years among the asymptomatic men). The race/ethnic composition of the clusters also differed significantly. Cluster 4 had lower proportions of Blacks and Hispanics than the other clusters; approximately half the proportion compared with the asymptomatic men. Hispanic men were also under-represented in the most symptomatic cluster (Cluster 5). A trend was noted for SES, with a large proportion (40.0%) of men in Cluster 5 being in the lowest SES category.

Risk factors and comorbidities were significantly associated with cluster assignment. Men in Clusters 3–5 (more symptomatic clusters) reported significantly more diabetes and cardiovascular symptoms, including hypertension and history of CVD, compared with asymptomatic men and those in Clusters 1 and 2. Almost 40% of the men in Cluster 5 reported a history of CVD, compared with <20% of men in Clusters 1 and 2. Similarly, past UTI and/or kidney infections were common in Cluster 5 (32.0%), compared with <10% of asymptomatic men or men in Cluster 1. Past bladder or prostate surgery was reported by a significantly higher percentage of men in the symptomatic clusters compared with those reporting fewer symptoms (16.0% in Cluster 5, compared with 1.1% in the asymptomatic group). Waist circumference was also significantly higher in the more symptomatic clusters (P < 0.01), and there was a trend towards higher rates of obesity in the more symptomatic men, with almost half of the men (43.2%) in Cluster 5 having a BMI of ≥30 kg/m2.

Among the lifestyle measures, the measure of interference with activities of daily living differed significantly across the clusters (P < 0.001). Men in clusters with more prevalent urological symptoms reported significantly higher levels of interference with daily living. Men in the most symptomatic cluster (Cluster 5) reported a mean (sem) interference score of 9.1 (1.0), compared with 0.6 (0.1) in the least symptomatic cluster (Cluster 1). Additionally, there were significant differences in physical activity levels across clusters. More than half (53.7%) of the men in Cluster 5 scored in the lowest range of physical activity, compared with 19.9% of asymptomatic men and 26.4% of those in Cluster 1.

There were no significant differences in smoking or alcohol consumption rates across the clusters, although a higher percentage of men in Cluster 5 (41.4%) reported abstinence from alcohol compared with asymptomatics (26.2%) or men in other Clusters (24.8–30.2%).

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix

We have used cluster analysis to categorize men in the BACH Survey into five cluster groups, based solely on the profile of urological symptoms reported. The resulting symptom clusters were stable and replicable within our sample, and bore similarities to the symptom clusters identified in an earlier Canadian study [19]. There was a mixed pattern of storage, voiding and postvoiding symptoms in several of the symptom clusters, which did not correspond to the symptom profiles typically identified with current diagnostic categories (i.e. overactive bladder syndrome, benign prostatic obstruction). Notably, men in the most symptomatic cluster (Cluster 5) reported a high frequency and prevalence of all urological symptoms, in conjunction with a high level of symptom impact and associated comorbidities. These findings were supported by an examination of the severity of symptoms, which was assessed by how often each symptom occurred over the period for symptom evaluation (data not shown). The men in the least symptomatic clusters were less affected by their urological symptoms as most men in these clusters reported that their symptoms occurred relatively infrequently. Conversely, men in Cluster 5 reported a higher frequency and intensity of all symptoms, including a broad range of storage, voiding and postvoiding symptoms. Ultimately, these findings could motivate a re-examination of the current nosology and diagnostic classification of LUTS, and alternative, symptom-based care could be considered. The cluster profiles identified in the present study may serve as a first step in this process.

The associations between the cluster profiles and other indicators of physical and mental health in the present sample provide clinical support of these clusters. Men in the more symptomatic clusters were older on average, less physically active, and had higher rates of hypertension, diabetes, depression and CVD. By contrast, men in the least symptomatic clusters (Clusters 1, 2) reported significantly fewer urological symptoms, less interference with activities of daily living, and a lower rate of past UTIs and bladder or prostate surgery. The trends for waist circumference and physical activity levels might indicate a potential role of central adiposity or the metabolic syndrome as an underlying determinant or aetiological mechanism. Further studies are needed to assess the role of other potential mechanisms, such as endothelial dysfunction or autonomic hyperactivity.

Major strengths of the present study are the empirical nature of the method and lack of a priori assumptions. The present findings are similar to those of a previous cluster analysis of urological symptoms in older men by Norman et al. [19]. In the earlier study, five clusters of LUTS were also identified, with similar distributions of storage and voiding symptoms as those in the present study. However, the present study evaluated a broader range of voiding symptoms in a population representative of Boston (MA, USA). We also showed significant associations between the cluster profiles and demographic characteristics, lifestyle risk factors, and common comorbidities in our sample. The present results replicate and extend previous findings in the BACH Survey and other large epidemiological studies showing a significant association between LUTS and common comorbidities and risk factors [25,33]. Identification of symptom cluster groups in the present study provides a potential basis for more in-depth investigation of these clinically important associations.

An additional strength of the present data is that its generalizability is known. For the comorbidities, it is known that (with the exception of asthma), BACH participants have a similar comorbidity profile when compared with USA estimates from the Behaviourial Risk Factor Surveillance System, the National Health Interview Survey, and the National Health and Nutrition Examination Survey, suggesting BACH participants are nationally representative on these variables [22,34].

Potential limitations of the present study deserve mention. First, our cluster results are based on self-reported symptoms, which might be influenced by reporting biases across individuals or groups. The lower proportion of Hispanic men in Cluster 5, for example, might be accounted for by a negative reporting bias in this group (i.e. reluctance to acknowledge urological symptoms). Alternatively, there may be other differences in Hispanic men that are protective in one way or another. Further studies are needed to investigate this hypothesis.

Another potential limitation is the selection of particular urological symptoms and symptom thresholds for inclusion in the analyses. For the present study, we included a broad range of 14 common urological symptoms, similar to those reported by Coyne in the accompanying paper [25]. In a separate paper [35], we address the effects of varying the number of symptoms included or altering symptom thresholds on the resulting cluster solutions. Furthermore, the cross-sectional nature of the study limits inferences that can be drawn about the potential direction of causality in our findings. For example, it is possible that higher rates of previous prostate or bladder surgery could be a cause or consequence of the symptom profile seen in Cluster 5. Finally, replication of the cluster findings in a clinic-based sample would strengthen the validity of the present findings.

Overall, the results of the present study provide evidence in support of a new diagnostic classification of LUTS in men that is more in line with the overlapping presentation of urological symptoms. The present findings are similar to those of Norman et al.[19], suggesting that men with urological symptoms can be categorized into five specific clusters, with increasing prevalence and frequency of symptoms and associated levels of interference across clusters. The high degree of association with other comorbidities and risk factors suggests the need for a more comprehensive approach to the management of urological symptoms in men, with greater attention to the diagnosis and management of these comorbid conditions. Cluster membership could influence prognosis or treatment outcome, e.g. lifestyle factors might influence treatment for highly symptomatic men. New treatment approaches may also be indicated that would be aimed at a specific profile or cluster of LUTS, rather than a putative diagnosis or single aetiological factor.

ACKNOWLEDGEMENTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix

David Henry, Anne Stoddard, Don Brambilla.

Funding: The BACH survey is supported by DK 56842 from the National Institute of Diabetes and Digestive and Kidney Diseases. Additional funding was provided as an unrestricted grant from Pfizer, Inc. for analyses presented in this paper.

CONFLICT OF INTEREST

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix

Steven Kaplan is a paid consultant to Pfizer, Sanofi-aventis, and Allergan. Zoe Kopp is an employee of Pfizer, and Claus Roehrborn is a paid consultant to GlaxoSmithKline, Sanofi-aventis, AMS, Spectrum, Aeterna Zentaris and Pfizer.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix
  • 1
    Abrams P. New words for old: lower urinary tract symptoms for ‘prostatism’. BMJ 1994; 308: 92930
  • 2
    Barry M, Cockett A, Holtgrewe H, McConnell JD, Sihelnik SA, Winfield HN. Relationship of symptoms of prostatism to commonly used physiological and anatomical measures of the severity of benign prostatic hyperplasia. J Urol 1993; 150: 3518
  • 3
    Bhargava S, Canda AE, Chapple CR. A rational approach to benign prostatic hyperplasia evaluation: recent advances. Curr Opin Urol 2004; 14: 16
  • 4
    Eckhardt MD, Van Venrooij GE, Boon TA. Symptoms, prostate volume, and urodynamic findings in elderly male volunteers without and with LUTS and in patients with LUTS suggestive of benign prostatic hyperplasia. Urology 2001; 58: 96671
  • 5
    Homma Y, Imajo C, Takahashi S, Kawabe K, Aso Y. Urinary symptoms and urodynamics in a normal elderly population. Scand J Urol Nephrol Suppl 1994; 157: 2730
  • 6
    Knutson T, Edlund C, Fall M, Dahlstrand C. BPH with coexisting overactive bladder dysfunction – an everyday urological dilemma. Neurourol Urodyn 2001; 20: 23747
  • 7
    Thomas AW, Cannon A, Bartlett E, Ellis-Jones J, Abrams P. The natural history of lower urinary tract dysfunction in men: minimum 10-year urodynamic follow-up of untreated detrusor underactivity. BJU Int 2005; 96: 1295300
  • 8
    Chapple CR, Wein AJ. The urgency of the problem and the problem of urgency in the overactive bladder. BJU Int 2005; 95: 2745
  • 9
    Andersson S, Rashidkhani B, Karlberg L, Wolk A, Johansson J. Prevalence of lower urinary tract symptoms in men aged 45–79 years: a population-based study of 40 000 Swedish men. BJU Int 2004; 94: 32731
  • 10
    Chapple CR, Artibani W, Cardozo LD et al. The role of urinary urgency and its measurement in the overactive bladder syndrome: current concepts and future prospects. BJU Int 2005; 95: 33540
  • 11
    Chapple C, Artibani W, Berges R et al. New medical developments in the management of LUTS in adult men. In McConnellJ, AbramsP, DenisL, KhourS, RoehrbornC eds, Male Lower Urinary Tract Dysfunction. Paris: Health Publications, 2006: 14394
  • 12
    Hong SJ, Ko WJ, Kim SI, Chung BH. Identification of baseline clinical factors which predict medical treatment failure of benign prostatic hyperplasia: an observational cohort study. Eur Urol 2003; 44: 94100
  • 13
    Porru D, Jallous H, Cavalli V, Sallusto F, Rovereto B. Prognostic value of a combination of IPSS, flow rate and residual urine volume compared to pressure-flow studies in the preoperative evaluation of symptomatic BPH. Eur Urol 2002; 41: 2469
  • 14
    Aldenderfer M, Blashfield R. Cluster Analysis. Newberry Park, CA: Sage Publications, 1984
  • 15
    Everitt B. Cluster analysis is a generic term for a wide range of numerical methods for examining data. Stat Methods Med Res 2004; 13: 3435
  • 16
    Hammer J, Howell S, Bytzer P, Horowitz M, Talley NJ. Symptom clustering in subjects with and without diabetes mellitus: a population-based study of 15 000 Australian adults. Am J Gastroenterol 2003; 98: 3918
    Direct Link:
  • 17
    Morana H, Camara F, Arboleda-Florez J. Cluster analysis of a forensic population with antisocial personality disorder regarding PCL-R scores: differentiation of two patterns of criminal profiles. Forensic Sci Int 2006; 164: 98101
  • 18
    Mulroy S, Gronley J, Weiss W, Newsam C, Perry J. Use of cluster analysis for gait pattern classification of patients in the early and late recovery phases following stroke. Gait Posture 2003; 18: 11425
  • 19
    Norman R, Nickel J, Fish D, Pickett S. ‘Prostate-related symptoms’ in Canadian men 50 years of age or older: prevalence and relationships among symptoms. Br J Urol 1994; 74: 54250
  • 20
    Kupelian V, Wei JT, O’Leary MP et al. Prevalence of lower urinary tract symptoms and effect on quality of life in a racially and ethnically diverse random sample: the Boston Area Community Health (BACH) survey. Arch Intern Med 2006; 166: 23817
  • 21
    Litman HJ, McKinlay JB. The future magnitude of urological symptoms in the USA: projections using the Boston Area Community Health survey. BJU Int 2007; 100: 8205
  • 22
    McKinlay JB, Link CL. Measuring the urologic iceberg: design and implemenation of the Boston Area Community Health (BACH) Survey. Eur Urol 2007; 52: 38996
  • 23
    Hall SA, Çinar A, Link CL et al. Do urological symptoms cluster among women? Results from the Boston Area Community Health Survey. BJU Int 2008; 101: 125766
  • 24
    Irwin DE, Milsom I, Hunskaar S et al. Population-based survey of urinary incontinence, overactive bladder, and other lower urinary tract symptoms in five countries: results of the EPIC study. Eur Urol 2006; 50: 130615
  • 25
    Coyne KS, Matza LS, Kopp ZS et al. Examining lower urinary tract symptom constellations using cluster analysis. BJU Int 2008; 101: 126773
  • 26
    O’Leary MP, Sant GR, Fowler FJ Jr, Whitmore KE, Spolarich-Kroll J. The interstitial cystitis index and problem index. Urology 1997; 49 (Suppl.): 5863
  • 27
    Research Triangle Institute. SUDAAN Language Manual Release 9.0. Research Triangle Park, NC: Research Triangle Institute, 2004
  • 28
    Washburn RA, Smith KW, Jette AM, Janney CA. The Physical Activity Scale for the Elderly (PASE): development and evaluation. J Clin Epidemiol 1993; 46: 15362
  • 29
    Green LW. Manual for scoring socioeconomic status for research on health behavior. Public Health Rep 1970; 85: 81527
  • 30
    Epstein RS, Deverka PA, Chute CG et al. Validation of a new quality of life questionnaire for benign prostatic hyperplasia. J Clin Epidemiol 1992; 45: 143145
  • 31
    SAS Institute. SAS 913 Help and Documentation. Cary, NC: SAS Institute, 2007: 20024
  • 32
    Kruskal W, Wallis W. Use of ranks in one-criterion variance analysis. J Am Stat Assoc 1952; 47: 583621
  • 33
    Litman HJ, Steers WD, Wei JT, Kupelian V, Link CL, McKinlay JB, Boston Community Health Survey Investigators. Relationship of lifestyle and clinical factors to lower urinary tract symptoms: results from Boston Area Community Health Survey. Urology 2007; 70: 91621
  • 34
    New England Research Institutes. Can the BACH data be generalized to the U.S. population? Comparison of BACH batch 1–4 data with data from three national surveys (NHIS, NHANES, and BRFSS). In New England Research Institutes, 2006. Available at: http://www.neriscience.com/web/documents/v2Combined%20Comparison%20of%20BACH%20Batch%201-5%20to%20NHIS%20NHANES%20BRFSS.pdf. Accessed January 2008
  • 35
    Rosen RC, Coyne KS, Henry DB et al. Beyond the cluster: methodological and clinical implications in the Boston Area Community Health survey and EPIC studies. BJU Int 2008; 101: 12748

Appendix

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. SUBJECTS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. ACKNOWLEDGEMENTS
  8. CONFLICT OF INTEREST
  9. REFERENCES
  10. Appendix
DETAILS OF CLUSTER ANALYSIS PROCEDURES
Step 1 – identification of an appropriate number of clusters

To identify an appropriate number of clusters for inclusion, we used a hierarchical clustering approach (PROC CLUSTER, SAS version 9.1.3 [31]) in which each subject begins as his own cluster and subjects are grouped into clusters according to their common symptoms. In this procedure, Ward’s linkage, a minimum variance method, was used to measure distances between responses. This method provides several measures of cluster coherence, such as the pseudo F and t2 statistics, which were used to select the appropriate number of clusters for the analysis. To ensure ample number of subjects for comparisons of comorbidities across clusters, cluster solutions with <5% of symptomatic respondents in any cluster were not considered.

Step 2 – assigning subjects to clusters

Once the number of clusters was determined, subjects were assigned to clusters by means of a nonhierarchical, k-means method (PROC FASTCLUS, SAS version 9.1.3 [31]), which minimizes within-cluster distances between individuals, relative to the between-cluster distances observed. Cluster seeds (a mathematical mid-point of the cluster), which had been generated by the corresponding hierarchical clustering solution were used to initiate the nonhierarchical procedure. Subjects were assigned to clusters iteratively, until the solution with the minimum squared deviance was obtained.

Step 3 – split-half validation

The internal validity of the cluster solution was assessed using a split-sample replication in which the cluster solution was generated in a two randomly selected subgroups of approximately half of the sample, using the seeds from the cluster solution determined in Step 2, above. The distribution of Euclidean distances (from each observed value to the cluster mean) for the two split-half samples used in the replication analysis were compared using the Kruskal–Wallis test [32], a nonparametric method for comparing distributions based on their median values. The null hypothesis being tested is that the cluster solutions from these two subsamples are not significantly different from each other.