Beyond the cluster:: methodological and clinical implications in the Boston Area Community Health survey and EPIC studies

OBJECTIVE To test the replicability and robustness of findings about urological symptoms in men and women, classified using an objective statistical method, cluster analysis, by planned sensitivity analyses conducted within and across two large, epidemiological studies of lower urinary tract symptoms. METHODS Sensitivity analyses were used to assess the effects of: (i) the number of urological symptoms included in the cluster analysis; (ii) the use of ordinal vs dichotomous scaling of responses; (iii) the type of cluster analysis used (hierarchical vs non-hierarchical; random vs nonrandom seeds); and (iv) the distance metric (median difference vs root mean square) of the resulting clusters. These sensitivity analyses were conducted independently in each of the two studies, with results systematically compared using Cramer's V statistic. Contingency tables were also used to assess the frequency of transitions or change in classification from one method to another. RESULTS There were marked similarities in the cluster profiles in men and women across the two studies. For both men and women, the largest clusters consisted of low-frequency, single-symptom profiles, with urinary frequency and urgency symptoms reported by both genders. There was a multiple, mixed and highly symptomatic cluster profile in both genders in the Boston Area Community Health (BACH) and EPIC studies. The sensitivity analyses showed stability across both BACH and EPIC studies, and varying cluster methods and solutions (Cramer's V, 0.37-0.93). CONCLUSION Sensitivity analyses show that cluster profiles are quite robust from EPIC to BACH, and that gender profiles within studies are relatively consistent across the methods and variables examined. Further studies are needed to investigate the mechanisms of action and clinical management implications of these findings.