Do overactive bladder symptoms exhibit a Gaussian distribution? Implications for reporting of clinical trial data

Amiri M1, Murgas S2, Michel M C3

Research Type

Clinical

Abstract Category

Overactive Bladder

Abstract 651
Overactive Bladder 2
Scientific Podium Short Oral Session 33
Friday 31st August 2018
15:35 - 15:42
Hall C
Overactive Bladder Outcomes Research Methods Prospective Study Pharmacology Incontinence
1. University of Duisburg-Essen, 2. Apogepha, 3. Johannes Gutenberg University
Presenter
M

Martin C Michel

Links

Abstract

Hypothesis / aims of study
To assess symptom severity of patients suffering from overactive bladder syndrome (OAB) voiding diary variables urgency, frequency, nocturia and incontinence are recorded at baseline and upon treatment. While some investigators report such variables as means with standard deviations (SD), others report medians with confidence intervals or inter-quartile ranges (IQ). However, this is not a discretionary question but should be determined based on whether the variable is exhibiting a Gaussian (normal) distribution. Large datasets could clarify this question but have not been reported, particularly based on real-life treatment settings. Therefore, we have examined distribution of OAB symptoms in two large non-interventional studies. Analyses were performed both for baseline symptoms and for treatment-associated changes thereof; the latter was assessed concomitantly as absolute and as relative changes (% reduction).
Study design, materials and methods
Two non-interventional studies of similar design were analyzed. Neither had specified inclusion or exclusion criteria other than the recommendations from the prescribing information. Rather patients were invited to participate if they started treatment with propiverine ER (30-45 mg/day; dose adjustment during the study was permitted) at the recommendation of their physician. Diary parameters of urgency, frequency, nocturia and incontinence were analyzed at baseline and after 12 weeks of treatment.

For the analysis of baseline symptoms, all patients were included except for those not exhibiting a given symptom at baseline. This exclusion was made because including patients e.g. being continent certainly would have skewed the analysis. Treatment-associated changes were only evaluated in patients with a recorded value after 12 weeks and had started and stayed on the 30 mg/day dose. The latter choice was made as dose and dose adjustment may be a confounding factor, and the 30 mg/day dose group was the largest in both studies.

Normality was tested by the D’Agostino and Pearson K2 omnibus test, which calculates the skewness and kurtosis and the values differ from the Gaussian distribution. Given the size of the underlying studies and the post-hoc character of the analysis, a P < 0.01 was defined as statistically significant. As true Gaussian distribution is rarely observed in large datasets, we also assessed whether deviation from a normal distribution was sufficiently substantial to have means and medians differ to a degree deemed to be of clinical relevance.
Results
Studies 1 and 2 included 1335 and 745 patients, respectively, but not all participants reported data for each of the four OAB symptoms in their voiding diary and some did not exhibit one or more of the symptoms. Specific numbers for each symptom as well as calculated mean and median are shown in Table 1. All deviated significantly from a Gaussian distribution, and mean values overestimated symptom severity, except for urgency in study 2, but differences were small in some cases.

Treatment-associated improvements of OAB symptoms also differed significantly from a Gaussian distribution, regardless whether expressed as delta of episodes (Table 2) or as % reduction (Table 3). Except for nocturia, means generally overestimated absolute improvements. However, differences between mean and median relative improvements were only small and not consistent across symptoms.
Interpretation of results
Two non-interventional studies consistently found that neither OAB symptoms nor their improvement upon treatment exhibit a Gaussian distribution. Ignoring this and reporting means leads to an overestimation of symptom severity at baseline and of absolute treatment responses. Relative (%) improvements are less affected.
Concluding message
Except for relative improvements, OAB symptoms should be reported as medians and non-parametric tests should be applied in their statistical analysis.
Figure 1
Figure 2
Disclosures
<span class="text-strong">Funding</span> The underlying non-interventional studies were funded by Apogepha, Dresden, Germany. <span class="text-strong">Clinical Trial</span> Yes <span class="text-strong">Public Registry</span> No <span class="text-strong">RCT</span> No <span class="text-strong">Subjects</span> Human <span class="text-strong">Ethics Committee</span> Sächsische Landesärztekammer EK-BR-14/12-1 and EK-BR-18/14-1 <span class="text-strong">Helsinki</span> Yes <span class="text-strong">Informed Consent</span> Yes