SAW PALMETTO FOR BENIGN PROSTATIC HYPERPLASIA OBIECTIVES AND OUTCOMES STATISTICAL ANALYSIS The primary objectives of the study were to deter- The study was designed to have a statistical power of 90 percent to ract r s the symptoms plas score AUASand a two-sided alpha value of 0.0 rates),as compared with placebo.The AUASI is a (set below 0.05 to allow for possible interim values require symptoms ref ee Der wa from o to cent cate mild symptoms;scores of to19,moderate The primary efficacy analyses weret score severe symp- parisons of the ch nation of chan the alit cific the sa and n We as to benign prostatic hyperplasia and overall qual sessed the significance of these differences in ects mode se moc els in Study 36-ttem Short-Form General Health Sur. measures gathered from each study participant vey [SF-36]);prostate size,measured by trans- as well as terms for the fixed effects of time,study esidua volume aft group,and the interacti on between tim e and ed side eff and char We fun time within each study group using likelihood. consists of atio tests and found that for most outcomes, wh. at AUAS fit the data well.However, eral and mental health vitalit social with quadratic time effects fit the data signifi function,and physical and emotional health. cantly better,and our models for these outcomes included these nonlinear effects of time RANDOMIZATION satisfied all including com letion of the run-in period.under tional form of the effect of time.before analyz went randomization in equal proportions tothe ing the study data.For each,we presen saw pametto and placebo groups.Randomiza stra ate 8 as est d treatment fects, hich we calcu late h D)and blocked with the use of randomly cho two study groups.The linear mixed-effects mod ered block sizes of less than 10 el analyses provide these estma according to o procedure i n stata, ndard err trolled trials The randomization list g We fit theinear mixed-effect ated by personnel who were not associated with model using the XTREG procedure in Stata soft the study.The study me ation was dispensed in are (version 8.0).The overall differences in b s (pre Dy e ma the tot al response ikei出 tween the tw study participants and all study personnel who and safety monitorin board (composed of ex administered in nterventions,assessed outcomes,perts selected by the National Institutes of Health or performed data analysi d the rm who were not affiliated with the udy)elected ed se 10 een N ENGLJ MED 354:6 559 Journal of Medici Downloaded from n use only.N saw palmetto for benign prostatic hyperplasia n engl j med 354;6 www.nejm.org february 9, 2006 559 Objectives and Outcomes The primary objectives of the study were to deter￾mine whether the daily use of saw palmetto ex￾tract reduces the symptoms of benign prostatic hyperplasia, as measured by the AUASI or objec￾tive measures of urinary obstruction (urinary flow rates), as compared with placebo. The AUASI is a validated seven-item, self-administered question￾naire that measures symptoms referable to urinary obstruction, with scores ranging from 0 to 35 ac￾cording to symptom severity: scores of 0 to 7 indi￾cate mild symptoms; scores of 8 to 19, moderate symptoms; and scores of 20 to 35, severe symp￾toms.22 Secondary objectives included an exam￾ination of changes in the quality of life specific to benign prostatic hyperplasia and overall qual￾ity of life (assessed by two self-administered questionnaires, the Benign Prostatic Hyperplasia [BPH] Impact Index23 and the Medical Outcomes Study 36-Item Short-Form General Health Sur￾vey [SF-3624]); prostate size, measured by trans￾rectal ultrasonography; residual volume after voiding, measured by BladderScan (Diagnostic Ul￾trasound); self-reported side effects; and chang￾es in levels of PSA, creatinine, testosterone, and other laboratory values. The SF-36 consists of 36 items, 35 of which are aggregated to evaluate eight dimensions of health: physical function, pain, general and mental health, vitality, social function, and physical and emotional health. Randomization Participants who satisfied all eligibility criteria, including completion of the run-in period, under￾went randomization in equal proportions to the saw palmetto and placebo groups. Randomiza￾tion was stratified according to the category of AUASI score (moderate [8 to 19] vs. severe [20 to 35])22 and blocked with the use of randomly cho￾sen even-numbered block sizes of less than 10 according to the ralloc.ado procedure in Stata, a software module used to design randomized, con￾trolled trials.25 The randomization list was cre￾ated by personnel who were not associated with the study. The study medication was dispensed in numbered bottles (provided by the manufacturer), according to the randomization sequence. All study participants and all study personnel who administered interventions, assessed outcomes, or performed data analysis were unaware of the treatment assignment and the randomized se￾quence list. Statistical Analysis The study was designed to have a statistical power of 90 percent to detect a difference between groups of 3.0 in the AUASI score,26 on the basis of a published standard deviation of 6.0 in the AUASI27,28 and a two-sided alpha value of 0.04 (set below 0.05 to allow for possible interim analyses). These calculations and values required the enrollment of 178 men, and the number was increased to a target enrollment of 224 to account for a potential dropout rate of up to 20 percent. The primary efficacy analyses were the com￾parisons of the change over time in the AUASI scores and the peak urinary flow rate between the saw palmetto and placebo groups. We as￾sessed the significance of these differences in changes in outcomes over time using linear mixed￾effects models.29 These models included ran￾dom intercepts to accommodate the repeated measures gathered from each study participant as well as terms for the fixed effects of time, study group, and the interaction between time and study group, the effects of interest in our analyses. We assessed the functional form of the effect of time within each study group using likelihood￾ratio tests and found that for most outcomes, linear time effects fit the data well. However, for the AUASI scores and testosterone levels, a model with quadratic time effects fit the data signifi￾cantly better, and our models for these outcomes included these nonlinear effects of time. We specified the linear mixed-effects model analytic strategy, including the assessment of the func￾tional form of the effect of time, before analyz￾ing the study data. For each outcome, we present estimated treatment effects, which we calculated as the difference in the predicted change in the response over a period of 12 months between the two study groups. The linear mixed-effects mod￾el analyses provide these estimates along with associated standard errors, which were used to construct 95 percent confidence intervals for treatment effects. We fit the linear mixed-effects model using the XTREG procedure in Stata soft￾ware (version 8.0).30 The overall differences in the total response curves between the two groups were tested with likelihood-ratio tests. The data and safety monitoring board (composed of ex￾perts selected by the National Institutes of Health who were not affiliated with the study) elected not to perform interim analyses of efficacy. Baseline variables were compared between The New England Journal of Medicine Downloaded from nejm.org on October 18, 2011. For personal use only. No other uses without permission. Copyright © 2006 Massachusetts Medical Society. All rights reserved