Development and Validation of a Risk Score for Somatic Erectile Dysfunction: Combined Results from Three Cross-Sectional Surveys

Transcription

1 European Urology European Urology 48 (2005) Erectile Dysfunction/Andrology Development and Validation of a Risk Score for Somatic Erectile Dysfunction: Combined Results from Three Cross-Sectional Surveys Martin Hellmich a, *, Thomas Evers b, Maria Kubin b, Sanjay Merchant c,1, Walter Lehmacher a, Udo Engelmann d, Moritz Braun d a Institute of Medical Statistics, Informatics and Epidemiology, University of Cologne, Joseph-Stelzmann-Straße 9, D Köln, Germany b Bayer HealthCare AG, Health Economics and Outcomes Research, Wuppertal, Germany c Bayer Pharmaceuticals Corporation, Health Economics and Outcomes Research, West Haven, CT, USA d Department of Urology, University of Cologne, Cologne, Germany Accepted 20 April 2005 Available online 12 May 2005 Abstract Objectives: Some men with erectile dysfunction (ED) have difficulties discussing their condition with their physicians. Existing screening and diagnostic tools for ED often require the administration of personal questions regarding the condition. We present a simple risk score to estimate the individual likelihood of somatic ED, based on age and existing health conditions. Methods: Data from the Cologne Male Survey (n = 4396) were used to develop a multivariable logistic regression model for the individual ED likelihood. The regression equation was both internally and externally validated using data from a national study (Berlin study) and a multinational cross-sectional study (MALES study). Results: A final regression equation including age, pelvic surgery, diabetes mellitus, arterial circulatory disorder, heart disease, smoking, and hypertension reached an area under the receiver operating characteristic curve of 0.84 (0.5 means random and 1.0 perfect discrimination). Internal validation did not indicate any relevant overfit and the external validation results (national data: AUC = 0.75; multinational data: AUC = 0.67) are similar to those of other popular risk scores. Conclusions: The validated ED risk score developed from the regression equation can be used as a screening tool to identify patients who are at a high risk of somatic ED. This tool can facilitate entering into discussions between physicians and patients regarding erectile function. # 2005 Elsevier B.V. All rights reserved. Keywords: Impotence; Penile erection; Questionnaires; Logistic models; Score 1. Introduction Erectile dysfunction (ED) has recently been defined by an expert committee as consistent or recurrent inability of a man to attain and/or maintain penile erection sufficient for sexual activity [1]. Reviews of the current epidemiological literature on ED show * Corresponding author. Tel ; Fax: address: martin.hellmich@uni-koeln.de (M. Hellmich). 1 Since 2004: Health Economics & Pricing, Johnson & Johnson, Raritan, NJ, USA. that the prevalence of ED differs considerably across various studies, from 1 9% below 40 years of age to 50 75% for men in their 70s and 80s [2,3]. These large differences may be attributed to various factors, including the age group reported on, how the data were selected and collected, the definition of ED and the questionnaires used to evaluate the presence of ED [4]. Though oral treatments to treat erectile dysfunction are available, sexual dysfunction remains a difficult topic for patients and providers to discuss [5]. Reasons for failure of discussion include patient embarrassment /$ see front matter # 2005 Elsevier B.V. All rights reserved. doi: /j.eururo

2 496 M. Hellmich et al. / European Urology 48 (2005) and/or unawareness of their physicians competence in sexual dysfunction [6,7]. Many patients would like their physicians to initiate a discussion of ED during routine visits, however, physicians may not be prepared (embarrassed and perhaps even dubiously competent) to talk about sexuality in a clinical setting [8]. Several cross-sectional surveys [9 11] have shown that the prevalence of ED is associated with various risk factors such as age, existing health conditions (e.g. diabetes mellitus and hypertension) as well as lifestyle habits (e.g. smoking). Thus, a novel means of approaching ED could be given by the evaluation of the individual likelihood of ED based on existing risk factors. Screening for ED is important, apart from lifestyle or quality of life issues, since ED may be a precursor of yet undetected serious health conditions such as heart disease [12,13]. The purpose of this research was to develop and validate a simple point scoring scheme for ED (called in short the ED Risk Score) which can be used as a screening tool to identify patients at high risk of somatic ED. The tool may be beneficial to patient management as a means to break the ice, by moving from general medical questions to the issue of sexual function. 2. Materials and methods 2.1. Survey settings Data from three cross-sectional surveys assessing the prevalence of ED and related health issues were used for this study. First, in the Cologne Male Survey, a representative age-stratified population of 8000 men, years of age and living in the Cologne region of Germany, was approached from June to November 1998 by means of a validated mail questionnaire [5]. A total of 4396 responses (55.0%, complete information on risk factors of interest) were used for development of the ED Risk Score. Secondly, in the Men s Health, Quality of Life and Sexuality study, carried out from August to November 2002, a questionnaire was mailed to a representative agestratified population of 6000 men, years of age and living in greater Berlin (Berlin study) [14]. A total of 1817 responses (30.3%, complete information on risk factors of interest) could be used for external validation of the ED Risk Score. Thirdly, the Multinational Men s Attitudes to Life Events and Sexuality (MALES) Phase I study, conducted from February to April 2001 in Brazil, France, Germany, Italy, Mexico, Spain, the UK, and the USA, involved men, years of age, who were either recruited via random digit dialing and interviewed via computer-assisted telephone interviewing (69.4%) or recruited by via random selection from a survey research database to participate in Internet-based interviews (30.6%) [11]. The proportion of dialed phone numbers that led to study participation was about 50%. A total of responses were used for external validation of the ED Risk Score(complete information on risk factors of interest); 6024 men aged were excluded since this project focused on men who were 30 years or older Self-report questionnaires/definition of variables For the Cologne Male Survey a validated 18-item questionnaire ( Cologne ED Questionnaire, German acronym KEED ) was used to identify symptoms of ED and its effects on quality of life [15]. Men were classified as having ED if they scored 17 or more points on the six-item ED rating scale (range 6 30). Variables screened for prognostic value were age (in years), alcohol intake, arterial circulatory disorder (essentially peripheral vascular disease), diabetes mellitus, heart disease (defined by heart failure or intake of nitrates), hypertension, lower urinary tract symptoms (LUTS, defined as present if at least two items of the International Prostate Symptom Score (IPSS) were applicable in more than 50% of urinary attempts), pelvic surgery (e.g. prostate, rectum/sigmoid or bladder), smoking, and slipped disk. All variables, except for age, were dichotomous. A more extensive questionnaire was used for the Berlin study with various items on socio-demographics, health history/status, quality of life, sexuality and lifestyle. Specifically, the definition of ED was based on the erectile function (EF) domain of the International Index of Erectile Function (IIEF) [16]. Men were classified as having ED if they scored 25 points or lower on the IIEF-EF (range 1 30) [17]. Except for the variables ED, which is defined differently, and LUTS, which is absent in the Berlin dataset, all variables taken from the Cologne dataset are both present and comparably defined in the Berlin dataset. A similar questionnaire was used for the MALES study to assess general demographic information, overall health rating, prevalence of selected diseases, current medication use, and erection difficulties. Men were classified as having ED if they answered yes to at least one of the questions (i) seen a doctor for erection difficulties?,(ii) tried any kind of remedy for erection difficulties?, or (iii) not done anything about erection difficulties? Thus, compared with the Cologne and Berlin datasets, ED is defined somewhat differently in the MALES study. No details are available on arterial circulatory disorder, LUTS, and pelvic surgery. The MALES study question ever been diagnosed with high blood pressure/heart trouble or angina/diabetes? was considered as roughly comparable in meaning to the corresponding definitions used in the Cologne Male Survey for hypertension, heart disease, and diabetes Statistical methods for model development and validation A multivariable logistic regression model for the individual ED likelihood was developed in a stepwise manner ( Wald-en backwards ) from a specific set of candidate predictors (all possible risk factors which showed bivariable association with ED) collected as part of the Cologne Male Survey. Model discrimination and calibration were assessed by calculation of the area under the corresponding receiver operating characteristic curve (AUC) and the Hosmer Lemeshow (HL) statistic [18], respectively. For model development, we chose the Cologne dataset instead of the Berlin or MALES data because of its higher fraction of responders/complete information on risk factors. Validation of the final model equation was performed, internally, by means of a bootstrap approach and, externally, using complete case records from the MALES study and the Berlin study (calculating AUC and HL). While external validation with the Berlin data was straightforward, missing variables in the MALES data were imputed by the corresponding prevalence estimates from the Cologne Male Survey, the best internal knowledge available. The validation would not be fully external if the MALES prevalence estimates were plugged in. Note that only calibration is affected by imputation, not discrimination. We decided against developing and validating a model based solely on common risk factors because a larger (still simple) model presumably yields better estimates and, using the same approach, other researchers are able to revalidate the model on different datasets, again (likely) with different subsets of common risk factors. Logistic regression analysis was done using SPSS for

3 Windows, Release The HL test used for external validation was calculated using the macro hl.sps available from The internal bootstrap validation was performed using the free software R (libraries design and hmisc ) [19]. Except for the p-value from the HL test, which is conventionally performed at an a level of 0.01, all other p-values were considered statistically significant if lower than For clarity of the presentation, no adjustment was made for multiple testing; however, even a conservative Bonferroni correction multiplier of 20 would not change the statistical significance of key findings. M. Hellmich et al. / European Urology 48 (2005) Results Fig. 1. Prevalence of ED in different age groups (with Clopper Pearson 95% confidence intervals) Bivariable association of ED and risk factors The participants of the Berlin study were on average 7.4 years older and had ED and hypertension more often than the Cologne Male Survey participants (see descriptive statistics in Table 1). The MALES participants were on average 3.5 years younger and had heart disease and hypertension more often than the Cologne participants. In all three datasets the prevalence of ED rises approximately linearly with age (see Fig. 1). Note that for the Berlin study even very mild EDs [17] were included whereas the KEED cutoff of 17 used for the Cologne Male Survey probably means that moderate to severe EDs were considered (data yet unpublished). All other risk factors listed in Table 1 seem to be associated with ED (at least in the Cologne data) since their unadjusted odds ratios obtained from univariable logistic regression analyses are significantly different from 1 (all p-values < 0.005; see Table 2). Note that smoking appears to be protective regarding ED in all three datasets; however, this association is reversed on multivariable analysis, presumably due to confounding by age. Obesity and the metabolic changes it induces, which have recently been demonstrated to be associated with ED [20], could not be considered for model development, since corresponding details were not queried in the Cologne Male Survey and only roughly assessed in the MALES study Model development using multivariable logistic regression A multivariable logistic regression model with all variables of interest included as explanatory variables (i.e. on the right side of the equation) was fitted to the Cologne dataset. The variables alcohol intake and Table 1 Descriptive statistics of ED and associated risk factors Cologne male survey (n = 4396) Berlin study (n = 1817) Males study (n = 20535) Erectile dysfunction, n (%) 849 (19.3) 983 (54.1) 3371 (16.4) Associated risk factors, n (%) Age in years, mean (SD) range 51.7 a (13.1) b (10.9) c (12.1) Arterial circulatory disorder 373 (8.5) 137 (7.5) NA Diabetes mellitus 284 (6.5) 143 (7.9) 1370 (6.7) Heart disease 304 (6.9) 196 (10.8) 2057 (10.0) Hypertension 757 (17.2) 515 (28.3) 4970 (24.2) Pelvic surgery 243 (5.5) 141 (7.8) NA Smoker 1479 (33.6) 529 (29.1) 6623 (32.3) Lower urinary tract symptoms d 1939 (44.3) Abbreviation: SD, standard deviation ; NA, not available. a Age was treated as a continuous variable in order to warrant efficient modeling and easy conversion of a final model into a simple scoring scheme. b Age was recorded in half decades; interval midpoints of age classes were chosen. c Age was recorded in decades; interval midpoints of age classes were chosen. d n = 4381.

4 498 M. Hellmich et al. / European Urology 48 (2005) Table 2 Details of the ED Risk Score model (bold entries) and the model with LUTS added to the regression equation (non-bold entries), and univariable associations of risk factors with ED in all three datasets Variable Cologne male survey Berlin study Males study Multivariable (without/with LUTS) Univariable Univariable Univariable B SE(B) OR 95% CI a OR 95% CI a OR 95% CI OR 95% CI Age (years) Pelvic surgery b NA NA Diabetes mellitus b Arterial circulatory disorder b NA NA Heart disease b Smoker b Hypertension b Constant LUTS b,c NA NA NA NA B, regression coefficient; SE, standard error; OR, odds ratio in favor of having ED; NA, not available. a All p-values < b Coding for dichotomous variables: no = 0, yes = 1. c n = Fig. 2. The ED Risk Score equation (a) and a worked example for practical use (b).

5 M. Hellmich et al. / European Urology 48 (2005) slipped disk were dropped from the model equation due to lack of statistical significance. The variable LUTS was excluded, though statistically significant, because physicians and patients may find the diagnosis of LUTS by means of the IPSS, a seven-item questionnaire, too complex. Moreover, if LUTS was included, the AUC would improve by only Table 2 gives the details on the model built from the remaining variables, the ED Risk Score model (also shown are details of the model including LUTS). In the ED Risk Score model the odds ratio in favor of having ED for a 1-year difference in age is exp(0.085) = 1.089, for a 10-year difference exp( ) = 2.340, and for a 20-year difference exp( ) = Thus Fig. 3. Point scoring scheme for the individual likelihood of ED.

6 500 M. Hellmich et al. / European Urology 48 (2005) the odds in favor of having ED, defined as p/(1 p), more than double every decade. The increase of the ED likelihood induced by pelvic surgery corresponds to the effect of about 21 additional years of age. The use of the ED Risk Score equation is exemplified in Fig. 2. For the corresponding scoring scheme shown in Fig. 3 the coefficients B from Table 2 were deliberately multiplied by 1/0.085 and rounded to the next integer. Thus one point in the scheme corresponds to one year of age. Due to rounding errors the results from the equation and the scoring scheme may differ about three percentage points at most. To continue the worked example shown in Fig. 2, a 63-year-old man with diabetes and heart disease gets = 86 points in total which corresponds to an estimated ED likelihood of 0.62 or 62%. The ED Risk Score model yielded a discrimination index (AUC) of 0.84 with a 95% confidence interval (CI) from 0.83 to 0.85 and did not show a statistically significant lack of calibration (HL statistic with p- value: 19.4, 0.013). The model explained 36.8% of the variation in the dependent variable (Nagelkerke s R 2 ). A threshold value of 0.25 for the estimated individual ED likelihood (p) yielded a classification rule which had a sensitivity of 67.3% and a specificity of 83.0% on the development sample (see Table 3). The derived scoring scheme yielded the same AUC and 95% CI as the original ED Risk Score model but showed slightly worse calibration (HL: 23.2, 0.003). Multicollinearity, meaning the presence of a high correlation between the explanatory variables, does not seem to be a problem for the ED Risk Score model, for example all variance inflation factors were below Validation of the ED risk score model To assess in an internal and rough manner how well the ED Risk Score model developed on the Cologne dataset might perform on other samples, the model was fitted to each of 1000 bootstrap samples (same risk factors but presumably different coefficients) and then applied without change to the original sample. This yielded an estimated optimism in the AUC estimate of 0.1% and a maximum error in predicted probabilities of 0.3% (estimate of unreliability). Additionally, model performance is usually assessed by using external data collected from appropriate populations in different settings. Specifically, application of the ED Risk Score equation to the national dataset (Berlin study) yielded an AUC of 0.75 (95% CI ). On the multinational dataset (MALES study) the equation yielded an AUC of 0.67 (95% CI , see Table 3). In both instances the HL test was statistically significant at level indicating some lack of calibration. Fig. 4 shows a scatterplot of the predicted vs. the observed likelihood, determined separately for each study sample using the ED Risk Score equation; calibration is good, if points are close to the bisecting line. Observations were grouped according to their predicted likelihood (0 to <0.05, 0.05 to <0.15,..., 0.95 to 1). Predicted and observed likelihood were then determined within these groups. Lack of calibration appears as non-linear curve progression (Berlin and MALES data). Table 3 Performance of the ED Risk Score model Development sample Validation samples Cologne male survey Berlin study Males study a AUC (95% CI) ( ) ( ) ( ) Best model b ( ) ( ) Hosmer Lemeshow statistic (p-value) 19.4 (0.013) (<0.001) (<0.001) Best model 1.0 (0.995) 29.6 (<0.001) Threshold value for p (< corresponds to no ED, to ED) Cologne male survey Berlin study Males study Sensitivity (%) Specificity (%) PPV (%) NPV (%) PPV, positive predictive value; NPV, negative predictive value. a Variables in the model equation that were unavailable in the MALES dataset were imputed by the prevalence of the corresponding condition as estimated from the Cologne dataset. b Multivariable logistic regression model with the same explanatory variables as ED Risk Score model fitted to respective dataset.

7 M. Hellmich et al. / European Urology 48 (2005) The AUC (model discrimination) results of the ED Risk Score model compare well with the performance of other prediction scores. For example, as applied in the management of coronary heart disease, the Framingham score [21] reached AUC values of 0.79 (men) and 0.83 (women) and the PROCAM score [22] yielded 0.83, and for osteoporosis, the simple calculated osteoporosis risk estimation (SCORE) [23] reached an AUC of 0.70 in older Caucasian women. Internal validation of the final model equation did not give any indication of relevant overfit in the obtained indices for discrimination and calibration. The external validation results compare well with those of the Framingham score which, applied to the data of multiple ethnic groups, reached AUC values within ranges of (men) and (women), respectively [24]. Similarly, external validation of the SCORE for patient selection for bone densitometry yielded AUC values between 0.67 (lumbar spine) and 0.73 (femoral neck) [25]. A low score on this somatic tool does not exclude the possibility that the patient has ED. Specifically, other etiologies/diseases such as central/peripheral nervous, latent cardiovascular, endocrinological, or psychogenic conditions are fully possible. Risk scores usually are rough and ready tools, hence many factors associated with increased likelihood of disease cannot be included, such as LUTS or extremes of risk factors (e.g. very heavy smoking). However, if the ED Risk Score indicates a high likelihood of ED in a patient, the physician may feel reassured to address the sensitive topic with patients to whom it really matters (in terms of high prevalence/incidence). He can then move on to confirm the diagnosis and the severity of ED with established instruments (e.g. IIEF, KEED). A simple question such as are you having any problems with sexual activity? is not an alternative to but may be a consequence of using the ED Risk Score or vice versa. The tool may prove beneficial to patient management both as a screening device (whom to ask) and as a means to facilitate discussion on the sensitive topic of sexual function (how to ask). Decision on treatment, however, will depend on many additional factors, including the burden that ED causes the patient, the presence of concomitant diseases, psychogenic and partner-related issues. Fig. 4. Scatterplot of predicted vs. observed likelihood (with Clopper Pearson 95% confidence intervals). 4. Discussion 4.1. Limitations Since none of the three datasets was designed to develop or validate an ED risk equation, the results of the external validation need to be interpreted cautiously. First, definitions of study variables, particularly ED and heart disease, were not standardized. In particular, the large difference in the prevalence of ED likely is a consequence of the use of different questionnaires/items and definitions (e.g. the MALES classification of ED used here is very rough). For example, in the Berlin study all men without sexual activity within a specified timeframe were supposed to have ED. Second, there may be socio-economic, cultural, and other differences (e.g. the validity of mailed questionnaires is dubious as compared with interviews) between study samples, which may partly be due to the fact that only 30.3% of the surveyed Berlin men gave complete responses as required for the validation. Such low response fractions always cast serious doubts on the validity and generalizabililty of study findings. From a clinical viewpoint, however, calibration for the multinational MALES sample was still acceptable because the difference between observed and predicted likelihood was small in the prevalent range from 50 to 73 total score points (see Fig. 4). Apart from calibration issues, the ED Risk Score discriminated well (as indicated by the AUC value) between the presence and absence of ED in all three studies and, hence, showed acceptable validity across the three samples. Further research activities should focus on longitudinal (cohort) studies to clarify whether the predictors identified so far also determine individual risk of developing ED in the future. Assessment of the incidence of ED is not possible with cross-sectional data. Finally, in the present article ED was assumed to be a condition that is either present or absent in men. In real

8 502 M. Hellmich et al. / European Urology 48 (2005) life, however, ED occurs with varying degrees of severity. Thus, future work should aim at the investigation of the association of disease severity and potential risk factors, as this is likely to yield a better statistical model. Unfortunately, the validation of such a model would require data that are currently not available. established methods is necessary, the ED Risk Score model can be used as a screening tool to identify patients who are at high risk of somatic ED. By moving from general medical questions to the issue of sexual function it can facilitate a more open discussion of this sensitive topic between the patient and the physician. 5. Conclusions The ED Risk Score model can calculate a patient s individual likelihood of somatic ED based on the patient s age and existing health conditions, which are normally collected as part of a routine health check. While confirmation of the clinical diagnosis with Acknowledgements This work was supported by research grants from Bayer HealthCare AG. The authors thank the expert reviewers and the Editorial Board for their constructive comments that helped to improve the presentation of the paper. References [1] Lue TF, Giuliano F, Montorsi F, Rosen RC, Andersson K-E, Althof S, et al. Summary of the recommendations on sexual dysfunctions in men. J Sex Med 2004;1:6 23. [2] Lewis RW, Fugl-Meyer KS, Bosch R, Fugl-Meyer AR, Laumann EO, Lizza E, et al. Epidemiology/risk factors of sexual dysfunction. J Sex Med 2004;1:35 9. [3] Laumann EO, Nicolosi A, Glasser DB, Paik A, Gingell C, Moreira E, et al., for the GSSAB Investigators Group. Sexual problems among women and men aged years: prevalence and correlates identified in the Global Study of Sexual Attitudes and Behaviours. Int J Impot Res 2005;17: [4] de Boer LJ, Bots ML, Lycklama a Nijeholt AAB, Moors JPC, Pieters HM, Verheij TJM. Impact of various questionnaires on the prevalence of erectile dysfunction. The ENIGMA study. Int J Impot Res 2004;16: [5] Braun M, Wassmer G, Klotz T, Reifenrath B, Mathers M, Engelmann U. Epidemiology of erectile dysfunction: results of the Cologne Male Survey. Int J Impot Res 2000;12: [6] Baldwin K, Ginsberg P, Harkaway RC. Under-reporting of erectile dysfunction among men with unrelated urologic conditions. Int J Impot Res 2003;15:87 9. [7] Costa P, Avances C, Wagner L. Erectile dysfunction: knowledge, wishes and attitudes: results of a French study of 5,009 men aged 17 to 70. Prog Urol 2003;13: [8] Bosinski HAG. Diagnosis and therapy of sexual disorders: an interdisciplinary challenge. Urologe A 2004;43:279 84, (in German). [9] Bacon CG, Mittleman MA, Kawachi I, Giovannucci E, Glasser DB, Rimm EB. Sexual function in men older than 50 years of age: results from the health professional follow-up study. Ann Intern Med 2003;139: [10] Nicolosi A, Moreira ED, Shirai M, Bin Mohd Tambi MI, Glasser DB. Epidemiology of erectile dysfunction in four countries: cross-national study of the prevalence and correlates of erectile dysfunction. Urology 2003;61: [11] Rosen R, Fisher W, Eardley I, Niederberger C, Nadel A, Sand M. The multinational Men s Attitudes to Life Events and Sexuality (MALES) study: I. Prevalence of erectile dysfunction and related health concerns in the general population. Curr Med Res Opin 2004;20: [12] El-Sakka AI, Morsy AM, Fagih BI, Nassar AH. Coronary artery risk factors in patients with erectile dysfunction. J Urol 2004;172: [13] Montorsi F, Briganti A, Salonia A, Rigatti P, Margonato A, Macchi A, et al. Erectile dysfunction prevalence, time of onset and association with risk factors in 300 consecutive patients with acute chest pain and angiographically documented coronary artery disease. Eur Urol 2003;44: [14] Schäfer GA, Englert HS, Ahlers CJ, Roll S, Willich SN, Beier KM. Erektionsstörung und Lebensqualität Erste Ergebnisse der Berliner Männer-Studie. Sexuologie 2003;10: [15] Braun M, Klotz T, Reifenrath B, Wassmer G, Engelmann U. KEED erster deutschsprachiger validierter Fragebogen zur Erfassung der männlichen sexuellen Funktion. Akt Urol 1998;29: [16] Rosen RC, Riley A, Wagner G, Osterloh IH, Kirkpatrick J, Mishra A. The international index of erectile function (IIEF). A multidimensional scale for assessment of erectile dysfunction. Urology 1997;49: [17] Cappelleri JC, Rosen RC, Smith MD, Mishra A, Osterloh IH. Diagnostic evaluation of the erectile function domain of the International Index of Erectile Function. Urology 1999;54: [18] Hosmer DW, Lemeshow S. Applied Logistic Regression. New York: John Wiley and Sons; [19] Harrell Jr FE. Regression Modeling Strategies with Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer; [20] Esposito K, Giugliano F, Di Palo C, Giugliano G, Marfella R, D Andrea F, et al. Effect of lifestyle changes on erectile dysfunction in obese men: a randomised controlled trial. JAMA 2004;291: [21] Wilson PWF, D Agostino RB, Levy D, Belanger A, Silbershatz H, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation 1998;97: [22] Assmann G, Cullen P, Schulte H. Simple scoring scheme for calculating the risk of acute coronary events based on the 10-year follow-up of the prospective cardiovascular Münster (PROCAM) study. Circulation 2002;105: [23] Von Mühlen D, Visby Lunde A, Barrett-Connor E, Bettencourt R. Evaluation of the simple calculated osteoporosis risk estimation (SCORE) in older Caucasian women: the Rancho Bernado Study. Osteoporos Int 1999;10: [24] D Agostino RB, Grundy S, Sullivan LM, Wilson P, for the CHD Risk Prediction Group. Validation of the Framingham coronary heart disease prediction scores. JAMA 2001;286: [25] Cadarette SM, Jaglal SB, Murray TM. Validation of the simple calculated osteoporosis risk estimation (SCORE) for patient selection for bone densitometry. Osteoporos Int 1999;10:85 90.