Referat zum Thema Epidemiologie SS2007 Anhaltspunkte
Transcription
Referat zum Thema Epidemiologie SS2007 Anhaltspunkte
Referat zum Thema Epidemiologie SS2007 Artikel: Klaz al. (2006): Severe acne vulgaris and tobacco smoking in young men. Journal of Investigative Dermatology 126, 1749-1752 Betreuer: PD. Dr. med. Michael Weichenthal (Dermatologie, Tel.: 597-1537, E-Mail: [email protected] ) Dr. rer. nat. Amke Caliebe (IMIS, Tel.: 597-3199, E-Mail: [email protected]) Anhaltspunkte Einleitung/Studie. Motivation, Zielsetzung Zielgröße, Einflussgröße(n) Studien-Design Statistische Verfahren. Prävalenz χ2-Test Odds-Ratio: Bedeutung und Aussage, Verhältnis zum Relativen Risiko Konfidenzintervalle für Odds-Ratios Logistische Regression: Interpretation der Ergebnisse, Berechnung der Odds-Ratios, Adjustierung Ergebnisse. Darstellung, Bewertung und Interpretation der Ergebnisse Diskussion. Kritische Bewertung der Studie: Repräsentativität, Optimalität des Studiendesigns Kausalität Statistische Signifikanz gegenüber klinischer Relevanz Störgrößen und Maßnahmen dagegen (Confounding, Bias) Anwendbarkeit und Limitierung der Resultate Folgestudien In allen Punkten sollte ein Bezug zur vorliegenden Studie hergestellt werden. Die aufgeführten Stichpunkte dienen nur der Orientierung. Die Setzung von Schwerpunkten, der Aufbau des Referats und das eventuelle Einbringen von zusätzlichen Aspekten ist den Referenten überlassen. ORIGINAL ARTICLE Severe Acne Vulgaris and Tobacco Smoking in Young Men Itay Klaz1,2, Ilan Kochba1, Tzipora Shohat1, Salman Zarka1 and Sarah Brenner2 As the relationship between tobacco smoking and acne remains unclear, we examined the relationship between cigarette smoking and severe acne in a large cohort of young men. Trained nurses interviewed subjects upon discharge from compulsory military service, regarding family history, habits, and tobacco smoking habits. Data was correlated with severe acne status, as diagnosed and coded by board-certified dermatologists. In total, 27,083 male subjects participated in the study from 1983 to 2003, of which 237 (0.88%) had severe acne, 11,718 (43.27%) were active smokers, and 15,365 (56.73%) were nonsmokers at the time of interviews. Active smokers showed a significantly lower prevalence of severe acne (0.71%) than nonsmokers (1.01%) (P ¼ 0.0078). An inverse dose-dependent relationship between severe acne prevalence and daily cigarette consumption became significant from 21 cigarettes a day (w2 and trend test: Po0.0001), odds ratio: 0.2 (95% CI: 0.06–0.63). The study did not aim to establish a temporal correlation, and passive smoking and acne treatments were not measured. Previous in vitro and clinical studies strongly support an association with nicotine. We suggest a trial with topical nicotine treatment for acne to further investigate this association. Journal of Investigative Dermatology (2006) 126, 1749–1752. doi:10.1038/sj.jid.5700326; published online 27 April 2006 INTRODUCTION Acne vulgaris affects over 80% of all individuals during childhood and early adult life, with male subjects more commonly affected than female subjects. Prevalence of the severe form, characterized by multiple nodular and postular-cystic lesions (Kligman and Plewig, 1976; O’Brien et al., 1998) that can leave permanent physical and psychological scars (Tan, 2004), ranges from 0.5% to 6% in males, depending on age and the clinical grading system used (Cunliffe and Gould, 1979; Rademaker et al., 1989; Stern, 1992; Lello et al., 1995; Bataille et al., 2002). In view of the controversial association between acne and smoking (Mills et al., 1993; Schafer et al., 2001; Jemec et al., 2002; Firooz et al., 2005), we studied the relationship in a large group of young men being discharged from military service. It is crucial to emphasize that any positive effects found must be traced to specific tobacco components that can be therapeutically used without smoking (e.g., nicotine patches or gums), to avoid any ‘‘legitimatizing’’ of smoking based on its beneficial effects on health. 1 Medical Corps, Israel Defense Forces, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel and 2Department of Dermatology, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel Correspondence: Dr Itay Klaz, Department of Dermatology, Tel Aviv Sourasky Medical Center, 6 Weizman Street, Tel Aviv 64239, Israel. E-mail: [email protected] Abbreviation: IDF, Israel Defense Force Received 18 September 2005; revised 8 March 2006; accepted 8 March 2006; published online 27 April 2006 & 2006 The Society for Investigative Dermatology RESULTS A total of 27,083 males participated in the study during the years 1983–2003. Approximately 20 military nurses and over 30 board-certified dermatologists participated in the interview process and on the medical profile committees, respectively. There was severe acne in 237 (0.88%) subjects (Table 1). At the time of interview, 11,718 (43.27%) were active smokers and 15,365 (56.73%) were nonsmokers; 0.37% (n ¼ 99) subjects did not report their smoking status. Categories of the daily cigarette smoking of the 26,984 subjects were 0 cigarettes (n ¼ 15,365), 110 (n ¼ 2,746), 1120 (n ¼ 3,766), 21–30 (n ¼ 1,645), 31–40 (n ¼ 2,086), and 441 (n ¼ 1,475). The prevalence of severe acne was significantly lower (P ¼ 0.0078) in active smokers (0.71%) than in nonsmokers (1.01%). There was an inverse, dosedependent relationship between severe acne prevalence and daily cigarette consumption. The prevalence of acne in the aforementioned categories was 0.99, 1.27, 1.04, 0.18, 0.24, and 0.20%, respectively. The inverse relationship became statistically significant from 21 cigarettes a day (w2 and trend test: Po0.0001). When the relationship between daily cigarette smoking and severe acne prevalence was controlled for father’s origin and number of siblings, there was still a significant dosedependent association between cigarette consumption and acne (see Table 2 and Figure 1). DISCUSSION The association between acne and smoking has been studied extensively with varying results. Mills et al. (1993) reported that 19.7% of 96 male acne vulgaris patients and 12.1% of 60 www.jidonline.org 1749 I Klaz et al. Acne and smoking in men Table 1. Prevalence of severe acne and demographic data on subjects Origin of subjects (% of acne subgroup: no, yes)a Severe acne No. of subjects subgroup (%) (n=27,073) 7s.d. West East Israel siblings 7s.d. No 26,846 (99.12%) 21.8471.52 40.85% 52.65% 6.49% 4.1272.29 21.8571.16 48.93% 43.78% 7.30% 3.5271.70 Yes 237 (0.88%) Age (years) Average no. of a Descendants of West European and North and South American-born fathers were categorized as ‘‘Western’’, and descendants of fathers born in Eastern Europe, Africa, and Arab countries were categorized as ‘‘Eastern’’. Table 2. Multiple logistic regression model for acne prevalence and cigarette consumption Variable Odds ratio 95% confidence interval Daily cigarette smoking 0 (reference) 1 1–10 1.28 0.88 1.86 11–20 1.06 0.74 1.52 20–30 0.20 0.06 0.63 30–40 0.28 0.11 0.68 440 0.25 0.08 0.78 Western vs Israeli 1.13 0.67 1.89 Western vs Eastern 0.89 0.67 1.19 0.90 0.84 0.98 Ethnic origin No. of siblings (ordinal variable) 0.6 Severe acne % 0.5 0.4 0.3 0.2 0.1 0 0 1−10 11−20 21−30 31−40 41−50 Cigarettes smoked per day 51−90 Figure 1. Relationship between prevalence of severe acne (n ¼ 237) and number of cigarettes smoked per day. female ones were smokers, which was significantly less than national statistics (Mills et al., 1993). Jemec et al. (2002) found that smoking was not significantly associated with acne in a random sample of 186 subjects (odds ratio: 0.54, 95% CI: 0.17–1.78). Schafer et al. (2001) found in 102 smokers compared with 184 nonsmokers that acne was significantly more prevalent in active smokers, but when only 15–40-yearolds were taken into account, there was no association 1750 Journal of Investigative Dermatology (2006), Volume 126 between acne and smoking. Firooz et al. (2005) compared smoking status of 293 acne patients to 301 patients suffering from other dermatological conditions. After accounting for acne’s higher prevalence and greater severity in men, no significant correlation was found. Our study in which all subjects were diagnosed by board-certified dermatologists is the largest sample to date. Although different dermatologists had participated during the study years, all used the same diagnosis criteria, thereby compensating in part for possible inter-observer variability. Our cohort consisted only of males, which probably skewed the results in view of the known differences in prevalence and clinical grading between male and female subjects, most probably owing to hormonal differences (both androgenic and oral contraceptive related). In this respect, a study carried out only on males has the advantage of eliminating certain confounding gender-specific factors, but any generalizing of results must be performed with caution. Although the prevalence of severe acne among smokers of 10–20 cigarettes per day was higher than the group of 0–10 per day, this finding was not statistically significant. The inverse relationship became significant from 21 cigarettes a day. A study measuring plasma cortisol levels in smokers found a similar effect (del Arbol et al., 2000). The differences between light and heavy smokers may be related to the effect of nicotine on nicotinic cholinergic receptors. At low doses, nicotine stimulates acetylcholine receptors, whereas high doses of nicotine selectively block them (Seyler et al., 1986). The aspect of passive smoking was not included in this study owing to the difficulty of accurately estimating exposure in such a large subject population and long study duration. All dermatologists were obliged to refer any patient with severe acne for official medical coding. Based on internal auditing, we estimate an under-diagnosis of up to 20% owing mostly to the large number of participating dermatologists, some of whom were civilian physicians less familiar with the military referral directives. The large sample size and the lack of any selective bias linked with smoking habits on the diagnosis or medical coding procedure probably compensated for this under-diagnosis. Another limitation of our study was the inclusion of only patients with severe acne, most of whom were referred to a dermatologist for retinoid treatment. However, as the clinico-pathological pathways of both moderate and severe acne involve inflammatory processes, I Klaz et al. Acne and smoking in men we assumed that the findings can be applied to a broad range of disease presentations, with possible exclusion of comedonic only grades. Social factors may also have contributed to the findings of our study. Individuals with higher self-awareness of health status, hence less prone to smoking, may have sought diagnosis and treatment for acne more vigorously than the smoking population. However, an earlier study of the same population found similar rates of smoking among subjects diagnosed with asthma as compared to healthy individuals (Zimlichman et al., 2004), supporting our assumption that health awareness was not a major reason for the present findings. Emotional stress is a known risk factor for smoking (Simantov et al., 2000). More smokers due to stress are expected among the severe acne group. Our finding to the contrary might have significance, but this remains to be verified. The adverse effects of tobacco on the skin are well known (Misery, 2004). Several studies suggest a possible protective action of nicotine against the development of inflammatory skin disorders. The nicotine constituent might even be beneficial to certain diseases. Positive effects were found in pemphigus, ulcerative colitis, pyoderma gangrenosum, aphthous stomatitis, and herpes simplex (Wolf et al., 2004). Nicotine enhanced keratinocyte adhesion, differentiation, and apoptosis and inhibited keratinocyte migration (Grando et al., 1995). Nicotine also inhibited inflammation through effects on the central and peripheral nervous systems (Sopori et al., 1998). Nicotine altered immune responses by directly interacting with T cells. Transdermal application of nicotine (patches) was followed by a decrease in response to sodium lauryl sulfate as well as the erythema response to UVB (Mills, 1998). Paradoxically, nicotine worsened buccal inflammation, in contrast to ameliorating small bowel and colonic inflammation (Eliakim and Karmeli, 2003). Using current acne status at the time of discharge from service time excluded subjects previously diagnosed with severe acne whose diagnosis changed owing to clinical improvement. However, owing to the cross-sectional nature of the study, it was not possible to delineate the time sequence of severe acne development and smoking. Some subjects may have started smoking after the onset, or even as a consequence, of acne, or vice versa. The dose-dependent relationship might indicate that smoking more than 20 cigarettes a day contributed somewhat in improving preexisting acne. Future prospective studies are needed to establish time sequence and therefore cause and effect. The following limitations of this large-scale study must be noted: the inclusion of only males and only severe acne; and the exclusion from the study protocol of acne therapy, data on the temporal relationship between smoking and development of acne, and the effect of passive smoking. The underlying causal mechanisms of the relationship between severe acne and smoking need further clarification, but previous in vitro and clinical studies strongly support an association with nicotine. We suggest a randomized controlled trial with topical nicotine treatment for acne to further investigate the significant inverse correlation between cigarette smoking and severe acne vulgaris observed in our study. MATERIALS AND METHODS The prospectively established database of the Israel Defense Forces (IDFs) Medical Corps during the years 1983–2003 served as the basis for this cross-sectional study. Subjects The study population came from a large-scale ongoing prospective survey of health behavior and attitudes, conducted among randomly selected soldiers of the IDFs. The survey systematically collected a representative sample of IDFs men at discharge from compulsory 3-year military service, ranging 21–22 years of age, as previously described (Kark and Laor, 1992). The IDFs Medical Corps Review Board approved the survey as well as the manner in which informed consent was obtained from the subjects. Each IDFs recruit went through medical tests at intake and in the event of a change in health status during service. Board-certified dermatologists made the diagnosis of severe acne according to the military criteria based on the Kligman and Plewig grading (1976) and the Leeds acne grading system (O’Brien et al., 1998); severe acne is defined as the presence of nodular and postular-cystic lesions. The diagnosing dermatologist then referred the soldiers to a military medical profiling committee for a review of the required clinical evidence and an official numerical encoding of the diagnosis. The data referred to the committee did not include the soldier’s smoking status. We used this code, called the medical military profile, to classify subjects with and without severe acne vulgaris. Data collection Subjects were asked to participate in the study on the day of discharge from military service. Trained nurses from the IDF Public Health Branch interviewed them about smoking status (current smoker, past smoker, or never smoked), average number of cigarettes smoked a day (0–10, 10–20, 20–30, 30–40, 440), father’s country of origin, found to correlate with smoking habits by Zimlichman et al. (2004), and number of siblings, found to be reversely associated with adult social class by Blane et al. (1999). Nurses did not know the current acne coding status of the soldier. Current acne status at discharge time as reflected by the medical profile was collected separately for all participants and analyzed against the above parameters. Statistical analysis Data were analyzed by the Statistical Analysis System version 9.2. (SAS Institute Inc., Cary, NC, USA). Proportions of smoking, family origin, and number of siblings were compared between subjects with severe acne and those without, using w2 test. Mean number of cigarettes smoked per day was compared using analysis of variance. Trend tests were performed in 2 N tables when appropriate. Multiple logistic and linear regression analyses were carried out taking severe acne prevalence as the dependent variable. All models included ethnic origin, number of siblings, and quantitative parameter of smoking status. The LOGISTIC and GLM procedures were used. Results were expressed as mean7s.d., or n (%); Po0.05 was considered significant. www.jidonline.org 1751 I Klaz et al. Acne and smoking in men CONFLICT OF INTEREST The authors state no conflict of interest. Kark JD, Laor A (1992) Cigarette smoking and educational level among young Israelis upon release from military service in 1988 – a public health challenge. Isr J Med Sci 28:33–7 ACKNOWLEDGMENTS Kligman AM, Plewig G (1976) Classification of acne. Cutis 17:520–2 We thank Dr Amir Tirosh and Dr Eyal Zimlichman of the IDFs Medical Corps for their helpful suggestions. We thank the nurses and medics of the Army Medical Corps Health Branch for administering the questionnaires. Funding was provided by the Medical Corps of the IDFs. No external funding was used. Lello J, Pearl A, Arroll B, Yallop J, Birchall NM (1995) Prevalence of acne vulgaris in Auckland senior high school students. N Z Med J 108:287–9 REFERENCES Bataille V, Snieder H, MacGregor AJ, Sasieni P, Spector TD (2002) The influence of genetics and environmental factors in the pathogenesis of acne: a twin study of acne in women. J Invest Dermatol 119: 1317–22 Blane D, Davey Smith G, Hart C (1999) Some social and physical correlates of intergenerational social mobility: evidence from the west of Scotland, collaborative study. Sociology 33:169–83 Cunliffe WL, Gould DJ (1979) Prevalence of facial acne vulgaris in late adolescence and in adults. BMJ 1:1109–10 del Arbol JL, Munoz JR, Ojeda L, Cascales AL, Irles JR, Miranda MT et al. (2000) Plasma concentrations of beta-endorphin in smokers who consume different numbers of cigarettes per day. Pharmacol Biochem Behav 67:25–8 Eliakim R, Karmeli F (2003) Divergent effects of nicotine administration on cytokine levels in rat small bowel mucosa, colonic mucosa, and blood. Isr Med Assoc J 5:178–80 Firooz A, Sarhangnejad R, Davoudi SM, Nassiri-Kashani M (2005) Acne and smoking: is there a relationship? BMC Dermatol 5:2–5 Grando SA, Horton RM, Pereira EF, Diethelm-Okita BM, George PM, Albuquerque EX et al. (1995) A nicotinic acetylcholine receptor regulating cell adhesion and motility is expressed on human keratinocytes. J Invest Dermatol 105:774–81 Jemec GBE, Linneberg A, Nielsen NH, Frolund L, Madsen F, Jorgensen T (2002) Have oral contraceptives reduced the prevalence of acne? A population-based study of acne vulgaris, tobacco smoking and oral contraceptives. Dermatology 204:179–84 1752 Journal of Investigative Dermatology (2006), Volume 126 Mills C (1998) Cigarette smoking, cutaneous immunity, and inflammatory response. Clin Dermatol 16:589–94 Mills CM, Peters TJ, Finlay AY (1993) Does smoking influence acne? Clin Exp Dermatol 18:100–1 Misery L (2004) Nicotine effects on skin: are they positive or negative? Exp Dermatol 13:665–70 O’Brien SC, Lewis JB, Cunliffe WJ (1998) The Leeds Revised Acne Grading System. J Dermatol Treat 9:215–20 Rademaker M, Garioch J, Simpson N (1989) Acne in schoolchildren: no longer a concern for dermatologists. Br Med J 298:1217–9 Schafer T, Niehnaus A, Vieluf D, Berger J, Ring J (2001) Epidemiology of acne in the general population: the risk of smoking. Br J Dermatol 145:100–4 Seyler LE, Pomerleau OF, Fertig JB, Hunt D, Parker K (1986) Pituitary hormone response to cigarette smoking. Pharmacol Biochem Behav 24: 159–62 Simantov E, Schoen C, Klein JD (2000) Health-compromising behaviors: why do adolescents smoke or drink? Identifying underlying risk and protective factors. Arch Pediatr Adolesc Med 154:1025–33 Sopori ML, Kozak W, Savage SM, Geng Y, Soszynski D, Kluger MJ et al. (1998) Effect of nicotine on the immune system: possible regulation of immune responses by central and peripheral mechanisms. Psychoneuroendocrinology 23:189–204 Stern R (1992) The prevalence of acne on the basis of physical examination. J Am Acad Dermatol 26:931–5 Tan JK (2004) Psychosocial impact of acne vulgaris: evaluating the evidence. Skin Therapy Lett 9:1–3, 9 Wolf R, Orion E, Matz H, Maitra S, Rowland-Payne C (2004) Smoking can be good for you. J Cosmet Dermat 3:107–11 Zimlichman E, Mandel D, Mimouni FB, Shochat T, Grotto I, Kreiss Y (2004) Smoking habits in adolescents with mild to moderate asthma. Pediatr Pulmonol 38:193–7 R. Bender1 Logistische Regression A. Ziegler2 St. Lange3 Mit Hilfe der linearen Regression lässt sich der Einfluss einer oder mehrerer erklärender Variablen X1,...,Xm (z.B. X1 = Alter, X2 = Geschlecht und X3 = Rauchen) auf eine stetige Zielvariable Y (z.B. Y = systolischer Blutdruck) statistisch untersuchen (3). Liegt nur eine erklärende Variable X vor, spricht man von der einfachen linearen Regression (engl.: simple linear regression) und verwendet die Geradengleichung (5) Y = α + βX . Im Fall mehrerer erklärender Variablen X1,...,Xm liegt das Modell der multiplen linearen Regression (engl.: multiple linear regression) vor, das durch die Gleichung Y = α + β1X1 + ... + βmXm beschrieben wird (3). Die Bedeutung der multiplen Regressionsmodelle in der medizinischen Statistik liegt zum einen darin, den gemeinsamen Einfluss mehrerer Variablen auf eine Zielvariable untersuchen zu können und zum anderen in der Möglichkeit, den interessierenden Effekt einer Variable bezüglich anderer Variablen zu adjustieren, um eine Verzerrung (engl.: bias) bei der Effektschätzung zu reduzieren (3). Logistische Regression Die logistische Regression (engl.: logistic regression) kommt als Auswertungsmethode in Frage, wenn man den Einfluss erklärender Variablen X1,...,Xm auf eine Zielvariable Y untersuchen möchte, und Y binäres Messniveau besitzt (z.B. Y = Krankheit ja/nein). Da Y nur die beiden Werte 1 = ja und 0 = nein annehmen kann, ist die Anwendung der linearen Regression in der Regel nicht sinnvoll. Betrachten wir zur Modellentwicklung zunächst den einfachen Fall von nur einer erklärenden Variable X. Der Schlüssel zur quantitativen Beschreibung eines Zusammenhangs zwischen Y und X liegt darin, anstelle von Y die Wahrscheinlichkeit für den Eintritt des Zielereignisses p = P(Y = 1) zu modellieren. In medizinischen Anwendungen ist die Wahrscheinlichkeit p meist ein Risiko für eine bestimmte Krankheit. Während Y nur die beiden Ausprägungen 1 und 0 besitzt, kann das Risiko p jede beliebige Zahl zwischen 0 und 1 annehmen. Die Chance (engl.: odds) p/(1-p) kann jede beliebige positive Zahl annehmen (2) und der Logarithmus der Chance log[p/(1-p)], genannt logit, besitzt die ganze reelle Zahlenmenge als Wertebereich. Damit ist es häufig sinnvoll, eine lineare Beziehung zwischen dem logit von p und X anzunehmen, d.h. T 11 logit (p) = log[p/(1-p)] = α+βX , was mathematisch äquivalent ist mit p= exp (α + βX) 1 + exp (α + βX) exp bezeichnet hierbei die Exponentialfunktion. Der rechte Term obiger Gleichung stellt die so genannte logistische Funktion dar, daher erklärt sich die Bezeichnung »logistische Regression«. Die Erweiterung auf ein multiples Modell mit mehreren erklärenden Variablen erhält man wie bei der linearen Regression, indem βX ersetzt wird durch die Linearkombination Institut AG Epidemiologie und Medizinische Statistik (Leitung: Prof. Dr. M. Blettner), Fakultät für Gesundheitswissenschaften , Universität Bielefeld 2 Institut für Medizinische Biometrie und Statistik (Direktor: Prof. Dr. A. Ziegler), Universitätsklinikum Lübeck, Medizinische Universität zu Lübeck 3 Abteilung für Medizinische Informatik, Biometrie u. Epidemiologie (Direktor: Prof. Dr. H.J. Trampisch), RuhrUniversität Bochum 1 Korrespondenz PD Dr. rer.biol.hum. Ralf Bender · AG Epidemiologie und Medizinische Statistik Fakultät für Gesundheitswissenschaften Universität Bielefeld · Postfach 100131 · 33501 Bielefeld · E-Mail: [email protected] Bibliografie Dtsch Med Wochenschr 2002; 127: T 11–T 13 · © Georg Thieme Verlag Stuttgart · New York · ISSN 0012-0472 it tsai ttSa t S Serie | Statistik Lineare Regression - Artikel Nr. 14 der Statistik-Serie in der DMW - Tab.1 Einfache logistische Regressionsanalyse für die Entwicklung einer diabetischen Nephropathie nach 6 Jahren bei 480 Typ 1 Diabetikern. Risikofaktor Regressionskoeffizient Achsenabschnitt HbA1c Tab.2 Standardfehler p-Wert – 5,089 0,731 0,0001 +0,457 0,089 0,0001 Differenz für Odds Ratio Odds Ratio 95% Konfidenzintervall 1% 1,58 1,33 – 1,88 Multiple logistische Regressionsanalyse für die Entwicklung einer diabetischen Nephropathie nach 6 Jahren bei 480 Typ 1 Diabetikern. Statistiken Risikofaktor Regressionskoeffizient Standardfehler p-Wert Differenz für Odds Ratio Achsenabschnitt – 8,980 1,736 0,0001 HbA1c +0,464 0,091 0,0001 1% 1,59 1,33 – 1,90 diast. Blutdruck +0,048 0,019 0,0148 5mm Hg 1,27 1,05 – 1,54 95% Konfidenzintervall Diabetesdauer +0,004 0,018 0,8220 5 Jahre 1,02 0,85 – 1,22 Geschlecht – 0,025 0,249 0,9212 männl. vs. weibl. 0,98 0,60 – 1,59 β1X1+...+βmXm. Zur Schätzung der logistischen Regressionskoeffizienten werden in der Praxis iterative Algorithmen eingesetzt. T 12 Odds Ratio Wie bei der linearen Regression muss auch bei der logistischen Regression die Modellgüte (engl.: goodness-of-fit) untersucht werden. Auf die entsprechenden Methoden können wir hier nicht eingehen. Der interessierte Leser sei auf die Literatur verwiesen (5). Außer der logistischen Regression für binäre Zielvariablen gibt es Modellerweiterungen für nominale und ordinale Daten. Das bekannteste Modell ist hierbei das proportionale Odds Modell für ordinale Zielvariablen (1). Beispiel Mit Hilfe der logistischen Regression wurde der Einfluss von Risikofaktoren auf die Entwicklung der diabetischen Nephropathie bei Typ 1 Diabetikern untersucht (7). Betrachten wir zunächst nur das glykierte Hämoglobin (HbA1c) als Risikofaktor. In Dtsch Med Wochenschr 2002; 127: T11–T13 · R. Bender, Logistische Regression Risiko für Nephropathie Da in der medizinische Forschung oftmals binäre Zielvariablen auftreten, wird die logistische Regression in der Praxis sehr häufig angewendet. Eine besondere Stellung erhält das logistische Regressionsmodell dadurch, dass man sowohl für prospektive Kohortenstudien als auch für retrospektive Fall-Kontroll Studien sinnvoll interpretierbare Effektschätzer erhält. Das gebräuchliche Effektmaß in der Epidemiologie ist das Odds Ratio (OR), das als Verhältnis der Chancen zwischen exponierten und nicht exponierten Personen definiert ist (2). Aus dem Regressionskoeffizient β einer logistischen Regression kann direkt das Odds Ratio berechnet werden durch OR = exp(β). In einem multiplen Modell kann für die Beziehung zwischen Y und einer erklärenden Variablen Xj das aus βj berechnete ORj =exp(βj) als das nach allen anderen erklärenden Variablen adjustierte Odds Ratio betrachtet werden. Bei stetigen erklärenden Variablen bezieht sich der Wert des Odds Ratios auf die Erhöhung der erklärenden Variablen um jeweils 1 Einheit bzw. auf den Anstieg einer vorher definierten klinisch relevanten Differenz (siehe Beispiel). 1,0 0,9 0,8 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0,0 4 5 6 7 8 9 10 11 12 13 14 15 16 HbA1c (%) 17 Abb.1 Risiko für die Entwicklung einer diabetische Nephropathie nach 6 Jahren in Abhängigkeit vom HbA1c bei Typ 1 Diabetes, berechnet mit Hilfe einfacher logistischer Regressionsanalyse (n = 480). der einfachen logistischen Regressionsanalyse ist das HbA1c ein signifikanter Risikofaktor (Tab.1). Die Stärke des Effekts lässt sich mit Hilfe des Odds Ratios angeben. Pro Einheit des HbA1c (1%) steigt die Chance nach 6 Jahren eine diabetische Nephropathie zu entwickeln um den Faktor von OR = 1.6 (95% Konfidenzintervall 1,3–1,9). Dieser Zusammenhang lässt sich auch grafisch veranschaulichen, indem das Risiko als Funktion des Risikofaktors dargestellt wird (Abb.1). Für HbA1c-Werte im Normalbereich (4,3– 6,1%) liegt das Risiko, eine diabetische Nephropathie zu entwickeln, unter 10%, während es bei extrem hohen HbA1c-Werten von 16% und höher auf über 90% ansteigt. Diese Ergebnisse verdeutlichen die starke Assoziation zwischen der Stoffwechseleinstellung und dem Risiko diabetischer Spätschäden bei Typ 1 Diabetes. Um zu zeigen, dass eine Reduktion des HbA1c auch zu einer Reduktion des Risikos für diabetische Spätschäden führt, benötigt man allerdings entsprechende Er- Literatur Tab.3 Übersetzung (deutsch – englisch). 1 Englisch erklärende Variable explanatory variable Zielvariable response variable einfache lineare Regression simple linear regression multiple lineare Regression multiple linear regression adjustieren adjust Verzerrung bias logistische Regression logistic regression binär binary Chance odds Kohortenstudie cohort study Fall-Kontroll Studie case-control study Regressionskoeffizient regression coefficient adjustiertes Odds Ratio adjusted odds ratio Modellgüte goodness-of-fit proportionales Odds Modell proportional odds model 2 3 4 5 6 7 Bender R, Grouven U. Ordinal logistic regression in medical research. J R Coll Physic London 1997; 31: 546–551 Bender R, Lange S. Die Vierfeldertafel. Dtsch Med Wochenschr 2001; 126: T36–T38 Bender R, Ziegler A, Lange S. Multiple Regression. Dtsch Med Wochenschr 2002; 127: T8–T10 The Diabetes Control and Complications Trial Research Group. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med 1993; 329: 977–986 Hosmer DW, Lemeshow S. Applied Logistic Regression. Wiley, New York, 1989 Lange S, Bender R. (Lineare) Regression/Korrelation. Dtsch Med Wochenschr 2001; 126: T33–T35 Mühlhauser I, Bender R, Bott U, Jörgens V, Grüsser M, Wagener W, Overmann H, Berger M. Cigarette smoking and progression of retinopathy and nephropathy in type 1 diabetes. Diabet Med 1996; 13: 536–543 Statistiken Deutsch gebnisse einer randomisierten klinischen Therapiestudie, wie z.B. den Diabetes Control and Complications Trial (DCCT, 4). Neben dem glykierten Hämoglobin gibt es noch weitere Risikofaktoren, die hier in Betracht gezogen werden müssen, vor allem Blutdruck, Diabetesdauer und möglicherweise das Geschlecht. Die Ergebnisse einer multiplen logistischen Regressionsanalyse zeigen, dass das HbA1c und der diastolische Blutdruck signifikante Risikofaktoren darstellen, während ein Effekt der Diabetesdauer und des Geschlechts nicht nachweisbar ist (Tab.2). T 13 Zur Darstellung des Odds Ratios wurde für den diastolischen Blutdruck eine Differenz von 5mm Hg und für die Diabetesdauer von 5 Jahren gewählt, da eine Erhöhung dieser Risikofaktoren um jeweils eine Einheit (1mm Hg bzw. 1 Jahr) nicht als klinisch relevante Änderung angesehen wird. Es lässt sich somit darstellen, dass bei einem Anstieg des diastolischen Blutdrucks um 5mm Hg die Chance, nach 6 Jahren eine diabetische Nephropathie zu entwickeln, um den Faktor von OR = 1,3 (95% Konfidenzintervall 1,1–1,5) erhöht ist. Für das HbA1c erhält man ähnliche Resultate wie im einfachen Modell, d.h. in diesem Fall gibt es kaum Unterschiede zwischen den rohen und den adjustierten Resultaten bezüglich des Zusammenhangs zwischen der Stoffwechseleinstellung und dem Risiko einer diabetischen Nephropathie. Die englischen Bezeichnungen der hier diskutierten Begriffe zeigt Tab.3. kurzgefasst: Mit Hilfe der multiplen logistischen Regression lässt sich der Einfluss erklärender Variablen (Risikofaktoren) auf eine binäre Zielvariable (z.B. Krankheit ja/ nein) untersuchen. Aus den Regressionskoeffizienten lassen sich adjustierte Odds Ratios als Maß für die Stärke des Zusammenhangs berechnen. Dtsch Med Wochenschr 2002; 127: T11–T13 · R. Bender, Logistische Regression