亚洲免费av电影一区二区三区,日韩爱爱视频,51精品视频一区二区三区,91视频爱爱,日韩欧美在线播放视频,中文字幕少妇AV,亚洲电影中文字幕,久久久久亚洲av成人网址,久久综合视频网站,国产在线不卡免费播放

        ?

        Binary outcome variables and logistic regression models

        2011-07-18 11:38:01XinhuaLIU
        上海精神醫(yī)學(xué) 2011年5期
        關(guān)鍵詞:醫(yī)學(xué)論壇心身疾病精神病學(xué)

        Xinhua LIU

        ·Research Methods·

        Binary outcome variables and logistic regression models

        Xinhua LIU

        Biostatistics in Psychiatry(5)

        Biomedical researchers often study binary variables that indicate whether or not a specific event,such as remission of depression symptoms,occurs during the study period.The indicator variable Y takes two values,usually coded as one if the event (remission)is present and zero if the event is not present(non-remission).Let p be the probability that the event occurs(Y=1),then 1-p will be the probability that the event does not occur(Y=0). The odds that an event will occur is a quantity defined as the ratio of the probabilities for the presence versus the absence of the event,that is,odds =p/(1-p).Unlike probability p which ranges between zero and one,the odds take non-negative values with values above one indicating the degree to which the target event is likely to be present and values below one indicating the degree to which the target event is likely to be absent.To compare the odds in two groups,one may use the odds ratio,OR=odds1/odds2,where odds1=p1/(1-p1)and odds2=p2/(1-p2)with p1and p2representing the probabilities that the target event is present in group 1 and group 2,respectively.When OR=1 the odds of the event occurring in the two groups are equal; OR>1 indicates that the event is more likely to occur in group 1 than in group 2,while OR<1 indicates the event is more likely to occur in group 2 than in group 1.

        In studies with binary outcome variables,to compare the occurrence of the outcome of interest in two groups investigators collect data to estimate the odds ratio and test the null hypothesis of H0:OR=1.Based on the estimates of p1and p2(proportions or rates of occurrence of the event of interest in the two groups),the odds ratio can be computed as a ratio of two estimated odds.As an example,consider a clinical trial of the efficacy of a new treatment for major depression with remission of depressive symptoms as the primary outcome.If the remission rate is 60%in the new treatment group and 42%in the standard treatment group,then the odds of remission is 1.5(0.60/0.40)with the new treatment and 0.7241(0.42/0.58)with the standard treatment.The odds ratio is therefore estimated as 2.0715(1.5/0.7241).Based on the remission rate and the sample size in each group,a 95%confidence interval of the odds ratio can be constructed and a test statistic can be calculated for testing significance.

        Logistic regression models are often used to predict the odds that a particular event will occur. The models assume that the odds of occurrence depend on a weighted sum of predictors,where the unknown weights or coefficients are the model parameters.The simplest logistic model has only one predictor X,with the form

        The odds of event occurrence are functions of predictor X with coefficients α and β.The model parameters can be interpreted as the intercept(that is,α)and the slope associated with X(that is,β). When the predictor variable X is dichotomous (i.e.,it onlytakes values of zero and one),then α is log(odds)for the group with X=0,and α+β is log(odds)for the group with X=1.Note that since y=log(z)is the natural logarithmic function of z,the inverse function is z=exp(y).Consequently exp(β)is the odds ratio comparing the odds of event occurrence between the two groups(X=1 vs.X=0).In contrast,when the predictor variable X is continuous,then

        is the odds ratio for one unit increase in X.The parameter β deserves special attention,as it indicatesthe association between predictor X and the outcome.Specifically,β=0 yields constant odds,indicating independence between event occurrence and the predictor X.The odds of event occurrence will increase with increasing X if β>0,while the odds of event occurrence will decrease with increasing X when β<0.

        The multiple logistic regression models may include more predictors,X1,…,Xm(m>1),which can be written as

        When none of the predictors is a function of other predictors(that is,the included variables are independent of each other),the predictors in the model are additive and the model is considered an additive model.The coefficient of each predictor(i.e.,β1,β2,…βm)is interpreted as the log(OR)for a unit change in the predictor while keeping the other predictors constant.As an example,consider a study of a new treatment for major depression.Suppose that the primary outcome is the remission of depressive symptoms in the six months after the treatment.Let X1be a group indicator with zero for standard treatment and one for a new treatment; and let X2be the baseline score of depressive symptoms.The logistic model with two main predictors has the form,

        The quantity exp(β1)will be the odds ratio comparing the odds of remission of depressive symptoms among subjects receiving the new treatment with that of subjects receiving the standard treatment for those subjects who have the same baseline score of depressive symptoms.Similarly,exp(β2)is the odds ratio comparing the odds of remission of depressive symptoms between those who received the same treatment but differed by one unit score on the baseline scale for depressive symptoms.

        When a predictor is created as a function of original predictor(s)or is closely correlated to another predictor in the model,caution has to be made when interpreting model parameters because the interaction of these variables changes the interpretation of the odds ratios.As an example,let X3= X1X2,which is the term for the X1by X2interaction. Then the logistic model with three predictors becomes

        Models like this that include interaction terms are called interactive models.In the above example,consider a patient with baseline score m for depressive symptoms,that is,X2=m.The treatment effect for this patient is given by the odds ratio comparing the odds of remission of depressive symptoms with the new treatment compared to that with standard treatment,which turns out to be exp (β1+m β3)according to the interactive model given above.To see this,we substitute X2=m into the model,resulting in two terms related to treatment effects:β1X1and β3X1m;combining the two terms,we have the overall treatment effect term given by (β1+m β3)X1.Thus the treatment effect for this patient on the odds ratio scale is exp(β1+m β3).

        Now consider another patient whose baseline score is one unit higher than the mean,i.e.,X2=m +1.The treatment effect for this patient is given by exp(β1+(m+1)β3).Therefore,under the interactive model,these two patients do not have the same treatment effect.How different are the treatment effects between these two patients?The ratio between the treatment for the second patient,exp(β1+(m+1)β3),and the treatment effect for the first patient,exp(β1+mβ3),is exp(β3),which shows that β3is the interaction between treatment and baseline score for depressive symptoms.More specifically,for every unit increase in baseline score,the treatment effect changes by a factor of exp (β3).

        The additive model with two main effects predictors is a special case of the interactive model with β3=0,which indicates that the association of the baseline depressive symptoms with the outcome does not differ between standard and new treatment groups.Therefore,testing the null hypothesis H0:β3=0 versus the alternative hypothesis H1:β3≠0 will determine whether or not the interaction term needs to be included in the model.(Editor’s note:Further discussions of interactive models will be given in a future column in this series.)

        In prospective studies logistic models may also help predict the probability of event occurrence for given values of the predictors,because p=1-{1/[1 +exp(η)]}with η=β0+β1X1+…+βmXmcan be calculated using the estimated parameters of β0,β1…βmfrom the study data.For example,using the previous additive model with two predictors of remission from depression after treatment where model parameters β0,β1and β2have values of-0.143,0.716 and-0.025,respectively,a patient with a baseline depressive symptom score of 10 who was receiving the new treatment would have a predicted probability of remission of 0.580(=1-{1/[1 +exp(-0.143+0.716-0.025x10)]})while a patient with a baseline score of 10 receiving standard treatment would have a predicted probability of remission of 0.403(=1-{1/[1+(exp(-0.143-0.025x10)]}).

        The major statistical software packages such as SAS,SPSS,STATA,R,S+all include procedures toimplement logistic regression analyses that fit models to the data collected.In addition to describing the specified model,the output usually includes estimated model parameters,95%confidence intervals,p-values for testing hypotheses on the model parameters of interest,goodness of fit measures and other important information.

        1. Hosmer DW Lemeshow S.Applied Logistic Regression.2nd ed. John Wiley&Sons,Inc.,NJ,USA,2000.

        2. Fleiss LJ,Levin B,Paik MC.Statistical Methods for Rates and Proportions.3rd ed,John Wiley&Sons,Inc.,NJ,USA,2003.

        Dr.Xinhua Liu is an associate professor at the Department of Biostatistics in the Mailman School of Public Health at Columbia University.Her research interests include 1)statistical analysis of various type of data in the epidemiological,clinical and laboratory studies related to metal health and environmental health sciences;2)development of statistical methods with applications in psychiatric research;3)modeling functional decline of patients with Alzheimer's disease;4)evaluating concordance correlation among raters for measurement subject to right censoring;5)non-parametric selection of items in the screening scales,etc.

        第九屆西部精神醫(yī)學(xué)論壇征文通知

        第九屆西部精神醫(yī)學(xué)論壇預(yù)定于2010年7月在四川省成都市召開,本次會(huì)議由西部精神醫(yī)學(xué)協(xié)會(huì)主辦。會(huì)議征文內(nèi)容:

        板塊一:精神病學(xué)

        1.精神疾病治療新技術(shù);

        2.精神病院的管理及相關(guān)法律問題;

        3.精神病患者的康復(fù)和回歸;

        4.司法精神病學(xué)的理論和實(shí)踐;

        5.精神病學(xué)教學(xué)法研究;

        6.精神科合理用藥。

        板塊二:聯(lián)絡(luò)會(huì)診醫(yī)學(xué)

        1.心身健康;

        2.心身疾病的教學(xué)法研究;

        3.心身疾病的危險(xiǎn)因素研究;

        4.心身疾病的臨床綜合干預(yù);

        5.心身疾病的全面康復(fù);

        6.健康管理。

        板塊三:社會(huì)心理

        1.女性及兒童青少年心理健康;

        2.災(zāi)后危機(jī)干預(yù);

        3.網(wǎng)絡(luò)與生活;

        4.戀愛、婚姻家庭及性心理;

        5.民眾心理教育。

        請將論文及摘要、正文、參考文獻(xiàn)以E-mail形式發(fā)往論壇秘書組張波老師郵箱zb_73@126.com(咨詢電話:18980602051),并請注明同意以何種形式(全文/摘要)在論文集(光盤)上登出。論文征集截稿時(shí)間:2012年5月15日前(以電子郵件日期為準(zhǔn))。

        西部精神醫(yī)學(xué)協(xié)會(huì)

        2011年10月8日

        10.3969/j.issn.1002-0829.2011.05.012

        Department of Biostatistics,Mailman School of Public Health,Columbia University New York,NY,USA.

        E-mail:xl26@columbia.edu

        猜你喜歡
        醫(yī)學(xué)論壇心身疾病精神病學(xué)
        基因決定了 你們會(huì)是好朋友
        心身疾病 宜先治心
        第九屆中國睡眠醫(yī)學(xué)論壇暨中國睡眠研究會(huì)睡眠障礙專業(yè)委員會(huì)學(xué)術(shù)年會(huì)圓滿落幕
        中國睡眠研究會(huì)社區(qū)與基層工作委員會(huì)成立大會(huì)暨第一屆社區(qū)與基層睡眠醫(yī)學(xué)論壇
        第九屆中國睡眠醫(yī)學(xué)論壇暨中國睡眠研究會(huì)睡眠障礙專業(yè)委員會(huì)學(xué)術(shù)年會(huì)會(huì)議通知(第二輪)
        您了解心身疾病嗎?
        第三屆中國睡眠研究會(huì)教育委員會(huì)學(xué)術(shù)年會(huì)暨第一屆南方精神醫(yī)學(xué)論壇圓滿落幕
        老年心身疾病診治中存在的幾個(gè)問題
        心身疾病
        ——心理導(dǎo)致的身體疾病
        保健與生活(2016年8期)2016-04-23 02:03:51
        四個(gè)字
        亚洲国际无码中文字幕| 久久99热精品这里久久精品| 亚洲产国偷v产偷v自拍色戒| 亚洲aⅴ无码日韩av无码网站| 日本色偷偷| 在线观看黄片在线播放视频 | 最新日本免费一区二区三区| 日本中文字幕有码在线播放| 国产av一区网址大全| 国产午夜免费一区二区三区视频| 99视频在线精品免费观看6| 女人被爽到高潮视频免费国产| 亚洲日韩av无码中文字幕美国| 中文字幕亚洲无线码| 中文字幕无码日韩欧毛| 不打码在线观看一区二区三区视频 | 99成人无码精品视频| 人妻少妇精品视频中文字幕国语| 亚洲国产成人av毛片大全| 精品激情成人影院在线播放| 亚洲一区自拍高清亚洲精品| 婷婷午夜天| 国产精品青草视频免费播放 | 久久精品国产亚洲AV高清wy| 日韩亚洲精选一区二区三区| 澳门蜜桃av成人av| 亚洲一区av在线观看| 免费a级毛片无码a∨免费软件| 最新国产拍偷乱偷精品| 国产精品女同学| 日本免费影片一区二区| 国产性自爱拍偷在在线播放| 亚洲av永久无码精品国产精品| 小12箩利洗澡无码视频网站| 久久狠狠髙潮曰十八女人| 日韩人妖视频一区二区| 久久久久av无码免费网| 疯狂做受xxxx高潮欧美日本| 天天干夜夜躁| 少妇人妻偷人中文字幕| 中字乱码视频|