SAS Support Communities

Got it. Thank you for your help. Koen. ... View more

Hello, sir or madam   Thank you for your reply and I think a scenario would help to express more clearly of my idea.   The maximum likelihood of Poisson regression requires the calculation of the joint likelihood function. For Poisson regression, observations are categorized by their characteristics into several groups and each has a likelihood function expressed via the probability mass function of Poisson distribution.   Saying, we have a model of log(λ) = a +bX and X is the independent variable. Assume we have 100 observations and we designed X for the representation of age groups in the unit of decade. Therefore, the observations are categorized by X into several groups (a discrete variable), and each group could have sample λ of integrals other than 0 and 1, which fits the Poisson distribution.   However, if we use designed X for the representation of age in the unit of second (similar to a continuous variable), for these 100 observations, it is very likely they will be categorized into 100 groups. For each group, the λ could be only 0 or 1, which violates the assumption of Poisson distribution. ... View more

I have read parts of this paper (https://pubmed.ncbi.nlm.nih.gov/35924051/). In the statistical method section, it described,   Daily medication exposure and processes of care (sedative, acid inhibitors, blood transfusion, mandatory ventilation, and head-of-bed elevation, gastrointestinal decompression, and rehabilitation exercise) and medications (sedatives, opioids, neuromuscular blockers, immunosuppressive agent, neuroleptic agents, antibiotics, expectorants, vasopressors, intestinal probiotics, and neuroleptic agents) were defined as time-varying variables. Time-varying variables were measured as daily exposure from initiation of MV to the event of interested for the model of VAEs and each day from ICU admission to ICU discharge for the model of ICU discharge and ICU mortality.   It looks like the authors included a dozen time-varying cofactors in their Cox regression model. So, I wonder how they handled the dataset. Each patient could have a dozen time-varying cofactors, and each such cofactor could have several different values. To record the data into a dataset must be a tough mission but I don't know how they did it. ... View more

And after some google later, I found the link = log is a model called "log-binomial regression" (https://stats.stackexchange.com/questions/581678/what-is-the-difference-between-a-logistic-regression-and-log-binomial-regression )   The thread says, Logistic regression always produces estimates of 𝑝𝑖 that are between 0 and 1. This is not true for log-binomial regression; I guess this is the reason why few people use the log-binomial regression to model their data because the risk of deriving a risk higher than 100%. Thanks. ... View more

Exactly! This is the issue. I specified serumALB instead of SplALB in the MODEL statement. After correcting this error, everything followed as expected. Thank you. ... View more

Hello, Dave.   Wow! That's a great reference. Thanks. I copied and edited the code to suit my project. But, so far, I have never get the Odd Ratio plot as shown in the first example. My code only generated one column for the variable of serumAlb. I am just not able to make it. Your investigation would be of great important. Thanks.   My code: %macro Logistic2; proc logistic data= &raw_data_name plots = oddsratio(type=vertical); class &class_var; effect SplALB=spline(serumAlb/naturalcubic); model &censor(event = '1') = &indep_list; effectplot fit(x=serumAlb); effectplot fit(x=serumAlb) / link; oddsratio serumAlb / at(serumAlb=34.8125 36.0425 37.35 39.0 40.92); store kmod; run; %mend; %Logistic2 ... View more

Thanks for these advice, ballardw.   Yep, I will follow your guide. ... View more