基于线性混合模型的连续时间下生活质量得分LS均值求解求助

Alright, let's tackle how to get those adjusted mean quality of life (QoL) scores at your 3, 6, 9, 12, etc. time points from your linear mixed model (LMM). You’re already on the right track with including fixed effects for gender and prognosis—here’s a practical, step-by-step approach tailored to your longitudinal data (with continuous time and irregular follow-up):

• First, create a target dataset for your time points
  To generate predictions at your specific intervals, you’ll need a dataset that includes every combination of your fixed effects (gender, prognosis) plus your target time points (3, 6, 9, 12,...). This ensures you’re estimating scores adjusted for all the clinical factors you included in the model. For example, if gender has 2 levels and prognosis has 3 levels, your target dataset will have 23N_time_rows entries (where N_time is the number of intervals you care about).

• Use marginal mean estimation (recommended) or model prediction
  The most reliable way to get adjusted average scores (accounting for your fixed factors) is to calculate marginal means (also called least-squares means). Here’s how to do this in common statistical tools:

  In R (using lme4 and emmeans packages) #

  # Load required libraries
  library(lme4)
  library(emmeans)
  
  # Assume your fitted model looks like this (adjust formula to match your actual model)
  lmm_model <- lmer(QoL_score ~ time + gender + prognosis + (1 | patient_id), data = your_patient_data)
  
  # Define your target time points
  target_times <- seq(3, 12, 3) # Adjust upper limit to your longest follow-up
  
  # Create the target dataset with all fixed effect combinations + target times
  new_data <- expand.grid(
    time = target_times,
    gender = unique(your_patient_data$gender),
    prognosis = unique(your_patient_data$prognosis)
  )
  
  # Get adjusted marginal means for each time point, grouped by gender/prognosis
  adjusted_means <- emmeans(lmm_model, ~ time | gender + prognosis, at = list(time = target_times))
  
  # If you want overall adjusted means (averaged across gender and prognosis)
  overall_adjusted_means <- emmeans(lmm_model, ~ time, at = list(time = target_times))
  
  # Convert to a data frame for easy plotting/analysis
  adjusted_means_df <- as.data.frame(adjusted_means)
  

  In Stata #

  # Assume your fitted model is:
  mixed QoL_score i.gender i.prognosis time || patient_id: , reml
  
  # Define target time points and create a temporary dataset
  tempfile target_data
  clear
  set obs 4 // Number of target time points (3,6,9,12)
  gen time = _n*3
  expand 2 // Number of gender levels
  bysort time: gen gender = _n
  expand 3 // Number of prognosis levels
  bysort time gender: gen prognosis = _n
  
  # Predict adjusted marginal means (marginal effects)
  margins, at(time=(3 6 9 12)) over(gender prognosis)
  marginsplot // Optional: Plot the trend
  

• Key considerations

  • If your model includes interaction terms (e.g., time*gender), make sure to include those in your marginal mean formula (e.g., ~ time*gender + time*prognosis in emmeans).
  • Marginal means give you the population-average adjusted scores (averaged over the distribution of your fixed factors), which is exactly what you need for reporting trends adjusted for clinical variables.
  • Avoid extrapolating beyond the maximum time point present in your dataset—predictions outside your observed time range are unreliable.

Once you have these adjusted means, you can plot them over time to visualize the QoL trend, along with their standard errors or confidence intervals for uncertainty.

内容的提问来源于stack exchange，提问作者user9563941