R语言JM包联合建模报错：optim函数传入非有限值问题求助

我在运行线性混合模型与Cox模型的联合建模时遇到报错，原代码及报错信息如下：

原代码 #

rbcLME <- lme(log(RBC.counts+1)~time+time:DAC, random = ~1|ID, 
          data=no.missing.rbc)
rbcSURV <- coxph(Surv(surv.time, death) ~ GroupforDCAtreatment, 
             data = rbc.df, x = TRUE)
rbc.JM <- jointModel(rbcLME, rbcSURV, timeVar = "time", 
                 method = "piecewise-PH-GH")

报错信息 #

Error in optim(thetas, opt.survPC, gr.survPC, method = "BFGS", control = list(maxit = if (it <  : 
      non-finite value supplied by optim

以下是针对该报错的修复方案：

1. 更换优化器并调整迭代参数 #

optim抛出的非有限值错误通常源于BFGS优化器在搜索过程中遇到目标函数不收敛的情况，尝试更换更稳健的优化方法并增加迭代次数：

# 使用Nelder-Mead优化器，提升最大迭代次数
rbc.JM <- jointModel(rbcLME, rbcSURV, timeVar = "time", 
                     method = "piecewise-PH-GH",
                     control = list(optim.method = "Nelder-Mead", maxit = 1000))

2. 检查并预处理数据 #

• 确保纵向数据no.missing.rbc与生存数据rbc.df的ID完全匹配，无缺失或不一致条目
• 对time变量做标准化处理，避免因数值范围过大导致优化不稳定：

# 标准化time变量
no.missing.rbc$time <- scale(no.missing.rbc$time)
rbc.df$time <- scale(rbc.df$time)

# 重新拟合LME模型
rbcLME <- lme(log(RBC.counts+1)~time+time:DAC, random = ~1|ID, 
              data=no.missing.rbc)

# 重新拟合联合模型
rbc.JM <- jointModel(rbcLME, rbcSURV, timeVar = "time", 
                     method = "piecewise-PH-GH")

3. 调整联合模型的拟合方法 #

尝试更换联合模型的基础方法，比如使用样条基函数替代分段比例风险模型：

# 使用样条PH模型，设置3个节点
rbc.JM <- jointModel(rbcLME, rbcSURV, timeVar = "time", 
                     method = "spline-PH-GH",
                     control = list(n.knots = 3))

4. 验证Cox模型的拟合有效性 #

确保Cox模型无分离问题或极端系数，必要时添加惩罚项稳定拟合：

# 查看Cox模型摘要，检查系数合理性
summary(rbcSURV)

# 若存在分离问题，使用弹性网惩罚拟合Cox模型
library(glmnet)
cox_pen <- glmnet(x = rbcSURV$x, y = Surv(rbc.df$surv.time, rbc.df$death), 
                  family = "cox", alpha = 0.5)

# 提取最优惩罚系数作为联合模型初始值
pen_coef <- coef(cox_pen, s = "lambda.min")
init_surv <- pen_coef[pen_coef != 0]

# 带初始值拟合联合模型
rbc.JM <- jointModel(rbcLME, rbcSURV, timeVar = "time", 
                     method = "piecewise-PH-GH",
                     inits = list(surv = init_surv))

内容的提问来源于stack exchange，提问作者user3561144