# 相加交互作用的通用函数，先运行这个函数 MUM_Interaction_svyglm <- function(model, coef.index){ theta1 <- coef(model)[coef[1]] # 0.05515819 theta2 <- coef(model)[coef[2]] # 0.1048913 theta3 <- coef(model)[coef[3]] # 0.06646235 theta1.se <- SE(model)[coef[1]] theta2.se <- SE(model)[coef[2]] theta3.se <- SE(model)[coef[3]] cov.mat <- vcov(model) h1 <- exp(theta1 + theta2 + theta3) - exp(theta1) h2 <- exp(theta1 + theta2 + theta3) - exp(theta2) h3 <- exp(theta1 + theta2 + theta3) reri.var <- (h1^2 * theta1.se^2) + (h2^2 * theta2.se^2) + (h3^2 * theta3.se^2) + (2 * h1 * h2 * cov.mat[coef[1], coef[2]]) + (2 * h1 * h3 * cov.mat[coef[1], coef[3]]) + (2 * h2 * h3 * cov.mat[coef[2], coef[3]]) reri.se <- sqrt(reri.var) reri.p <- exp(theta1 + theta2 + theta3) - exp(theta1) - exp(theta2) + 1 # 注意，这里是 OR reri.l <- reri.p - (z * reri.se) reri.u <- reri.p + (z * reri.se) reri <- data.frame(est = reri.p, lower = reri.l, upper = reri.u) mult.p <- as.numeric(exp(theta3)) mult.ci <- suppressMessages(confint(object = model, parm = coef[3])) mult.l <- as.numeric(exp(mult.ci[1])) mult.u <- as.numeric(exp(mult.ci[2])) multiplicative <- data.frame(est = mult.p, lower = mult.l, upper = mult.u) cov.mat <- vcov(model) h1 <- ((exp(theta1 + theta2 + theta3) - exp(theta1))/(exp(theta1 + theta2 + theta3))) - ((exp(theta1 + theta2 + theta3) - exp(theta1) - exp(theta2) + 1)/(exp(theta1 + theta2 + theta3))) h2 <- ((exp(theta1 + theta2 + theta3) - exp(theta2))/(exp(theta1 + theta2 + theta3))) - ((exp(theta1 + theta2 + theta3) - exp(theta1) - exp(theta2) + 1)/(exp(theta1 + theta2 + theta3))) h3 <- 1 - ((exp(theta1 + theta2 + theta3) - exp(theta1) - exp(theta2) + 1)/exp(theta1 + theta2 + theta3)) apab.var <- (h1^2 * theta1.se^2) + (h2^2 * theta2.se^2) + (h3^2 * theta3.se^2) + (2 * h1 * h2 * cov.mat[coef[1], coef[2]]) + (2 * h1 * h3 * cov.mat[coef[1], coef[3]]) + (2 * h2 * h3 * cov.mat[coef[2], coef[3]]) apab.se <- sqrt(apab.var) apab.p <- (exp(theta1 + theta2 + theta3) - exp(theta1) - exp(theta2) + 1)/exp(theta1 + theta2 + theta3) apab.l <- apab.p - (z * apab.se) apab.u <- apab.p + (z * apab.se) apab <- data.frame(est = apab.p, lower = apab.l, upper = apab.u) s.p <- (exp(theta1 + theta2 + theta3) - 1)/(exp(theta1) + exp(theta2) - 2) cov.mat <- vcov(model) if (class(model)[1] == "glm" & class(model)[2] == "lm" & s.p < 0) { warning(paste("Point estimate of synergy index (S) is less than zero (", round(s.p, digits = 2), ").\n Confidence intervals cannot be calculated using the delta method. Consider re-parameterising as linear odds model.", sep = "")) } if (class(model)[1] == "clogit" & class(model)[2] == "coxph" & s.p < 0) { warning(paste("Point estimate of synergy index (S) is less than zero (", round(s.p, digits = 2), ").\n Confidence intervals cannot be calculated using the delta method. Consider re-parameterising as linear odds model.", sep = "")) } if (class(model)[1] == "geeglm" & class(model)[2] == "gee" & s.p < 0) { warning(paste("Point estimate of synergy index (S) is less than zero (", round(s.p, digits = 2), ").\n Confidence intervals cannot be calculated using the delta method. Consider re-parameterising as linear odds model.", sep = "")) } if (class(model)[1] == "glmerMod" & s.p < 0) { warning(paste("Point estimate of synergy index (S) is less than zero (", round(s.p, digits = 2), ").\n Confidence intervals cannot be calculated using the delta method. Consider re-parameterising as linear odds model.", sep = "")) } h1 <- ((exp(theta1 + theta2 + theta3))/(exp(theta1 + theta2 + theta3) - 1)) - (exp(theta1)/(exp(theta1) + exp(theta2) - 2)) h2 <- ((exp(theta1 + theta2 + theta3))/(exp(theta1 + theta2 + theta3) - 1)) - (exp(theta2)/(exp(theta1) + exp(theta2) - 2)) h3 <- exp(theta1 + theta2 + theta3)/(exp(theta1 + theta2 + theta3) - 1) lns.var <- h1^2 * theta1.se^2 + h2^2 * theta2.se^2 + h3^2 * theta3.se^2 + (2 * h1 * h2 * cov.mat[coef[2], coef[1]]) + (2 * h1 * h3 * cov.mat[coef[3], coef[1]]) + (2 * h2 * h3 * cov.mat[coef[3], coef[2]]) lns.se <- sqrt(lns.var) lns.p <- log(s.p) lns.l <- lns.p - (z * lns.se) lns.u <- lns.p + (z * lns.se) s.l <- exp(lns.l) s.u <- exp(lns.u) s <- data.frame(est = s.p, lower = s.l, upper = s.u) rval <- list(reri = reri, apab = apab, s = s, multiplicative = multiplicative) return(rval) } model <- F10 coef.index <- c(2,3,10) # coef.index 是指，关注的交互变量在模型中的序号，以 F10，为例，则是 2，3，10 MUM_Interaction_svyglm(F10, coef.index = c(2,3,10)) #### 既往代码-可以不采用了 #### interaction_coef <- coef(F10)["depression.group3Depression:sleep.disorderSleep Disorder"] interaction_se <- sqrt(diag(vcov(F10)))["depression.group3Depression:sleep.disorderSleep Disorder"] # 获取depression.group3和sleep.disorder的系数估计值和标准误差 depression_coef <- coef(F10)["depression.group3Depression"] depression_se <- sqrt(diag(vcov(F10)))["depression.group3Depression"] sleep_disorder_coef <- coef(F10)["sleep.disorderSleep Disorder"] sleep_disorder_se <- sqrt(diag(vcov(F10)))["sleep.disorderSleep Disorder"] # 计算相对过量风险（RERI） RERI <- interaction_coef - depression_coef - sleep_disorder_coef # 计算RERI的95%置信区间 RERI_se <- sqrt(interaction_se^2 + depression_se^2 + sleep_disorder_se^2) RERI_CI <- RERI + qnorm(c(0.025, 0.975)) * RERI_se # 计算相互作用归因比例（API） API <- (interaction_coef - depression_coef - sleep_disorder_coef) / interaction_coef # 计算API的95%置信区间 API_se <- sqrt((interaction_se / interaction_coef)^2 + (depression_se / depression_coef)^2 + (sleep.disorder_se / sleep_disorder_coef)^2) API_CI <- API + qnorm(c(0.025, 0.975)) * API_se # 计算协同作用指数（SI） SI <- (interaction_coef - depression_coef - sleep_disorder_coef) / (depression_coef + sleep_disorder_coef - interaction_coef) # 计算SI的95%置信区间 SI_se <- sqrt((interaction_se / (depression_coef + sleep_disorder_coef - interaction_coef))^2 + (depression_se / (depression_coef + sleep_disorder_coef - interaction_coef))^2 + (sleep_disorder_se / (depression_coef + sleep_disorder_coef - interaction_coef))^2) SI_CI <- SI + qnorm(c(0.025, 0.975)) * SI_se