### Initialize riskRegression Package

Source: https://github.com/tagteam/riskregression/blob/master/simulation/k-fold-CV.org

Load the riskRegression package and verify the installed version.

```R
library(riskRegression)
packageVersion("riskRegression")
```

--------------------------------

### Install riskRegression Package

Source: https://github.com/tagteam/riskregression/blob/master/README.md

Install the riskRegression package from GitHub using the devtools library. Ensure devtools is installed first.

```r
library(devtools)
install_github("tagteam/riskRegression")
```

--------------------------------

### Setup Simulation Data and Models

Source: https://github.com/tagteam/riskregression/blob/master/simulation/k-fold-CV.org

Simulate active surveillance data and fit logistic regression models for performance comparison.

```R
  set.seed(18)
  astrain <- simActiveSurveillance(278)
  astest <- simActiveSurveillance(208)
  astrain[,Y1:=1*(event==1 & time<=1)]
  astest[,Y1:=1*(event==1 & time<=1)]
  lrfit.ex <- glm(Y1~age+lpsaden+ppb5+lmax+ct1+diaggs,data=astrain,family="binomial")
  lrfit.inc <- glm(Y1~age+lpsaden+ppb5+lmax+ct1+diaggs+erg.status,data=astrain,family="binomial")
  ## Score(list("Exclusive ERG"=lrfit.ex,"Inclusive ERG"=lrfit.inc),data=astest,formula=Y1~1,se.fit=0L,metrics="brier",contrasts=FALSE)
```

--------------------------------

### Simulate Example Data for Binary, Survival, and Competing Risks Outcomes

Source: https://context7.com/tagteam/riskregression/llms.txt

Generates simulated datasets for testing and examples with binary, survival, or competing risks outcomes. Creates 10 covariates (5 binary, 5 continuous) and response variables. Requires the 'riskRegression' library.

```r
library(riskRegression)

# Binary outcome
set.seed(10)
binary_data <- sampleData(n = 100, outcome = "binary")
head(binary_data)
# Contains: Y (binary outcome), X1-X5 (binary), X6-X10 (continuous)
```

```r
# Survival outcome
survival_data <- sampleData(n = 100, outcome = "survival")
head(survival_data)
# Contains: time, event, X1-X5 (binary), X6-X10 (continuous)
```

```r
# Competing risks outcome
cr_data <- sampleData(n = 100, outcome = "competing.risks")
head(cr_data)
# Contains: time, event (0=censored, 1=cause1, 2=cause2), X1-X10
```

```r
# Custom formula for regression coefficients
custom_data <- sampleData(n = 100, outcome = "binary",
                          formula = ~ f(X1, 3) + f(X6, 0.5) + f(X8, -0.5),
                          intercept = -1)
```

--------------------------------

### Setup and Fit Logistic Regression Models

Source: https://github.com/tagteam/riskregression/blob/master/simulation/k-fold-CV.html

Initializes training and testing datasets and fits two logistic regression models for comparison.

```R
set.seed(18)
astrain <- simActiveSurveillance(278)
astest <- simActiveSurveillance(208)
astrain[,Y1:=1*(event==1 & time<=1)]
astest[,Y1:=1*(event==1 & time<=1)]
lrfit.ex <- glm(Y1~age+lpsaden+ppb5+lmax+ct1+diaggs,data=astrain,family="binomial")
lrfit.inc <- glm(Y1~age+lpsaden+ppb5+lmax+ct1+diaggs+erg.status,data=astrain,family="binomial")
```

--------------------------------

### Check Package Versions

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.org

Prints the installed versions of the 'data.table', 'survival', 'riskRegression', and 'Publish' packages. This is useful for ensuring compatibility and reproducibility.

```R
library(data.table)
library(survival)
library(riskRegression)
library(Publish)
cat("data.table:")
packageVersion("data.table")
cat("\nsurvival:")
packageVersion("survival")
cat("\nriskRegression:")
packageVersion("riskRegression")
cat("\nPublish:")
packageVersion("Publish")
```

--------------------------------

### Load and Prepare Data for Risk Regression

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.org

Simulates learning and validation datasets for a prostate cancer study and defines a binary outcome variable for 1-year progression. Ensure the 'riskRegression' package is installed.

```R
set.seed(18)
astrain <- simActiveSurveillance(278)
astest <- simActiveSurveillance(208)
astrain[,Y1:=1*(event==1 & time<=1)]
astest[,Y1:=1*(event==1 & time<=1)]
```

--------------------------------

### Define and Synthesize Lava Model for Simulation

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Defines a latent variable model (lvm) using the `lava` package, specifies distributions and regression relationships, synthesizes the model, and then simulates data. This setup is for survival analysis with censored data.

```R
library(lava)
u <- lvm()
# specify covariate
distribution(u,~sex) <- binomial.lvm() #specify binary covariate
distribution(u,~age) <- normal.lvm() #specify continuous covariate
distribution(u,~logthick) <- normal.lvm() #note: remove paranthesis around log
categorical(u,K=3) <- "invasion" #specify categorical covariate
#include event time, so it knows it is the survival case
u <-eventTime(u,time~min(time.cens=0,time.death=1,time.other=2), "status")
#specify regression relationships
lava::regression(u,logthick~sex+age+invasion) <- 1
lava::regression(u,age~invasion) <- 1
#specify relationships for censored variables (could also change them)
lava::regression(u,time.other~logthick+sex+age+invasion) <- 1
lava::regression(u,time.death~logthick+sex+age+invasion) <- 1
lava::regression(u,time.cens~logthick+sex+age+invasion) <- 1
#finally synthesize object 
u_synt <- synthesize(u,data=Melanoma)

#do a simulation study
d <- sim(u_synt,10000)
fit.rec.alt <- lm(log(thick)~sex+age+invasion,data=Melanoma)
fit.rec.alt.s <- lm(logthick ~ sex + age + invasion, data= d)
res <- as.data.frame(cbind(coef(fit.rec.alt),coef(fit.rec.alt.s)))
colnames(res) <- c("Parameters of the model from original data", "Parameters of the model from simulated data")
knitr::kable(res)
```

--------------------------------

### Simulate and Fit CoxPH Model

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Simulates data from a model and fits a Cox proportional hazards model to both original and simulated data to compare parameter estimates. Ensure the 'survival' and 'knitr' packages are installed.

```R
set.seed(15)
d <- sim(ms,10000)
fit <- coxph(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=Melanoma)
fit.s <- coxph(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=d)

res <- as.data.frame(cbind(coef(fit),coef(fit.s)))
colnames(res) <- c("Parameters of the model from original data", "Parameters of the model from simulated data")
knitr::kable(res)
```

--------------------------------

### Load riskRegression package

Source: https://github.com/tagteam/riskregression/blob/master/examples/demo.org

Initializes the environment by loading necessary libraries and checking the package version.

```R
library(riskRegression)
library(survival)
packageVersion("riskRegression")
```

--------------------------------

### Load and Prepare Melanoma Data

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Initializes the environment and subsets the Melanoma dataset for survival analysis.

```R
library(riskRegression)
#> riskRegression version 2021.08.27
library(survival)
library(lava)
data("Melanoma")
Melanoma <- subset(Melanoma, select = c("time","status","thick","sex","age","invasion"))
knitr::kable(head(Melanoma))
```

--------------------------------

### Load riskRegression Package and Session Info

Source: https://github.com/tagteam/riskregression/blob/master/simulation/k-fold-CV.html

Loads the riskRegression package and displays session information, including attached packages and R version. This is useful for ensuring the correct environment for analysis.

```R
library(riskRegression)
sessionInfo()
```

--------------------------------

### Simulate and Fit Alternative Recursive Model

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Simulates data from a recursive model and fits an alternative linear model to compare parameters with the original data. Requires the `lava` package.

```R
set.seed(15)
d <- sim(ms.rec,10000)
fit.rec.alt <- lm(log(thick)~sex+age+invasion,data=Melanoma)
fit.rec.alt.s <- lm(log(thick) ~ sex + age + invasion, data= d)

res <- as.data.frame(cbind(coef(fit.rec.alt),coef(fit.rec.alt.s)))
colnames(res) <- c("Parameters of the model (with formula: logthick ~ sex + age + invasion) from original data", "Parameters of the model (with formula: logthick ~ sex + age + invasion) from simulated data")
knitr::kable(res)
```

--------------------------------

### Simulate Data from Linear Model

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Simulate a dataset based on a fitted linear model. Use set.seed for reproducibility. The simulated data can then be used to fit a new model.

```R
set.seed(15)
d <- sim(m.cars,10000)
fit <- lm(dist ~ speed, data = cars)
fit.s <- lm(dist ~ speed, data = d)

res <- as.data.frame(cbind(coef(fit),coef(fit.s)))
colnames(res) <- c("Parameters of the model from original data", "Parameters of the model from simulated data")
knitr::kable(res)
```

--------------------------------

### Prepare Binary Event Variable

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Converts the status variable into a binary event indicator for survival modeling.

```R
Melanoma$eventBin <- 1*(Melanoma$status!=0)
```

--------------------------------

### Fit GLMnet and Predict Survival Risks

Source: https://context7.com/tagteam/riskregression/llms.txt

Demonstrates fitting a GLMnet model with unpenalized terms and predicting survival risks.

```r
fit_mixed <- GLMnet(Surv(time, status == 1) ~ unpenalized(age) +
                    invasion + epicel + ulcer + logthick,
                    data = Melanoma)

# Predict survival risks
surv_risks <- predictRisk(fit_surv, newdata = Melanoma[1:10,], times = c(1000, 2000))
```

--------------------------------

### Linear Regression with synthesize

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Fit a simple linear regression model using the synthesize function. Ensure the 'cars' dataset is loaded. The output provides model summary statistics.

```R
data(cars)
m.cars <- synthesize(dist ~ speed, data=cars)
summary(m.cars)
```

--------------------------------

### Generate Random Numbers and Predict Risk

Source: https://github.com/tagteam/riskregression/blob/master/examples/useScore.org

Demonstrates generating sample data and comparing risk predictions between two random forest models.

```R
set.seed(437)
rnorm(10)
library(randomForestSRC)
library(riskRegression)
d1 <- sampleData(400,outcome="binary")
d2 <- sampleData(8,outcome="binary")
d1[,Y1:=factor(Y)]
d2[,Y1:=factor(Y)]
f1 <- rfsrc(Y~X6+X8,data=d1,ntree=20,seed=8)
f2 <- rfsrc(Y1~X6+X8,data=d1,ntree=20,seed=8)
cbind(predictRisk(f1,newdata=d2),
      predictRisk(f2,newdata=d2))
```

--------------------------------

### Simulate and Visualize Risk Regression Models

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

This snippet simulates data for recursive and non-recursive models, fits Cox proportional hazards models, and generates a boxplot to compare the coefficients.

```R
colnames(df) <- column.names
      
      #simulate
      for (sample.size in sample.sizes){
        for (i in 1:n.sim){
          d <- sim(object, sample.size)
          coeff <- coef(coxph(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=d))
          df[nrow(df)+1,] <- c(coeff,sample.size,"non recursive")
          
          d.rec <- sim(object.rec, sample.size)
          coeff.rec <- coef(coxph(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=d.rec))
          df[nrow(df)+1,] <- c(coeff.rec,sample.size,"recursive")
        }
      }
    
      #make variables of the correct type; otherwise ggplot doesn't understand
      for (c in names(df)[0:length(coef.compare)]){
        df[[c]] <- as.numeric(df[[c]])
      }
      df$sample.size <- as.factor(df$sample.size)
      df$recursive <- as.factor(df$recursive)
      
      ggplot(df, aes(x=sample.size, y=age, fill=recursive)) +geom_boxplot() + geom_hline(yintercept =coef.compare[["age"]])
      
      #plots <- c()
      
      #for (c in names(df)[0:length(coef.compare)]){
      #  plots <- c(plots,ggplot(df, aes(x=sample.size, y=c, fill=recursive)) +geom_boxplot() + geom_hline(yintercept =coef.compare[[c]]))
      #}
      #do.call("grid.arrange", c(plots, ncol=length(coef.compare)))
    }
    
    make.boxplot.simulation(ms,ms.rec, Melanoma, c(100,200,500,1000),1000)
```

--------------------------------

### Methodological References in BibTeX

Source: https://github.com/tagteam/riskregression/blob/master/README.org

A collection of BibTeX entries documenting the statistical methods and research papers underlying the riskRegression package.

```LaTeX
@article{gerds2006consistent,
  title =	 {Consistent Estimation of the Expected {B}rier Score
                  in General Survival Models with Right-Censored Event
                  Times},
  author =	 {Gerds, T.A. and Schumacher, M.},
  journal =	 {Biometrical Journal},
  volume =	 48,
  number =	 6,
  pages =	 {1029--1040},
  year =	 2006,
  publisher =	 {Wiley Online Library}
}

@article{gerds2007efron,
  title =	 {Efron-Type Measures of Prediction Error for Survival
                  Analysis},
  author =	 {Gerds, T.A. and Schumacher, M.},
  journal =	 {Biometrics},
  volume =	 63,
  number =	 4,
  pages =	 {1283--1287},
  year =	 2007,
  publisher =	 {Wiley Online Library}
}

@article{gerds2008performance,
  title =	 {The performance of risk prediction models},
  author =	 {Gerds, T.A. and Cai, T. and Schumacher, M.},
  journal =	 {Biometrical Journal},
  volume =	 50,
  number =	 4,
  pages =	 {457--479},
  year =	 2008,
  publisher =	 {Wiley Online Library}
}

@Article{mogensen2012pec,
  title =	 {Evaluating random forests for survival analysis
                  using prediction error curves},
  author =	 {Mogensen, U B and Ishwaran, H. and Gerds, T A},
  journal =	 {Journal of Statistical Software},
  year =	 2012,
  volume =	 50,
  number =	 11
}

@article{Blanche2013statmed,
  title =	 "{Estimating and comparing time-dependent areas under
                  receiver operating characteristic curves for
                  censored event times with competing risks}",
  author =	 {Blanche, P. and Dartigues, J-F and Jacqmin-Gadda,
                  H.},
  journal =	 {Statistics in Medicine},
  volume =	 32,
  number =	 30,
  pages =	 {5381--5397},
  year =	 2013
}

@article{blanche2015,
  title =	 {Quantifying and comparing dynamic predictive
                  accuracy of joint models for longitudinal marker and
                  time-to-event in presence of censoring and competing
                  risks},
  author =	 {Blanche, Paul and Proust-Lima, C{'e}cile and
                  Loub{'e}re, Lucie and Berr, Claudine and Dartigues,
                  Jean-Fran{'c}ois and Jacqmin-Gadda,
                  H{'e}l{'e}ne},
  journal =	 {Biometrics},
  volume =	 71,
  number =	 1,
  pages =	 {102--113},
  year =	 2015,
  publisher =	 {Wiley Online Library}
}

@article{ozenne2017,
  title =	 {riskRegression: Predicting the Risk of an Event
                using Cox Regression Modelss},
  author =	 {Ozenne, Brice and Sørensen, Anne Lyngholm 
                and Scheike, Thomas and Torp-Pedersen, Christian
                and Gerds, Thomas Alexander},
  journal =	 {The R Journal},
  volume =	 9,
  number =	 2,
  pages =	 {440--460},
  year =	 2017
}
```

--------------------------------

### Publish Head of Test Data (Org Mode)

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.html

Publishes the first few rows of the 'astest' data frame, excluding specific columns, in Org mode format with specified digit precision. Useful for displaying sample predictions.

```R
publish(head(astest[,-c(8,9)]),digits=1,org=TRUE)
```

--------------------------------

### Evaluate Models with Bootstrap Cross-Validation

Source: https://github.com/tagteam/riskregression/blob/master/simulation/k-fold-CV.org

Perform bootstrap cross-validation to evaluate model performance using Brier score and IPA.

```R
  X1 <- Score(list("Exclusive ERG"=lrfit.ex,"Inclusive ERG"=lrfit.inc),data=astest,
	      formula=Y1~1,summary="ipa",se.fit=0L,metrics="brier",contrasts=FALSE,
	      split.method = "bootcv", B=100)
```

```R
 X1
```

--------------------------------

### Simulate and Fit Competing Risk Model

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Simulates data using a competing risk model and fits a Cox proportional hazards model to both original and simulated data. Requires the `survival` package.

```R
set.seed(15)
d <- sim(ms.comp,10000)
fit.comp <- coxph(Surv(time,status==2)~log(thick)+sex+age+invasion,data=Melanoma)
fit.comp.s <- coxph(Surv(time,status==2)~log(thick)+sex+age+invasion,data=d)
```

--------------------------------

### Fit Penalized Regression Models (LASSO, Ridge, Elastic Net) with Formula Interface

Source: https://context7.com/tagteam/riskregression/llms.txt

Fits elastic net (LASSO/ridge) regression models using a formula interface, with support for survival outcomes, penalty weights, and restricted cubic splines. Requires 'riskRegression', 'glmnet', and 'survival' libraries.

```r
library(riskRegression)
library(glmnet)
library(survival)

# Binary outcome with penalization
set.seed(42)
bin_data <- sampleData(200, outcome = "binary")
bin_test <- sampleData(50, outcome = "binary")

fit_lasso <- GLMnet(Y ~ X1 + X2 + X3 + X6 + X7 + X8 + X9 + X10,
                    data = bin_data,
                    alpha = 1)  # LASSO

# Predict
risks <- predictRisk(fit_lasso, newdata = bin_test)
head(risks)
```

```r
# Survival outcome with mixed penalties
data(Melanoma, package = "riskRegression")

fit_surv <- GLMnet(Surv(time, status == 1) ~ pen(invasion, 4) + epicel +
                   ulcer + logthick + pen(sex, 0) + age,
                   data = Melanoma,
                   alpha = 0.5)  # Elastic net
```

--------------------------------

### Plot Survival Calibration Curves with Quantile Groups

Source: https://context7.com/tagteam/riskregression/llms.txt

Visualize survival calibration curves using quantile-based groups. This method divides the data into quantiles to assess calibration. Requires 'riskRegression', 'survival', and 'prodlim' libraries.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Survival outcome calibration
surv_data <- sampleData(200, outcome = "survival")

cox1 <- coxph(Surv(time, event) ~ X1 + X5 + X7, data = surv_data, x = TRUE)
cox2 <- coxph(Surv(time, event) ~ X1 + X3 + X6 + X7, data = surv_data, x = TRUE)

xs <- Score(list(Cox1 = cox1, Cox2 = cox2),
            Surv(time, event) ~ 1,
            data = surv_data,
            times = 10,
            plots = "cal")

# With quantile-based groups
plotCalibration(xs, method = "quantile", q = 10)
```

--------------------------------

### Fit Fine-Gray Regression Model with riskRegression

Source: https://context7.com/tagteam/riskregression/llms.txt

Illustrates fitting Fine-Gray subdistribution hazard models for competing risks analysis using the FGR function. This approach directly models the cumulative incidence function for a specific cause of interest. Includes model summary and prediction.

```R
library(riskRegression)
library(prodlim)

# Generate competing risks data
set.seed(123)
train <- sampleData(300, outcome = "competing.risks")
test <- sampleData(50, outcome = "competing.risks")

# Fit Fine-Gray model for cause 1
fgr_fit <- FGR(Hist(time, event) ~ X1 + X3 + X6 + X8,
               data = train,
               cause = 1)

# View model summary
summary(fgr_fit)

# Predict cumulative incidence at specific times
pred <- predict(fgr_fit, newdata = test, times = c(5, 10, 15))
head(pred)
```

--------------------------------

### Simulate and Compare Model Coefficients

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Simulates data from a recursive model and compares the resulting Cox proportional hazards model coefficients with those from the original data.

```R
set.seed(15)
d <- sim(ms.rec,10000)
fit.rec.s <- coxph(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=d)

res <- as.data.frame(cbind(coef(fit),coef(fit.rec.s),coef(fit.s)))
colnames(res) <- c("Parameters of the model from original data", "Parameters of the model from simulated data of the recursive model","Parameters of the model from simulated of the non-recursive model")
knitr::kable(res)
```

--------------------------------

### Score Logistic Regression Models for IPA and Brier

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.org

Calculates IPA and Brier metrics for logistic regression models. Ensure the 'Score' function and necessary models (lrfit.ex, lrfit.inc) are available. Results are presented in percentages.

```R
X1 <- Score(list("Exclusive ERG"=lrfit.ex,"Inclusive ERG"=lrfit.inc),data=astest,
            formula=Y1~1,summary="ipa",se.fit=1L,metrics="brier",contrasts=FALSE)
X1
```

--------------------------------

### Plot Calibration Curves with Bars for Binary Outcomes

Source: https://context7.com/tagteam/riskregression/llms.txt

Visualize calibration curves with bars for a specific binary outcome model. This provides a segmented view of calibration performance. Requires 'riskRegression', 'survival', and 'prodlim' libraries.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Binary outcome calibration
set.seed(10)
bin_data <- sampleData(300, outcome = "binary")

lr1 <- glm(Y ~ X1 + X5 + X7, data = bin_data, family = binomial)
lr2 <- glm(Y ~ X1 + X3 + X6 + X7, data = bin_data, family = binomial)

xb <- Score(list(model1 = lr1, model2 = lr2),
            Y ~ 1,
            data = bin_data,
            plots = "cal")

# Calibration with bars (for single model)
plotCalibration(xb, bars = TRUE, models = "model1")
```

--------------------------------

### Load Data for Logistic Regression

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Read a CSV file from a URL into a data frame for logistic regression analysis. The 'head' function displays the first few rows.

```R
UCLA.dat <- read.csv("https://stats.idre.ucla.edu/stat/data/binary.csv")
knitr::kable(head(UCLA.dat))
```

--------------------------------

### Simulate Data from Logistic Model

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Simulate data from a fitted logistic regression model. Use set.seed for reproducibility. The simulated data is then used to fit a new logistic regression model.

```R
set.seed(15)
d <- sim(UCLA.syn,10000)
fit <- glm(admit ~ gre + gpa + factor(rank), data = UCLA.dat)
fit.s <- glm(admit ~ gre + gpa + rank, data = d)

res <- as.data.frame(cbind(coef(fit),coef(fit.s)))
colnames(res) <- c("Parameters of the model from original data", "Parameters of the model from simulated data")
knitr::kable(res)
```

--------------------------------

### Evaluate risk models with Score

Source: https://github.com/tagteam/riskregression/blob/master/examples/useScore.org

Compares multiple risk prediction models by calculating AUC and Brier scores on a test dataset. Requires the riskRegression, randomForestSRC, and survival packages.

```R
library(riskRegression)
library(randomForestSRC)
library(survival)
set.seed(437)
d1 <- sampleData(400)
d2 <- sampleData(400)
f1 <- CSC(Hist(time,event)~X1+X6+X8,data=d1)
## sum(predictRisk(f1,newdata=d2,cause=1,times=4)>1)
## max(predict(f1,newdata=d2,cause=1,times=4,product.limit=TRUE)$absRisk)
f2 <- rfsrc(Surv(time,event)~X6+X8,data=d1,ntree=20)
x <- Score(list(f1,f2),data=d2,formula=Hist(time,event)~1,times=4,cause=1)
x$AUC$score
x$AUC$contrast
x$Brier$score
x$Brier$contrast
summary(x)
```

--------------------------------

### Cross-Validation with Bootstrap for Binary Outcomes

Source: https://context7.com/tagteam/riskregression/llms.txt

Perform cross-validation using bootstrap for binary outcome models. This helps in assessing the generalization performance of the models. Requires the 'riskRegression', 'survival', and 'prodlim' libraries.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Cross-validation with bootstrap
score_cv <- Score(list("Simple" = lr1, "Complex" = lr2),
                  formula = Y ~ 1,
                  data = bin_data,
                  split.method = "bootcv",
                  B = 100,
                  metrics = "auc")

print(score_cv)
```

--------------------------------

### Evaluate Models with K-Fold Cross-Validation

Source: https://github.com/tagteam/riskregression/blob/master/simulation/k-fold-CV.org

Perform 5-fold cross-validation to evaluate model performance.

```R
X1 <- Score(list("Exclusive ERG"=lrfit.ex,"Inclusive ERG"=lrfit.inc),data=astest,
            formula=Y1~1,summary="ipa",se.fit=0L,metrics="brier",contrasts=FALSE,
            split.method = "cv5", B=100)
```

```R
 X1
```

--------------------------------

### Synthesize Survival Model

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Creates a lava object to synthesize survival outcomes based on the provided formula and dataset.

```R
ms <- synthesize(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=Melanoma)
summary(ms)
```

--------------------------------

### Synthesize a Recursive Event History Model

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Creates a recursive model using the synthesize function with the recursive parameter set to TRUE.

```R
ms.rec <- synthesize(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=Melanoma,recursive = TRUE)
ms.rec
```

--------------------------------

### Plot Calibration Curves for Binary Outcomes

Source: https://context7.com/tagteam/riskregression/llms.txt

Visualize calibration curves for binary outcome models, showing agreement between predicted probabilities and observed outcomes. Supports multiple smoothing methods. Requires 'riskRegression', 'survival', and 'prodlim' libraries.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Binary outcome calibration
set.seed(10)
bin_data <- sampleData(300, outcome = "binary")

lr1 <- glm(Y ~ X1 + X5 + X7, data = bin_data, family = binomial)
lr2 <- glm(Y ~ X1 + X3 + X6 + X7, data = bin_data, family = binomial)

xb <- Score(list(model1 = lr1, model2 = lr2),
            Y ~ 1,
            data = bin_data,
            plots = "cal")

# Calibration curve plot
plotCalibration(xb)
```

--------------------------------

### Compute IPA for Model

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.org

Computes the IPA for the full model on the validation set.

```R
IPA(cscfit.inc,newdata=astest,times=3)
```

--------------------------------

### Fit Cause-Specific Cox Model with riskRegression

Source: https://context7.com/tagteam/riskregression/llms.txt

Demonstrates fitting cause-specific Cox proportional hazards models for competing risks data using the CSC function. It shows how to fit models with a single formula or different formulas per cause, and how to predict absolute risk.

```R
library(riskRegression)
library(survival)
library(prodlim)

# Generate competing risks data
set.seed(42)
train <- sampleData(200, outcome = "competing.risks")
test <- sampleData(50, outcome = "competing.risks")

# Fit cause-specific Cox model for both causes
csc_fit <- CSC(Hist(time, event) ~ X1 + X3 + X6 + X8,
               data = train,
               cause = 1)  # Primary cause of interest

# Fit with different formulas per cause
csc_fit2 <- CSC(list(Hist(time, event) ~ X1 + X3 + X6,      # Cause 1 model
                     Hist(time, event) ~ X2 + X8 + X9),      # Cause 2 model
                data = train)

# Predict absolute risk at specific times
pred <- predict(csc_fit, newdata = test, times = c(5, 10, 15), cause = 1)
print(pred)

# Predict with confidence intervals
pred_ci <- predict(csc_fit, newdata = test, times = c(5, 10),
                   cause = 1, se = TRUE, band = TRUE)
print(pred_ci)
```

--------------------------------

### Evaluate Survival Outcome Models

Source: https://github.com/tagteam/riskregression/blob/master/examples/useScore.org

Compares Cox proportional hazards and random forest models for survival outcomes at a specific time point.

```R
library(riskRegression)
library(survival)
library(randomForestSRC)
set.seed(437)
d1 <- sampleData(400,outcome="survival")
d2 <- sampleData(400,outcome="survival")
f1 <- coxph(Surv(time,event)~X6+X8,data=d1,x=1,y=1)
f2 <- rfsrc(Surv(time,event)~X6+X8,data=d1,ntree=20)
x <- Score(list(f1,f2),data=d2,formula=Hist(time,event)~1,times=4,conservative=1L)
x$AUC$score
x$AUC$contrast
x$Brier$score
x$Brier$contrast
summary(x)
```

--------------------------------

### Compare Model Coefficients

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Extracts coefficients from fitted Cox models and presents them in a data frame for comparison between original and simulated data. Uses `knitr::kable` for table formatting.

```R
res <- as.data.frame(cbind(coef(fit.comp),coef(fit.comp.s)))
colnames(res) <- c("Parameters of the model from original data", "Parameters of the model from simulated data")
knitr::kable(res)
```

--------------------------------

### Evaluate censored survival models

Source: https://github.com/tagteam/riskregression/blob/master/examples/demo.org

Compares Cox proportional hazards models using the Score function with survival data.

```R
set.seed(11)
train=sampleData(n=474,outcome="survival")
test=sampleData(n=736,outcome="survival")
f1=coxph(Surv(time,event)~X1+X2+X8+X9,data=train,x=TRUE,y=TRUE)
f2=coxph(Surv(time,event)~X3+X4+X7+X10,data=train,x=TRUE,y=TRUE)
x=Score(list(f1=f1,f2=f2),data=test,formula=Surv(time,event)~X1+X2+X3+X4+X8+X9+X10,times=5)
summary(x)$score
x1=Score(list(f1=f1,f2=f2),data=test,formula=Surv(time,event)~1,times=5)
summary(x1)$score
```

--------------------------------

### Calculate and Publish Risk Regression Results

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.org

This snippet calculates the 'risk.inc' column using a logistic regression model and then publishes a subset of the results.

```R
astest[,risk.inc:=100*predictRisk(lrfit.inc,newdata=astest)]
publish(head(astest[,-c(8,9)]),digits=1,org=TRUE)
```

--------------------------------

### Compute Average Treatment Effects (ATE) with G-formula, IPTW, or AIPTW

Source: https://context7.com/tagteam/riskregression/llms.txt

Estimates average treatment effects (risk differences and ratios) for survival and competing risks data using G-formula, IPTW, or augmented IPTW (doubly robust) estimators. Requires 'riskRegression', 'survival', and 'prodlim' libraries.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Generate survival data with treatment
set.seed(10)
n <- 200
dt <- sampleData(n, outcome = "survival")
dt$time <- round(dt$time, 1)
dt$X1 <- factor(rbinom(n, prob = 0.5, size = 1), labels = c("Control", "Treatment"))

# G-formula estimator (outcome model only)
fit_outcome <- coxph(Surv(time, event) ~ X1 + X6 + X8, data = dt, x = TRUE)

ate_gformula <- ate(event = fit_outcome,
                    treatment = "X1",
                    data = dt,
                    times = c(5, 10, 15),
                    se = TRUE)
print(ate_gformula)
```

```r
# AIPTW (doubly robust) estimator
fit_treatment <- glm(X1 == "Treatment" ~ X6 + X8, data = dt, family = binomial)
fit_censoring <- coxph(Surv(time, event == 0) ~ X1 + X6, data = dt, x = TRUE)

ate_aiptw <- ate(event = fit_outcome,
                 treatment = fit_treatment,
                 censor = fit_censoring,
                 data = dt,
                 times = c(5, 10),
                 estimator = "AIPTW",
                 se = TRUE)
print(ate_aiptw)
```

```r
# Competing risks with CSC model
dt_cr <- sampleData(n, outcome = "competing.risks")
dt_cr$X1 <- factor(rbinom(n, prob = 0.5, size = 1), labels = c("A", "B"))

fit_csc <- CSC(Hist(time, event) ~ X1 + X6 + X8, data = dt_cr)

ate_cr <- ate(event = fit_csc,
              treatment = "X1",
              data = dt_cr,
              times = c(5, 10),
              cause = 1,
              se = TRUE)
print(ate_cr)
```

--------------------------------

### Competing Risk Model Synthesis in R

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Use this function to synthesize a competing risk model. It requires a survival formula with time and status, and a data frame. The summary provides detailed model structure and parameters.

```r
ms.comp <- synthesize(Surv(time,status)~log(thick)+sex+age+invasion,data=Melanoma)
summary(ms.comp)
```

--------------------------------

### Predict Survival Outcomes with predictCox

Source: https://context7.com/tagteam/riskregression/llms.txt

Demonstrates using predictCox to compute survival probabilities, cumulative hazards, and hazard rates from fitted Cox regression models. Supports standard errors and influence functions, and requires specifying the prediction type.

```R
library(riskRegression)
library(survival)

# Fit Cox model
set.seed(50)
train <- sampleData(200, outcome = "survival")
test <- sampleData(20, outcome = "survival")

cox_fit <- coxph(Surv(time, event) ~ X1 + X6 + X8, data = train, x = TRUE)

# Predict survival probabilities
pred_surv <- predictCox(cox_fit, newdata = test, times = c(5, 10, 15),
                        type = "survival")
head(pred_surv$survival)

# Predict cumulative hazard
pred_cumhaz <- predictCox(cox_fit, newdata = test, times = c(5, 10, 15),
                          type = "cumhazard")
head(pred_cumhaz$cumhazard)
```

--------------------------------

### Extract Risk Predictions with predictRisk

Source: https://context7.com/tagteam/riskregression/llms.txt

Shows how to use the generic predictRisk function to extract predicted event probabilities from various fitted models (logistic, Cox, CSC) for binary, survival, and competing risks outcomes. Requires specifying outcome type and optionally prediction times and cause.

```R
library(riskRegression)
library(survival)
library(prodlim)

# Binary outcome example
set.seed(7)
bin_data <- sampleData(100, outcome = "binary")
bin_test <- sampleData(30, outcome = "binary")

# Fit logistic regression
lr_fit <- glm(Y ~ X1 + X2 + X6 + X8, data = bin_data, family = binomial)

# Get predicted risks for binary outcome
risks_binary <- predictRisk(lr_fit, newdata = bin_test)
head(risks_binary)

# Survival outcome example
surv_data <- sampleData(100, outcome = "survival")
surv_test <- sampleData(30, outcome = "survival")

# Fit Cox model
cox_fit <- coxph(Surv(time, event) ~ X1 + X6 + X8, data = surv_data, x = TRUE)

# Get predicted risks at specific times
risks_surv <- predictRisk(cox_fit, newdata = surv_test, times = c(5, 10, 15))
head(risks_surv)

# Competing risks example
cr_data <- sampleData(200, outcome = "competing.risks")
cr_test <- sampleData(30, outcome = "competing.risks")

csc_fit <- CSC(Hist(time, event) ~ X1 + X6 + X8, data = cr_data)
risks_cr <- predictRisk(csc_fit, newdata = cr_test, times = c(5, 10), cause = 1)
head(risks_cr)
```

--------------------------------

### Logistic Regression with synthesize

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Fit a logistic regression model using the synthesize function with multiple predictor variables. The 'admit' variable is modeled based on 'gre', 'gpa', and 'rank'.

```R
UCLA.syn <- synthesize(admit ~ gre + gpa + rank, data = UCLA.dat)
```

--------------------------------

### Predict and Publish Risk Scores in R

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.org

Calculates and publishes 3-year risk scores using Cox regression models. Requires pre-defined coxfit.ex and coxfit.inc models and astest dataset.

```R
astest[,risk.ex:=100*predictRisk(coxfit.ex,newdata=astest,times=3)]
astest[,risk.inc:=100*predictRisk(coxfit.inc,newdata=astest,times=3)]
publish(head(astest[,-c(8,9)]),digits=1,org=TRUE)
```

--------------------------------

### Define Simulation Boxplot Function

Source: https://github.com/tagteam/riskregression/blob/master/simulation/synthesize.html

Initializes a function to compare model coefficients across different sample sizes and recursion settings.

```R
library(ggplot2)
#> 
#> Attaching package: 'ggplot2'
#> The following object is masked from 'package:lava':
#> 
#>     vars
#library(gridExtra)

make.boxplot.simulation <- function(object, object.rec, data, sample.sizes, n.sim){
  #Copare to model from original data
  coef.compare <- coef(coxph(Surv(time,eventBin)~log(thick)+sex+age+invasion,data=data))
  
  # make dataframe with columnnames of the form: parameters, sample.size, recursion
  df <- data.frame(matrix(ncol = length(coef.compare)+2, nrow = 0))
  column.names <- c(names(coef.compare),"sample.size", "recursive")
```

--------------------------------

### Plot Calibration Curves for Survival Outcomes

Source: https://context7.com/tagteam/riskregression/llms.txt

Visualize calibration curves for survival outcome models at a specific time point. Handles censored data using local Kaplan-Meier/Aalen-Johansen estimators. Requires 'riskRegression', 'survival', and 'prodlim' libraries.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Survival outcome calibration
surv_data <- sampleData(200, outcome = "survival")

cox1 <- coxph(Surv(time, event) ~ X1 + X5 + X7, data = surv_data, x = TRUE)
cox2 <- coxph(Surv(time, event) ~ X1 + X3 + X6 + X7, data = surv_data, x = TRUE)

xs <- Score(list(Cox1 = cox1, Cox2 = cox2),
            Surv(time, event) ~ 1,
            data = surv_data,
            times = 10,
            plots = "cal")

plotCalibration(xs)
```

--------------------------------

### Evaluate Competing Risks Prediction Performance

Source: https://context7.com/tagteam/riskregression/llms.txt

Evaluate competing risks models using AUC at specified time points. Requires the 'riskRegression', 'survival', and 'prodlim' libraries. Set a seed for reproducibility.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Competing risks evaluation
cr_data <- sampleData(300, outcome = "competing.risks")

csc1 <- CSC(Hist(time, event) ~ X1 + X6, data = cr_data)
csc2 <- CSC(Hist(time, event) ~ X1 + X6 + X8 + X9, data = cr_data)

score_cr <- Score(list("CSC1" = csc1, "CSC2" = csc2),
                  formula = Hist(time, event) ~ 1,
                  data = cr_data,
                  cause = 1,
                  times = c(5, 10),
                  metrics = "auc")

print(score_cr)
```

--------------------------------

### Publish Model Results (Org Mode)

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.html

Publishes the results of the fitted logistic regression model 'lrfit.inc' in Org mode format, including odds ratios, confidence intervals, and p-values. Useful for generating reports.

```R
publish(lrfit.inc,org=TRUE)
```

--------------------------------

### Calculate IPA (R-squared) for Binary, Survival, and Competing Risks Outcomes

Source: https://context7.com/tagteam/riskregression/llms.txt

Computes the Index of Prediction Accuracy (IPA), also known as R-squared, by comparing the Brier score of a model to a null model. Supports binary, survival, and competing risks outcomes. Requires loading 'riskRegression', 'survival', and 'prodlim' libraries.

```r
library(riskRegression)
library(survival)
library(prodlim)

# Binary outcome
set.seed(18)
bin_learn <- sampleData(200, outcome = "binary")
bin_val <- sampleData(100, outcome = "binary")

lr1 <- glm(Y ~ X1 + X2 + X7 + X9, data = bin_learn, family = binomial)

# IPA on training data
ipa_train <- IPA(lr1)
print(ipa_train)

# IPA on validation data
ipa_val <- IPA(lr1, newdata = bin_val)
print(ipa_val)
```

```r
# Survival outcome
data(pbc, package = "survival")
pbc <- na.omit(pbc)

train_idx <- sample(1:nrow(pbc), size = 0.632 * nrow(pbc))
pbc_train <- pbc[train_idx, ]
pbc_test <- pbc[-train_idx, ]

cox1 <- coxph(Surv(time, status != 0) ~ age + sex + log(bili) + log(albumin),
              data = pbc_train, x = TRUE)

# IPA at specific time
ipa_surv <- IPA(cox1, formula = Surv(time, status != 0) ~ 1,
                newdata = pbc_test, times = 1000)
print(ipa_surv)
```

```r
# Competing risks outcome
data(Melanoma, package = "riskRegression")

csc_fit <- CSC(Hist(time, status) ~ age + sex + epicel + ulcer,
               data = Melanoma)

ipa_cr <- IPA(csc_fit, times = 1000, cause = 1)
print(ipa_cr)
```

--------------------------------

### Evaluate Binary Outcome Models

Source: https://github.com/tagteam/riskregression/blob/master/examples/useScore.org

Compares GLM and random forest models for binary outcomes using the Score function to calculate AUC and Brier scores.

```R
library(riskRegression)
library(randomForestSRC)
set.seed(437)
d1 <- sampleData(400,outcome="binary")
d2 <- sampleData(400,outcome="binary")
d1[,Y:=factor(Y,levels=c("0","1"),labels=c("0","1"))]
f1 <- glm(Y~X6+X8,data=d1,family="binomial")
f2 <- rfsrc(Y~X6+X8,data=d1,ntree=20)
x <- Score(list(f1,f2),data=d2,formula=Y~1)
x$AUC$score
x$AUC$contrast
x$Brier$score
x$Brier$contrast
summary(x)
```

--------------------------------

### Fit Logistic Regression Models for Binary Outcome

Source: https://github.com/tagteam/riskregression/blob/master/vignettes/IPA.org

Fits two logistic regression models to predict 1-year progression status using the simulated training data. One model includes the 'erg.status' biomarker, while the other excludes it. The 'Publish' package is used for displaying model results.

```R
lrfit.ex <- glm(Y1~age+lpsaden+ppb5+lmax+ct1+diaggs,data=astrain,family="binomial")
lrfit.inc <- glm(Y1~age+lpsaden+ppb5+lmax+ct1+diaggs+erg.status,data=astrain,family="binomial")
publish(lrfit.inc,org=TRUE)
```