supplement.Rmd

---
title: "A new model to predict survival in patients with end-stage liver disease"
subtitle: >
    Supplement
author:
    - Sebastian Gibb^[mail@sebastiangibb.de]
    - Thomas Berg
    - Daniel Seehofer
    - Berend Isermann
    - Thorsten Kaiser
bibliography:
  - "`r rprojroot::find_rstudio_root_file('analysis', 'bibliography', 'bibliography.bib')`"
output:
  bookdown::html_document2:
    toc: false
  bookdown::word_document2:
    keep_md: true
    pandoc_args: # pandoc_args doesn't support r evaluation
      - --csl=pandoc/csl/vancouver.csl
---

```{r knitr_setup, include = FALSE}
```

```{r knitr_setup_word, eval = !knitr::is_html_output(), include = FALSE}
knitr::opts_chunk$set(dpi = 1200, echo = FALSE)
```

```{r supplement_setup, echo = FALSE, message = FALSE}
library("ameld")
library("data.table")
library("targets")
library("mlr3")
library("mlr3misc")
library("mlr3proba")
library("mlr3viz")
library("viridis")
library("ggplot2")
```

# Survival plot

```{r survival.plot, fig.cap = "Survival plot. Kaplan-Meier survival estimate for the first 90 days of the analysed population.", fig.width = 8, fig.height = 5}
tar_load(zlog_data)
srv <- Surv(zlog_data$DaysAtRisk, zlog_data$Deceased)
col <- palette.colors(nlevels(zlog_data$MeldCategory))

srvfit <- survfit(srv ~ 1)
times <- c(2, 7, 14, 30, 90)

## calculate risk tables
sm <- summary(srvfit, times = times)
nrisk <- as.matrix(sm$n.risk)
ncumevents <- as.matrix(cumsum(sm$n.event))
rownames(nrisk) <- rownames(ncumevents) <- times

## keep old graphic parameters and restore them afterwards
old.par <- par(no.readonly = TRUE)

layout(matrix(1:3, nrow = 3), height = c(5, 1, 1))
plot_surv(
    srvfit,
    main = "Kaplan-Meier survival estimate to day 90",
    times = times,
    xmax = 90,
    col = col[2],
    cex = 1.2
)
par(mar = c(1.1, 5.1, 1.1, 2.1))
plot_table(
    nrisk, at = times, main = "Number at risk",
    xaxis = FALSE, cex.text = 1, ylabels = FALSE
)
par(mar = c(1.1, 5.1, 1.1, 2.1))
plot_table(
    ncumevents, at = times, main = "Cumulative number of events",
    xaxis = FALSE, cex.text = 1, ylabels = FALSE
)
par(old.par)
```

# Benchmark results of machine learning algorithms

```{r benchmark-plot, fig.cap = "Benchmark results of machine learning algorithms.", echo = FALSE, fig.width = 18, fig.height = 12}
tar_load(bmrk_results)
r <- bmrk_results$clone()
id <- grep("^scale", x = r$learners$learner_id, invert = TRUE, value = TRUE)
r$filter(task_id = "zlog_eldd", learner_id = id)
autoplot(r) +
    geom_boxplot(aes(fill = learner_id)) +
    geom_jitter(position = position_jitter(0.2)) +
    scale_fill_viridis(discrete = TRUE) +
    theme(axis.text.x = element_text(angle = 90, hjust = 1, vjust = 0.5))
```

```{r benchmark-table, echo = FALSE}
agg <- bmrk_results$aggregate(msr("surv.cindex", id = "harrell"))
agg[, `:=`
    (nr = NULL, resample_result = NULL, resampling_id = NULL, iters = NULL)
]
agg <- subset(agg, task_id == "zlog_eldd" & !grepl("^scale", learner_id))
agg[, task_id := NULL]
setorder(agg, -harrell)
knitr::kable(
    agg,
    col.names = c("Learner ID", "Harrell C"),
    caption = paste0("Benchmark results. Ranking of the ", nrow(agg), " tested machine learning algorithms.")
)
```

# AUROC trend

```{r roc-plot, echo = FALSE, fig.width = 18, fig.height = 12, fig.cap = "Trend in AUROC. Trend in the area under the time-dependent receiver operating characteristic curve (AUROC) based on the nonparametric inverse probability of censoring weighting estimate (IPCW) for AMELD, MELD, MELD-Na and MELD-Plus7 as described in [@blanche2013]."}
tar_load(timeROC_MELD)
tar_load(timeROC_MELDNa)
tar_load(timeROC_MELDPlus7)
tar_load(timeROC_RCV)

m <- list(
    AMELD = timeROC_RCV,
    MELD = timeROC_MELD,
    "MELD-Na" = timeROC_MELDNa,
    "MELD-Plus7" = timeROC_MELDPlus7
)
plot_surv_roc_trend(m, xlab = "time [days]", col = viridisLite::viridis(5)[-2])
```

# Variable importance

```{r vimpbs, fig.width = 8, fig.height = 10, fig.cap = "Variable importance by frequency of bootstrap selections.", echo = FALSE, message = FALSE}
tar_load(bootrcv)
plot(bootrcv, what = "selected", cex = 0.5)
```

```{r vimprf, fig.width = 8, fig.height = 10, fig.cap = "Variable importance by logrank in random forest.", echo = FALSE, message = FALSE}
tar_load(rngr)
plot_dots(
    sort(rngr$variable.importance),
    main = "Variable importance (random forest)",
    xlab = rngr$splitrule
)
```

# References