Time Dependent Roc

Source: https://github.com/aipoch/medical-research-skills

Time-Dependent ROC Analysis

Use this skill to compute and plot time-dependent ROC curves for survival outcomes.

Use This Skill When

• You have survival data with futime and fustat columns.
• You need time-specific ROC curves and AUC values for a numeric risk marker.
• You want a command-line workflow with parameter validation and standardized outputs.
• Your file contains a numeric marker column named risk_score, or you will provide another marker explicitly with --marker_col.

Primary Command

Rscript scripts/main.R \
  --data_file <input_file> \
  --times <comma_separated_times> \
  --output_dir <output_dir>

Prerequisites

• Rscript is available in the shell.
• Required R packages: optparse, timeROC, survival, ggplot2.
• Optional package for Excel input: openxlsx.
• Install missing packages with Rscript -e 'install.packages(c("optparse", "timeROC", "survival", "ggplot2", "openxlsx"), repos="https://cloud.r-project.org")'.

Core Arguments

Input Requirements

• The input file must contain futime and fustat columns.
• The selected marker column must exist and be numeric or coercible to numeric.
• If --marker_col is omitted, the input file must contain risk_score.
• futime must be numeric or coercible to numeric.
• fustat must be numeric or coercible to numeric.
• Rows with missing futime, fustat, or marker values are excluded.
• At least 5 complete rows and at least 1 event for the requested cause are required.

Example input:

id	fustat	futime	risk_score	risk_group	GPR161	RIBC2
TCGA-C5-A1M5	1	5.6219	-1.1070	low	2.8252	5.3532
TCGA-EA-A5O9	0	2.1589	-0.8964	low	2.6046	4.4293
TCGA-C5-A3HL	0	1.7014	-0.9044	low	4.1248	4.9166

Minimal Workflow

1. Confirm the input file exists and includes futime, fustat, and risk_score, or plan to pass --marker_col.
2. Run scripts/main.R with the requested time points.
3. Check the output directory for the AUC table and PDF figure.

If you omit --data_file, the script exits with SKILL_MISSING_INPUT.

Outputs

Expected output structure:

<output_dir>/
├── data/
│   └── time_roc_points.<csv|txt>
├── session_info.txt
├── figure/
│   └── time_roc.pdf
└── table/
    └── time_roc_auc.<csv|txt>

Primary output files:

• data/time_roc_points.csv or data/time_roc_points.txt
• table/time_roc_auc.csv or table/time_roc_auc.txt
• figure/time_roc.pdf
• session_info.txt

Exported result fields include:

• time
• time_unit
• auc
• marker_col
• n_complete
• n_events

Exported ROC point fields include:

• false_positive_rate
• sensitivity
• time
• time_unit
• auc
• curve_label

Read These Files When Needed

Quick Examples

Basic analysis:

Rscript scripts/main.R \
  --data_file tests/data/time_roc_sample1.txt \
  --times 1,3,5 \
  --output_dir tests/output_basic

Specify marker column and plot styling:

Rscript scripts/main.R \
  --data_file tests/data/time_roc_sample2.txt \
  --marker_col GPR161 \
  --times 1,2,4 \
  --line_colors "#4DBBD5,#E64B35,#00A087" \
  --legend_position bottom \
  --output_dir tests/output_gpr161

Validation

Rscript scripts/main.R --help

Rscript scripts/main.R \
  --data_file tests/data/time_roc_sample1.txt \
  --times 1,3,5 \
  --output_dir tests/validation_output

After running analysis, verify that these files exist:

• tests/validation_output/data/time_roc_points.csv
• tests/validation_output/table/time_roc_auc.csv
• tests/validation_output/figure/time_roc.pdf
• tests/validation_output/session_info.txt

Common Errors

• SKILL_FILE_NOT_FOUND: Input file path is wrong or inaccessible.
• SKILL_MISSING_COLUMNS: futime or fustat is missing.
• SKILL_INVALID_DATA: The marker column is non-numeric, no complete rows remain, event coding is unsuitable, or requested time points yield unusable AUC values.
• SKILL_INVALID_PARAMETER: An argument value is invalid.
• SKILL_INSUFFICIENT_DATA: Too few complete rows or no target events remain.
• SKILL_DEPENDENCY_MISSING: A required R package such as optparse, timeROC, survival, ggplot2, or openxlsx is unavailable.

If the issue is not obvious, read references/troubleshooting.md.