Longitudinal Monitoring - Usage Guide

Overview

Tracks ctDNA across serial liquid-biopsy timepoints for molecular residual disease (MRD) and treatment-response monitoring. Treats MRD as a binary integrated detected/not-detected call across the patient's full variant set rather than a per-timepoint VAF threshold, handles undetectable samples as left-censored at the per-sample limit of detection, and estimates clearance kinetics and molecular relapse with that censoring respected.

Prerequisites

pip install pandas numpy scipy matplotlib

Quick Start

Tell your AI agent what you want to do: - "Build a detected/not-detected MRD trajectory from my serial samples with per-sample LoD carried through" - "Estimate ctDNA clearance half-life, treating undetectable draws as censored not zero" - "Call molecular relapse only on a confirmed rising trend over consecutive draws" - "Generate a monitoring report with baseline, nadir, and below-LoD points flagged"

Example Prompts

MRD Trajectory and Detection

"Integrate these serial draws into a binary MRD call per timepoint, keeping the per-sample LoD as a covariate."

"Mark every undetectable draw as left-censored at its limit of detection rather than plotting it at zero."

Clearance Kinetics

"Estimate the clearance half-life from the on-treatment decay phase, excluding the post-op surge and any rebound."

"Fit log-linear decay only over the uncensored points and report the slope and R-squared."

Relapse and Response

"Call molecular relapse only if ctDNA rises above nadir on at least two consecutive confirmed draws."

"Classify molecular response using my assay's validated cutoff, not a generic 2-log rule."

Interpretation

"Flag where lead time is being reported so I do not confuse it with a survival benefit."

"Check whether any rising signal could be CHIP drift rather than tumor in this tumor-naive panel."

What the Agent Will Do

Assemble serial measurements with timepoint, tumor fraction/VAF, and per-sample LoD (genome-equivalents)
Flag below-LoD draws as left-censored and compute baseline/nadir-referenced change without using zero
Estimate clearance half-life over the genuine monotonic-decay phase only
Call molecular relapse from a confirmed rising trend, annotated with each draw's LoD
Surface caveats: non-harmonized response definitions, lead-time bias, CHIP drift, low-shedder negatives

Tips

MRD is a binary integrated call across the full variant set, not a per-locus VAF threshold
Undetectable is left-censored at the per-sample LoD, never a true zero; log(0) breaks decay fits
Per-sample LoD depends on cfDNA input; a low-input negative is weak evidence and not comparable to a high-input one
Use the same assay and tube type across timepoints; mixing them confounds the trajectory
Plot tumor fraction on a log y-axis with a shaded below-LoD band; draw censored points at the LoD, not at zero
Molecular-response cutoffs ("2-log" vs "90%" vs molecular CR) are non-harmonized; use the assay's own validated definition
Require a confirmed trend over >=2 consecutive draws before calling relapse; repeated surveillance is a multiple-testing problem
For tumor-naive panels, control for CHIP drift with matched WBC sequencing before reading a rise as relapse
Lead time over imaging is analytic sensitivity, not proven benefit; clinical utility needs an interventional design

ctdna-mutation-detection - detect the variant set that is then tracked
tumor-fraction-estimation - per-timepoint tumor burden
analytical-validation - per-timepoint LoD and left-censoring of undetectable samples
fragment-analysis - fragmentomic trends as a complementary monitoring signal
clinical-biostatistics/survival-analysis - relapse, lead-time, and endpoint analysis