# Patient-specific rubric-based evaluation of large language models across longitudinal cancer care

Source: [arXiv](https://arxiv.org/abs/2610.35847v1)  
Feed7 permalink: https://feed7.dev/p/2610-35847v1-0se27hb  
Published: 2026-10-26T17:56:26.000Z  
Trust: Needs Review (needs_review)

## Why Included

NPC-RubricBench finds that clinical LLM failure modes shift across cancer-care phases. For agent builders, it shows why a strong score on one task may not predict reliable performance across a workflow.

## Source Summary

NPC-RubricBench evaluates **30 LLMs** across six cancer-care phases, using records from **405 patients** and 17,688 patient-specific criteria. Errors dominated anatomical staging, while omissions were more common in initial work-up and follow-up.

## Practical Implication

For agents that operate across a workflow, evaluate each phase and decision point instead of relying on a single diagnostic score. The study also finds a prompt trade-off: checklist-like requests raised decision quality but reduced evidence faithfulness and safety adherence.

## Agent-Ready Context

NPC-RubricBench evaluates **30 LLMs** across six cancer-care phases, using records from **405 patients** and 17,688 patient-specific criteria. Errors dominated anatomical staging, while omissions were more common in initial work-up and follow-up.

For agents that operate across a workflow, evaluate each phase and decision point instead of relying on a single diagnostic score. The study also finds a prompt trade-off: checklist-like requests raised decision quality but reduced evidence faithfulness and safety adherence.

This benchmark covers nasopharyngeal cancer care at two centers. Its results do not establish performance in other diseases or clinical settings.

## Connected Context

Feed7 judgment across 928 accumulated Signals:

This adds a patient-specific, phase-by-phase view of clinical agent quality: staging errors and omissions vary across care stages, and checklist prompting trades better decision quality for weaker evidence faithfulness and safety adherence. It supports measuring consequential errors at each workflow step, while its two-center nasopharyngeal cancer scope limits broader claims.

- [Measure Before You Train: Typed Decision Models vs. a Classical Baseline vs. a Generative LLM for Routing a Personal Multilingual Inbox](https://feed7.dev/p/2610-32955v1-18vy54w) — Both make action-relevant errors visible beyond a single accuracy score; the inbox study tracks missed actionable mail, while this benchmark distinguishes clinical errors and omissions by phase.
- [Domain-Specific Hallucination Detection in Large Language Models](https://feed7.dev/p/2609-11878v1-1ltrudx) — The prompt trade-off here reinforces the need to evaluate evidence faithfulness and safety directly; the hallucination study likewise cautions that generic evaluation safeguards may not transfer across domains.

## Context Map

- Layer: benchmark
- Domains: research, data
- Topics: agent-evals, agent-reliability

## Uncertainty

- This benchmark covers nasopharyngeal cancer care at two centers. Its results do not establish performance in other diseases or clinical settings.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.