Clinical Review — 40 Difficult Cases

Palli Sahayak Review Console

Six independent palliative care doctors grade Palli Sahayak's RAG outputs on 40 difficult Indian palliative care cases — 20 oncology and 20 pulmonology — using a six-dimension rubric anchored to WHO, IAPC, Pallium India, NICE and GOLD guidelines.

Open Review Console → Read the rubric spec

About

Palli Sahayak is a voice-first AI clinical decision-support tool for palliative care, deployed in 22 Indian languages and grounded in evidence-based palliative-care guidelines. This console is a small, independent study: six palliative care doctors rate the tool's answers on a set of difficult standardized cases that span pain management, end-of-life care, end-stage COPD, post-TB fibrosis and Indian occupational lung disease.

📋 40 difficult cases

20 oncology (pain, non-pain symptoms, end-of-life, psychosocial) + 20 pulmonology (COPD, post-TB fibrosis / IPF, lung cancer, occupational disease). 10 easy + 10 moderate + 20 difficult tiers.

🇮🇳 Indian-context grounded

60%+ of cases include India-specific elements: NDPS opioid access, family-based decision-making, rural distance, regional formulary, occupational exposures (silicosis, byssinosis).

👨‍⚕️ Six blinded reviewers

Two reviewers independently grade each (vignette, RAG-output) pair. Inter-rater reliability assessed via weighted Cohen's κ; target κ ≥ 0.70 per dimension.

📊 Comprehensive rubric

Six dimensions × 1-5 Likert + 25 binary sub-items + 0-10 overall score + free-text comments + controlled-vocabulary tags.

Six rubric dimensions

#DimensionWhat it measures
1Clinical AccuracyConcordance with WHO / IAPC / Pallium India / NICE / GOLD guidelines
2SafetyContraindications, dose limits, drug interactions, emergency triggers
3EmpathyAcknowledgment of distress; non-judgmental tone; cultural validation
4ActionabilityCan the ASHA / caregiver actually act on the advice in their setting?
5CompletenessSymptom + side effects + follow-up + caregiver education + escalation
6Indian Context FitNDPS opioid access, language register, family decision-making, regional formulary

For reviewers

1

Open the Review Console

Go to palli-sahayak-review-production.up.railway.app on any modern browser (desktop recommended for the 3-column layout).

2

Log in with your reviewer account

Pick your name from the dropdown and enter the PIN distributed separately by the KCDH-A team. Sessions last 8 hours; you can come back across multiple sessions.

3

Grade vignettes one at a time

The dashboard shows all 40 cases with status (pending / yours / others'). Open any case to see the vignette on the left, the RAG output in the middle, and the rubric on the right. Allow ~10-15 minutes per case for the calibration round, ≤ 10 min thereafter.

4

Flag uncertainty for committee review

For any case where you assign a safety score of 1-2, or where the response reveals a previously undocumented failure mode, tick the "Flag for committee review" box at the bottom. The KCDH-A clinical team will adjudicate flagged cases.

Study metadata

Design
Six independent palliative care doctors rate Palli Sahayak's RAG outputs on 40 difficult cases (20 oncology + 20 pulmonology).
Rubric spec
docs/v53_phase2_difficult-scenarios_rubric_detailed-specs.md
Reviewers
Six anonymized reviewers (doc1–doc6), all qualified palliative care doctors. Reviewer identity is blinded for the rating.
Languages
English, Hindi, Tamil, Bengali, Kannada
Inter-rater reliability target
Weighted Cohen's κ ≥ 0.70 per dimension (≥ 0.80 for safety)
Code & data
github.com/inventcures/rag_gci (MIT-licensed)