Dr. Avinash kumar gupta

VibeRounds Directory 2: Lightweight Disease-Specific Protocol

Tuberculosis as the Index Disease — From Acute Decision-Making to Critical Appraisal

The core idea: Rather than running all 20 modules on every case, a lightweight model serialises only the high-yield modules in a fixed sequence — starting with rapid action selection (Module 20: RPD), then cross-matching against guidelines and research, and ending with adversarial stress-testing to prevent anchoring bias.

1. Why Build a Lightweight Model?

The full VibeRounds 20-module pipeline produces deep, comprehensive clinical knowledge — but at significant analytical cost. For a disease-specific directory, a streamlined pathway is needed that:

Preserves clinical rigour without requiring full-stack analysis on every case
Moves from fast action to slow verification in a structured sequence
Can be deployed across the scale of a real PubMed corpus (hundreds of thousands of records)
Remains teachable for students while remaining useful for clinicians

The proposed model — Directory 2: Disease-Specific Scaffolds — serialises five targeted modules into a fast-to-slow reasoning loop.

2. Selecting the Index Disease: Why Tuberculosis?

TB is an optimal choice for the index disease in a medicine-focused directory. While CKD is the primary tutorial example in the VibeRounds protocol, TB offers superior clinical complexity across multiple framework dimensions.

Criterion	Why TB Excels
Differential complexity	TB is the “great mimicker” — it can present identically to malignancy, autoimmune disease, sarcoidosis, or lymphoma, making Module 12 (Devil’s Advocate) essential rather than optional
Multisystem network reasoning	TB involves interacting systems — pulmonary, renal, neurological, disseminated — making Module 18 (Causal Networks) deeply applicable
Conditional weighting	The significance of a single finding (e.g., night sweats) is conditional on other factors (geographic exposure, HIV status, contact history), which trains probabilistic thinking
Social determinants	TB is inseparable from poverty, overcrowding, malnutrition, and healthcare access — making Module 19 (Social Medicine) structurally necessary, not supplementary
Resource constraints	TB management often requires decision-making under resource ceilings, directly testing Module 14
Acute emergencies	Massive haemoptysis and TB meningitis triage require the RPD model (Module 20) for time-critical action selection
Guideline complexity	Rapidly evolving drug-resistant TB regimens (e.g., BPaLM) mean the evidence frontier is active and contested

3. The Data Corpus for TB in Directory 2

Source Type	Volume	Notes
TB case reports (PubMed free full-text)	36,950	High value for understanding atypical presentations
All TB research publications	319,050	Requires strategic pruning; cannot be processed uniformly
TB-specific guidelines	6,934	Many likely outdated; requires evidence frontier filtering

Processing priority: Given this volume, the lightweight protocol focuses on clinical utility over comprehensiveness. The goal is to identify the highest-yield items at each stage and discard the rest.

4. The Lightweight Protocol: Five-Step Sequence

Step 1 — Sorting and Pattern Recognition (Module 6: Registry-Level Analytics)

Purpose: With 319,050 research items, the first task is to identify which TB phenotypes are clinically most relevant. Processing everything uniformly is neither feasible nor useful.

Task:

Perform horizontal analytics to identify recurring TB phenotypes: Multi-Drug Resistant TB (MDR-TB), TB Meningitis, Miliary TB, Extrapulmonary TB, TB-HIV co-infection
Match these clusters against local de-identified patient logs to determine which phenotypes most commonly present in the target clinical environment
Use this as the primary filter for which case reports and research papers enter the deeper pipeline

Module: 6 (Level 6 — Phenotype Detection)

Step 2 — Acute Action Selection (Module 20: Recognition-Primed Decision)

Purpose: TB produces time-critical emergencies that cannot wait for systematic analysis. This step trains rapid, pattern-driven decision-making for acute presentations.

Task:

For each high-risk phenotype identified in Step 1, define the critical recognition pattern (e.g., “patient with known TB contact, fever, worsening headache, neck stiffness” → TB meningitis)
Retrieve one plausible immediate course of action — not a differential list — and simulate it forward to identify failure points
Example failure point check: “If I give a fluid bolus to this patient with possible TB-related cor pulmonale, what happens to right heart function?”

The RPD discipline here is deliberate: the learner is not given a menu of options. They must commit to one action and defend it against simulated failure — the same cognitive demand placed on a senior clinician in an emergency.

Module: 20 (RPD — Naturalistic Decision-Making)

Step 3 — Guideline and Evidence Frontier Pruning (Module 21: Evidence Frontier Search)

Purpose: Of the 6,934 TB guidelines in the corpus, many will be outdated. This step retrieves the current frontier, not the historical archive.

Task:

Formulate a PICO question anchored to the acute decision made in Step 2
Example: “In adults with MDR-TB (P), does a 6-month BPaLM regimen (I) compared to standard 18–24 month care (C) achieve non-inferior treatment success and lower toxicity (O)?”
Use live, verified search to identify the most recent WHO or national guidelines and active clinical trials relevant to that PICO question
Flag any conflicts between retrieved guidelines (e.g., national protocol versus WHO recommendation)

Module: 21 (Steps 21.1 and 21.3 — Frontier Identification and Trial Mapping)

Step 4 — Deep Case Report Digestion (Module 10: Layered Article Reading)

Purpose: The 36,950 TB case reports are the highest-value source for understanding TB as the great mimicker — but only if analysed against a real clinical anchor, not read in isolation.

Task:

Take a de-identified patient vignette from the local clinical environment
Apply layered digestion: first extract the methods and presentation from a matched case report, then compare findings to the local patient
Focus specifically on atypical features — presentations that do not fit the standard TB illness script and which required a diagnostic detour before the correct diagnosis was reached
Document which “mimicker” diagnoses were considered and how TB was ultimately confirmed

Module: 10 (Layer 1 — Methods Translation; Layer 2 — Findings Mapping)

Step 5 — Adversarial Stress-Testing (Module 12 + Framework D)

Purpose: In a TB-endemic environment, clinicians are at high risk of “anchoring” — defaulting to a TB diagnosis without adequately excluding other serious diseases. This step structurally forces the opposite.

Task:

Apply Devil’s Advocate mode: generate three arguments against the TB diagnosis and demand a clinical rebuttal for each
Example challenges:
- “Why is this not sarcoidosis? The lymphadenopathy pattern, the calcium level, and the chest imaging are all consistent.”
- “Why is this not lymphoma? The B symptoms, the age, and the absence of contact history support it.”
- “Why is this not a fungal infection? The patient is immunocompromised and the culture takes weeks.”
Apply Framework D (Critical Awareness) to audit for automation bias — the risk of trusting the guideline or the AI output without interrogating the underlying assumptions

Module: 12 (Step 12.1 — First Assault); Framework D (Bias and Hallucination Audit)

5. The Full Protocol at a Glance

Priority	Step	Task	Module	Reasoning Mode
1 — Fast	Phenotype Sorting	Identify which TB subtypes are locally relevant	Module 6	Analytical
2 — Action	Acute Triage	RPD simulation for time-critical emergencies	Module 20	Naturalistic / Intuitive
3 — Verify	Guideline Pruning	PICO-structured evidence frontier search	Module 21	Traditional / Systematic
4 — Digest	Case Report Anchoring	Layered digestion of atypical presentations	Module 10	Slow / Comparative
5 — Audit	Adversarial Stress-Test	Devil’s Advocate + bias check	Module 12 + Framework D	Adversarial

6. Design Principles of This Lightweight Model

Fast-to-Slow Sequencing

The protocol is deliberately sequenced so that fast, intuitive reasoning (RPD) happens before slow, systematic reasoning (article digestion, critical appraisal). This mirrors how expert clinicians actually reason under time pressure — and makes the pedagogical sequence more realistic than starting with a literature review.

Socratic Constraint Throughout

At no step does the protocol give the learner an answer directly. Each module forces the learner to:

Construct their own pattern recognition (Step 1)
Commit to their own action and simulate it (Step 2)
Formulate their own PICO question (Step 3)
Apply evidence to their own patient (Step 4)
Attack their own conclusion (Step 5)

This prevents automation bias — the risk that a powerful knowledge system produces passive, uncritical users.

Manageable at Scale

Because the protocol runs only five targeted modules rather than all twenty, it can realistically be applied to:

Every TB phenotype cluster identified in Step 1
High-priority case reports in the 36,950-item corpus
Emerging guideline conflicts as new evidence arrives

The full 20-module pipeline remains available for complex index cases. The lightweight model handles the high-volume, disease-specific layer.

7. Suggested Extension Diseases by Specialty

Once the TB model is validated, the lightweight protocol can be adapted by specialty using the same five-step sequence:

Specialty	Case Reports Available	Suggested Index Disease
Medicine	745,738	Tuberculosis (index)
Surgery	283,585	Acute abdomen with diagnostic uncertainty
Paediatrics	66,278	Kawasaki disease (great mimicker; acute decision critical)
Gynaecology	21,947	Ectopic pregnancy with atypical presentation
Psychiatry	14,453	First-episode psychosis with organic cause
Otorhinolaryngology	12,737	Neck mass with malignancy versus infection differential

Framework: VibeRounds Master Case Analysis Protocol v1.0 | Dr. Avinash Kumar Gupta | June 2026 Directory 2 Concept: Lightweight Disease-Specific Scaffolds — TB as Index Disease