The Protocol

From Question to Search String

How do you translate a PECO-F question into a reproducible search strategy — one you can log, hand off, and defend?

Why reproducibility matters in a policy context

A literature search that cannot be reproduced is not a search — it is a reading list. For WHO India work, reproducibility is not academic rigour for its own sake. It matters because a health financing recommendation that reaches a Joint Secretary or a state health ministry may be challenged on methodological grounds. "We searched PubMed" is not a protocol. A documented, replicable search string with date ranges, database names, and inclusion logic is.

AI tools make it easier to search — and easier to search inconsistently. This session gives you a five-step protocol that produces a defensible, repeatable search every time, including a log you can attach to any evidence brief.

The five-step protocol

Lock your PECO-F elements

Before opening any tool, write out all five PECO-F elements in plain language. Do not start searching until you can state the population, intervention, comparator, outcome, and feasibility/equity constraint in one sentence each. A vague P or a missing F will produce a vague search. This is the session 3 output — if you skipped it, go back.

Expand each element into synonym clusters

Every PECO-F element has multiple valid terms. "Catastrophic health expenditure" is also "out-of-pocket payments," "financial catastrophe," "impoverishment," "financial hardship." "Health insurance" is also "social health insurance," "community-based health insurance," "CBHI," "PM-JAY," "government health scheme." List every synonym before you write a search string. AI tools are good at this expansion step — ask Claude or GPT to "generate all synonyms and related terms for [element]" before building the string.

Build the Boolean string

Combine synonym clusters using Boolean logic: OR within clusters (to catch all synonyms), AND between clusters (to require all PECO elements). Wrap each cluster in parentheses. Add geographic and date filters last. The structure is always: (P synonyms) AND (E synonyms) AND (O synonyms) AND (geography filter) — with C and F as optional refiners rather than mandatory ANDs, since over-constraining kills recall.

Run in sequence across tools

No single database covers the full literature. The minimum defensible sequence for WHO India health financing work: PubMed / Medline (indexed biomedical), Embase (broader European coverage), Cochrane (systematic reviews), Dimensions.ai (LMIC-filtered), plus a targeted grey literature search of HTAIn, NSSO, NHA, and WHO IRIS. Log the date, database, string, and result count for each run.

Log everything in a search protocol table

A one-page protocol log attached to every brief makes the search auditable. It takes five minutes to complete and eliminates the most common methodological challenge to evidence synthesis: "How did you decide what to include?" See the template below.

A worked example: PM-JAY and financial protection

Starting question: Does PM-JAY reduce catastrophic health expenditure among below-poverty-line households in India?

Step 2 — synonym expansion

P: "below poverty line" OR "BPL" OR "low income" OR "poor households" OR "informal sector" OR "uninsured"
E: "PM-JAY" OR "Pradhan Mantri Jan Arogya Yojana" OR "Ayushman Bharat" OR "government health insurance" OR "social health insurance" OR "publicly funded health insurance"
O: "catastrophic health expenditure" OR "out-of-pocket payment" OR "financial protection" OR "financial hardship" OR "impoverishment" OR "financial risk protection"
Geography: "India" OR "South Asia"

Synonym clusters — one per PECO element, before Boolean assembly

Step 3 — Boolean string

("below poverty line" OR "BPL" OR "low income" OR "poor households" OR "informal sector")
AND
("PM-JAY" OR "Pradhan Mantri Jan Arogya Yojana" OR "Ayushman Bharat" OR "government health insurance" OR "social health insurance" OR "publicly funded health insurance")
AND
("catastrophic health expenditure" OR "out-of-pocket payment" OR "financial protection" OR "financial risk protection" OR "impoverishment")
AND
("India" OR "South Asia")

PubMed / Embase format — paste directly into database advanced search

AI prompt version (for Elicit, Claude, Consensus)

Paste-ready AI prompt "Search for systematic reviews and economic evaluations published after 2015 on whether publicly funded health insurance schemes — specifically PM-JAY or Ayushman Bharat — reduce catastrophic health expenditure or out-of-pocket payments among low-income or below-poverty-line households in India. Include studies that report equity outcomes disaggregated by income quintile, gender, or geography. Exclude studies focused solely on clinical outcomes without financial protection measures."

The search protocol log

Every search should generate a one-page log. Here is the minimum structure — fill one row per database run:

Database / Source	Date run	Search string / query	Filters applied	Results	Included after screening
PubMed	2024-11-12	(PM-JAY OR Ayushman Bharat…) AND (catastrophic…) AND (India…)	2015–2024 · English	147	18
Cochrane	2024-11-12	Same string, simplified	Systematic reviews only	9	4
Dimensions.ai	2024-11-13	Natural language query + LMIC filter	India · 2015–2024	203	22
HTAIn website	2024-11-13	Manual browse · PM-JAY category	—	6 reports	3
WHO IRIS	2024-11-13	"financial protection" AND "India"	2015–2024	31	5

Build your own search string

⚗️ String builder — enter your PECO-F elements

Fill in your elements using plain language or synonyms separated by commas. The builder will assemble a Boolean database string and a copy-paste AI prompt for you.

P — Population e.g. below poverty line households, informal sector workers

E — Intervention / Exposure e.g. community health insurance, PM-JAY, tax-funded coverage

O — Outcome e.g. catastrophic expenditure, financial risk protection, utilisation

F — Equity / Feasibility constraint e.g. gender equity, income quintile, willingness to pay

Geography Leave blank for global; add countries or regions to narrow

Boolean string — for PubMed / Embase / Cochrane

AI prompt — for Elicit, Claude, Consensus

Grey literature checklist — run manually

🎯 Key takeaway

A search protocol is not overhead — it is the thing that makes your evidence brief defensible when challenged. The five steps are: lock PECO-F, expand synonyms, build Boolean string, run in sequence across databases, and log everything. The protocol log from this session becomes the methodology appendix of the brief you write in Session 5. Build it once; it takes ten minutes and saves ten hours of justification later.