Living Review & automated surveillance

The observatory is designed to remain current through scheduled, transparent, human-in-the-loop updates.

1. Step 1

   Quarterly trigger

   Every 3 months a backend job calls the surveillance endpoint automatically (1st of Jan/Apr/Jul/Oct, 03:00 UTC).

2. Step 2

   Multi-source retrieval

   PubMed/MEDLINE (E-utilities), Europe PMC, Crossref, Semantic Scholar, OpenAlex. Scopus, Web of Science, Embase and IEEE Xplore remain manual / API-dependent where institutional access is required.

3. Step 3

   Deduplicate

   Records merged across sources by DOI, PMID, and a title+year+first-author similarity key.

4. Step 4

   AI screening

   Each record is screened against inclusion criteria and labelled likely include / maybe / likely exclude with a written rationale.

5. Step 5

   Human approval

   Likely-include and maybe records enter the admin review queue marked pending review. Nothing publishes without human sign-off; a summary email goes to the editorial lead after every run.

Search strategy

Boolean query combined with AND across all configured sources:

("vestibular schwannoma" OR "acoustic neuroma" OR "acoustic neurinoma" OR "eighth cranial nerve tumor" OR "8th cranial nerve tumor")
AND
(prediction OR predictive OR prognostic OR diagnostic OR "risk model" OR "prediction model" OR "prognostic model" OR nomogram OR "machine learning" OR "deep learning" OR "neural network" OR "random forest" OR SVM OR XGBoost OR LASSO OR "logistic regression" OR "Cox regression" OR "multivariable analysis")

Update history

Transparent log of every surveillance run.