Root Cause Analysis: Multivariate Failures, HPLC Productivity, and Why the First Root Cause Is Almost Never Right
A Six Sigma expert on why batch failures are multivariate like car accidents, how AI found humidity as a root cause humans missed, and how RPA cut the reviewer-to-analyst ratio from 1:5 to 1:10.
Vivek Gera Host
Co-founder · Leucine
Ganadhish Kamat
Ex-Quality Head · Dr. Reddy's & Lupin
About this episode
Vivek Gera speaks with Ganadhish Kamat, former Quality Head at Dr. Reddy's and Lupin, about the structural reasons why root cause investigations fail in pharmaceutical manufacturing. Kamat argues that QbD studies are done for submission compliance rather than genuine process understanding — so when a multivariate failure hits, the product knowledge base is too thin to find the cause. Using a principal component analysis that identified ambient humidity as the root cause of a dissolution failure — a parameter that was within spec but still causing the failure — the episode makes a compelling case for multivariate AI analysis over linear parameter correlation. The episode also covers practical wins: using HPLC software injection logs to show actual equipment utilization at 25% versus claimed 80%, and deploying RPA for chromatographic audit trail review to double the reviewer-to-analyst ratio.
Topics
Key takeaways
- Root cause failures start at development — QbD studies are done to meet submission timelines, not to build genuine process knowledge, leaving critical parameter relationships undiscovered until commercial failures surface them
- Batch failures are multivariate like car accidents: slippery road AND darkness AND distraction together cause the crash; a single parameter correlation almost never reveals the true root cause
- AI-powered principal component analysis identified ambient humidity — within specification, but still causally driving dissolution failure — something linear evaluation of CPPs would never have found
- HPLC equipment utilization at most labs is 25–30% actual versus 70–80% claimed; the gap only becomes visible when you measure using the chromatography software's own injection logs, not manual estimates
- RPA for chromatographic audit trail review cut the reviewer-to-analyst ratio from 1:5 to 1:10 — by automating objective compliance checks that humans perform inconsistently and with fatigue
- Product robustness — defined by lot acceptance rate, complaint rate, and process capability above 1.3 Cpk — is the only reliable path to drug supply reliability and regulatory confidence; compliance is just the licence to operate
Powered by
LeucineOS
See how an AI-native quality platform unifies QMS, compliance, and manufacturing operations.
Explore LeucineOSSee how Leucine's AI-native platform is transforming pharmaceutical quality and manufacturing operations.
