Behind the Paper

What a "good" laboratory found when it went looking for its own errors

Our laboratory already met conventional quality benchmarks. A simple risk assessment, run by our own staff, still found weak points. Targeted fixes cut documented analytical errors by 76% in a year, for 45 staff-hours of effort.

Why we looked at the "solved" phase

When people talk about laboratory errors, the conversation usually starts with the pre-analytical phase: wrong tubes, mislabelled samples, delayed transport. That focus is justified, because most errors happen before a sample ever reaches an analyser. The analytical phase, by contrast, is often treated as the solved part of testing. Analysers are automated, quality control runs every day, and accreditation audits check the paperwork.

Yet errors in this phase matter. A missed calibration or a misread quality-control chart changes the number a doctor sees on the report. We wanted to test the "solved" assumption in our own laboratory.

How I came to this question

I am not an academic researcher. I trained in chemistry, have worked in laboratory medicine since 2000, and founded our laboratory in Rajkot, western India, in 2012. My research comes from applying quality-control rules to our own laboratory's data. This paper is one of a series in which we have examined each phase of testing in our laboratory: before, during and after analysis.

Our laboratory is medium-sized: about 51,000 tests a year, 12 technical staff, accredited by NABL, India's national accreditation board for laboratories. By conventional measures we were already doing well. Our analytical error rate worked out to 4.8 on the Sigma scale, above the 3-Sigma level usually treated as the minimum. The easy conclusion was that there was nothing left to fix. The more useful question was what we would find if we looked systematically, and whether a laboratory of our size could act on it without a dedicated quality department.

What we did

In December 2023 we brought together a small team: the laboratory director, the quality assurance manager, two senior bench technologists and our laboratory information system administrator. They mapped every step between loading a sample on the analyser and verifying the result.

We used Failure Modes and Effects Analysis (FMEA), a method from engineering that asks what could go wrong, how badly, and how often. We deliberately simplified it. Each failure mode was scored only for severity and probability, each from 1 to 3, giving a Hazard Score out of 9. Classical FMEA adds a third factor, detectability, but scoring it well needs validated data we did not have, and we wanted a method other laboratories like ours could realistically repeat. Team members scored independently before discussing their scores.

Five failure modes came out: quality-control interpretation, reagent storage, external quality assessment (EQA) reporting, instrument calibration and water quality. Three scored as high risk. We observed our baseline through 2024 while designing corrective and preventive actions (CAPA), installed them as one bundle in the last three months of that year, and measured again through 2025.

What we found

Documented analytical errors fell by 76%, from 21 in 2024 to 5 in 2025, moving us from 4.8 to 5.2 Sigma. Designing and rolling out the changes took 45 staff-hours. None of the errors in either year reached a patient report in a form that altered a clinical decision; all were caught at a quality-control, calibration or verification step. Fewer errors upstream still means fewer chances for one to slip through.

The details taught us more than the headline.

Automation beat retraining. The cleanest gains came from changes that removed the need for someone to remember. Calibration reminder alerts took calibration lapses from 6 to 0, and digital temperature monitoring with alarms meant we recorded no out-of-range temperature events after the change. All technical staff also completed an 8-hour course on quality-control rules, with a competency test, but we treated training as a supplement, not the main defence.

The remaining errors pointed to the next fix. The three reagent incidents left after the changes were not temperature problems; they were physical handling problems.

A risk score is not an error count. Water quality scored as medium risk yet produced no incidents in either year. Its score reflected what could go wrong silently, which is exactly what a proactive risk assessment is meant to capture.

What peer review added

Five reviewers commented on the manuscript, and the paper is better for it. They asked us to report rates per 10,000 tests with confidence intervals, to define "medium-volume" in numbers and to justify the two-factor model. One reviewer spotted a dating error in our project timeline, which we corrected in the revision. They also pushed us to discuss honestly whether staff behaved differently simply because they knew they were being watched. We address that in the paper, but it remains a real limitation of a single-site before-and-after study, along with the modest absolute numbers.

Why it matters beyond our laboratory

Most published FMEA work comes from large hospital or academic laboratories. Yet much of the world's diagnostic testing is done in smaller laboratories without quality engineers or large budgets. Our experience suggests the tools are within reach: a cross-functional team, existing incident records, a simple scoring grid and a preference for engineered controls over reminders to be careful. A smaller laboratory could start with only its highest-scoring failure modes and expand from there.

The next steps are clear: replication across several laboratories, blinded review of reported errors, and a formal cost analysis, which our study did not include.

This was a team effort. My co-authors, Dr Kinjal Butani and Hetal Sakhiya, carried out the investigation; Dr Butani validated the results and Hetal Sakhiya curated the incident data. Our bench staff took part in the risk assessment itself. We would welcome hearing from colleagues who have tried similar approaches, especially in resource-constrained settings.

Read the paper: https://doi.org/10.1007/s12291-026-01449-x

Free view-only version: https://rdcu.be/fzeoK

AI use: This post used Claude (Anthropic) to help draft the text and check it against the published article. The authors reviewed every statement and take responsibility for the content.