Flow Cytometry QC: Tracking Instrument Performance Across Runs

The BD LSRFortessa in core room 3 passed its CS&T report every morning for two weeks. Then a batch of 40 samples came back with the CD4/CD8 ratio shifted by 18% compared to the previous run. No alarm fired. The CS&T report was green. What changed was the detector voltage on the PE channel—a 32V drift that fell inside the ±50V tolerance window but was enough to shift the compensation matrix and pull the CD8-FITC gate left. Without longitudinal tracking, that drift was invisible.

Passing a daily QC check is not the same as tracking flow cytometry QC metrics across instrument runs. The first is a binary gate; the second is a trend. This post covers which metrics to pull, how to build a Levy-Jennings chart that actually catches drift, and what to do when a run passes QC but the numbers look wrong.

Which Metrics to Track

Not every value in the CS&T or equivalent QC bead report is worth charting. These four are the signal—the rest is mostly noise for routine monitoring purposes.

Peak MFI (bead populations). The median fluorescence intensity of each bead peak, per channel. This is your primary instrument stability readout. A stable instrument holds its bead MFI within ±5% of the monthly mean across channels. Drift beyond that needs an explanation before you run samples. On a Cytek Aurora, track the per-detector intensity of each spectral peak from the 8-peak Rainbow bead; on a conventional instrument, track the CS&T or equivalent per-channel bead position.

%rCV (resolution, aka peak CV). The percent coefficient of variation of the bead peak distribution in each channel. For single-peak calibration beads, %rCV under 3% is ideal; 3–6% is acceptable; above 6% should trigger investigation before running a biological experiment. A rising %rCV—even if still under 6%—can precede detector or laser failure by days to weeks if you’re tracking it.

Detector efficiency / optical background. BD CS&T reports this explicitly. It reflects the ratio of signal to instrument noise at a standardized input. Gradual optical background increases usually point to laser power decay or contaminated optics. Sudden jumps often mean a filter shift or fiber coupling problem.

PMT voltage (or gain setting). The voltage that the QC bead protocol sets to reach its target MFI. Track this number, not just whether the report passed. On BD instruments, CS&T will adjust PMT voltage each run to compensate for signal drift—that voltage history tells you how much compensation is accumulating. A ±50V window is the action threshold; continuous drift of 20–30V in one direction over two weeks means your detector or laser is degrading, even if you’re not at the threshold yet.

Setting Up a Levy-Jennings Chart

A Levy-Jennings chart is a simple run-order plot with control limits drawn at ±2 SD (warning) and ±3 SD (action). Build one for each tracked channel. The calculation requires a baseline: run QC beads on 20 consecutive sessions without instrument changes, then compute the mean and SD for each metric. Those become your center line and control limits.

In practice:

1. Export QC bead results to CSV at the end of each session. Most instruments let you do this automatically; BD FACSDiva writes a CS&T report you can parse.
2. Maintain a running spreadsheet (or use Cytomaton’s batch statistics export) with date, operator, and per-channel values. This is the same spreadsheet you’d use for exporting flow cytometry statistics for publication, just applied to bead data instead of sample data.
3. Plot each channel’s MFI and %rCV as a time series. Add horizontal reference lines at mean ±2 SD and mean ±3 SD.
4. Recalculate baseline after any scheduled maintenance (laser replacement, detector service, full alignment). Using pre-maintenance limits post-maintenance will trigger false alarms or, worse, miss real drift as the instrument settles to a new equilibrium.

The 20-run baseline is a minimum. Sixty runs gives tighter limits and catches seasonal or lot-to-lot bead variation that shorter baselines can’t distinguish from instrument drift.

Westgard Rules for Flow: Which Ones Actually Matter

Clinical chemistry labs run full Westgard rule sets. For flow cytometry instrument QC, three rules cover most actionable failures:

1_3S: One observation outside ±3 SD. Action: do not run samples. Re-run QC after cycling instrument; if still outside limits, escalate to facility manager.

2_2S: Two consecutive observations outside ±2 SD in the same direction. Warning: this is early drift, not a single-point outlier. Investigate laser power, check for filter contamination, review recent maintenance log. Samples can run with notation if the metric is between 2S and 3S, but document the warning.

10_X: Ten consecutive observations on the same side of the mean. This is the most commonly missed rule. The individual values may all be within ±2 SD, so nothing flags—but the consistent direction indicates systematic drift. In flow cytometry, this pattern often precedes a 1_3S event by one to three weeks.

Connecting QC Metrics to Gating Outcomes

Instrument drift does not affect all assays equally. A 15V PMT voltage shift on the PE channel matters more for a panel using PE-conjugated antibodies near their expression threshold than for one using PE-Cy7 on a bright population. When evaluating a QC trend, ask: which channels are drifting, and what markers are on those channels in the biological assays running on this instrument?

If you’re running templated analysis across multiple operators or sessions, the same channel drift that shifts bead MFI will shift your biological populations. That interaction is covered in detail in reducing gating variability between operators—the short version is that instrument drift and operator variation compound, and you can’t separate them without baseline bead data.

For batch experiments spanning multiple instrument sessions, flag any session where a QC metric crossed a warning threshold and re-examine the sample data from that session specifically. A batch processing workflow should include a column for QC status in the sample manifest so out-of-tolerance sessions are traceable after the fact.

When the Report Says Pass but Something Looks Wrong

CS&T and equivalent automated QC tools set pass thresholds conservatively to minimize false positives. They will not catch gradual drift that stays inside limits or systematic bias that developed slowly over months. Three situations where the report passes but you should dig deeper:

Population shift without QC alarm. Your biological CD4 gate is consistently landing in the same place, but MFI is trending down week over week. Check PMT voltage history: if voltage has been rising by 5–10V per session to compensate for signal loss, your detector is degrading. The pass threshold is fixed; the voltage correction is the tell.

CV creeping toward the 6% ceiling. A %rCV of 5.8% on the APC channel that was 3.1% six months ago has not yet tripped the action threshold, but it will. This is the longitudinal chart catching what the daily threshold misses.

Lot-to-lot bead variation. New bead lot, same instrument, same settings—but the MFI values jump 12%. That’s not instrument drift; it’s between-lot variation. Establish a new baseline or use a correction factor by running old and new lots in parallel for at least 5 sessions before switching fully.

Practical Implementation: Minimum Viable QC Log

If your facility doesn’t have a QC tracking system and you need to start one today, the minimum viable version is a spreadsheet with these columns: date, operator, instrument ID, and per-channel bead MFI + %rCV. Export from the instrument after each session, paste into the sheet, and review the last 10 rows before releasing results.

That’s not a Levy-Jennings chart with statistically derived limits—but it’s enough to catch the most common failure modes: a sudden MFI drop (detector or laser problem), a rapid %rCV increase (dirty optics or alignment problem), and PMT voltage creep (gradual detector aging). Start there and add statistical limits after you have 20 sessions of baseline data.

The value of tracking is proportional to the frequency of the experiments that depend on the instrument. A core facility running 200 samples per week needs tighter tracking than a lab running one panel per month. Calibrate your QC effort to your assay throughput—but do not skip it entirely, because by the time a problem is visible in your biological data, the instrument has usually been drifting for longer than you think.