When lab results can be compared
How CHAI handles results from different labs and methods, and determines which results can support a trend.
The problem
A patient on a protocol draws at a primary lab every quarter and at a direct-to-consumer lab every month. Estradiol comes back by LC/MS from one and by immunoassay from the other. Total testosterone comes back as a number from one and as ">1500" from the other. Put them on one line and you get a trend that never happened.
Decision grade
Every value CHAI stores carries its lab, its assay method and a decision grade. For the markers where method decides the answer (estradiol, total testosterone, free testosterone), a value is confirmed when the method meets the grade (LC/MS or equilibrium dialysis) and a screen otherwise. A screen can raise a question. It never answers one.
The current value of a marker is the newest confirmed point if one exists within 45 days of the newest point from any source; otherwise the newest point, graded screen and labelled as such.
Trend within one basis
A trend is computed on one series only: the confirmed series when it has at least two points, otherwise the series from the current point's source. The other lab's points are drawn on the chart, marked with a different symbol, and never fitted. The marker sheet says so in words: "trend within Rythm Health only (4 points); 3 other-lab points shown, not fitted."
Compare results collected close together
Two results are treated as a pair only when they are within three days of each other. A fortnight apart, a disagreement may simply be a real change.
Within a pair, each marker has an analytic tolerance: 25% for estradiol, 15% for total testosterone, 3 points absolute for hematocrit, 0.3 absolute for HbA1c, and so on. Beyond tolerance the pair is discordant, not noise. If one side is confirmed, it wins. If neither is, nobody acts until a confirmed value lands.
Establishing agreement for each marker
A direct-to-consumer panel becomes decision-grade for a marker one marker at a time. Three consecutive same-blood pairs within tolerance and that source is trusted for that marker. From three pairs on, a median ratio can map the screen source onto the confirmed scale for display and trend, marked as adjusted, and never used for dosing.
The requisition engine schedules these calibration draws: venipuncture pairs for the markers the monthly panel has not yet earned.
Results beyond the assay’s reporting limits
A result reported as ">1500 ng/dL" or "<12 pg/mL" is stored as censored: above or below the assay's reportable limit. It is excluded from statistics and from the trend, and every surface says so rather than pretending the value is 1500. A rising free testosterone that ends in ">518" is described as beyond the assay limit, with a dilution or LC/MS run suggested before any dose decision.
Where it shows up
- The marker sheet: value, source, method, decision grade, trend basis.
- Ask CHAI: statistics computed only from usable points, with censored values named.
- The cycle review: censored labs called censored, not "not in this panel".
- The requisition: which markers still need a confirmed draw, and why.