Pushing the method detection limits to minimize bias in calculated disinfection byproduct toxicity due to left-censored data
Tanveer, H.U.; Adusei, K.B.; Furst, K.E.
In preparation for Environmental Science & Technology. Planned submission: early 2027
What the paper does
When water is disinfected with chlorine, the chlorine reacts with natural organic matter and forms hundreds of disinfection byproducts (DBPs). Regulations watch two families, the trihalomethanes (THMs) and the haloacetic acids (HAAs), because they are the most abundant. Abundance and harm are different things. Some byproducts that exist only in traces, the haloacetonitriles (HANs), haloacetaldehydes (HALs) and especially the haloacetamides (HAMs), are hundreds to thousands of times more toxic to cells than the regulated ones. Toxicity is concentration times potency, so a compound you can barely detect can carry more of the harm than one you can measure easily.
That creates a measurement problem. Every laboratory method has a method detection limit (MDL) below which it reports a compound as a non-detect (ND), and conventional methods for these byproducts report at around 0.1 to 0.3 micrograms per liter. The most toxic compounds are present at and below that level. I developed a method on a triple-quadrupole gas chromatograph mass spectrometer (GC-MS) with a high-efficiency ion source (Agilent 7010B). It quantifies about 60 volatile and semi-volatile byproducts with reporting limits near 0.001 micrograms per liter, about a hundred times lower than before, using the same 50 mL liquid-liquid extraction that laboratories already run. Then we measured real tap water from ten distribution systems in the mid-Atlantic and asked how much of the calculated toxicity a conventional method would have missed.
The answer, in these ten waters, is between 7 and 41 percent of the calculated toxicity, while missing only 1 to 8 percent of the mass. The regulated trihalomethanes made up 29 to 58 percent of the mass and under 1.5 percent of the toxicity. The haloacetamides, invisible to most methods, carried 21 to 74 percent of it. Which number you substitute for a non-detect, zero or half the limit, changes the answer again, in the other direction.
Mass and toxicity, ranked
The ten taps are identified by code. The slider starts at 0.2 µg/L, roughly where conventional methods report. Pull it down to 0.001 µg/L, where this method reports, to see which compounds appear. Censored values are treated as zero here. The paper also tests other conventions.
Why it matters
Calculated toxicity is increasingly what engineers and regulators use to compare treatment options, because running cell assays on every sample is impossible. If the inputs are censored at the wrong place, the comparison is biased towards whatever happens to be easy to measure, and the compounds doing the damage drop out of the sum. The main finding from my two years at George Mason is that improving a method can change which water looks safer. The paper proposes reporting practices that let any laboratory quantify and limit the bias whatever its detection limits.
What a detection limit does to an average
Show the numbers
| Estimate | Value (ng/L) | Error vs true |
|---|
Hollow markers are values the instrument would report as non-detects. The blue row is the true mean of all sixty values. The two amber rows are what a laboratory reports when it substitutes zero, or half the detection limit, for each non-detect. At a low detection limit all three agree.