Nucleus of the disinfection-byproducts cluster and the wiki’s opening on drinking-water chemical
contaminants as a modifiable exposure. Chlorinating water to kill pathogens generates by-products;
trihalomethanes (THMs) are the regulated marker of the mixture. The question is whether lifetime THM
exposure raises bladder-cancer risk, and — since the answer is a small relative risk — whether it is
a lever anyone can pull. Two sources land it: a gold pooled/meta-analysis of the per-exposure
risk (Costet 2011) and a population-burden translation (Evlampidou 2020) that borrows that
risk estimate. The Layer-1 verdict is upfront: a small per-person effect on near-universal low-level
exposure — population-material, individually negligible, and mostly a water-treatment lever, not a
behaviour.
Unit note: both source PDFs’ OCR dropped the micro sign, rendering µg/L as “mg/l” (Costet) and “lg=L” (Evlampidou). The correct unit is µg/L throughout — confirmed by the EU/WHO THM regulatory limit of 100 µg/L. Verbatim «…» spans below preserve the OCR artifact; prose uses µg/L.
The risk: a modest male dose-response, null in women, null via ingestion (Costet 2011)
Costet pooled three European case-control studies (France, Finland, Spain; 5,467 subjects) and, in a two-stage meta-analysis, combined them with the North American studies (global base 4,351 cases / 7,055 controls). Exposure is an individual 40-year residential average TTHM level, reconstructed from residential history x modelled/measured tap-water THM x personal consumption (Costet et al., 2011).
- Men, top vs bottom band (meta-analysis): «A significant odds-ratio was observed for men exposed to an average residential TTHM level > 50 mg/l (OR¼1.47 (1.05; 2.05)) when compared to men exposed to levels # 5 mg/l.» (Costet et al., 2011) (read “mg/l” as µg/L). The total international meta-analysis gives OR 1.51 (1.26-1.82) at >50 µg/L and 1.31 (1.16-1.49) at >5 µg/L vs <=5; linear trend p=0.01 (meta) (Costet et al., 2011).
- Women: no association. «No significant association was found among women or with cumulative exposure through ingestion.» (Costet et al., 2011) The women’s ORs are inconsistent and non-significant across bands (e.g. >50 µg/L OR 0.70, 0.43-1.14).
- Studied range bounds the curve. Among the ~75% who were exposed, the median residential TTHM was 30 µg/L (max 145 µg/L); ~25% were unexposed (Costet et al., 2011). So the >50 µg/L “band” sits in the upper-middle of the observed data, not at an extrapolated edge, but there is no evidence above ~145 µg/L. No knee is located — the display is a categorical band-contrast, weak evidence of curve shape (-> Measurement Error in Dietary Assessment).
The route matters and cuts against the obvious lever. Cumulative ingested THM showed no association (trend p=0.18 men) — yet the male signal tracked residential level, and in the Spanish data «Significant increasing risks of bladder cancer among men with bathing, showering and swimming-pool attendance were found.» (Costet et al., 2011) Costet reads the level as an «indicator of global exposure, regardless of the route (ingestion, dermal absorption or inhalation)» (Costet et al., 2011). So dermal/inhalation exposure (showering, bathing, pools), not drinking, carries the male signal — a point-of-use drinking-water filter would target the arm that was null. Caveat: the ingestion null may partly be measurement error (daily intake was held constant over the 40-year window), so the route contrast is directional, not settled.
The burden: 4.9% of EU bladder cancers, borrowing Costet’s risk (Evlampidou 2020)
Evlampidou translated risk into population burden across EU28. It did not estimate an independent effect: it re-derived a continuous OR = 1.004 (95% CI 1.002, 1.006) per 1-µg/L THM increment «derived using pooled data for men and women age 30-80 from Costet et al. 2011» (Evlampidou et al., 2020), then applied it to current national THM means x incidence x population.
- Exposure: population-weighted mean THM 11.7 µg/L (median 10, IQR 3.1-24.2), range 0 to 301 µg/L (Portugal), covering 75% of the EU26 population (Evlampidou et al., 2020) — i.e. current EU means sit well below the 100 µg/L regulatory limit and near the bottom of Costet’s studied range.
- Burden: PAF 4.9% (95% CI 2.5, 7.1), accounting for 6,561 (3,389, 9,537) bladder-cancer cases per year (Evlampidou et al., 2020). Country PAFs range 0-23% (highest Cyprus, Malta, Ireland; ~0% Denmark, Netherlands).
- The avoidable slice: bringing the 13 above-mean countries down to the EU mean (11.7 µg/L) cuts attributable cases by 2,868/year (95% CI 1,522, 4,060), a 43.7% reduction to 3,693 (Evlampidou et al., 2020). So more than half the burden is the ubiquitous background that no attainable treatment removes; the actionable slice is the above-average tail (largest absolute reductions: Romania 891, Spain 860, UK 685 — dominated by a few high-THM systems).
Synthesis
The two sources are a risk-to-burden decomposition, not independent corroboration
Before any cross-source claim, the parameter table (op-weave 2a) — every row’s “same quantity?” is NO, and the two share authors, so this is type-F decomposition + G-flavor (needs aggregation), NOT independent-E:
| Parameter | Costet 2011 | Evlampidou 2020 | Same quantity? |
|---|---|---|---|
| Effect metric | categorical OR by band, men: >50 vs <=5 µg/L = 1.47 (1.05-2.05) | continuous 1.004 (1.002-1.006) per 1 µg/L, both sexes | NO — band-contrast (men) vs per-unit slope (pooled sexes) |
| Underlying data | 3 EU + 3 N.Am/Can case-control studies pooled | re-pools the same «largest international pooled analysis» = Costet’s data | NO independent data — the ER function IS Costet’s |
| Output quantity | relative risk per exposure level | PAF + attributable cases (RR x incidence x population) | NO — relative risk vs population burden |
| Exposure construct | individual 40-y residential THM history | country population-weighted current mean | NO — individual lifetime vs ecological present |
Independence is defeated at the cheapest test: Costet, Villanueva, Jaakkola and Kogevinas are authors on both papers, and Evlampidou explicitly builds its exposure-response on Costet’s pooled data. Evlampidou is therefore a burden translation of Costet, not a second causal witness — its 6,561 cases are only as sound as the borrowed OR and the assumed causality . Evlampidou states this itself: «An underlying assumption of this research study is the causal relationship between THMs and bladder cancer.» (Evlampidou et al., 2020)
Layer-1 sizing — a small environmental lever, dwarfed by the big rock in the same table
The emergent move is the sizing, visible only across the pair. In Costet’s own Table 1 the dominant bladder-cancer exposure is smoking: current-smoker OR 3.91 (3.32-4.60), former 2.31 (Costet et al., 2011) — against which THM’s OR ~1.3-1.5 at high exposure (1.004 per µg/L (Evlampidou et al., 2020), Evlampidou’s continuous estimate above) is a small effect. For an individual, THM is far down the Layer-1 ranking, below smoking and below the big lifestyle rocks. The exposure is also near-universal and low-variance (EU mean 11.7 µg/L, mostly under the 100 µg/L limit), so:
- The lever is structural, not behavioural. The material burden lives at the population level (water-treatment optimization, keeping high-THM systems down toward the mean), where Evlampidou puts it. Individual behaviour barely moves it, and the one plausible individual act (filtering drinking water) targets the ingestion route that was null in Costet — the male signal ran through dermal/inhalation exposure (showering, bathing, pools).
- The ceiling is the finding. For a reasonably-healthy non-smoker on regulated municipal water at ~EU-mean THM, this is a small, uncertain, mostly-unactionable lever — reportable as small, not a gap to keep digging.
Certainty and the binding constraint
Causation is not established: IARC/WHO judged the epidemiological evidence insufficient for a
causal relation, and THM is a surrogate for a >600-compound DBP mixture — «the three studies …
used TTHM levels as a proxy for the global DBP level» (Costet et al., 2011), with the
same TTHM level carrying different underlying mixtures across countries. Retrospective exposure over a
40-year window with a 10-40-year latency is measured with large error; as with dietary intake, a
modest or attenuated dose-response is weak evidence of no gradient, and the categorical displays
here cannot show a knee (-> Measurement Error in Dietary Assessment, Surrogate Outcomes).
confidence: low reflects the gold design of the pooled risk estimate held against unresolved
causality, surrogate exposure, and low individual decision-relevance.
Open threads / gaps
- G (needs aggregation): a pooled SFA-style effect on bladder-cancer incidence that folds in the US studies omitted here (Beane Freeman 2017, 1,213 cases — which Evlampidou notes it did not include, and which reported higher risk in women, against the male-only European pattern) (Evlampidou et al., 2020).
- The route question (ingestion vs dermal/inhalation) is decision-relevant and only one study (Spanish) could address it — a gap, since it determines whether any individual lever exists.