Chloroform Epidemiology and Liver and Kidney Toxicity

July 2026

RoseyFields evaluates the human epidemiological evidence for chloroform as follows.

1. Overview of Study Populations and Data Sources

The human evidence base for chloroform is unusually broad for a single chemical, drawing on several cat egories of study populations that differ in exposure route, exposure intensity, and susceptibility to bias. Weighing these populations together, rather than relying on any single design, is central to how RoseyFields approaches causal inference for this chemical.

Occupational cohorts. The earliest systematic human data come from workers in chemical, pharmaceu tical, and manufacturing settings where chloroform vapor concentrations could be measured or recon structed. Bomski and colleagues followed chloroform-production workers in Poland with concurrent un exposed comparison groups (Bomski et al. 1967); Challen and colleagues studied a small cohort of long and short-term factory workers in England with historical and contemporary air measurements (Challen et al. 1958); and Li and colleagues conducted a Chinese factory cohort with both exposed and unexposed workers and a defined geometric-mean exposure concentration (Li et al. 1993). These cohorts offer relatively well-characterized, sustained inhalation exposure, but are small, often lack chloroform-specific exposure reconstruction, and involve co-exposure to other solvents.

Surgical/anesthesia exposure series. Because chloroform was used as a general anesthetic through the mid-twentieth century, case series of surgical patients provide a second, distinct population: individuals with well-documented, short-duration, very high-level inhalational exposure and directly measured arte rial blood chloroform concentrations (Smith et al. 1973; Whitaker and Jones 1965). These series are not designed as etiologic studies, but they provide some of the only quantitative human exposure-response data available for chloroform.

Community and contamination-site cohorts. A large share of the modern literature examines general pop ulations exposed to chloroform as a disinfection byproduct in chlorinated municipal drinking water, some times combined with inhalation and dermal uptake from bathing, showering, and swimming. These include large prospective and retrospective cohorts such as the Iowa Women’s Health Study (Doyle et al. 1997; Jones et al. 2019; Medgyesi et al. 2022), Massachusetts birth-outcome cohorts (Wright et al. 2004; Rivera Núñez and Wright 2013; Rivera-Núñez et al. 2018), and the Spanish and Italian multicase-control studies conducted under the MCC-Spain framework (Villanueva et al. 2017; Font-Ribera et al. 2018; Donat-Vargas et al. 2023, 2024). These studies typically assign exposure at the community or water-utility level rather than to individuals.

General-population biomonitoring studies. Since the 2000s, national biomonitoring surveys — principally the U.S. National Health and Nutrition Examination Survey (NHANES) — have measured chloroform di rectly in blood in nationally representative samples, allowing cross-sectional analyses of associations with renal function (Liu et al. 2023a), reproductive hormones (Wei et al. 2023), prostate-specific antigen (Wei et al. 2022), and cancer mortality (Min and Min 2016). Blood chloroform has a short biological half-life, so these measurements reflect only recent exposure, a significant limitation discussed further below.

Birth cohorts. Prospective pregnancy cohorts — including PELAGIE in France (Costet et al. 2011), INMA in Spain (Villanueva et al. 2011), a Kaunas, Lithuania cohort (Grazuleviciene et al. 2013), and multi-site U.S. cohorts (Savitz et al. 2006) — recruit women early in pregnancy and follow them prospectively, reducing recall bias for birth outcomes and permitting trimester-specific exposure windows to be examined.

Occupational cancer case-control studies. A distinct cluster of case-control studies uses job-exposure matri ces to estimate historical occupational chloroform exposure among cancer cases and population or hospital controls, covering kidney cancer (Purdue et al. 2017), glioma and meningioma (Neta et al. 2012; Ruder et al. 2013), non-Hodgkin lymphoma (Callahan et al. 2018), multiple myeloma (Gold et al. 2011), and multiple cancer sites simultaneously (Christensen et al. 2013).

Relying on this diversity of populations — rather than any single cohort or design — strengthens causal inference in several ways. Occupational and surgical series characterize the high end of the exposure range with reasonably direct measurement, while community and biomonitoring studies characterize effects (or their absence) at the low, environmentally relevant exposures that concern most plaintiffs. Findings that recur across geographically and demographically distinct populations (Poland, England, China, the United States, Canada, Spain, France) and across different exposure media (occupational air, drinking water, blood) are less easily explained by a single population’s idiosyncratic confounders, selection pressures, or measurement artifacts. Conversely, where an association appears in only one population or one study design, that pattern itself is informative: it suggests the finding may be attributable to chance, uncontrolled confound ing, or a study-specific artifact rather than a true, generalizable effect of chloroform. As described below, this is in fact the dominant pattern in the chloroform literature.

2. Human Health Outcomes Investigated, by Body System

Human studies of chloroform have examined outcomes across the hepatobiliary, neurological, renal, repro ductive, developmental, and carcinogenic domains. No human epidemiological outcome for chloroform meets the classic criteria for “strong, consistent, quantitatively modeled” evidence in the sense of a dose response relationship replicated across independent populations at environmentally relevant exposures — a fact that ATSDR (2024), EPA IRIS (2001), and IARC (1999) all acknowledge explicitly, each stating that the human epidemiological database is “inadequate” or of only “low” evidentiary weight for essentially every endpoint. The evidence is better described on a graded spectrum, and RoseyFields organizes it accordingly below.

Hepatobiliary Effects — Strongest Available Human Signal (Acute, High-Dose)

The liver is the body system for which human evidence is most internally consistent, but that consistency is confined to acute, high-concentration exposure from anesthesia and heavy occupational contact; it does not extend to chronic, low-level community exposure.

In case series of surgical patients anesthetized with chloroform, Smith et al. (1973) documented repro ducible elevations in bromosulfalein retention, serum total bilirubin, lactate dehydrogenase, and prothrom bin time following exposures reaching 20,000 ppm, with mean arterial blood concentrations of 9.8 mg/100 mL; transaminases and alkaline phosphatase were, notably, unchanged. Whitaker and Jones (1965), in a series of 1,502 anesthesia patients, reported jaundice in a small fraction of cases. In occupational cohorts, Bomski et al. (1967) found a dose-related excess of clinically diagnosed liver disease (enlarged liver, toxic hepatitis, fatty liver) and abnormal bromosulfalein retention among currently exposed Polish chloroform workers relative to unexposed controls, and Phoon et al. (1983) documented two independent outbreaks of toxic hepatitis with jaundice among Singaporean factory workers at exposures exceeding 400 ppm and 14– 50 ppm, respectively — an internally replicated finding within a single report. By contrast, lower-exposure occupational cohorts found no hepatic signal: Challen et al. (1958) observed no excess jaundice, enlarged liver, or abnormal liver-function tests among English factory workers, and Li et al. (1993) found no change in hepatomegaly or serum ALT among Chinese workers exposed at a geometric mean of only 4.19 ppm. Community-level exposure produces no detectable hepatic effect: competitive swimmers in chlorinated pools showed no change in ALT, AST, or GGT relative to unexposed athletes (Aiking et al. 1994).

Taken together, this is the one outcome for which chloroform shows a reasonably consistent, threshold-like pattern across independent populations and decades: clear hepatotoxicity at high acute or heavy occupa tional exposure, and no detectable effect at low occupational or general-population exposure levels. This pattern is also the one most directly corroborated by the extensive animal and mechanistic literature (see Section 5).

Neurological Effects — Suggestive, Limited to Occupational Exposure

Human neurological data are limited to two small occupational cohorts, both suggestive of an effect but neither adequate to establish a dose-response relationship at low exposure. Li et al. (1993) reported a dose related excess of self-reported dizziness, fatigue, somnolence, insomnia, impaired memory, anorexia, de pression, and anger among Chinese workers, together with objective deficits in simple visual reaction time, symbol-digit substitution, digit span, visual retention, and pursuit aiming in the higher-exposure group; the lower-exposure group showed only an isolated deficit in pursuit aiming. Challen et al. (1958) found sub jectively reported lassitude and drowsiness in both long- and short-term English workers, but decreased concentration, slowness, depression, and irritability only among long-term workers. Both studies relied on self-report or subjective symptom checklists, used small samples, and could not exclude co-exposure to other workplace solvents as a contributor.

Renal Effects — Null to Inadequate

Human renal data are sparse and largely null. Li et al. (1993) found no association between chloroform expo sure and serum BUN; Aiking et al. (1994) found no change in serum creatinine or urinary β2-microglobulin among chlorinated-pool swimmers. The one signal in the modern literature is a cross-sectional NHANES analysis reporting an inverse association between blood chloroform and both albumin-to-creatinine ratio and estimated glomerular filtration rate in 6,070 adults (Liu et al. 2023a) — a single cross-sectional find ing, unreplicated, and vulnerable to reverse causation given that impaired renal clearance could itself alter blood chloroform half-life.

Reproductive Effects — Inconsistent, Predominantly Null

A substantial body of literature has examined chloroform (usually as one of several trihalomethanes in tap water) against stillbirth, preterm birth, menstrual cycle characteristics, sperm parameters, and reproductive hormones, with no consistent pattern across studies.

For stillbirth, King et al. (2000) reported an increased risk — including specifically asphyxia-related stillbirth — at municipal water concentrations of 100 µg/L or more in a Canadian cohort of nearly 50,000 births, but a larger Massachusetts study of 2,460 stillbirths and 24,460 controls found no association (Rivera-Núñez et al. 2018). For preterm birth, the large majority of cohort and case-control studies found no association (Kramer et al. 1992; Costet et al. 2011; Villanueva et al. 2011; Wright et al. 2004; Savitz et al. 2006 for pregnancy loss generally), while a Chinese cohort found an increased risk in the first and third trimesters but a decreased risk in the second (Zhu et al. 2022) — a pattern more consistent with statistical noise than a biologically coherent exposure-response relationship.

Sperm-quality studies are similarly inconsistent. Chen et al. (2020) reported inverse associations between blood chloroform and sperm count, total motility, and progressive motility in 1,199 Chinese men, though most associations did not persist at three-month follow-up. Zeng et al. (2013, 2014) found no association with sperm concentration, count, or motility, but paradoxically reported improved sperm motion parameters (straight-line and curvilinear velocity) at higher exposure. Iszatt et al. (2013), in an English case-control study of impaired sperm quality, found no association with tap-water chloroform at all. Reproductive hor mone studies — NHANES data on serum estradiol, testosterone, and sex hormone-binding globulin (Wei et al. 2023) and a Chinese cross-sectional study of serum testosterone (Zeng et al. 2013) — found no as sociations. Menstrual cycle length and follicular/luteal phase length showed no relationship to tap-water chloroform in a Californian cohort (Windham et al. 2003).

Developmental Effects — Inconsistent, With One Modest Meta-Analytic Signal

Birth-defect studies are predominantly null: no associations were found between chloroform in tap water and cardiac, musculoskeletal, urogenital, neural tube, craniofacial, or limb-reduction defects in cohorts from Lithuania, Canada, and Massachusetts (Grazuleviciene et al. 2013; Dodds and King 2001; Kaufman et al. 2018), nor with hypospadias in a ten-state U.S. case-control study (Zaganjor et al. 2020). One exception is a six- to seven-fold increase in diaphragmatic hernia risk at higher tap-water concentrations (Kaufman et al. 2020) — a single, unreplicated finding for a rare outcome.

Birth-weight and small-for-gestational-age (SGA) findings are more heterogeneous but yield the closest thing to a quantitatively pooled developmental signal in the chloroform literature: a meta-analysis of eleven studies found a small but statistically significant increase in SGA risk with maternal drinking-water chlo roform exposure (odds ratio 1.05, 95% CI 1.01–1.08) (Summerhayes et al. 2021). Individual contributing studies are inconsistent in direction and magnitude — some retrospective cohorts found increased SGA risk or reduced birth weight at higher exposure (Summerhayes et al. 2012; Wright et al. 2004), one found increased birth weight at higher exposure (Rivera-Núñez and Wright 2013), and several prospective cohorts using individual-level intake estimates found no association at all (Villanueva et al. 2011; Costet et al. 2011; Bonou et al. 2017; Levallois et al. 2012; Hoffman et al. 2008; Hinckley et al. 2005; Porter et al. 2005). The meta-analytic point estimate is small enough, and the underlying study-level heterogeneity substan tial enough, that RoseyFields characterizes this as a suggestive but weak signal rather than an established effect.

Carcinogenicity — Suggestive at Individual Cancer Sites, but Inconsistent Across Studies and Not Confirmed for Chloroform Specifically

Cancer is the outcome most heavily studied in humans and most relevant to litigation, and also the outcome where the tension between individual positive findings and overall inconsistency is most pronounced.

Bladder cancer carries the longest history of a suspected association with chlorinated drinking water gen erally. A cluster of 1990s case-control studies (McGeehin et al. 1993; King and Marrett 1996; Freedman et al. 1997; Cantor et al. 1998) prompted EPA to calculate a population-attributable-risk range of 2–17% for chlorinated-water exposure and bladder cancer, and a later New York case-control study found increased bladder cancer risk in the highest quartile of chloroform exposure (Bove et al. 2007). However, other studies of similar design and comparable populations found no association (Salas et al. 2013), an ecological study of 76 water supplies found no association with bladder cancer mortality (Cantor et al. 1978), and a 2024 dose-response meta-analysis of three qualifying studies found no significant pooled association specifically for chloroform (relative risk 1.50, 95% CI 0.72–3.14) (Shi et al. 2024).

Colorectal cancer shows a similarly inconsistent, and in places inverse, pattern. Hildesheim et al. (1998) and the original Iowa Women’s Health Study (Doyle et al. 1997) reported increased colon cancer risk in the highest quartile of drinking-water chloroform, but the same cohort’s extended follow-up found no colon cancer association, only an isolated rectal cancer signal (Jones et al. 2019). Two Spanish/Italian case-control studies found decreased colorectal cancer risk with higher chloroform exposure (Villanueva et al. 2017, 2021), and a 2024 meta-analysis found no significant pooled association (relative risk 0.97, 95% CI 0.46– 1.91) (Shi et al. 2024).

Leukemia, the outcome examined in the most methodologically current review available to RoseyFields, illustrates the same pattern at the level of a formal meta-analysis. Soleimani et al. (2025) pooled four case control studies (Infante-Rivard et al. 2001; Heck et al. 2014, contributing two leukemia-subtype analyses; Donat-Vargas et al. 2024) and found no overall association between chloroform exposure and leukemia (pooled odds ratio 0.75, 95% CI 0.25–2.27), with very high between-study heterogeneity (I² = 95%). Sub group analysis found a significant increased risk for acute myeloid leukemia specifically (pooled OR 1.98, 95% CI 1.31–3.00, driven by Heck et al. 2014), a significant decreased risk for chronic lymphocytic leukemia (pooled OR 0.15, 95% CI 0.09–0.25, driven by Donat-Vargas et al. 2024), and no association for acute lym phoblastic leukemia (pooled OR 0.95, 95% CI 0.63–1.27, drawing on Infante-Rivard et al. 2001 and Heck et al. 2014). Because each subtype estimate rests on a single contributing study, these subtype-specific findings — including the AML signal — require independent replication before they can be considered es tablished. Consistent with this, other childhood-leukemia studies using ambient air exposure found no association with acute lymphoblastic or myeloid leukemia or neuroblastoma (Heck et al. 2013, 2014), ma ternal occupational exposure showed no association with childhood ALL (Infante-Rivard et al. 2005), while one Chinese case-control study found increased childhood acute leukemia risk with higher indoor-air chlo roform (Gao et al. 2014).

Other cancer sites are, on the whole, null. Occupational job-exposure-matrix studies found no association between chloroform and kidney cancer (Purdue et al. 2017), glioma or meningioma (Neta et al. 2012), non Hodgkin lymphoma (Callahan et al. 2018), or multiple myeloma (Gold et al. 2011); Ruder et al. (2013) found a decreased glioma risk in women and no association in men. A multi-site occupational case-control study found no association with bladder, prostate, colon, stomach, rectal, kidney, esophageal, liver, or melanoma cancers, and only a small, statistically fragile increase in pancreatic cancer based on two exposed cases (Christensen et al. 2013) — a finding not replicated in a separate pancreatic-cancer case-control study (Do et al. 2005) or in the ecological Cantor et al. (1978) analysis. Single positive findings exist for endometrial cancer, with a dose-related increase across drinking-water quartiles in the Iowa Women’s Health Study, most pronounced among hormone-replacement-therapy users (Medgyesi et al. 2022), breast cancer in a Spanish case-control study (Font-Ribera et al. 2018, not replicated in Doyle et al. 1997), and prostate cancer, which was associated with drinking-water concentration but not with calculated lifetime ingested dose in the same study population (Donat-Vargas et al. 2023) — an internal inconsistency that undercuts confidence in the finding. A cross-sectional NHANES analysis found no association between blood chloroform and total cancer mortality (Min and Min 2016), and the most recent dose-response meta-analysis found no significant association between chloroform in drinking water and all-cancer risk overall (relative risk 1.16, 95% CI 0.83– 1.62) (Shi et al. 2024).

3. Study Quality Considerations

Several recurring methodological issues temper confidence across the outcomes above, and RoseyFields weighs each finding against them individually rather than treating the literature as uniform.

Confounding by co-exposure. The great majority of the general-population cancer, reproductive, and developmental literature examines chlorinated drinking water as a whole rather than chloroform in isolation. Chlorinated water contains a mixture of trihalomethanes (bromodichloromethane, dibro mochloromethane, bromoform) and other disinfection byproducts, several of which have independently been associated with adverse reproductive and carcinogenic outcomes in epidemiological or animal studies. Very few of the studies summarized above adjust for co-occurring disinfection byproducts, meaning that even a robust, replicated association with “chloroform” in these designs cannot be attributed to chloroform per se. This is precisely the limitation regulators have identified in concluding that the drinking-water bladder-cancer literature, even if causal, could not establish chloroform-specific carcino genicity. Occupational cohorts carry an analogous problem: workers in chloroform-producing or -using facilities were frequently exposed to other solvents simultaneously (e.g., Wennborg et al. 2000, reporting spontaneous-abortion risk in laboratory workers without chloroform-specific exposure data).

Exposure misclassification. Most community-level studies assign exposure using water-utility monitoring data linked to residential address, rather than individual biomonitoring, and do not account for variation in individual water consumption, bathing habits, or use of bottled or filtered water. Several studies attempt to model total uptake (ingestion plus inhalation and dermal absorption from bathing and showering), but these models rest on assumptions drawn from small validation studies and are themselves a source of estimation error. Blood chloroform, used in NHANES-based studies, has a short elimination half-life, so a single measurement reflects exposure over the preceding hours to days, not cumulative or lifetime dose — a significant limitation for any study relating a spot blood level to a chronic disease like cancer or chronic kidney disease.

Reverse causation and temporality. Most cancer and reproductive case-control studies rely on retrospective self-report of residential history or water source, raising the possibility that recall differs between cases and controls. Cross-sectional biomonitoring analyses (e.g., blood chloroform and glomerular filtration rate, or blood chloroform and cancer mortality) cannot establish that exposure preceded the outcome, and are additionally vulnerable to reverse causation if disease processes alter chemical absorption, metabolism, or excretion.

Small numbers and multiplicity. Several of the more striking positive findings — pancreatic cancer in Christensen et al. (2013), diaphragmatic hernia in Kaufman et al. (2020), the AML- and CLL-specific pooled esti mates in Soleimani et al. (2025) — rest on very small numbers of exposed cases, and many studies examine numerous outcomes or subgroups without correction for multiple comparisons, increasing the likelihood that some statistically significant findings reflect chance.

Publication and study-selection bias. The formal meta-analysis available for leukemia found no evidence of publication bias by Begg’s or Egger’s tests (Soleimani et al. 2025), but the underlying dataset comprised only four studies, giving these tests limited power. The consistency of null findings across most other endpoints, and the presence of null and even protective associations in some of the largest and most rigorously con ducted cohorts (e.g., Jones et al. 2019; Villanueva et al. 2017, 2021), argues against wholesale publication bias suppressing null results in this literature.

4. Agency and Expert-Panel Weight-of-Evidence Conclusions

Multiple independent bodies have formally evaluated the human and animal evidence for chloroform and reached carcinogenicity classifications that are informative but, notably, are all driven predominantly by the animal and mechanistic database rather than by human epidemiology, which each body characterizes as inadequate to establish causation on its own.

The International Agency for Research on Cancer classified chloroform as Group 2B, “possibly carcinogenic to humans,” based on sufficient evidence in experimental animals and inadequate evidence in humans. EPA’s Integrated Risk Information System likewise concluded that chloroform is “likely to be carcinogenic to humans by all routes of exposure” but only “under high-exposure conditions that lead to cytotoxicity and regenerative hyperplasia in susceptible tissues,” and explicitly “not likely to be carcinogenic to humans” at exposures below that cytotoxic threshold — a nonlinear, threshold-based determination grounded in the mode-of-action data rather than in human epidemiology. The U.S. National Toxicology Program similarly lists chloroform as “reasonably anticipated to be a human carcinogen” based on sufficient evidence from animal studies. Separately, EPA’s Science Advisory Board reviewed the drinking-water epidemiology directly and concluded that the available human data are “insufficient to establish a causal relationship between exposure to chloroform in drinking water and increased risk of cancer,” applying standard causal-inference criteria — strength, consistency, specificity, temporality, dose-response, and biological plausibility — and finding them unmet.

RoseyFields reads these determinations as converging on a common structure: the animal toxicology and mechanistic data (chloroform’s metabolism via CYP2E1 to the reactive metabolite phosgene, producing cytotoxicity, regenerative hyperplasia, and eventually tumors in the liver and kidney at sufficiently high, sustained doses) are considered adequate to support a hazard classification, while the human epidemi ological evidence, considered on its own, is not adequate to establish that chloroform causes cancer or any other chronic disease in humans at real-world exposure levels. This is an important distinction for any downstream causation analysis: the “probable” or “possible” carcinogen classifications above are weight of-evidence judgments resting heavily on animal data, not epidemiological proof of human causation.

5. Use of Epidemiological Evidence in Litigation: General and Specific Causation

In toxic tort practice, epidemiological evidence and toxicological (animal and mechanistic) evidence serve complementary, not interchangeable, roles in supporting the two distinct causation opinions courts require.

General causation asks whether chloroform is capable of causing a given disease in humans at all, at expo sure levels comparable to those at issue. Where human epidemiological evidence is strong, consistent, and quantitatively modeled — as it is, within its acute, high-dose domain, for hepatotoxicity — it can indepen dently support general causation, particularly when reinforced by a coherent dose-response relationship and known mechanism. Where the epidemiology is inconsistent or null, as is true for the great majority of chloroform’s suggested cancer, reproductive, and developmental associations, general causation opinions necessarily rely more heavily on the animal toxicology, mode-of-action data, and structure-activity reasoning that underlie the IARC, EPA, and NTP classifications discussed above, applying Bradford Hill-type considerations (strength of association, consistency, biological gradient, plausibility, coherence, temporality) to the full evidentiary record rather than to epidemiology in isolation. Experts and courts alike recognize that regulatory hazard classifications such as “Group 2B” or “reasonably anticipated to be a human carcino gen” are themselves precautionary, hazard-identification judgments and are not equivalent to, and do not substitute for, a general-causation finding under the applicable legal standard.

Specific causation asks whether chloroform in fact caused a particular plaintiff’s disease. Even where gen eral causation is supported, specific causation requires an individualized analysis: reconstructing the plain tiff’s actual dose and duration of exposure (occupational history, water source and consumption patterns, or biomonitoring data) and comparing it against the exposure levels at which human or animal evidence demonstrates an effect; considering background disease incidence and the plaintiff’s other risk factors and alternative etiologies (family history, smoking, other occupational or environmental exposures, unre lated medical conditions); and, for outcomes like chloroform-associated hepatotoxicity where a mechanistic threshold (cytotoxicity followed by regenerative hyperplasia) has been characterized, assessing whether the plaintiff’s exposure plausibly exceeded that threshold. Because chloroform’s human evidence base is heavily populated with null, inconsistent, or single-study findings outside the hepatotoxicity domain, specific causation opinions for outcomes such as leukemia, colorectal cancer, bladder cancer, or adverse birth outcomes require particular care: an expert relying on a single positive study (for example, the AML-specific or pancreatic-cancer findings discussed above) should be prepared to address the surrounding literature’s inconsistency, the small numbers underlying that specific estimate, and the co-exposure and exposure misclassification limitations described above, all of which bear directly on the reliability and weight the finding can support.

References