Study design and population

CLL cases were recruited within the multicentric multicase-control (MCC) study in Spain (http://www.mccspain.org) in collaboration with the International Cancer Genome Consortium on Chronic Lymphocytic Leukemia Project (ICGC-CLL, www.cllgenome.es and www.icgc.org). Briefly, cases and controls were recruited between January 2010 and January 2013. CLL cases were identified in 11 hospitals from different Spanish regions, together with a set of frequency-matched controls by sex, age, and region. Controls were selected from the general population, identified from the lists of randomly selected family practitioners in primary health centers located within the catchment area of the hospitals recruiting CLL [13]. Response rates were 87% for cases and 53% for controls. Inclusion criteria required participants to be 20–85 years old, be able to understand and answer the questionnaire, and have lived for at least 6 months in the study area.

Data collection

At recruitment, participants answered a structured computer-assisted questionnaire administered by trained personnel in a face-to-face interview to gather information on anthropometrics (self-reported), socio-demographics, lifestyle factors, and personal and family medical history. Cases and controls provided full address, year start and stop living in all the residences where they lived for at least 12 months since age 18 until the time of the interview, and the type of water consumed in each residence (municipal, bottled, well, other). Amount (glasses/day) of water ingested on average lifetime at home, work and other places was ascertained. Information related to swimming pool attendance included ever attendance (defined as more than 10 times in the lifetime), year (or age) when swimming pool attendance started and ended, average attendance frequency in summer and the rest of the year separately, average time in the water, and type of swimming pool (indoor, outdoor, both). Dietary habits the year before the interview were collected through a self-administered semiquantitative food frequency questionnaire, including a total of 140 food items, previously validated in Spain [14]. Questionnaires used are available online (http://mccspain.org). A final section evaluating the reliability of the interview was completed by the interviewer.

Outcome definition

CLL cases were diagnosed according to the criteria of the International Workshop on Chronic Lymphocytic Leukemia [15]. CLL and small lymphocytic lymphoma are considered the same underlying disease. All diagnoses were morphologically and immunologically confirmed using flow cytometry immunophenotype and complete blood count. In the vast majority of CLL cases, we could differentiate between rai stage 0 or I–IV tumors. Given the generally indolent course of CLL, with most of the patients having a slow progression disease and with no treatment need, CLL cases were invited to participate independently of their date of diagnosis. Patients who received a CLL diagnosis within three years prior to the start of the study (January 2010) were eligible for inclusion. Incident cases were defined as patients diagnosed at during the study recruitment period (January 2010 – January 2013), while prevalent cases were defined as participants diagnosed ≤ 3 years prior to the start of the study (January 2010).

THM and nitrate levels in municipal drinking water

We designed a structured questionnaire aimed at water utilities, local and/or health authorities to collect drinking water source (surface/ground water proportion) and treatment in the study areas back to 1940, as well as available THMs and nitrate measurements. We targeted data collection among study municipalities that contributed up to 80% of person-years. In addition, centralized routine monitoring data on THMs (chloroform, bromodichloromethane, dibromochloromethane and bromoform) and nitrate was provided by the SINAC (Spanish acronym of the National Information System on Water for Consumption) for the years 2004–2010. This data corresponded to samples collected at drinking water treatment plants and the distribution network, and included information at the water zone level fed by water supply operators from public or private companies or municipalities, and public or private laboratories. The water zone, that mostly corresponds to municipality, was defined as a geographical area supplied by water with a homogeneous source and treatment, and whose quality in the water distributed in the networks can be considered homogeneous. We linked each postal code from the residence to the corresponding water zone. The distribution of the sampling points and the sampling frequency varied greatly depending on the population served, extension of the water zone and the year, and could be more than once a day (e.g., Madrid), up to once every three months, or once a year in less populated areas. Measurements below the analytical limit of quantification (QL) were substituted with half the QL (QL/2) [16]. If the QL was missing, we imputed half of the most frequently reported QL.

THM and nitrate levels in non-municipal drinking water

We measured nitrate in the 9 most-consumed bottled water brands in Spain using UV spectrophotometry, with 0.5/0.1 mg/L detection/quantification limit. Nitrate concentrations were in the range of 2.3–15.6 mg/l [17]. THMs were previously measured in 15 popular bottled water brands in Spain, through purge-and-trap and gas chromatography. Mean concentrations for chloroform and brominated THMs were ≤0.1 μg/L [18]; and limits of detection were 0.015 (chloroform), 0.004 (bromodichloromethane), 0.005 (dibromochloromethane) and 0.011 μg/L (bromoform). We used THM data from 56 measurements in different Spanish areas that were supplied by chlorinated ground water (e.g. wells). Average concentrations were 0.3, 0.3, 0.8, and 1.8 μg/L for chloroform, bromodichloromethane, dibromochloromethane, and bromoform, respectively. Nitrate data in private wells was not available.

Estimation of long-term levels in municipal drinking water

We calculated the annual average level of nitrate and THMs at the water zone level. Years without measurements were assigned the average of all available measurements in the water zone if the water source and treatment did not change over the years. In the case of changes in the water source and/or treatment, procedures to back extrapolate were applied.

For THMs, since their concentrations in surface water are generally higher than in ground sources [19], we used surface water percentage as a weight to back extrapolate individual THM concentrations when water source changed through linear interpolation, assuming that concentrations increased proportionately to the percentage of surface water. Likewise, water zones with changes in treatment over the years and THM measurements were used to estimate the change percentage of THM concentrations after introducing such treatments. These percentages were applied as a weight to back-extrapolate THM concentrations in areas with changes in these specific treatments when measurements were unavailable. Before chlorination started, THMs concentrations were assumed to be zero. Total THMs (TTHM) levels were calculated by adding up chloroform, bromodichloromethane, dibromochloromethane, and bromoform concentrations.

For years without nitrate measurements in water zones where water source changed over the years, the ground water percentage was used as a weight to back-extrapolate concentrations using linear interpolation, assuming that nitrate levels increased with ground water proportion [9]. In municipalities without any nitrate measurement (covering ~0.5% of the total person-years), we imputed the levels of neighboring municipalities supplied with similar ground water proportion ± 10%.

Individual indices of THM and nitrate exposure in the study populationAverage THMs and nitrate levels in residential tap water

We used municipality and year to link levels in drinking water with residential histories of study participants from age 18 years to 2 years before the interview. We estimated the average concentration of nitrate (mg/L) and THMs (µg/L) for this period, henceforth referred to as “lifetime” or “long-term exposure”. In general, participants were assigned to different water zones based on their residence patterns. On average, individuals lived in three different residences during the exposure window period, with the residence at the time of the interview being the longest, spanning approximately 30 years.

Average ingested nitrate

To calculate waterborne ingested nitrate (mg/day), we assigned levels in drinking water by year according to the water type consumed at home, including municipal (tap), bottled, and private well/other water. Nitrate levels in municipal water were assigned for tap water consumption. Nitrate levels in the sampled bottled waters (range 2.3–15.6 mg/l) [17] were averaged using the sales frequency of each brand as a weight, leading to 6.1 mg/L of nitrate, that was assigned to study participants consuming bottled water. Since nitrate levels in well water were not available, waterborne ingested nitrate was considered missing for years when well water consumption was reported (~2%). The annual nitrate estimates were averaged from age 18 to 2 years before the interview and multiplied by the average daily water intake at the residence. Total amount of ingested water was ascertained as the number of water glasses per day consumed on average by the participant at home (L/day, assuming 200 mL/glass). Water intakes = 0 and above the 99th percentile (4 L/day), considered implausible, were treated as missing values in the analyses.

Covariables

We collected information on self-reported age, education, weight and height 1 year before the interview to compute body mass index (BMI, kg/m2), family history of hematologic malignancy, smoking and physical activity. Smokers were defined as those smoking at least one cigarette/day for ≥6 months. Former smokers were defined as those who quit smoking ≥1 year before the interview. Physical activity was ascertained through open questions on any type of physical activity practiced in life, years, and frequency (hours/ week), to calculate metabolic equivalents (METs) from age 16 to 2 years before the interview.

Statistical Analysis

The initial sample consisted of 1,842 participants, comprising 246 CLL cases (92 incident, 154 prevalent) and 1,596 controls. Number of controls was greater than the number of cases because controls were selected to also be matched to other cancer sites. We excluded 1 subject with unreliable interview as qualified by the trained interviewers, and, in order to have a similar geographical distribution of cases and controls, only municipalities with at least one case and one control were included (n = 229 excluded). For the residential tap water analyses, participants with THMs or nitrate estimates covering less than 70% of the years between age 18 to 2 years before the interview were also excluded (n = 238), resulting in a sample of 1374 (144 cases, 1230 controls). For the nitrate waterborne ingestion analyses the sample was made up of 1245 (131 cases, 1114 controls) due to missing data on water consumption in 129 participants. Finally, out of 1612 there were 372 participants without complete data on swimming pool attendance and the sample for these analyses it was made up of 1240 (157 cases, 1803 controls) (Fig. 1).

Fig. 1: Flow chart showing exclusions of study participants from the Multicase-Control Study in Spain (MCC-Spain).

Cases were defined as patients with CLL recruited within three years from diagnosis to interview (diagnosed ≤ 3 y prior to recruitment). * From 18 years of age to 2 years before the interview.

As exposure to nitrate in drinking water is exclusively through the oral route, waterborne nitrate ingestion was evaluated as nitrate exposure. However, since inhalation and dermal absorption are suggested as relevant routes for THMs (volatile and permeable to the skin), residential tap water THM levels were used as main exposure. As a supplementary analysis, waterborne THM ingestion was additionally explored. Average lifetime residential tap water of total THMs, brominated THMs, chloroform (μg/L), and waterborne nitrate ingestion (mg/day), were treated as continuous and as tertiles defined according to the distribution among controls.

Our study also encompassed an evaluation of swimming pool attendance. Firstly, we conducted a comparative analysis between individuals classified as ‘users’ (defined as those who have attended the pool ≥10 times in their lifetime) and ‘non-users’ (defined as those who have attended the pool <10 times in their lifetime). Subsequently, we performed a secondary analysis by categorizing participants based on the average annual frequency of pool attendance. Given that over half of the participants fell into the non-user category, the reference category was designated as the group of ‘non-users, while users were further divided based on their attendance frequency, both below and above the median of the yearly average times of attendance. To achieve a finer-grained measurement, we further evaluated each successive 25th percentile increment of yearly pool attendance frequency, distinguishing between summer and rest of the year.

We estimated odds ratios (OR) and 95% confidence intervals (CI) of CLL using mixed models with recruitment area (Barcelona, Asturias, Cantabria) as random effect. To test for linear trends (P-trend) across increasing categories of exposure, the median concentration within each category was treated as a continuous variable in the model. Smoothed spline with three degrees of freedom from general additive models (GAM) were used to visually display the exposure–response relationships on continuous variables.

The selection of adjustment variables was made based on prior knowledge about CLL risk or preventive factors and variables potentially correlated with exposures of interest. Thus, all models were adjusted for recruitment area (introduced as random effect in the mixed model), age (<55, 55–64, 65–69, 70–74, ≥75 y) sex, and education (primary school or lower, secondary school, university). Further adjustment included family history of hematologic malignancy (yes/no), BMI (kg/m2), smoking (never, former, current), physical activity (inactive, low, moderate, very active), ever worked in farming or agriculture (yes/no) and alcohol consumption (never, former, current moderate consumption [(≤20 g/day men; ≤10 g/day women)], current high consumption [>20 g/day men; >10 g/day women]). An additional model was reported with mutual adjustments between nitrate waterborne ingestion, residential tap water chloroform and brominated THMs levels. Concerning the analysis of swimming pool attendance, the second model was not adjusted for physical activity, while the subsequent third model did include adjustments for physical activity. We used stochastic regression (which adds a random error term that appropriately reproduces the correlation between X and Y) to impute missing values in BMI (4%), physical activity (<1%) and alcohol consumption (18%). Treating the missing of the alcohol variable as a separate category had no impact on any estimate.

We explored the stratification by sex and by rai stage (0 and I–IV) and some sensitivity analyzes were further conducted: (i) excluding participants with family history of hematologic malignancy, to completely avoid any genetic influence and (ii) only including incident cases (i.e., CLL diagnosed after recruitment), to minimize the possibility of reverse causation. For this supplementary analysis, exposures were exclusively investigated in their continuous form, as this approach was necessitated by a considerable reduction in the sample size. All p values presented are two-tailed; <0.05 was considered statistically significant. Analyses were performed using STATA version 16.0 (Stata Corp, College Station, TX).