Editor,— We read with interest the article by Law et al.1 This ecological study examined the incidence of non-Hodgkin's lymphoma (NHL) in Yorkshire and North Humberside in 1984–1993 in relation to nitrate concentrations in drinking water. Nitrate exposure was estimated for 148 water supply zones by calculating a mean of monthly means based on six years of monitoring data (1990–1995) that largely post-dated the time period of NHL incidence. The results showed a relation between NHL incidence during 1984–1989 and nitrate concentrations in the early 1990s but not for incidence during 1990–1993.

The authors compared the distribution of their recent nitrate average with the distribution of long term average nitrate values from our population-based case-control study of NHL in the state of Nebraska, USA.2 They also described the long term average nitrate metric we calculated as “an imprecise estimate of nitrate exposure assessed from single annual measurements for a city or town”. The characterisation of our historical nitrate estimates as “imprecise” is inappropriate; rather, our nitrate metric had most elements required for an accurate estimate of historical nitrate exposure. An accurate estimate of historical nitrate requires individual level information on water source for many decades preceding cancer incidence, water consumption amount, and historical data on nitrate levels in water supplies. We had all this information, as well as individual information on potential confounders such as pesticide use and dietary intake of nitrosation inhibitors.

We collected complete residential and water source histories so that we could determine who drank community water and so that we could compute average nitrate exposure over multiple residences for an approximately 40 year period, pre-dating cancer incidence. We also obtained information on the amount of tap water intake, although incorporating information on usual consumption did not change our risk estimates. Nitrate measurement data were available for 138 Nebraska municipalities during 1947–1987. The population size of the towns in 1980 ranged from 18 to 314 000; the median size was 1300 and only two municipalities exceeded 50 000. Although measurement data before the late 1960s were sparse, most towns had multiple years of measurement data from the late 1960s to the late 1980s. In this latter period, many towns had multiple measurements within a year that we averaged to compute an annual mean. Seasonal variation in nitrate concentrations was not common because most communities used ground water from deep wells (>35 metres). The number of years with missing data was similar for cases and controls, thus misclassification of exposure would be non-differential and tend to bias the odds ratios towards the null. We calculated a long term average level for each person in the study by linking the water source history information to the historical municipal nitrate database. To reduce misclassification, we excluded people for whom we had no information on nitrate concentrations in their drinking water supplies for more than 10% of their person years after 1947. Exposure in the five years before diagnosis of cases and interview of controls was not counted in a person's average exposure level because recent exposures are unlikely to be related to risk of cancer.

Thus, the average nitrate concentrations for people in our study, calculated over approximately 35 years, are not directly comparable to the six year average population levels calculated by the authors. Furthermore, it is important to consider the calendar time period over which average nitrate concentrations are computed, because several factors affect nitrate values in community supplies, including changes in nitrogen fertilizer use over time and changes in the source of a community's drinking water. The residential stability of the study population has also been shown to be an important factor in the correlation between recent and long term average nitrate exposures.3 Substantial misclassification can occur if recent measurement data are used to estimate long term average exposure. To illustrate this, we recalculated nitrate concentrations for people in our study using the last five years of nitrate measurements, data that were excluded from the original average nitrate calculation. For cases and controls, we compared quartile categories of long term average nitrate (our original metric) with the average nitrate concentrations computed using the last five years of exposure (recent five year average nitrate). We determined the per cent agreement between the exposure metrics. The results of the cross categorisation are shown in table 1; the per cent agreement was similar for cases and controls so the combined results are presented.

Only 22 per cent of people were classified into the same quartile of both exposure metrics. In our study, we observed a significantly increase risk of NHL among those in the highest quartile of long term average nitrate compared with those in the lowest quartile (odds ratio 2.0, 95% confidence intervals 1.1, 3.6). Even if we ignore the misclassification for those in the middle quartiles of long term average nitrate (382 people), the per cent agreement between the exposure metrics for those in the first and fourth quartile of long term average nitrate was only 30 per cent. Much of the misclassification was attributable to the assignment of those in the highest quartile of long term nitrate to the lowest two quartiles of recent average nitrate. These results illustrate the potential for substantial misclassification when recent data are used to estimate historical exposures.

Nitrate concentrations in water sources are influenced by agricultural, animal husbandry, and waste treatment practices, and these practices can change substantially over time. To further investigate the association between drinking water nitrate and cancer risk, well designed studies with individual information on historical nitrate exposure are needed.