Spatial-temporal analysis of non-Hodgkin lymphoma in the NCI-SEER NHL case-control study

Abstract

Background: Exploring spatial-temporal patterns of disease incidence through cluster analysis identifies areas of significantly elevated or decreased risk, providing potential clues about disease risk factors. Little is known about the etiology of non-Hodgkin lymphoma (NHL), or the latency period that might be relevant for environmental exposures, and there are no published spatial-temporal cluster studies of NHL.

Methods: We conducted a population-based case-control study of NHL in four National Cancer Institute (NCI)-Surveillance, Epidemiology, and End Results (SEER) centers: Detroit, Iowa, Los Angeles, and Seattle during 1998-2000. Using 20-year residential histories, we used generalized additive models adjusted for known risk factors to model spatially the probability that an individual had NHL and to identify clusters of elevated or decreased NHL risk. We evaluated models at five different time periods to explore the presence of clusters in a time frame of etiologic relevance.

Results: The best model fit was for residential locations 20 years prior to diagnosis in Detroit, Iowa, and Los Angeles. We found statistically significant areas of elevated risk of NHL in three of the four study areas (Detroit, Iowa, and Los Angeles) at a lag time of 20 years. The two areas of significantly elevated risk in the Los Angeles study area were detected only at a time lag of 20 years. Clusters in Detroit and Iowa were detected at several time points.

Conclusions: We found significant spatial clusters of NHL after allowing for disease latency and residential mobility. Our results show the importance of evaluating residential histories when studying spatial patterns of cancer.

Figure 1

Akaike information criterion (AIC) and span values of the crude spatial model for Los Angeles at a residential time lag of 20 years.

Figure 2

Crude and adjusted local odds ratios (OR, scale at right) for NHL at a residential lag time of 20 years in the Detroit study area. Clusters of statistically significant elevated odds ratios are identified with a solid white line and statistically significant lowered odds ratios are identified with a dashed black line. Crude model: span = 0.6 (p-value = 0.07); Adjusted model: span = 0.6 (p-value = 0.07). Model adjusted for age, gender, race, education, and home termite treatment before 1988.

Figure 3

Crude and adjusted local odds ratios (OR, scale at right) for NHL at a residential lag time of 20 years in Iowa. Clusters of statistically significant elevated odds ratios are identified with a solid white line and clusters of statistically significant lowered odds ratios are identified with a dashed black line. Crude model: span = 0.625 (p-value = 0.21); Adjusted model: span = 0.625 (p-value = 0.14). Model adjusted for age, gender, race, education, and home termite treatment before 1988.

Figure 4

Crude and adjusted local odds ratios (OR, scale at right) for NHL at a residential lag time of 20 years in the Los Angeles study area. Clusters of statistically significant elevated odds ratios are identified with a solid white line and statistically significant lowered ORs are identified with a dashed black line. Crude model: span = 0.275 (p-value = 0.003); Adjusted model: span = 0.275 (p-value = 0.03). Model adjusted for age, gender, race, education, and home termite treatment before 1988.

Figure 5

Crude and adjusted local odds ratios (OR, scale on right) for NHL at a residential lag time of 10 years in the Seattle study area. Crude model: span = 1 (p-value = 0.20); Adjusted model: span = 1 (p-value = 0.15). Model adjusted for age, gender, race, education, and home termite treatment before 1988.

Figure 6

Adjusted local odds ratios for NHL at time of diagnosis and four residential lag times in Los Angeles. Clusters of statistically significant elevated odds ratios are identified with a solid white line and statistically significant lowered odds ratios are identified with a dashed black line. The optimal span size was used for each residential lag time. Models adjusted for age, gender, race, education, and home termite treatment before 1988.

Figure 7

Adjusted local odds ratios for NHL for study participants (left) and all eligible cases and controls (right) in Detroit at the time of study enrollment. Clusters of statistically significant elevated odds ratios are identified with a solid white line and statistically significant lowered odds ratios are identified with a dashed black line. Participants: span = 0.6 (p-value = 0.05); All eligible: span = 0.325 (p-value = 0.08). Models adjusted for age and gender.

Figure 8

Adjusted local odds ratios for NHL for study participants aged 65 years or more (left) and eligible cases and controls aged 65 years or more (right) in Los Angeles at the time of study enrollment. Participants: span = 1 (p-value = 0.62); Eligible: span = 1 (p-value = 0.16). Models adjusted for age and gender.