Spatial analysis of bladder, kidney, and pancreatic cancer on upper Cape Cod: an application of generalized additive models to case-control data

Abstract

Background: In 1988, elevated cancer incidence in upper Cape Cod, Massachusetts prompted a large epidemiological study of nine cancers to investigate possible environmental risk factors. Positive associations were observed, but explained only a portion of the excess cancer incidence. This case-control study provided detailed information on individual-level covariates and residential history that can be spatially analyzed using generalized additive models (GAMs) and geographical information systems (GIS).

Methods: We investigated the association between residence and bladder, kidney, and pancreatic cancer on upper Cape Cod. We estimated adjusted odds ratios using GAMs, smoothing on location. A 40-year residential history allowed for latency restrictions. We mapped spatially continuous odds ratios using GIS and identified statistically significant clusters using permutation tests.

Results: Maps of bladder cancer are essentially flat ignoring latency, but show a statistically significant hot spot near known Massachusetts Military Reservation (MMR) groundwater plumes when 15 years latency is assumed. The kidney cancer map shows significantly increased ORs in the south of the study area and decreased ORs in the north.

Conclusion: Spatial epidemiology using individual level data from population-based studies addresses many methodological criticisms of cluster studies and generates new exposure hypotheses. Our results provide evidence for spatial clustering of bladder cancer near MMR plumes that suggest further investigation using detailed exposure modeling.

Figure 1

Upper Cape Cod study area. Cape Cod is located in Massachusetts in the northeast United States.

Figure 2

Spatial distribution of participants. Each point represents the residence of one participant. Locations have been geographically altered to preserve confidentiality.

Figure 3

Bladder cancer results. Odds ratios are relative to the whole study area. a) Adjusted, no latency. b) Adjusted, 15 years of latency. Assuming 15 years of latency increases the magnitude of hot and cold spots. Black contour lines denote areas of significantly increased and decreased risk at the 0.05 level. c) Crude, 15 years of latency, created using the optimal span (0.30) of the adjusted map. Little difference from the adjusted map suggests spatial confounding is not an issue. d) Adjusted, 15 years of latency. Restriction to residences of longest duration has little effect when the same span (0.30) is used as for all residences.

Figure 4

Kidney cancer results. Odds ratios are relative to the whole study area. a) Crude, no latency, created using the optimal span (0.90) of the adjusted map. b) Adjusted, no latency, optimal span. Black contour lines denote areas of significantly increased and decreased risk at the 0.05 level. Lack of important differences between the crude and adjusted maps suggests spatial confounding is not an issue. c) Adjusted, no latency, span of 0.40. Results are similar to optimal span. d) Adjusted, no latency, span of 0.15. Small span size results in more spatial variation in risk.

Figure 5

Pancreatic cancer results. Odds ratios are relative to the whole study area. a) Crude, no latency, created using the optimal span (0.40) of the adjusted map. b) Adjusted, no latency, optimal span. Black contour lines denote areas of significantly increased and decreased risk at the 0.05 level. Difference of the crude and adjusted maps indicates spatial confounding.

Figure 6

Groundwater plumes, the Massachusetts Military Reservation (MMR), and significant bladder cancer hot spots. a) Bladder cancer map (Fig. 3b), adjusted, 15 years of latency with an optimal span of 0.30. Odds ratios are relative to the whole study area. b) Location of the MMR and groundwater plumes from the MMR and other sources including landfills.