Oct 21, 1998 - To date, only arsenic and arsenical compounds, which cause lung and ... ceptable cancer research conditions, as is the case in Costa Rica.
© International Epidemiological Association 1999
International Journal of Epidemiology 1999;28:365–374
Printed in Great Britain
Geographical differences of cancer incidence in Costa Rica in relation to environmental and occupational pesticide exposure Catharina Wesseling,a,b,c Daniel Antich,a Christer Hogstedt,c,d Ana Cecilia Rodrígueza,e and Anders Ahlbomb,f
Background This study describes geographical differences in cancer incidence in Costa Rica, and investigates if some of these differences may be related to pesticides. Methods
Data were combined from the cancer registry (1981–1993), the 1984 population census, the 1984 agricultural census, and a national pesticide data set. The 81 counties of Costa Rica were the units for the ecological analyses. Adjacent counties were grouped into 14 regions (3 urban and 11 rural) with relatively similar socioeconomic characteristics. County indices for population density and agricultural variables were constructed and categorized. Differences across regions and categories were assessed by comparing observed numbers of incident cases to expected values derived from national rates. Within the tertile of most rural counties, rate ratios between categories of high and low pesticide use were calculated.
Results
In urban regions, excesses were observed for lung, colorectal, breast, uterus, ovary, prostate, testis, kidney, and bladder cancers; and in rural regions for gastric, cervical, penile, and skin cancers. Skin cancers (lip, melanoma, non-melanocytic skin and penile cancer) occurred in excess in coffee growing areas with extensive use of paraquat and lead arsenate. In the most rural subset, heavy pesticide use was associated with an increase of cancer incidence overall and at a considerable number of specific sites, including lung cancer (relative risk [RR] 2.0 for men and 2.6 for women) and all female hormone-related cancers (RR between 1.3 and 1.8).
Conclusions Regions and populations at high risk for specific cancers were identified. Several hypotheses for associations between pesticides and cancer emerged. The findings call for studies at the individual level. Keywords
Cancer, developing country, environmental exposure, epidemiology, occupational exposure, pesticides
Accepted
21 October 1998
Excess cancer risks have been reported in relation to various pesticides,1–4 but few associations between pesticides and cancer have been sufficiently studied in human populations. To date, only arsenic and arsenical compounds, which cause lung and skin cancers, are classified as carcinogens in humans
a Central American Institute of Studies on Toxic Substances, Universidad
Nacional, Box 86-3000, Heredia, Costa Rica. b Institute of Environmental Medicine, Karolinska Institute, Stockholm, Sweden. c National Institute of Working Life, Solna, Sweden. d Department of Occupational Health, Karolinska Hospital, Stockholm, Sweden. e National Tumor Registry, Ministry of Health, San José, Costa Rica. f Department of Epidemiology, Karolinska Hospital, Stockholm, Sweden.
Reprint requests to: Catharina Wesseling, Division of Epidemiology, Institute of Environmental Medicine, Karolinska Institute, Box 210, 171–77 Stockholm, Sweden.
(Class 1) by the International Agency for Research on Cancer.5,6 Epidemiological studies of associations between pesticides and cancer could be particularly informative in those Third World countries where high exposures coexist with acceptable cancer research conditions, as is the case in Costa Rica. In Costa Rica, working conditions and environmental emissions from traffic, industry, and agriculture did not receive serious attention until recently. Considerable occupational and environmental exposures to carcinogens may have occurred. Agriculture was the principal economic activity for decades, and pesticides are the only widespread chemical pollutants in rural areas. Average pesticide use during the 1980s was 16 kg/ha of arable land, higher than in areas with intensive agriculture in developed countries.7 Inappropriate handling of pesticides and contamination of workers8 as well as the environment7 have been documented. On the other hand, Costa Rica (population
365
366
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
2.4 million in 1984 and 3.2 million in 1993) has reasonable infrastructures, a health care system that covers the entire population, good health indicators (life expectancy 76 years in 1993), population registries, and a cancer registry (National Tumor Registry, RNT). The RNT has operated since 1978 with many of the limitations characteristic of a developing country. In the early 1990s, a Swedish-Costa Rican research collaboration was initiated. The programme addressed, among other matters, the improvement of RNT with the long-term goal of meeting international quality standards. The use of RNT for research purposes during the improvement process is essential for its continued evaluation and consolidation. In a recent study, an excess of certain cancers was observed in a cohort of banana plantation workers, a group highly exposed to pesticides.9 So far, no data on cancer risk are available for any other subgroup of the population of Costa Rica. This study describes geographical differences in cancer incidence in Costa Rica during 1981–1993, and explores if some of these differences may be related to pesticide use.
Materials and Methods Four data sources were used: the cancer registry, the population census, the agricultural census, and a data set on pesticide use in Costa Rica.
Cancer data Over 50 000 cases of cancer were reported to the RNT during 1981–1993. The standard RNT record contains personal identification, gender, age at diagnosis, cancer site and histology (International Classification of Diseases for Oncology, first edition), year of diagnosis, and residence at diagnosis. Basal cell skin and in situ cervical cancers were excluded from the study, and data on residence or age were missing for 834 other cancers (2.1%). The remaining 39 647 cancers were included.
Population data Population data for geographical regions by gender and 5-year age strata for the year 1987 were used. These were the projections of the Latin American Center of Demography (CELADE), based on the 1984 census. The numbers of workers by economic branch were obtained from the 1984 census files.
Agricultural data The agricultural census of 1984 provided data on total land surface, cropping in hectares, and number of hectares treated with pesticides per crop in geographical areas.
Pesticide use and exposure Data on types of pesticides, frequency of applications, and spray methods for specific crops according to geographical areas were obtained from the Central American Institute of Studies on Toxic Substances at the Universidad Nacional, Costa Rica. A pesticide exposure indicator (PEI) was calculated for each county: k PEIcounty = ∑ hiniai / population i =1
PEI = pesticide exposure indicator i = crop (1,2,...k) hi = hectares treated with pesticides (1984)
ni = estimated average number of applications per year (1980–1984) ai = aerial spraying or presence of large fraction of workers in sprayed fields: low = 1; medium = 2; high = 4 (1980–1984). The numerator of the formula is an estimate of the total pesticide load. With the size of the total population of the county in the denominator, the PEI provides an estimate of pesticide used per inhabitant. The structure of the databases corresponded to the administrative geographical division of Costa Rica. The 81 counties of the country were the basic units for the ecological analyses. Computer programs were developed to enumerate cancer cases by population and agricultural variables within the counties, as well as in aggregations of counties. Adjacent counties were aggregated into 14 regions with a reasonable population size (.74 000), and relatively similar socioeconomic attributes, and geological and climatic characteristics. Furthermore, county indices were constructed for population density, proportion of workers by economic branch, total arable land, specific crops, and pesticide use. The indices were categorized into quartiles. Table 1 displays socioeconomic and agricultural characteristics of the 14 regions. Differences in cancer incidence across the 14 regions and across categories of the various indices were assessed by comparing the observed number of all cancers and of specific sites to the expected numbers. The expected numbers were derived for gender- and 5-year age-specific strata from the person-years in the regions or categories (1987 population times 13 years) and the national incidence rates. The standardized incidence ratios (SIR) were calculated as ratios of observed and expected numbers of cases. The 95% CI were calculated with the normal approximation of the Poisson distribution; for observed numbers below 20, the exact distribution was used. Overrepresentation of cases in counties with a hospital was anticipated. Preliminary analyses to evaluate the influence of the hospitals on the risk estimates showed that in the central county with the capital San José, where the four most specialized hospitals of Costa Rica are concentrated, SIR were increased for both genders, for all sites combined as well as for nearly all specific sites. In the remaining counties with hospitals, the SIR for all cancers combined correlated weakly with the total number of specialist services at the rural hospitals (r = 0.5 for male cancers and 0.3 for female cancers), but no pattern was observed for SIR of specific cancers and the existence or absence of specific medical services. Consequently, we excluded the central county of San José from all analyses. Many counties had both urban (unexposed to agricultural use of pesticides) and rural (potentially exposed to agricultural use of pesticides) populations. To diminish the within-county heterogeneity of exposure, and to achieve a more comparable distribution of extraneous risk factors across the counties,10,11 we performed analyses restricted to the tertile of most rural counties, i.e. the 27 counties with the lowest population density. After ranking these counties according to their PEI-value (range 7–243), we divided them into categories of low and high PEI with roughly the same population size (cutpoint 50; mean PEI 16 [SD 8.2] and 137 [SD 69.2]). Relative risks (RR) along with the 95% CI were calculated for men and women as ratios of the age-adjusted incidence rates in high versus low PEI categories.
119 466
192 643 164 086 225 813
12. North-west Costa Rica
13. Mid-south Costa Rica
14. East Costa Rica (Atlantic Region)
17
27
19
17
30
17
87
41
167
116
227
333
1002
2432
Rural
Rural
Rural
Rural
Rural
Rural
Rural
Rural
Rural
Rural
Rural
Urban
Urban
Urban
Population Urban/ density rural (per km2) status
59
68
50
59
39
65
33
55
34
30
18
9
6
3
4
4
5
3
10
5
14
7
12
15
19
22
20
23
92 538
67 296
145 081
108 723
44 130
62 159
36 903
35 976
10 255
41 983
5773
4404
5039
2120
Agricultural Industrial workers workers Arable (% of EAPb) (% of EAPb) land (ha)
Coffee
Coffee, sugarcane
Coffee
Coffee
Coffee
Coffee
Main crops (1984)
Banana,corn, beans, rice
Corn, beans, coffee
Rice, sugarcane
Rice, banana, corn, beans
Rice
Corn, beans, rice
Coffee, sugarcane
Coffee, corn, beans
c Pesticide exposure indicator (mean pesticide load per inhabitant, see Method). d OP: organophosphates; CB: carbamates; OC: organochlorines. Fungicides: copper compounds, mancozeb, chlorothalonil, benomyl, tridemorph.
a Central county of San José excluded. b Economically active population.
136 031
11. Southern Central Pacific and South Pacific Region
137 225
9. Mid-north Costa Rica
10. Northern Central Pacific Region
320 884
8. Eastern Central Valley and east mountain chain (Cartago)
297 363
5. Western Central Valley
105 924
112 537
4. Northern mountain chain of Central Valley
7. Southern mountain chain of Central Valley
138 579
3. Northern Central Valley
74 574
207 723
2. Southern Central Valley
6. South-west Central Valley
296 319
1. Central Valleya
Region
Population (1987)
Table 1 Regional socioeconomic and agricultural characteristics
143
15
48
92
18
10
10
16
9
12
6
4
2
1
PEIc
OP, DBCP, paraquat, fungicides, formaldehyde, OC, phenoxyacids
Paraquat, lead arsenate, copper, OP, CB
OC, OP, phenoxyacids, propanil, pyrethroids
OC, OP, phenoxyacids, propanil, DBCP, paraquat, fungicides, formaldehyde
OP, OC, phenoxyacids, propanil
Paraquat, OP, CB, OC, phenoxyacids, propanil
Paraquat, lead arsenate, copper, OP, CB, 2,4-D, pyrethroids
Paraquat, lead arsenate, copper, OP, CB
Paraquat, lead arsenate, copper, OP, CB
Paraquat, lead arsenate, copper, OP, CB, 2,4-D, pyrethroids
Paraquat, lead arsenate, copper, OP, CB
Paraquat, lead arsenate, copper, OP, CB
Paraquat, lead arsenate, copper, OP, CB
Paraquat, lead arsenate, copper, OP, CB
Main pesticidesd
CANCER INCIDENCE IN COSTA RICA
367
368
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Results For the 14 regions, SIR for all sites ranged from 73 to 123 in men and from 84 to 116 in women (Figure 1). With the exception of lymphohaematopoietic cancers, the geographical differences were more than twofold for all major tumour sites (Tables 2 and 3). Standardized incidence ratios .120 with
narrow CI were observed in at least two of the three most urban regions (regions 1–3) for lung, colorectal, breast, prostate, testis, kidney, and bladder cancers. Standardized incidence ratios .120 with narrow CI were observed, in at least two of the 11 rural regions, for gastric, cervical, penile, and skin cancers (Tables 2 and 3). The southern suburbs of San José (region 2) showed the highest risk for all cancers combined for men and
Figure 1 Regional distribution of cancer risk (*SIR: standardized incidence ratio), all sites, Costa Rica, 1981–1993
Site
122 113–132
8. Eastern Central Valley and east mountain chain (Cartago)
137 122–153
13. Mid-south Costa Rica
c 95% confidence interval. d Lymphohaematopoietic system.
a Central county of San José excluded. b Standardized incidence ratio.
73 65–82
61 53–69
12. North-west Costa Rica
14. East Costa Rica (Atlantic Region)
72 61–84
90 79–103
11. Southern Central Pacific and South Pacific Region
10. Northern Central Pacific Region
83 71–96
129 113–146
7. Southern mountain chain of Central Valley
9. Mid-north Costa Rica
95 80–111
6. South-west Central Valley
119 104–136
4. Northern mountain chain of Central Valley 94 87–103
105 92–118
3. Northern Central Valley
5. Western Central Valley
140 127–154
89b 81–98c
Stomach
2. Southern Central Valley
1. Central Valleya
Region
78 51–107
78 48–121
65 40–91
72 42–118
76 44–110
85 52–132
79 57–103
118 70–167
93 52–154
139 110–170
85 50–137
128 86–172
113 78–148
135 104–167
Colon
62 36–89
265 199–332
38 21–66
81 47–133
74 44–118
109 65–155
91 65–118
98 57–158
143 87–221
106 79–135
39 16–81
103 62–145
114 77–152
107 78–137
Liver
63 37–90
144 95–195
100 67–134
121 73–170
85 49–122
77 44–126
119 89–149
86 49–146
115 67–185
97 71–123
74 40–127
110 68–153
116 78–154
90 63–118
Pancreas
127 104–150
56 37–76
68 52–86
91 65–117
109 84–135
46 28–65
118 100–137
55 34–77
60 34–85
72 59–87
100 72–130
128 100–157
137 112–163
146 125–168
Lung
82 54–111
122 81–165
58 36–91
70 39–116
49 26–87
89 51–128
106 79–134
148 92–204
127 74–204
112 84–141
100 61–157
118 75–163
115 79–151
106 77–135
Brain
92 78–107
106 88–125
79 62–98
94 76–113
97 81–113
82 65–100
106 94–120
81 62–101
104 79–130
111 98–125
94 74–116
97 79–117
107 91–124
107 94–122
LHd
99 69–129
80 47–114
34 19–56
57 31–96
67 39–96
50 27–87
107 82–133
105 63–149
69 37–120
95 73–119
127 80–175
128 89–169
137 101–175
171 138–205
Bladder
61 44–79
135 103–167
76 57–96
54 33–77
149 117–181
77 52–103
79 63–96
71 45–98
153 112–196
121 102–141
50 31–78
169 134–205
87 66–110
114 94–135
Skin
121 105–139
82 65–101
90 77–105
78 60–97
101 83–119
66 50–83
76 65–87
76 58–96
86 65–108
97 86–109
99 78–121
141 119–163
110 93–128
137 122–154
Prostate
Table 2 Standardized incidence ratios and 95% confidence intervals of all and selected cancer sites by region, male population of Costa Rica, 1981–1993
123 72–198
216 130–337
103 56–173
118 54–224
82 36–163
85 34–176
54 27–97
175 91–307
34 4–126
88 53–138
99 40–206
96 44–184
105 56–180
87 50–142
Penis
49 26–84
43 19–86
71 41–117
79 40–142
79 42–135
62 30–121
116 82–151
55 22–115
134 69–234
103 71–136
143 87–224
80 43–137
154 105–205
164 121–207
Testis
87 82–92
111 104–118
73 69–78
79 73–85
97 91–103
77 71–83
102 98–107
102 95–110
98 91–107
102 98–107
98 91–105
117 110–124
123 117–129
114 110–120
All sites
CANCER INCIDENCE IN COSTA RICA
369
Sites
86 71–102 85 67–104 100 88–113 105 82–129 147 122–174 114 102–128
3. Northern Central Valley
4. Northern mountain chain of Central Valley
5. Western Central Valley
6. South-west Central Valley
7. Southern mountain chain of Central Valley
8. Eastern Central Valley and east mountain chain (Cartago)
c 95% confidence interval. d Lymphohaematopoietic system.
a Central county of San José excluded. b Standardized incidence ratio.
72 58–87
164 137–192
13. Mid-south Costa Rica
14. East Costa Rica (Atlantic Region)
73 60–87
96 72–121
82 66–99
12. North-west Costa Rica
11. Southern Central Pacific and South Pacific Region
10. Northern Central Pacific Region
83 63–104
120 105–137
2. Southern Central Valley
9. Mid-north Costa Rica
90b 80–101c
Stomach
1. Central Valleya
Region
91 59–124
104 61–149
57 34–82
64 31–124
80 47–115
69 37–118
67 46–88
84 49–135
89 51–146
150 119–181
57 31–98
135 97–175
114 83–146
126 101–153
Colon
85 47–144
190 116–294
80 45–133
102 44–202
83 42–149
95 44–182
95 60–131
50 17–119
69 26–152
123 83–164
99 50–178
110 65–175
132 83–181
77 49–107
Liver
78 45–128
103 55–177
92 55–130
169 97–276
85 48–141
78 36–148
118 84–153
92 48–162
84 41–165
88 59–118
61 28–118
76 45–123
89 54–125
137 104–170
Pancreas
150 105–195
107 65–168
53 32–85
75 36–147
134 88–181
49 22–98
101 74–129
73 39–127
63 31–123
92 66–118
45 21–87
97 62–133
128 93–165
125 99–153
Lung
51 26–90
100 59–162
69 39–114
108 58–185
84 45–144
92 49–158
93 62–125
93 46–167
122 61–219
94 61–127
99 53–169
127 74–181
127 84–172
130 95–167
Brain
78 61–95
101 79–124
86 62–112
93 71–115
94 76–113
74 53–95
90 76–104
128 99–159
83 57–111
106 91–123
104 79–130
87 67–107
133 112–154
112 97–128
LHd
93 40–185
76 21–195
57 21–124
0 0–94
80 30–176
0 0–77
100 57–163
54 11–160
60 12–175
116 70–183
114 46–235
127 66–223
136 78–222
150 98–203
Bladder
49 31–69
214 164–265
38 23–54
120 77–164
137 104–172
78 46–110
49 35–63
154 112–196
132 92–173
142 119–166
60 36–86
173 140–208
81 61–103
82 67–98
Skin
78 66–91
63 50–77
80 69–93
62 46–78
90 75–105
57 44–72
88 79–99
83 66–100
100 80–121
117 105–129
87 71–103
111 96–126
120 107–134
142 131–154
Breast
Table 3 Standardized incidence ratios and 95% confidence intervals of all and selected cancer sites by region, female population of Costa Rica, 1981–1993
138 124–154
123 107–141
131 117–146
197 172–223
124 108–141
111 94–129
77 69–86
85 70–102
61 47–77
79 70–88
81 68–96
75 64–87
100 89–112
85 77–94
Cervix
127 94–161
69 40–99
61 38–84
63 35–107
84 53–116
80 46–115
91 70–113
56 31–95
112 67–159
108 85–133
107 69–146
115 81–149
113 85–142
129 105–153
Ovary
93 88–99
110 103–118
84 79–89
111 103–120
100 94–107
84 78–91
87 83–91
103 96–111
92 85–100
103 100–108
88 82–95
102 96–108
116 111–122
110 106–114
All sites
370 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
CANCER INCIDENCE IN COSTA RICA
women; men had SIR .100 for 22 out of 29 sites and women for 26 out of 30 sites; a particularly high risk for cancer of the nasopharynx was detected in women (SIR = 236, 95% CI : 114–459). Several rural regions had markedly high risks for specific cancers. In one area with predominantly smallholder agriculture growing grains, coffee, and sugarcane (region 9), excesses were found for cancer of the eye (men: SIR = 175, 95% CI : 83–337; women: SIR = 262, 95% CI : 131–470) and of the oropharynx among women (SIR = 395, 95% CI : 145–860, based on six cases). In an aggregation of seven small counties south of the Central Valley (parts of regions 6 and 7, contiguous to region 2), with mainly coffee production, elevated SIR (between 123 and 281) with rather narrow CI were observed for cancer of the tongue and oral cavity (men), oesophagus (particularly women), stomach (both genders), and liver (men). Similarly elevated rates were also observed for lymphohaematopoietic cancers (women), skin melanoma (men), penile cancer, eye (men), and brain cancers (particularly men). Excess cancer of the lip occurred in both genders in the four regions with the largest coffee production areas (use of, for example, lead arsenate and paraquat), and excess cancer of the oesophagus and stomach in three of them. High risks for penile cancer, non-melanocytic skin cancer and skin melanoma were also observed in several of these coffee regions. In the rice producing north-west area of Costa Rica (use of, for example, chlorophenoxy herbicides), the rates for most sites, except cervical cancer, were remarkably low. In the banana producing Atlantic Region (use of, for example, dibromochloropropane, chlorothalonil, and mancozeb) SIR were high for several cancers usually not seen in excess in rural areas: respiratory cancers in both genders, and ovary and prostate cancers. With regard to population density, excesses were predominantly observed for the highest and deficits for the lowest quartile. Risks of invasive cervical and penile cancer were inversely related to population density. The analyses by quartiles of proportions of industrial workers were similar to those by population density, while those by agricultural workers showed an opposite tendency. No or irregular trends were obtained for areas of arable land, specific crops and PEI. In the tertile of most rural counties, the RR of all sites combined for counties with high versus low PEI was slightly elevated in men, and more clearly in women (Table 4). The risk of lung cancer was twofold in both genders. Increases between roughly 25% and 80% were observed for hormone-related cancers in women, but in men the RR of prostate cancer was just over unity (RR = 1.1) and of testicular cancer well below (RR = 0.6). Relative risks >1.5 with at least 15 cases in the high PEI category were observed for skin melanoma and cancers of the oesophagus, gallbladder, larynx, bone, and bladder in men; and for colorectal, liver, gallbladder and brain cancer in women.
Discussion This study showed considerable geographical differences in the incidence of several cancers. Cancer patterns contrasted between rural and urban areas, and population density was related to many sites. Within the subset of most rural counties, heavy pesticide use was associated with an increase in overall cancer occurrence and at a substantial number of specific sites,
371
including a twofold risk of lung cancer among both genders, and excesses of all hormone-related cancers in women. The results may have been affected by a number of potential systematic errors. The quality of the cancer data can be considered in general acceptable.12 Cancer reporting is compulsory for public and private health and pathology facilities, where trained technicians are in charge, but the completeness of coverage of RNT has not been evaluated. Before 1994 cancers reported on death certificates that could not be traced in any health centre were not included. Any underregistration is more likely in remote areas, and would have caused a dilution of the SIR in the most rural regions. Registration of the county of treatment as the county of residence inflated the incidence rates in the hospital counties. This bias was reduced by excluding the central county of San José with its specialized hospitals, but it may partly explain the overall increased cancer risk in the nearby, densely populated, suburbs (region 2). The incidence rates in the other regions are not affected by a concentration of cases in the hospital counties, since patients come from surrounding areas. Some confidence in the validity of the results can be derived as the findings are in accordance with current epidemiological knowledge. Excesses of nasopharyngeal cancers (SIR = 175, 95% CI : 105–247, based on 23 cases) and nasal cancers (SIR = 149, 95% CI : 77–261, based on 12 cases) occurred in men in the west part of the Central Valley (region 5), where the district of Sarchí with the main wood carving and furniture industry of Costa Rica is located. The residence of most cases was registered as the hospital county, Alajuela. Conversely, in categories of non-adjacent counties, overrepresentation of cases in hospital counties may have biased the results into either direction. The impact of this potential bias on the risk estimates in the tertile of most rural counties was examined by excluding the counties with a hospital of a substantial size (one in the high and one in the low PEI category), one or both at a time. Concentration of cases in hospital counties did not seriously affect the results for this subset; the high and low PEI hospital counties respectively increased and decreased the rate ratios somewhat, for all as well as most specific sites, but when eliminating both counties the net changes were very small. Population data for 1987 were used. Recent estimates from the Central American Population Program at the University of Costa Rica show that in 1995 the largest population change over the 14 regions was a 1.6% increase in the Atlantic Region, due to expansion of banana plantations after 1990. The data on pesticide use have several limitations. The complexity of pesticide and crop patterns did not allow assessment of specific pesticides, and the PEI is an indicator for the combined use of different pesticides with varying carcinogenic properties. Furthermore, the PEI is based on pesticide use in 1984, whereas the relevant exposure period for cancers may extend one to two decades further back. In the rural subset, the high and low PEI categories distinguished well between counties with a commonly known history of early, heavy pesticide use and other areas. Finally, migration, particularly related to the banana plantations in the Atlantic Region, may make geographical comparisons of cancer risk more difficult. The use of aggregated data instead of the joint distributions of exposure, outcome, and covariates at the individual level, may
372
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Table 4 Risk of all cancers and specific sites for category of high pesticide use versus category of low pesticide use, tertile of most rural counties (,30 inhabitants/km2), Costa Rica, 1981–1993 Men
Women
Cases in high PEIa category
RRb
95% CI c
Cases in high PEIa category
RRb
95% CI c
1427
1.15
1.07–1.23
1453
1.32
1.23–1.42
Mouth (140–149)
51
1.19
0.81–1.74
27
1.18
0.70–2.00
Oesophagus (150)
32
1.69
0.98–2.93
5
1.20
0.37–3.93
325
1.01
0.87–1.16
133
1.01
0.80–1.27
Colon (153)
42
1.19
0.78–1.81
31
1.65
0.98–2.78
Rectum (154)
35
1.61
0.96–2.69
28
1.86
1.06–3.25
Liver (155)
22
0.61
0.36–1.02
23
2.20
1.16–4.17
Gallbladder (156)
19
1.49
0.77–2.90
41
1.73
1.07–2.78
Pancreas (157)
43
1.28
0.84–1.97
23
0.97
0.57–1.68
Larynx (161)
33
1.88
1.10–3.23
4
0.48
0.11–1.57
107
2.01
1.49–2.72
37
2.63
1.50–5.62
Hodgkin’s disease (965–966)
35
1.23
0.76–1.98
13
1.14
0.55–2.35
Non-Hodgkin lymphoma (959–964, 969, 975)
44
1.17
0.77–1.77
23
1.32
0.73–2.41
101
1.16
0.88–1.52
78
1.17
0.85–1.60
Multiple myeloma (973)
15
0.97
0.50–1.90
10
1.29
0.51–3.28
Bone (170)
15
1.81
0.83–3.95
6
1.31
0.44–3.91
Soft-tissue sarcoma (171)
18
0.58
0.32–1.03
13
1.11
0.52–2.35
Skin (non-melanocytic) (173)
73
1.16
0.85–1.59
35
1.23
0.79–1.92
Skin melanoma (872–879)
17
1.77
0.83–3.79
4
0.29
0.10–0.87
Breast (174)
174
1.25
1.02–1.54
Cervix (invasive) (180)
Site (ICDO-1) All (140–199)
Stomach (151)
Lung (162)
Leukaemia (980–994)
445
1.32
1.16–1.51
Corpus uteri (182)
28
1.78
1.02–3.11
Ovary (183)
50
1.78
1.17–2.71
Prostate (185)
167
1.12
0.90–1.38
Testis (186)
19
0.60
0.33–1.09
Penis (187)
20
1.05
0.57–1.93
Bladder (188)
34
1.71
1.01–2.90
5
0.94
0.29–3.08
Kidney (189)
10
0.58
0.27–1.24
18
1.09
0.58–2.04
Eye (190)
14
1.39
0.67–2.89
6
0.70
0.26–1.83
Brain (191)
35
1.17
0.73–1.86
29
1.64
0.94–2.88
Thyroid (193)
14
1.47
0.65–3.31
50
1.26
0.85–1.86
a Pesticide exposure indicator. b Relative risk. c Confidence interval.
lead to severe bias in ecological analyses.10,11 Restriction of the analyses to the tertile of least densely populated counties reduced this bias. The populations of this subset are predominantly agricultural and, on the county level, rather homogeneous with regard to farming practices. This considerably diminished the within-county heterogeneity of pesticide exposures, while large differences between the counties remained. Confounding from the grouping variable, due to the differential distribution of other risk factors over the counties, was at least partly solved by this stratification. Population density is an indirect measure of degree of urbanization and a collective
indicator for a complex set of potentially confounding factors related to socioeconomic status and lifestyle. However, residual confounding is feasible. Overall, it is difficult to assess the net impact of the ecological bias. The apparent gradient of cancer occurrence between urban and rural areas is consistent with previous observations.13,14 Lung, colorectal, and breast cancers occurred more frequently in the urban regions where lifestyles in terms of smoking, alcohol intake, diet, and calorie expenditure resemble more western industrialized societies than in rural Costa Rica. The incidence rates of cervix and penile cancers were high in rural areas with low
CANCER INCIDENCE IN COSTA RICA
socioeconomic status, as expected.13,15 There are no easy interpretations of regional differences with regard to crops or specific pesticides, because of large overlap and because most background cancer rates are low. However, a few results stand out. For example, coffee growing regions seemed to be related to increased risks for various skin type cancers. Solar radiation is one plausible explanation, but pesticides could also play a role. Lead arsenate was widely used on coffee during the 1970s and 1980s, and arsenical compounds are acknowledged skin carcinogens.5 An alternative hypothesis is an association of paraquat with skin cancers. Paraquat is extensively used on coffee and banana, and dermal contamination and chemical burns with this herbicide are common.8 An increased risk for melanoma and penile cancer, observed in a cohort study of banana plantation workers, is consistent with this hypothesis.9 Hormone-related cancers occurred in excess among women in the high PEI category, but not among men. The possibility of associations with environmental hormone-disrupting chemicals continue to be a public health concern.16–18 A recent review of occupational and environmental exposures in relation to ovarian cancers concluded that there is little or no evidence for excess risk among agricultural- and pesticide-exposed women, but the authors also pointed out the lack of good quality epidemiological studies.17 A similar recent breast cancer review considered that exposure to chlorinated hydrocarbon pesticides is possibly a risk factor, but that evidence was insufficient and studies among agricultural populations with high exposure would be of interest.18 Lung cancer rates were decreased or close to unity in most rural regions, while a twofold increase was observed for rural counties with high PEI as compared to low PEI. Both findings are consistent with previous studies. Deficits for lung cancer among farmers have been attributed to a lower prevalence of smoking in this population.1,19 Increased risk for lung cancer among pesticide exposed populations has been observed in ecological as well as cohort and case-control studies.20–24 A doubled risk for lung cancer in both genders appears high, even if confounding by smoking occurred. Assuming a 50% higher prevalence of smokers in the high PEI counties and, on top of that, that all are heavy smokers with a tenfold risk, an RR of 1.4 (0.6 × 10 + 0.4 × 1)/(0.4 × 10 + 0.6 × 1) would be obtained due to smoking alone, which is well below the observed excesses. The excesses of lung cancer, as well as a number of other cancers, among both men and women suggest that, if these are true associations, population wide exposures may have occurred, although mainly men work directly with pesticides. Aerial spraying occurred in all counties in the high PEI category, on rice or banana, and drifts and contamination of water supplies have been documented.7,25 In conclusion, there are substantial geographical differences in the incidence of most major cancer sites in Costa Rica. Although ecological bias and other types of systematic error preclude aetiological interpretations, some of the observed excesses may be due to occupational or environmental exposures to pesticides. The excess risks of skin melanoma, lung cancer, and female hormone-related cancers seem particularly interesting topics for epidemiological studies at the individual level. This investigation also permitted an improved insight into quality aspects of the cancer registry and increases the potential of cancer research in Costa Rica.
373
Population and agricultural characteristics of the 81 counties of Costa Rica, grouped by region are available from the author on request.
Acknowledgements This investigation was financially supported by the Department of Research Cooperation (SAREC) and the Health Division of the Swedish International Development Cooperation Agency (Sida). The authors thank Dr Timo Partanen for useful comments on the manuscript.
References 1 Dich J, Zahm SH, Hanberg A, Adami HO. Pesticides and cancer. Cancer
Causes Control 1997;8:420–43. 2 Pearce N, Matos E, Vainio H, Boffetta P, Kogevinas M. Occupational
Cancer in Developing Countries. IARC Scientific Publications No. 129. Lyon: IARC, 1994. 3 Donna A, Crosignani P, Robutti F et al. Triazine herbicides and ovarian
epithelial neoplasms. Scand J Work Environ Health 1989;15:47–53. 4 Garabrandt DH, Held J, Langholz B, Peters JM, Mack TM. DDT and
related compounds and risk of pancreatic cancer. J Natl Cancer Inst 1992;80:764–71. 5 International Agency for Research on Cancer. Overall evaluation of
carcinogenicity: an updating of IARC monographs volumes 1–42. IARC monographs on the evaluation of the carcinogenic risks to humans, supplement 7. Lyon: IARC, 1987. 6 International Agency for Research on Cancer. Occupational Exposures in
Insecticide Application, and Some Pesticides. IARC monographs on the evaluation of the carcinogenic risks to humans, volume 53. Lyon: IARC, 1991. 7 Castillo LE, de la Cruz E, Ruepert C. Ecotoxicology and pesticides in
tropical aquatic ecosystems of Central America. Environ Toxicol Chem 1997;16:41–51. 8 van Wendel de Joode BN, de Graaf IA, Wesseling C, Kromhout H.
Paraquat exposure of knapsack applicators on banana plantations in Costa Rica. Int J Occup Environ Health 1996;2:294–304. 9 Wesseling C, Ahlbom A, Antich D, Rodríguez AC, Castro R. Cancer in
banana plantation workers in Costa Rica. Int J Epidemiol 1996;25: 1125–31. 10 Morgenstern H. Uses of ecologic analyses in epidemiologic research.
Am J Public Health 1982;72:1336–44. 11 Greenland S, Robins J. Invited commentary: ecological studies—
biases, misconceptions, and counterexamples. Am J Epidemiol 1994; 139:747–60. 12 Parkin DM, Whelan SL, Ferlay J, Raymond L, Young J (eds). Cancer
Incidence in Five Continents, Vol. VII. Ed. IARC Scientific Publications No. 143. Lyon: IARC, 1997. 13 Tomatis L, Aitio A, Day NE et al. (eds). Cancer: Causes, Occurrence and
Control. Ed. IARC Scientific Publications No. 100. Lyon: IARC, 1990. 14 Blot WJ, Fraumeni JF. Geographic epidemiology of cancer in the
United States. In: Schottenfeld D, Fraumeni JF (eds). Cancer Epidemiology and Prevention. Philadelphia: WB Saunders Company, 1982. 15 Herrero R, Brinton LA, Hartge P et al. Determinants of the geographic
variation of invasive cervical cancer in Costa Rica. Bull Pan Am Health Organ 1993;27:15–25. 16 Colborn T, vom Saal FS, Soto AM. Developmental effects of endocrine
disrupting chemicals in wildlife and humans. Environ Health Perspect 1993;101:378–84.
374
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
17 Shen N, Weiderpass E, Anttila A et al. Epidemiology of occupational
22 Pesatori AC, Sontag JM, Lubin JH, Consonni D, Blair A. Cohort
and environmental risk factors related to ovarian cancer. Scand J Work Environ Health 1998;24:175–82.
mortality and nested case-control study of lung cancer among structural pest control workers in Florida (United States). Cancer Causes Control 1994;5:310–38.
18 Welp EA, Weiderpass E, Boffetta P et al. Environmental risk factors of
breast cancer. Scand J Work Environ Health 1998;24:3–7. 19 Blair A, Zahm HS, Pearce NE, Heineman EF, Fraumeni JF. Clues to
cancer etiology from studies of farmers. Scand J Work Environ Health 1992;18:209–15. 20 Brooks SM, Stockwell HG, Pinkham PA, Armstrong AW, Witter DA.
Sugarcane exposure and the risk of lung cancer and mesothelioma. Environ Res 1992;58:195–203.
23 Sathiakumar N, Delzell E, Austin H, Cole P. A follow-up study of
agricultural chemical production workers. Am J Indust Med 1992; 21:321–30. 24 Stokes C S, Brace K D. Agricultural chemical use and cancer mortality
in selected rural counties in the USA. J Rural Stud 1988;4:239–47.
21 Brownson RS, Alavanja MCR, Chang JC. Occupational risk factors for
25 Hilje L, Castillo LE, Thrupp LA, Wesseling I. Pesticide Use in Costa
lung cancer among nonsmoking women: a case-control study in Missouri (United States). Cancer Causes Control 1993;4:449–54.
Rica (in Spanish). San José, Costa Rica: Editorial Heliconia/EUNED, 1987.
