Table 5.
Heterogeneity of the oral cancer smokeless tobacco relative risk, unadjusted and adjusted for cigarette smoking (based on the estimates given in Table 3).
| Unadjusted for any factor | Adjusted for smoking | |||||||||
| Heterogeneityb | Heterogeneityb | |||||||||
| Factor | N | OR/RR (95%CI)a | χ2 | I2 | p(χ2)c | N | OR/RR (95%CI)a | χ2 | I2 | p(χ2)c |
| All estimatesd | 19 | 1.58 (1.39–1.78) | 232.1 | 92.2 | < 0.0001 | 15 | 1.31 (1.13–1.53) | 28.7 | 51.3 | 0.0114 |
| Type | ||||||||||
| Chewing tobacco | 3 | 3.20 (2.12–4.82) | 23.8 | 91.6 | < 0.0001 | 4 | 1.29 (0.99–1.69) | 5.6 | 46.2 | 0.1342 |
| Snuff | 3 | 27.9 (17.0–45.9) | 19.6 | 89.8 | 0.0001 | 4 | 1.62 (1.24–2.11) | 13.4 | 77.7 | 0.0038 |
| Smokelesse | 13 | 1.18 (1.03–1.35) | 31.5 | 61.9 | 0.0017 | 7 | 1.09 (0.85–1.42) | 5.4 | 0.0 | 0.4967 |
| Between levels | 157.3 | 67.7 | < 0.0001 | 4.3 | 15.1 | 0.1142 | ||||
| Sex | ||||||||||
| Female | 4 | 20.3 (12.6–32.5) | 35.0 | 91.4 | < 0.0001 | 2 | 2.51 (1.73–3.64) | 3.5 | 71.5 | 0.0610 |
| Male | 7 | 1.27 (1.07–1.51) | 60.6 | 90.1 | < 0.0001 | 10 | 1.15 (0.97–1.37) | 10.8 | 16.4 | 0.2925 |
| Both | 8 | 1.35 (1.11–1.64) | 16.5 | 57.5 | 0.0213 | 3 | 1.19 (0.66–2.15) | 0.6 | 0.0 | 0.7573 |
| Between levels | 120.1 | 51.7 | < 0.0001 | 13.9 | 48.4 | 0.0010 | ||||
| Study design | ||||||||||
| Case-control, hospital controls | 13 | 1.90 (1.64–2.21) | 200.4 | 94.0 | < 0.0001 | 7 | 1.41 (1.18–1.68) | 22.0 | 73.7 | 0.0009 |
| Case-control, population controls | 6 | 1.01 (0.81–1.27) | 11.0 | 54.6 | 0.0514 | 3 | 0.99 (0.69–1.42) | 0.0 | 0.0 | 0.9935 |
| Case-control, mixed controls | 0 | -- | -- | -- | -- | 2f | 1.35 (0.44–4.13) | -- | -- | 0.1337 |
| Cohort | 0 | -- | -- | -- | -- | 3 | 1.24 (0.65–2.36) | -- | -- | 0.7062 |
| Between levels | 20.7 | 8.9 | < 0.0001 | 3.0 | 10.4 | 0.3928 | ||||
| Study location | ||||||||||
| USA | 15 | 1.76 (1.54–2.02) | 215.6 | 93.5 | < 0.0001 | 10 | 1.39 (1.17–1.64) | 23.7 | 62.1 | 0.0047 |
| Scandinavia | 2 | 0.93 (0.67–1.30) | 1.1 | 5.3 | 0.3041 | 3 | 0.99 (0.68–1.45) | 0.2 | 0.0 | 0.9054 |
| Other | 2g | 0.99 (0.53–1.84) | 1.3 | 24.6 | 0.2495 | 2h | 1.35 (0.44–4.13) | 2.2 | 55.5 | 0.1337 |
| Between levels | 14.2 | 6.1 | 0.0008 | 2.5 | 8.8 | 0.2810 | ||||
| Study period | ||||||||||
| -1969 | 5 | 4.49 (3.55–5.66) | 79.8 | 95.0 | < 0.0001 | 4 | 2.02 (1.28–3.20) | 3.6 | 15.8 | 0.3129 |
| 1970–1979 | 5 | 1.05 (0.84–1.33) | 33.0 | 87.9 | < 0.0001 | 1 | 2.67 (1.83–3.90) | 0.0 | 0.0 | -- |
| 1980–1989 | 7 | 1.01 (0.83–1.23) | 10.4 | 42.4 | 0.1080 | 2 | 0.97 (0.71–1.31) | 0.0 | 0.0 | 0.8743 |
| 1990+ | 2 | 1.37 (0.64–2.96) | 0.0 | 0.0 | 0.8321 | 8 | 1.10 (0.88–1.37) | 1.8 | 0.0 | 0.9719 |
| Between levels | 108.9 | 46.9 | < 0.0001 | 23.4 | 81.4 | < 0.0001 | ||||
a Fixed-effect estimates
b For the "Between levels" rows, the I2 column contains the percentage of variation explained by the factor levels rather than the heterogeneity I2 statistic
c The probability for the heterogeneity χ2 between studies within levels is shown, except that for the "Between levels" rows, the probability is for the between levels χ2
d For each study/sex, the OR/RR for smokeless tobacco from Table 3 was included if available, otherwise that for chewing tobacco or snuff was used. For study 25 where estimates for snuff and chewing tobacco, but not smokeless tobacco, were available, that for chewing tobacco was included as it was for smokers and nonsmokers combined
e This includes all the OR/RR estimates from Table 3 where the exposure was smokeless, both from studies where only estimates for unspecified exposure were available and from studies where it had been estimated from the separate data for chewing tobacco and snuff
f In study 7 (two estimates), both hospital and population controls were used
g Studies 9 (UK) and 18 (Brazil)
h Study 7 (conducted in Puerto Rico, two estimates)
I Since only one study with an effect estimate adjusted for smoking was conducted between 1970 and 1979 (study 13), an additional assessment of heterogeneity was carried out based on two periods (-1979 vs. 1980+). The corresponding overall effect estimates, based on 5 and 10 studies respectively, were 2.39 (1.78–3.19) and 1.05 (0.88–1.26), with p < 0.0001 for between levels heterogeneity