The authors reported that nine relevant studies were identified: three cohort studies and six nested case-control studies. Two of the nested-control studies were from one cohort study. Five studies (two cohort studies and three nested case-control studies) were from a second cohort study. Therefore, overall the studies were derived from four cohort studies (n=70,739, range 4,581 to 26,677).
The pooled analysis found an increased risk of secondary malignant neoplasms after radiotherapy using a fixed-effect model (ERR per Gy 0.50, 95% CI 0.20 to 1.21) and a random-effects model (ERR per Gy 0.53, 95% CI 0.22 to 1.31). There was significant heterogeneity (Q=40.4, p<0.001).
The estimates were similar when ERRs were determined using a random-effects model for cohort and nested case-control studies separately, for the individual analyses of the two cohort studies that contributed more than one study to the analysis (analysis for two and five studies) and for the primary cancer sites of all cancers (three studies), all cancers other than leukaemia (two studies) and solid cancers (two studies). Confidence intervals were very wide for some of the analyses. The final pooled analysis that excluded two studies that may have included the same participants gave larger ERR estimates for both the fixed-effect (ERR per Gy 0.73, 95% CI 0.27 to 1.93) and random-effects models (ERR per Gy 0.69, 95% CI 0.25 to 1.92).
The overall ERR estimate was only slightly higher than solid cancer ERR estimates averaged over age and sex in atomic bomb survivors (Preston et al. 2007, see Other Publications of Related interest) and much lower than the ERR of solid cancer incidence in atomic bomb survivors exposed as young children (Preston et al. 2008, see Other Publications of Related Interest).
Egger’s test gave no evidence for publication bias, but the funnel plot results were less clear.