Nineteen studies were included: 8 cohort studies (although one of these appears to be a randomised clinical trial) and 11 case-control studies.
Ever use versus never use of HRT (n=19 studies).
There was substantial heterogeneity across studies, this remained for subsets of studies defined by design, year of publication and adjustment for confounding (p ranged from greater than 0.001 for all studies to 0.36 for studies published before 1990). The summary RR obtained using a fixed-effect meta-analysis was 0.82 (0.76, 0.89) indicating a protective effect of the use of HRT on colorectal cancer. The summary RR from the random-effects model was 0.81 (95% CI: 0.70, 0.93). Summary measures of effect varied little across subgroups (fixed-effect RRs ranged from 0.76 for studies which controlled for confounding to 0.93 for studies published before 1990).
Current verus past use of HRT - n= 6 studies.
Studies of current use showed evidence of heterogeneity (p=0.07), studies of past use appeared to be fairly homogeneous (p=0.48). The summary estimate of the RR obtained for current use was 0.65 (95% CI: 0.54 to 0.79) this was less than that for past use (RR=0.78, 95% CI: 0.69, 0.88). Estimates for sub-groups within these categories (cohort studies only, case-control studies only and studies published after 1990) were similar to the overall estimates.
Long (greater than or equal to 5 years) versus short-term (greater than 5 years) use of HRT.
Both sub-sets of studies showed evidence of heterogeneity (p=0.04 for those of short-term use and p=0.02 for long-term use). The summary estimate of the RR obtained for short-term use was 0.84 (95% CI: 0.73, 0.97), this was greater than that for long-term use (RR=0.69, 95% CI: 0.58, 0.82). All RR estimates for the subgroup (cohort studies only, case-control studies only and studies published after 1990) analyses were greater for studies of short-term duration compared to those of long-term duration when the estimates for similar sub-groups were compared.