31 Duration of the fertility enhancing effect of HSG with oil-based contrast 2 When censoring for IVF/ICSI, the estimated hazard ratio for oil versus water over 3 years was 1.29 (95% CI 1.11–1.50). Results from plots and tests using scaled Schoenfeld residuals were very similar to those in the primary analysis, as were the estimated hazard ratios at sequential time points (results not shown). When adjusting for baseline characteristics, the estimated hazard ratio for oil versus water over 3 years was 1.30 (95% CI 1.13–1.50). Results from plots and tests using scaled Schoenfeld residuals were very similar to those in the primary analysis, as were the estimated hazard ratios at sequential time points (results not shown). Supplementary analysis There was found no evidence that the effect of oil versus water was different for expectant management, IUI/IUI-OS or IVF/ICSI: the interaction between oil versus water allocation and MAR treatment was not significant (P = 0.39) and did not lead to a better fit in terms of AIC. DISCUSSION Evidence was found that the hazard ratio for ongoing pregnancy after an HSG with oilbased contrast versus water-based contrast was highest shortly after HSG and then gradually decreased. This change was best described as linear with log-time, decreasing from a hazard ratio of 1.71 to 1, i.e. no effect, after approximately 2 years. In the subgroup of women who experienced pain during HSG, which might be because flushing dislodged debris or mucus plugs in their Fallopian tubes, there was found no evidence for a change in hazard ratio over time. A decreasing hazard ratio over time could be due to three potential mechanisms. The first is that, for each woman, the hazard ratio for the effect of oil-based versus waterbased contrast diminishes over time. Second, a decreasing hazard ratio may also be explained by heterogeneity of treatment effect, meaning that the oil contrast may have a beneficial effect that is stable over time in only a subgroup of women. These women quickly conceived and, due to selection, at later time points the treatment effect was only evaluated in women for whom it was not beneficial, leading to a lower hazard ratio. A third possible explanation is unobserved heterogeneity, meaning that pregnancy chances varied between couples due to factors unknown to the authors. Even with a treatment effect that is constant over time and similar for all women, unobserved heterogeneity may lead to attenuation of the hazard ratio towards 1 over time (17, 18). When adjusting for baseline characteristics that are known prognostic factors, the results did not differ
RkJQdWJsaXNoZXIy MTk4NDMw