While Cahill et al are to be credited for considerable work addressing an important problem in their decision and economic analysis, the numbers don’t add up, and the conclusion is unsubstantiated.
Table 1 proposes the probability of the cervix measurement ≤15 mm is 0.0119 (0.0100-0.0168). Based upon the model hypothetical cohort population of 4 million, the number of women with cervix measurement ≤15 mm would be 47,600. If 100% of these women received progestogen, then based upon the chosen reference study (relative risk, 0.56; 95% confidence interval, [CI], 0.36–0.86 ), 7235 preterm births (PTB) would be prevented, with a 95% CI of 2292–10,479–not the 95,920 proposed in Table 4. The overestimate of PTB prevention by a factor of 13 skews the model in favor of universal sonographic screening and produces a total cost saved >13 times the reasonable optimal scenario. Furthermore, assuming that 100% of women would be screened and that 100% of women with a cervix ≤15 mm would receive progestogen would be unreasonable in a real-world effectiveness model. The CI demonstrates that, even under the optimized model, as few as 2292 PTB might be prevented.
The purpose of modeling cervical length screening combined with progestogen treatment for short cervix would be a reasonable belief that treating pregnant women with short cervix ≤15 mm with progestogen would decrease PTB–this is based upon 67 total events from 1 study.
The authors chose not to include the study of O’Brien et al in Table 1, despite the fact that it is the largest single trial of progestogens to prevent PTB and has a risk of bias equivalent or superior to the studies of Meis et al or Fonseca et al. The authors infer that they excluded the study of O’Brien et al on the basis that the progestogen formulation was different; yet O’Brien et al used vaginal progesterone and the model of Cahill et al used vaginal progestogen.
Finally, the cost-effectiveness model is heavily influenced by cost of neonatal severe morbidity, which is estimated to be >19,000 times more expensive than the advocated test (ultrasound), and advocated intervention (progestogen). Therefore, it is important to get numbers right, since slight changes in the number of severe morbidity cases would swing the cost dramatically. The trials used to support the intervention of progestogen to prevent PTB do not provide evidence of reduction in neonatal severe morbidity; therefore the model layers presumption based upon inference and the gestalt conclusion is unsupportable.