Interventions:
The Prostate Px test, nomogram, and usual care were well described. No reason was given for the selection of a 50% threshold for the nomogram, rather than the threshold used in practice or that maximised its sensitivity and specificity as for the Prostate Px test. A sensitivity analysis of the different high-risk probability predictions was conducted and mimicked the different thresholds for sensitivity and specificity for each test, given that true- and false-positives were not distinguished in this analysis. Ideally, the best thresholds would be determined using cost-effectiveness criteria. The usual care was appropriately analysed.
Effectiveness/benefits:
The authors did not report a systematic review of the literature and it was unclear whether the best available evidence was used. The effectiveness and utility values were presented, but they were not fully described. No reasons were given for the selection of the different transition probabilities. A major assumption was that the treatment effectiveness and natural history were the same (except with watchful waiting) regardless of the patient's risk profile.
Costs:
The costs were relevant to the perspective adopted. A breakdown of the unit costs and their sources were provided. No cost was given for the nomogram test and no explanation was given for this; the clinicopathologic tests were common to both procedures, but the costs for analysing these data might have differed. Only a few details of the resource composition of the cost estimates were given.
Analysis and results:
The analytic approach was satisfactorily reported, with the model structure in full and a diagram. A Markov model to combine the effectiveness and cost data from a variety of sources was appropriate for the disease. The model did not include the specificity and sensitivity of either test, nor the false-positives and false-negatives, which effectively assumes that the tests were perfect. This means that there were more low-risk patients in the cohort when using the Prostate Px test than when using the nomogram.
The effectiveness of the active treatments, such as local radiation and hormonal therapy, did not differ depending on whether the patient was a true or false low risk and whether active treatment was delayed until after progression, with watchful waiting; if the patient was a false low-risk and received delayed active treatment, it might be less effective than if it was provided earlier. While a specialist is likely to judge the patient's risk and provide appropriate treatment, the probability of receiving local radiation, hormonal therapy, or watchful waiting for patients receiving usual care did not differ by patient risk. High-risk patients received one treatment, which was the most cost-effective one calculated by the model, not the usual care nor the best practice.
The incremental analysis was appropriate for determining the cost-effectiveness of the Prostate Px test compared with the nomogram and usual care. The impact of uncertainty was appropriately explored through both one-way and probabilistic sensitivity analyses. The uncertainty in the results indicated by probabilistic sensitivity analysis is always an underestimate. Given the considerable assumptions made in this model, the sensitivity analyses could have been more informative. The results of both the base-case and the sensitivity analyses were satisfactorily reported and extensively discussed.
Concluding remarks:
There were considerable assumptions made in this analysis and the authors' conclusions do not seem to adequately account for this uncertainty, regardless of the sensitivity analyses conducted.
