Purpose: Comparison of the results of a complex decision analytic model developed in Microsoft Office Excel® 2007 versus an implementation of the same model in TreeAge® Pro Suite 2009. The aim was to identify any difference in the performance and validity between models implemented with the two software packages in terms of incremental cost-effectiveness ratio (ICER) estimates and probabilistic sensitivity analysis (PSA) results.
Method: Decision analytic models were developed in Excel and TreeAge to assess the cost-effectiveness of a hypothetical, new therapy for a hypothetical, chronic disease. The models consisted of a decision tree (22 branches) for the first year of therapy and of a Markov model to capture long-term costs and effects. The Excel model comprised 24 decision nodes and 6 Markov models. The TreeAge implementation comprised 129 decision nodes and 22 Markov models (one Markov model for each branch). All costs, effectiveness and utility parameters were hypothetical. Relevant parameters were discounted. Probabilistic sensitivity analysis was used to assess decision uncertainty by performing 10 000 Monte Carlo simulations.
Result: The incremental cost-effectiveness ratio (ICER) of the new therapy was $75 962.164371494/QALY when calculated with TreeAge and $75 962.164371494/QALY when calculated with Excel. At a threshold value of $50 000 per QALY, the probability that the hypothetical treatment is cost-effective was 14.5% when calculated with Excel and 13.6% when using TreeAge. At the higher thresholds of $100 000 and $150 000 per QALY, the probabiliy estimates increased to 80.4% (Excel) and 80.6% (TreeAge) and 98.0% (Excel) and 98.1% (TreeAge), respectively. Computation time for the PSA was less than 5 minutes for both models.
Conclusion: Excel and TreeAge allow building and analyzing complex decision models. Although the practical implementation of a model does differ between the two software packages, the underlying decision analytic theory that has to be applied is still the same. As we showed, both model implementations yield the same results when calculating an ICER, up to the maximum precision of the TreeAge estimate (limited to 9 decimals). The difference in PSA results might have been due to their probabilistic nature (here: 10 000 iterations). Both packages have been used extensively for medical decision making. Choosing one package over the other does not need to be performance-based and can be left to personal preference.
Candidate for the Lee B. Lusted Student Prize Competition