OBJECTIVES: Typically, in PSA, we use the pseudo-random (PR) number generator embedded in the software used to develop the model. The impact of the PR generator has no noticeable influence if the simulation requires a small data package (below 10,000 records). Once this threshold is crossed, a certain disbalance of the generated values is possible. We aimed to perform a comparison between PSA output from PR and natural random (NR) values.

METHODS: We developed a random seed functionality that operates on pre-defined values set for every calculation. Thus, we can import random values from any generator. For the NR numbers, we used the service of Random.org, providing values processed observation of natural phenomena and processing it into the requested format. A similar simulation was performed for the PSA using a NR numbers and PR generator. All the functionalities were developed within the Excel environment with the use of a recently developed model for NASH.

RESULTS: Comparison and density analysis of the PSA outcomes have shown that there is a difference between the two methods of data generation. The spread of the points is more uniform in NR-generated numbers. However, the reached averaged incremental QALYs and costs between the PR and NR generator were similar (PR:0.156 QALY, -9198.34$; NR: 0.158, -9105.49$) while comparing the same treatments in PSA. PR outcomes fitted in a range between QALY(0.038,0.278) and cost(-12033.83; -6501.20); and for the NR, it fitted in range QALY(0.040; 0.290) and cost (-11969.12; -6531.87). No significant difference in the probability of being cost-effective was observed between PSA outcomes with PR and NR generators.

CONCLUSIONS: The impact of pseudo-random numbers seems to be negligible from the perspective of the results of analyzed cost-effectiveness model. However, it has a noticeable effect on the density aggregation of the PSA outcomes and therefore should be explored in different model types.