Journal of NeuroEngineering and Rehabilitation

Data sources Literature review PubMed, Embase, CINAHL, PsycINFO, Web of Science, and Google Scholar were searched for relevant peer-reviewed articles. References of the identified litera- ture were manually searched for additional publications. Non-peer reviewed literature such as technical reports produced by government agencies or industry associa- tions was also examined. For each input parameter, a range of estimates was compiled from the literature whenever possible, where the median value served as the base case in the simulation model, while the upper and lower bounds were used in the sensitivity analysis. Expert panel process An expert panel was convened to supplement the litera- ture review, to validate the assumptions made, to ensure adequate and complete understanding of the prosthetics literature, and to ensure appropriate model development and construction. In addition, when the model parame- ters were not available in the published literature, ex- perts were asked to provide estimates for such parameters. Fifteen experts were selected based on their publication record in the various topics that informed the simulation model. Telephone-based panel discus- sions and one-on-one interviews were conducted on an as-needed basis. Cost of device acquisition The cost of device acquisition is approximated using the current Medicare payment amount. Therefore, it does not necessarily represent the manufacturer list price. The base case value was based on expert input and the upper and lower bounds were derived from the 2016 Medicare fee schedule allowed payments [ 20 ] for the two most frequent combinations of L codes, which were identified among the new unilateral transfemoral ampu- tees in the 2011 – 2014 Medicare claims data. The me- dian of the Medicare allowed payments in the 2 years after the device fitting in the same Medicare population was used as the cost of device repair and physical ther- apy. Dobson & DaVanzo LLC conducted all the Medi- care claims data analyses. Simulation model A cohort-level Markov model [ 21 , 22 ] was developed to simulate the clinical and economic outcomes for a uni- lateral transfemoral amputee population. This hypothet- ical cohort was assigned to two different treatment strategies, NMPK or MPK, with all other prosthetic components being the same. The simulation was limited to 10 years because the existing evidence comes from relatively short-term studies, meaning that longer-term predictions can be subject to large uncertainty. All health and cost outcomes were discounted to the present time using a 3% discount rate. Because the data available from the literature permit- ted the conversion of only two clinical conditions, falls and osteoarthritis, into economic benefits, two modules were constructed for the model respectively. The lack of data, however, prevented the quantification of potential benefit derived from other medical conditions, such as obesity and cardiovascular diseases. In the fall module, patients can experience three health states: fall, no fall, and death. Falls can be either medical, i.e., require medical attention, or non-medical. Medical falls can be minor, major, or fatal. Major injurious falls are associated with an admission to a medical facility. A patient may enter the “ death ” state from the “ no fall ” state due to causes other than falling. While Markov models are “ memoryless, ” meaning the health state at a subsequent step depends only on the state at the previ- ous step, the model updates the annual probability of falling to simulate the effect of learning. The osteoarth- ritis module has three states as well: no osteoarthritis, osteoarthritis, and death. Patients can move from one state to another until the end of the 10-year time period or death. After implementing the model, validation testing was performed to assure that the computations were done correctly, and the outputs responded as ex- pected to changes in key parameter input values. The model was programmed in Visual Basic for Applica- tions for Microsoft Excel and followed the modeling guidelines of the International Society for Pharmacoe- conomics and Outcomes Research [ 23 ] . Model parameters were compiled from the literature review, expert consultation, and the analysis of Medi- care claims data. When parameters were not available from the published literature, expert opinion was used and, if needed, assumptions were made. The model parameters, assumptions, and data sources are listed in Table 1 . One-way sensitivity analyses were conducted where one input parameter was changed at a time to inspect the sensitivity of model results to changes in key in- put parameter values as they were varied individually. Probabilistic sensitivity analyses on model inputs with 1000 replications, assuming uniform distributions for all variables, were also conducted. Results Clinical benefits Physical function A number of studies assessed the effects of advanced prosthetics by measuring biomechanical and physical performances when subjects wore NMPK and after sub- jects were fitted with MPK. Overall, there is strong Chen et al. Journal of NeuroEngineering and Rehabilitation 2018, 15 (Suppl 1):62 Page 51 of 72