Probabilistic Sensitivity Analysis
We conducted a probabilistic sensitivity analysis (PSA) in which we varied all major parameters in the analysis based on random draws from each parameter’s respective distribution in a Monte Carlo simulation. Probabilities and proportions were varied based on beta distributions, large cost estimates were based on lognormal distributions, and small costs and multiplier parameters were varied based on a normal distribution as shown in Tables 17.1, 17.2, and 17.3. Distribution parameters were imputed based on mean values and confidence intervals where known. Where no confidence interval was known, we assumed a 50% change in the point estimate represented a deviation of 2 standard deviations from the mean. We re-sampled all parameters simultaneously while recalculating the costs for 10,000 iterations. We calculated a 95% credible interval of results representing the 2.5 and 97.5 percentile values of the output. Credible intervals, sometimes referred to as probability intervals, are the Bayesian statistical alternative to the more widely referenced confidence interval from frequentist statistics.
Table 17.4 and Figure 14 show the results of the PSA. The 95% credible interval of total costs is $112 billion to $174 billion. The preponderance of uncertainty in the results is attributable to the productivity losses of vision loss, which in turn is a function of uncertainty in the prevalence of vision loss and the average reduction in productivity for persons with vision loss. The cost of diagnosed disorders, nonmedical vision aids, and lost productivity due to informal care were the next most uncertain values, with credible intervals spanning over $1 billion each.