With interest I read the article on the expected-value decision analysis of operative vs nonoperative treatment after primary traumatic anterior glenohumeral dislocation by Bishop et al.1Bishop J.A. Crall T.S. Kocher M.S. Operative versus nonoperative treatment after primary traumatic anterior glenohumeral dislocation: expected-value decision analysis.J Shoulder Elbow Surg. 2011; 20: 1087-1094https://doi.org/10.1016/j.jse.2011.01.031Abstract Full Text Full Text PDF PubMed Scopus (44) Google Scholar Their article concludes that, in general, after a primary anterior glenohumeral dislocation, management with arthroscopic stabilization (AS) is preferable over nonoperative treatment (NOT). Although I do not question the validity of the analysis performed by the authors, I do believe it should have been extended with a full analysis of the uncertainty in model outcomes. Ultimately, uncertainty in model outcomes—here the utility values for the AS and NOT strategies—may lead to incorrect decision making, such as adopting a suboptimal strategy. It is this risk of making an incorrect decision that the authors failed to address. The best way to assess uncertainty in decision analytic models is with probabilistic sensitivity analysis (PSA).2Briggs A.H. Statistical approaches to handling uncertainty in health economic evaluation.Eur J Gastroenterol Hepatol. 2004; 16: 551-561https://doi.org/10.1097/00042737-200406000-00007Crossref PubMed Scopus (50) Google Scholar, 3Briggs A.H. Sculpher M. Claxton K. Decision modelling for health economic evaluation. Oxford University Press, Oxford2006Google Scholar, 4Doubilet P. Begg C.B. Weinstein M.C. Braun P. McNeil B.J. Probabilistic sensitivity analysis using Monte Carlo simulation. A practical approach.Med Decis Making. 1985; 5: 157-177https://doi.org/10.1177/0272989X8500500205Crossref PubMed Scopus (700) Google Scholar, 7Hunink M.G.M. Glasziou P.P. Siegel J.E. Weeks J.C. Pliskin J.S. Elstein A.S. et al.Decision making in health and medicine: integrating evidence and values. Cambridge University Press, Cambridge2001Google Scholar In PSA, distributions are defined for all parameters, based on current evidence, to reflect the uncertainty in parameter estimates. Then, the effect of simultaneously varying all parameters is evaluated. It is also known that 1-way and 2-way sensitivity analyses, such as reported by the authors, provide only limited insight into the extent of decision uncertainty. Often, the effect of changing 1 or 2 individual parameters is far more limited than the effect of simultaneously changing all parameters.2Briggs A.H. Statistical approaches to handling uncertainty in health economic evaluation.Eur J Gastroenterol Hepatol. 2004; 16: 551-561https://doi.org/10.1097/00042737-200406000-00007Crossref PubMed Scopus (50) Google Scholar, 3Briggs A.H. Sculpher M. Claxton K. Decision modelling for health economic evaluation. Oxford University Press, Oxford2006Google Scholar Here, assessing uncertainty with PSA actually is straightforward, because all of the evidence required to define the distributions is readily available. The 3 transition probabilities for stiffness, infection, and recurrence after AS are based on, respectively 5, 3, and 36 events observed in 329 patients. The probability of no recurrence after NOT is based on 25 – 3 = 22 events observed in 54 – 4 = 50 patients.6Hovelius L. Olofsson A. Sandstrom B. Augustini B.G. Krantz L. Fredin H. et al.Nonoperative treatment of primary anterior shoulder dislocation in patients forty years of age and younger. A prospective twenty-five-year follow-up.J Bone Joint Surg Am. 2008; 90: 945-952https://doi.org/10.2106/JBJS.G.00070Crossref PubMed Scopus (292) Google Scholar For the utilities associated with the outcomes, the authors already provide the standard deviation (incorrectly indicated with “95% CI”) in their Table II.1Bishop J.A. Crall T.S. Kocher M.S. Operative versus nonoperative treatment after primary traumatic anterior glenohumeral dislocation: expected-value decision analysis.J Shoulder Elbow Surg. 2011; 20: 1087-1094https://doi.org/10.1016/j.jse.2011.01.031Abstract Full Text Full Text PDF PubMed Scopus (44) Google Scholar When appropriate distributions, in this case β distributions and normal distributions, are defined for these parameters, and PSA is performed, the utilities for the 2 strategies become probabilistic and are distributed as shown in Figure 1 (based on 500,000 samples). Figure 1 supports the notion that the expected value is higher for the AS strategy than for the NOT strategy because the mean values for respective distributions are 7.7 and 5.9. However, the distribution of the difference in utility, which is almost completely but not entirely positive, indicates that there is a non-zero probability that the NOT strategy actually has the highest expected value. Here, the size of the shaded area is the probability of making an incorrect decision by preferring the AS strategy and equals 0.023. In other words, given the combined uncertainty in all input parameters for the decision tree, there is a 97.7% probability that the AS strategy is optimal, that is, has higher utility than the NOT strategy. This result therefore considerably strengthens the main conclusion of the original paper. The need for a probabilistic approach can also be illustrated by investigating the original 1-way and 2-way sensitivity analyses. In their 1-way analysis, the authors conclude that NOT is the preferred strategy when the rate of recurrence after NOT falls below 32% or when the utility value for successful AS falls below 6.6. This analysis actually is a threshold analysis indicating at which (hypothetical) threshold value of a parameter the optimal strategy, based on expected-value analysis, changes. This threshold value is unrelated to the actual evidence available. Indeed, here the available evidence indicates that the probability of a recurrence rate of less than 32% after NOT is less than 0.05%. Moreover, given a recurrence rate of less than 32%, there still is a probability of approximately 47% that AS would yield a higher utility than NOT. In other words, this 32% recurrence rate is highly unlikely to occur in clinical practice, and if it were to occur, then it would be completely unclear which strategy would be optimal. Similar results hold for the utility value for successful AS, which has a 0.5% probability of falling below 6.6, and for that specific situation, a 30% probability that AS would still be optimal. The authors present a 2-way sensitivity analysis in their Figure 3. There, the optimality threshold is just a straight line and the ranges over which the parameters varied were predefined and were again not based on available evidence. Conversely, Figure 2 displays the distribution of parameter values, with larger circles indicating a larger probability of occurrence as well as the uncertainty in the optimality of strategies, with the grey part of the circles depicting the probability that AS is optimal and the black part depicting the probability that NOT is optimal. This figure captures more aspects of the uncertainty underlying the analysis and hence is more informative. As expected, around the threshold line (reproduced from the original graph), the optimal strategy is not apparent at all. Around this line, the difference in utility values is very small, and although one strategy may have a slightly higher utility value, both strategies have a substantial probability of being optimal. In fact, on the line itself, both strategies have a 50% probability of being optimal. From Figure 2 it can also be seen that, for example, based on current evidence, there is zero probability of observing a utility of uncomplicated surgery of less than 4. In addition, the sizes of the circles indicate that the rate of recurrent instability after NOT is likely to range from 0.4 to 0.7. In this range, the optimal strategy is clear only when the utility of uncomplicated surgery exceeds 9 (rendering AS optimal) or is less than 5 (rendering NOT optimal). If, on the other hand, this utility falls between 5 and 9, then both strategies may be optimal, and picking the one with the highest expected value may induce a substantial risk of making an incorrect decision. Evaluating how such a risk can be decreased by collecting additional evidence, and based on PSA, is known as value of information analysis.5Eckermann S. Karnon J. Willan A.R. The value of value of information: best informing research design and prioritization using current methods.Pharmacoeconomics. 2010; 28: 699-709https://doi.org/10.2165/11537370-000000000-00000Crossref PubMed Scopus (76) Google Scholar, 8Yokota F. Thompson K.M. Value of information literature analysis: a review of applications in health risk management.Med Decis Making. 2004; 24: 287-298https://doi.org/10.1177/0272989X04263157Crossref PubMed Scopus (124) Google Scholar Performing a full analysis of all uncertainty provides additional support for the main conclusion that AS is the preferred strategy after a primary anterior glenohumeral dislocation, thereby enhancing the interpretation and usefulness of the original article. However, for specific (combinations of) values for the rate of recurrence after NOT and utility value for successful AS, the preferred strategy is currently unclear, and informed decision making would require additional data collection. Evidently, “Nothing in life is certain, except death and taxes” (Benjamin Franklin), so appropriate and explicit assessment of all uncertainty remains paramount.