A Compositional Simulation Framework for Estimating the Potential Mortality Benefit of an Integrated Precision Surgical Protocol (IPSP) in High-Risk Surgery

Background. High-risk neurosurgical, cardiovascular, and orthopedic procedures carry substantial preventable mortality: 30-day mortality of 20.3% has been reported for craniotomy after traumatic brain injury [1], in-hospital mortality of 1.62–4.97% for coronary artery bypass grafting [3, 4], and 1-year mortality of 4.22% (5-year, up to 21%) following periprosthetic joint infection [5]. Enhanced Recovery After Surgery (ERAS) protocols reduce total complications (odds ratio [OR] 0.50) [7] but remain largely static and non-adaptive. Objective. To specify the Integrated Precision Surgical Protocol (IPSP), a compositional simulation framework in which the population-level relative risk reduction (RRR) in mortality is not assigned in advance but is instead computed from independently sourced, cause-specific intervention effects and their attribution to baseline mortality pathways, so that the result cannot trivially reproduce a target value chosen by the investigator. Methods. Baseline perioperative mortality (Beta(50,950) prior, mean 5.00%, an illustrative pooled rate consistent with the range of cause-specific rates in the cited high-risk-surgery literature) is decomposed into three cause-specific pathways — neuro-cerebrovascular, cardiovascular, and infection/technical-surgical — each carrying its own literature-anchored or explicitly flagged assumption-based relative risk (RR) for the corresponding IPSP intervention. Attribution of baseline mortality across pathways is treated as an uncertain quantity (Dirichlet-distributed) rather than a fixed input, and its influence on the result is stress-tested directly. Monte Carlo simulation (200 replications, n = 10,000 patients/replication), global sensitivity analysis (Sobol/Saltelli indices via the Jansen 1999 estimator), a five-scenario robustness sweep on the attribution assumption, a manually implemented Metropolis–Hastings Bayesian analysis, and logistic regression fitted (not asserted) on simulated patient-level data are reported. Results. The simulation yields a baseline mortality of 5.04% (200-replication 95% interval 3.77–6.54%) and an IPSP-arm mortality of 3.41% (2.29–4.79%), corresponding to a mean RRR of 32.2% (95% interval 15.7–47.0%; mean Z = 5.75; mean p = 1.5 × 10−3). Replacing clipped log-normal sampling with properly truncated log-normal sampling for the pathway-specific relative risks (removing an artificial boundary pile-up) reveals that 1 of 200 replications does not reach conventional significance (p = 0.235), a genuine tail-risk finding that the earlier clip-based implementation obscured. A five-scenario sweep of the attribution assumption shifts the mean RRR between 25.0% and 37.5%, and a complementary continuous sweep of the Dirichlet concentration parameter (holding the mean attribution fixed) shows the point estimate is stable (31.3–33.1%) across a more than 100-fold range of attribution-certainty, indicating that the earlier scenario-to-scenario variation is driven by the assumed mean attribution, not by how confidently it is specified. Global sensitivity analysis identifies the infection/technical-pathway RR as the dominant driver of output variance (total-order Sobol index ST = 0.52), followed by the neuro-pathway RR (ST = 0.25) and the attribution assumption itself (ST = 0.15); these indices are unaffected by the truncation correction, since the Sobol analysis samples each factor independently over its declared range rather than via the clipped/truncated log-normal mechanism. A dedicated pessimistic-bridging stress test, fixing the two weakest-evidence pathway effects at their least favorable plausible values, yields a mean RRR of 23.0% (95% interval 9.9–36.8%); because this stress test targets exactly the two inputs identified by the Sobol analysis as dominating output uncertainty, 23.0%, not the base-case 32.2%, is treated throughout this manuscript as the more defensible current working estimate. A Bayesian re-analysis of the sampling uncertainty alone (Metropolis–Hastings, 39.9% acceptance) gives a posterior mean RRR of 32.2% (95% credible interval 22.4–41.0%) — narrower than the full Monte Carlo interval because it does not include the structural parameter uncertainty that the latter propagates; this distinction is reported explicitly rather than conflated. Conclusion. Under the assumptions disclosed and tested here, IPSP’s components are compatible with a perioperative mortality reduction on the order of 16–47% under full propagated uncertainty, but the base-case mean of 32.2% is an optimistic figure that assumes the two weakest, least-verified pathway assumptions perform at their literature-anchored median; a dedicated pessimistic-bridging stress test that instead fixes those two assumptions at their least favorable plausible values yields a mean RRR of 23.0% (95% interval 9.9–36.8%), and this lower, stress-tested figure, not the base-case estimate, is the more defensible current projection. In all cases this remains a simulation-derived, assumption-dependent projection, not an empirical finding, and several of its inputs are explicitly acknowledged as unverified bridging assumptions rather than direct clinical evidence. The framework is falsifiable: its primary criterion is refuted if an adequately powered randomized trial fails to demonstrate a mortality or major-complication RRR of at least 22% (set at the pessimistic-bridging estimate itself, rather than at the more optimistic base case), with a secondary, more permissive floor of 15% and a directional criterion described in full in Section 7. This work does not constitute evidence of clinical efficacy and should not be interpreted as supporting deployment of IPSP prior to the prospective validation described there.