Surgical Decision-Making in Cubital Tunnel Syndrome Stratified by Electrodiagnostic Severity: A Reproducible Markov Model With Probabilistic Sensitivity Analysis Authors: Michael J Franco Published: 2026 ABSTRACT BackgroundCubital tunnel syndrome is the second most common compressive neuropathy of the upper extremity. Existing decision analyses generally use simplified decision-tree frameworks. They do not stratify by electrodiagnostic severity and do not model medial antebrachial cutaneous (MABC) nerve morbidity as a durable health state. We developed a reproducible Markov model comparing simple decompression (SD) and transmuscular anterior transposition (TMT), stratified by demyelinating versus axonal disease.MethodsA five-cycle annual Markov cohort model was constructed from the societal perspective with 3% annual discounting. Patients entered the model as mild/moderate demyelinating or severe/axonal cubital tunnel syndrome. MABC injury was modeled as a permanent state with an annual utility decrement. Probabilistic sensitivity analysis (PSA) used 10,000 Monte Carlo iterations, beta distributions for probabilities and utilities, and gamma distributions for costs. A complete Python implementation is provided as supplementary material.ResultsIn the reproducible implementation, SD remained the preferred strategy in mild/demyelinating disease, with higher mean five-year quality-adjusted life year (QALYs) and lower mean cost than TMT. In axonal disease, TMT produced a small mean QALY advantage but remained close to equipoise, with overlapping credible intervals and a higher mean cost. The MABC risk differential narrowed in axonal disease because the SD pathway has greater downstream revision exposure. These findings support SD for demyelinating disease without subluxation and support individualized discussion of TMT in axonal disease rather than a categorical claim of superiority.ConclusionsThis model supports a stratified approach based on electrodiagnostic severity while emphasizing the uncertainty in axonal disease. The strongest conclusion is not that TMT is universally superior in axonal cubital tunnel syndrome, but that the tradeoff changes when revision risk and MABC morbidity are modeled over time. The model is hypothesis-generating and should be validated with prospective electromyography (EMG)-stratified outcome data. Introduction Cubital tunnel syndrome (CuTS) produces a clinical spectrum ranging from intermittent paresthesias to fixed motor weakness, intrinsic atrophy, and clawing [1]. Surgical options include simple in situ decompression, medial epicondylectomy, and anterior transposition [2]. Randomized trials and meta-analyses have generally not demonstrated clear superiority of transposition over decompression in the broad CuTS population, so simple decompression is often favored because it is less invasive and less costly [3,4]. That broad conclusion may not apply equally to all electrodiagnostic subgroups. Demyelinating neuropathy can recover by remyelination, whereas axonal neuropathy requires regeneration and is more vulnerable to delay, recurrence, and reoperation [1]. A decision model that treats all CuTS as one disease may therefore obscure clinically meaningful heterogeneity. The medial antebrachial cutaneous (MABC) nerve is another under-modeled variable. It crosses the operative field and may be injured during primary or revision cubital tunnel surgery [5-7]. In a revision setting, painful MABC neuroma can be a dominant source of morbidity and can complicate interpretation of persistent or recurrent symptoms [8]. For that reason, the present model treats MABC injury as a durable health state rather than as a transient perioperative complication. In this analysis, we attempt to present a transparent, reproducible Markov model comparing simple decompression (SD) and transmuscular anterior transposition (TMT) across demyelinating and axonal disease subgroups, with explicit code-based methods that can be inspected, rerun, and modified. Prior decision analyses have addressed cubital tunnel surgery but generally used simplified decision-tree frameworks without electrodiagnostic stratification or durable modeling of MABC morbidity [9,10]. Materials and methods Model structure We constructed a cohort Markov model with five annual cycles and a 3% annual discount rate applied to costs and quality-adjusted life years (QALYs) [11,12]. The model compared SD and TMT in two disease-severity strata: mild/moderate demyelinating disease and severe/axonal disease. Ulnar nerve subluxation was considered a pre-model triage criterion favoring transposition and was not treated as a probabilistic subgroup within the SD arm. The model states were symptomatic CuTS at entry, postoperative improvement, postoperative good outcome, postoperative partial recovery/plateau, failed or recurrent symptoms undergoing revision, post-revision good outcome, post-revision partial/poor outcome, and MABC injury as a permanent state. Primary MABC injury could occur during the index procedure. Secondary MABC injury could occur through the revision pathway. Patients with MABC injury remained in that state, with an annual utility decrement applied over subsequent cycles. Model implementation and reproducibility The model was implemented in Python using NumPy and pandas. The probabilistic sensitivity analysis (PSA) uses 10,000 iterations and random seed 42. Probabilities and utility values are sampled from beta distributions approximated from manuscript point estimates and plausible 95% ranges. Costs are sampled from gamma distributions. Cumulative revision probabilities were converted to annual probabilities using: p_annual = 1 - (1 - p_cumulative)^(1/4), applied across cycles two through five. The complete Python file supplied with this revision writes four outputs: raw PSA iterations, summary outputs by arm and severity, incremental TMT-versus-SD outputs, and cost-effectiveness acceptability curve (CEAC) points across willingness-to-pay thresholds from $0 to $200,000/QALY. Model inputs Table 1 summarizes the model health states and assigned utility ranges. Table 1Health-state utility inputs used in the reproducible modelCuTS: cubital tunnel syndrome.Health stateArmUtility value/rangeBasisSymptomatic CuTS - mild/moderateBoth0.72-0.74Song et al. [5]; Teunissen et al. [13]; demyelinating entry stateSymptomatic CuTS - severe/axonalBoth0.62-0.65Song et al. [5]; lower bound plus motor deficit/atrophy decrementPostoperative improvingBoth0.68-0.76 in codeEarly recovery state; intentionally conservative to avoid overestimating first-year QALYsPostoperative good outcomeBoth0.88-0.91Brauer and Graham [4]; normalized to generic utility scalePostoperative partial recovery/plateauBoth0.75-0.78Residual hand dysfunction analogFailed primary/recurrent symptomsBoth0.65-0.68Persistent postoperative symptomsMABC injury permanent stateBoth0.68-0.72 with annual decrement 0.05-0.08Chronic neuropathic pain/painful neuroma analogPost-revision good outcomeBoth0.82-0.85Revision success statePost-revision partial/poor outcomeBoth0.68-0.72Poor surgical outcome analog Health-state utility inputs used in the reproducible model CuTS: cubital tunnel syndrome. These values represent the health-related quality-of-life weights applied to each Markov state during each annual cycle. The values were derived from published CuTS decision-analysis literature, time trade-off data, hand-surgery utility literature, and modelled estimates where direct cubital tunnel-specific data were unavailable. Table 2 summarizes the transition probabilities used in the model, including primary MABC injury risk, revision risk, complication rates, and probabilities of good outcome after primary or revision surgery. Table 2Transition probabilities by surgical armMABC: medial antebrachial cutaneous.ParameterSDTMTSource/basisMABC injury - primary surgery1-2%5-8%Lowe et al. [7]; Benedikt et al. [6]; Mackinnon/Novak [9] contextOverall complication rate - primary surgery3.60%9.60%Wade et al. [14] network meta-analysisRevision - mild disease, 5-year cumulative~8%~4%Modeled from long-term revision literatureRevision - axonal disease, cumulative >=48 months11.90%3.20%Reichenbach et al. [10] meta-regressionGood outcome after primary surgery - mild disease83-85%81-83%Song et al. [5]; Brauer and Graham [4]Good outcome after primary surgery - axonal disease~60%~70%Modeled assumption; key sensitivity parameterGood outcome after revision surgery~67%~67%Revision series estimateMABC neuroma found at revision73%73%Mackinnon and Novak [9] revision series Transition probabilities by surgical arm MABC: medial antebrachial cutaneous. These values determine how patients move between Markov states over the five-year time horizon. Table 3 summarizes the cost inputs used in the model from the societal perspective. Table 3Cost inputs in 2024 USD, societal perspectiveUSD: United States dollars; OPPS: outpatient prospective payment system; CPT: current procedural terminology; MABC: medial antebrachial cutaneous; OT: occupational therapy; QALY: quality-adjusted life year. Cost parameterValueBasisSimple decompression operative cost$2,800-$4,200CMS OPPS/professional fee estimate for CPT 64718Transmuscular anterior transposition operative cost$4,500-$6,800CMS OPPS/professional fee estimate for CPT 64719 plus longer OR timeRevision cubital tunnel surgery$6,200-$9,500Revision neurolysis/transposition complexity estimatePostoperative rehabilitation$800-$1,400Standard OT course estimateMABC neuroma management$1,800-$4,200Injection series and potential operative excision estimateIndirect productivity cost - mild disease$1,200-$3,200Conservative first-year productivity/recovery estimate used in codeIndirect productivity cost - axonal disease$3,200-$7,600Return-to-work literature and societal perspective estimate Cost inputs in 2024 USD, societal perspective USD: United States dollars; OPPS: outpatient prospective payment system; CPT: current procedural terminology; MABC: medial antebrachial cutaneous; OT: occupational therapy; QALY: quality-adjusted life year. These include primary operative costs, revision surgery costs, MABC neuroma treatment costs, postoperative rehabilitation, indirect productivity costs, and willingness-to-pay thresholds. Together, Tables 1-3 define the inputs used in the PSA and Monte Carlo simulation. Results The reproducible Python implementation generated the following PSA outputs. To maximize reproducibility, the reported values are generated directly from the accompanying Python implementation (Table 4). Table 4 Reproducible PSA outputs by severity subgroup and arm (N=10,000 iterations)MABC: medial antebrachial cutaneous; SD: simple decompression; TMT: transmuscular anterior transposition; QALY: quality-adjusted life year; PSA: probabilistic sensitivity analysis; CrI: credible interval.SeverityArm5-year QALYs, mean (95% CrI)5-year cost, mean (95% CrI)Lifetime MABC probabilityMildSD3.941 (3.880-4.003)$6,490 ($5,444-$7,660)2.40%MildTMT3.872 (3.809-3.932)$9,282 ($7,762-$10,972)7.00%AxonalSD3.812 (3.742-3.883)$15,392 ($11,262-$20,788)4.90%AxonalTMT3.817 (3.747-3.885)$15,713 ($11,955-$20,452)7.20% Reproducible PSA outputs by severity subgroup and arm (N=10,000 iterations) MABC: medial antebrachial cutaneous; SD: simple decompression; TMT: transmuscular anterior transposition; QALY: quality-adjusted life year; PSA: probabilistic sensitivity analysis; CrI: credible interval. In mild/demyelinating disease, SD generated higher QALYs and lower costs than TMT, making SD the dominant strategy in the reproducible model. This finding is consistent with current practice patterns favoring decompression for patients without subluxation or advanced axonal loss [2-4] (Table 5). Table 5Incremental analysis of TMT versus SDCE: cost-effective; ICER: incremental cost-effectiveness ratio; QALY: quality-adjusted life-year; SD: simple decompression; TMT: transmuscular anterior transposition. In the mild/demyelinating stratum, TMT yielded fewer QALYs at a higher cost and is therefore dominated, so a positive ICER is not defined. In the axonal stratum, the incremental QALY difference straddled zero across iterations, rendering the ICER ratio uninterpretable; the probability of cost-effectiveness and the dominance frequencies are reported instead. Values are means across 10,000 Monte Carlo iterations.SeverityDelta QALYDelta costMedian ICERP(TMT CE) at $50k/QALYP(TMT CE) at $100k/QALYTMT dominantTMT dominatedMild-0.069$2,793Dominated0.60%1.80%0.10%93.60%Axonal0.005$321Not meaningful50.00%51.70%28.80%29.00% Incremental analysis of TMT versus SD CE: cost-effective; ICER: incremental cost-effectiveness ratio; QALY: quality-adjusted life-year; SD: simple decompression; TMT: transmuscular anterior transposition. In the mild/demyelinating stratum, TMT yielded fewer QALYs at a higher cost and is therefore dominated, so a positive ICER is not defined. In the axonal stratum, the incremental QALY difference straddled zero across iterations, rendering the ICER ratio uninterpretable; the probability of cost-effectiveness and the dominance frequencies are reported instead. Values are means across 10,000 Monte Carlo iterations. In axonal disease, TMT generated a small mean QALY advantage but at higher mean cost. The credible intervals overlapped substantially, and the probability of cost-effectiveness depended on the willingness-to-pay threshold and on the modeled good-outcome differential between TMT and SD in axonal disease. This result should therefore be interpreted as equipoise with a possible advantage for TMT in selected axonal patients rather than as definitive evidence of superiority. The MABC comparison changed across severity strata. In mild disease, SD had lower lifetime MABC exposure. In axonal disease, the difference narrowed because SD had greater downstream revision exposure, and revision surgery carries a high probability of encountering or treating MABC neuroma in the published revision series [8,15,16]. Discussion This analysis expresses the clinically important idea of electromyography (EMG)-severity stratification, and aims to create a reproducible Markov model. The most stable conclusion is that SD remains the preferred operation for mild-to-moderate demyelinating CuTS without subluxation. It is less costly, less invasive, and associated with lower cumulative MABC exposure in this subgroup. The axonal subgroup is more uncertain and should be presented that way. The rationale for considering TMT in axonal disease is not that the procedure is uniformly superior. Rather, the rationale is that axonal disease changes the balance of risks: recurrent compression and revision surgery may be more consequential, and the long-term MABC advantage of SD narrows once revision exposure is modeled [16]. In this setting, TMT may function as a more definitive operation for selected patients [13], but this hypothesis requires prospective validation. Limitations The model relies on published aggregate data and several modeled assumptions rather than patient-level EMG-stratified comparative trial data. The axonal subgroup results are especially sensitive to the assumed good-outcome differential between TMT and SD. The utility decrement for MABC injury is estimated from neuropathic pain analogs rather than directly measured in patients with confirmed MABC neuroma. The 73% MABC neuroma figure comes from a revision population and should not be interpreted as the primary-procedure iatrogenic injury rate [8]. Generic utility instruments may underestimate meaningful hand-specific improvement after cubital tunnel surgery [16]. Finally, the transmuscular technique modeled here may not generalize to all anterior transposition techniques or all surgeon experience levels [13,17]. Conclusions This reproducible Markov model supports SD as the dominant strategy for mild/demyelinating CuTS without subluxation. In axonal disease, the model suggests clinical equipoise with a possible role for TMT in selected patients because revision risk and cumulative MABC exposure change the long-term tradeoff. Subluxation remains an independent indication for transposition. The analysis should be interpreted as a quantitative framework for shared decision-making and as a rationale for prospective EMG-stratified comparative research.