Clotting Trajectories After Aortic Dissection Surgery Reveal Hidden Mortality Patterns

When the wall of the body’s largest artery tears, every hour counts. Acute type A aortic dissection (ATAAD) is one of cardiovascular medicine’s most feared emergencies: the dissection propagates along the ascending aorta, threatens branch vessels and the heart itself, and demands emergency surgical repair. Yet even in contemporary practice, early mortality remains substantial, and the perioperative course is complicated by a hemostatic storm that begins before the first incision. A new study from the First Affiliated Hospital of Shantou University Medical College in Guangdong, China, published in Frontiers in Medicine, offers a fresh longitudinal perspective on this storm by tracking how one routine coagulation marker evolves from the last preoperative measurement through the first postoperative week — and by asking whether the shape of that evolution carries information about subsequent survival.

The marker in question is prothrombin time activity, or PTA, a prothrombin-time result expressed as a percentage of normal activity. PTA reflects the extrinsic and common coagulation pathways rather than any single factor, and its interpretation depends on the analytical system, including the thromboplastin reagent and calibration used by each laboratory. In patients with ATAAD, preoperative coagulation abnormalities have previously been linked to disease severity and adverse outcomes. But the dissection process itself exposes tissue factor and activates coagulation and fibrinolysis before surgery, while cardiopulmonary bypass, hypothermia, circulatory arrest, operative trauma, hemorrhage, and transfusion all further perturb hemostasis. An abnormal result may reflect the presenting dissection, physiological stress, operative exposure, administered treatment, or a combination of these influences — which makes coagulation status both clinically relevant and difficult to summarize with a single measurement.

The research team, led by Hongle Yan, Xuejun Shen, Huozhen Weng, Weiping Li, and Shengli Yang, reasoned that serial PTA measurements might follow different longitudinal morphologies. Patients who begin with similar postoperative values may have arrived there from markedly different preoperative levels, and patients with comparable preoperative values may diverge after surgery. A single value cannot represent these paths. To capture them, the investigators turned to group-based trajectory modeling, a descriptive statistical framework for identifying latent groups of patients with similar within-person patterns, provided that the number and shape of groups are selected with attention to fit, classification, stability, parsimony, and clinical interpretability.

The study was a single-center retrospective cohort of consecutive adults with computed tomography angiography-confirmed ATAAD who underwent surgical repair between January 2021 and December 2023. Hospital records identified 1,002 patients with aortic dissection; 522 had Stanford type A dissection, 444 met the acute definition, and 285 surgically treated patients met clinical eligibility criteria. One was excluded because essential perioperative laboratory data were unavailable, leaving 284 patients in the final cohort. The dataset contained 4,037 PTA measurements, with a median of 12 measurements per patient. Fourteen patients died during postoperative days 0 through 7 and one was censored on postoperative day 6, leaving 269 patients observable immediately after postoperative day 7. One patient had insufficient longitudinal information for trajectory assignment, yielding a trajectory cohort of 268 patients contributing 1,895 observed values across the last-preoperative and postoperative-day-1-to-7 window.

The outcome-blind modeling identified four reproducible perioperative PTA trajectory groups. The moderate-stable group, comprising 87 patients or 32.5 percent, began with moderately reduced PTA and remained comparatively stable. The low-declining group, 50 patients or 18.7 percent, entered surgery with lower PTA and declined further. The preserved-rapidly-declining group, 61 patients or 22.8 percent, started at a relatively preserved level but showed a marked postoperative fall. The high-preserved group, 70 patients or 26.1 percent, maintained higher PTA throughout the observation window. Classification diagnostics were strong: entropy was 0.830, group-specific average posterior probabilities ranged from 0.897 to 0.922, and all 30 multistart fits reached the same optimum. Although a five-group model had marginally better information criteria, it added a level split and showed poorer bootstrap stability, with 23 unstable and three collapsed solutions among 200 successful repetitions compared with five unstable and no collapsed solutions for the four-group model. In a high-completeness subgroup of 235 patients, agreement with the primary four-group assignment was 98.7 percent.

Mortality analyses began immediately after postoperative day 7 to preserve temporal ordering between the exposure window and subsequent outcome — a landmark design that avoids immortal time bias. During follow-up, 27 deaths occurred among the 268 trajectory participants: 9 of 87 in the moderate-stable group, 11 of 50 in the low-declining group, 6 of 61 in the preserved-rapidly-declining group, and 1 of 70 in the high-preserved group. Kaplan-Meier cumulative mortality by postoperative day 365 differed globally across groups, with a log-rank P value of 0.00210. The estimated cumulative mortality was 10.5 percent in the moderate-stable group, 24.0 percent in the low-declining group, 10.0 percent in the preserved-rapidly-declining group, and just 1.4 percent in the high-preserved group. Trajectory group remained globally associated with subsequent mortality in the age- and sex-adjusted Cox model, with a global P value of 0.0322, and in a Firth-penalized sensitivity analysis designed to address sparse events, with a global P value of 0.0481.

But the story is more nuanced than a simple risk ranking. The high-preserved reference group contained only one death, making class-specific hazard ratios imprecise and unreliable. The investigators had prespecified a direct comparison between the two declining morphologies — the low-declining and preserved-rapidly-declining groups — after the trajectory solution was selected but before survival outcomes were examined. That comparison did not demonstrate a clear difference: the age- and sex-adjusted hazard ratio was 1.543 with a 95 percent confidence interval of 0.544 to 4.378, and the Firth hazard ratio was 1.503 with a confidence interval of 0.558 to 4.378. Both groups reached low PTA levels during the postoperative window, but they arrived there through different paths — one from an already reduced preoperative level and the other after a pronounced postoperative decline. This distinction illustrates how a longitudinal trajectory retains information about temporal morphology that a single postoperative measurement cannot fully represent, yet it does not establish different mortality tiers.

Perhaps the most sobering finding came from the incremental analysis. Adding the four-group trajectory classification to age, sex, and the last preoperative PTA did not meet the prespecified likelihood-ratio criterion, with a P value of 0.304 and a Firth penalized-likelihood P value of 0.380. Optimism-corrected performance differences between models with and without trajectory membership were small and uncertain: the difference in Harrell’s C statistic was 0.008 with a bootstrap interval spanning negative to positive values, and the difference in the area under the time-dependent curve at day 358 was less than 0.001. In other words, the last preoperative PTA already contained prognostic information that was not clearly augmented by trajectory-group membership in this sample. The simple PTA analyses reinforced this boundary: postoperative-day-1 PTA was associated with subsequent mortality, with a hazard ratio of 0.800 per 10 percentage points, but the preoperative-to-postoperative-day-1 change itself did not show clear incremental information beyond the preoperative value.

The clinical context of the trajectory groups was equally revealing. Aortic cross-clamp duration differed across groups, and Bentall procedures and mechanical valves were descriptively more frequent in the two declining trajectory groups than in the moderate-stable and high-preserved groups. Because mechanical-valve replacement entails a distinct postoperative antithrombotic context, the investigators refitted the trajectories after outcome-blind exclusion of mechanical-valve recipients. The broad four-pattern structure persisted among 197 patients, with high agreement with the primary assignment, although the rapid decline was attenuated. This indicates that the overall structure was not confined to mechanical-valve recipients, but it does not establish independence from postoperative antithrombotic treatment, because patient-level treatment exposure was unavailable. The trajectories are therefore best understood as composite perioperative patterns reflecting the combined influence of baseline disease state, anatomy, operative exposure, bleeding and transfusion, and postoperative treatment — not as discrete readouts of intrinsic coagulation biology.

The authors are candid about the study’s limitations. PTA was obtained during routine care rather than on a research-mandated schedule, and measurement frequency varied with early death, continuous renal replacement therapy, delayed sternal closure, and the intensity of intensive care — meaning the observation process may have been informative. A complementary flexible joint longitudinal-survival model that could have accommodated irregular measurements failed to meet predefined MCMC convergence criteria, so no association estimate from that model was reportable. Nonconvergence, the authors stress, should not be interpreted as evidence of no association. Residual confounding by baseline disease severity cannot be excluded, and external reproducibility remains to be established. The clinical implications are deliberately modest: characterizing perioperative PTA trajectories may provide a useful way to conceptualize heterogeneous longitudinal patterns and to structure postoperative reassessment or research stratification, but trajectory-group assignment requires observations through postoperative day 7 and is unavailable at surgery. The study did not evaluate a real-time classifier, treatment guidance, or PTA-guided anticoagulation. Prospective multicenter studies with standardized sampling, recoverable assay metadata, time-aligned treatment data, and sufficient events are needed before longitudinal PTA morphology can be considered for clinical implementation. Until then, the four trajectories stand as a hypothesis-generating map of the perioperative coagulation landscape — one that shows where patients travel, even if it cannot yet predict who survives the journey.

Subject of Research: Perioperative coagulation trajectories and mortality after acute type A aortic dissection repair

Article Title: Perioperative prothrombin time activity trajectories and subsequent mortality after surgical repair of acute type A aortic dissection

Article References: Yan, H., Shen, X., Weng, H., Li, W., & Yang, S. (2026). Perioperative prothrombin time activity trajectories and subsequent mortality after surgical repair of acute type A aortic dissection. Frontiers in Medicine, 13, Article 1885240. https://doi.org/10.3389/fmed.2026.1885240

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1885240

Keywords: acute type A aortic dissection, prothrombin time activity, group-based trajectory modeling, landmark analysis, cardiac surgery, coagulopathy, cardiopulmonary bypass, all-cause mortality, retrospective cohort study, Frontiers in Medicine, aortic repair, hemostasis