Hidden Patient Profiles Revealed Behind Serious Bleeding After Dental Surgery

A routine tooth extraction is supposed to end with a bit of soreness and a follow-up appointment. For a small but clinically important group of patients, however, the aftermath is far more dramatic: bleeding that refuses to stop and ends with an emergency admission to a hospital ward. A new retrospective study from University Hospital Schleswig-Holstein in Kiel, Germany, has taken one of the most detailed looks yet at who these patients actually are, and the results paint a picture of hidden complexity lurking behind what many consider minor surgery.

The research, published in Frontiers in Systems Biology, analyzed 700 patients hospitalized for postoperative intraoral bleeding after dental or oral surgery between January 2011 and December 2021. Rather than treating the cohort as a single uniform group, the team applied a statistical technique called latent class analysis to uncover recurring patterns of chronic disease hiding within the data. What emerged were five distinct multimorbidity classes, each with its own constellation of cardiovascular, metabolic, renal, and gastrointestinal conditions, and each with a tellingly different pattern of blood-thinning medication use.

The demographic profile of the cohort is striking on its own. The median age of hospitalized bleeding patients was 74 years, and nearly two-thirds, 65.7 percent, were men. Median hospital stay was just two days, suggesting that most episodes were stabilized quickly, but the underlying medical burden was enormous: 74.9 percent had documented cardiac disease, 48.0 percent had hypertension, 17.6 percent had diabetes, and roughly one in seven had undergone heart valve replacement or suffered from kidney disease. Critically, 83.0 percent were taking at least one antithrombotic medication at admission, meaning these patients were deliberately kept in a state of reduced clotting ability to protect them from heart attacks and strokes.

The study also captured a quiet revolution in prescribing practice. Over the eleven-year window, the balance shifted decisively from vitamin K antagonists, the older warfarin-style anticoagulants that require regular blood monitoring, toward direct oral anticoagulants, or DOACs, the newer agents that have swept into routine care for atrial fibrillation and venous thromboembolism. VKA monotherapy accounted for 36 hospitalized bleeding cases in 2011 and fell to just seven by 2021, while DOAC monotherapy rose from zero cases in 2011 to 25 in 2020. Combined antiplatelet and VKA therapy disappeared from the records entirely after 2016. The authors are careful to stress that these are counts among patients who bled, not medication-specific bleeding rates, because the number of patients taking each drug and undergoing dental procedures in the wider population was unknown.

The analytical heart of the paper lies in the latent class analysis. Of the 700 patients, 458 had at least two documented comorbidities and were subjected to a probabilistic modeling exercise using eleven binary indicators covering conditions from cardiac disease and hypertension to deep venous thrombosis, heart valve replacement, stroke, peripheral arterial disease, diabetes, renal disease, gastrointestinal disease, rheumatic disease, and hepatic disease. Latent class analysis works by assuming that observed patterns of diagnoses arise from a small number of unobserved patient subtypes, and it estimates both the characteristics of those subtypes and the probability that each individual belongs to one. The researchers compared models containing one to six classes using information criteria, entropy, a bootstrap likelihood ratio test, class sizes, and clinical interpretability.

The five-class solution emerged as the winner, though not without statistical tension. It had the lowest Akaike information criterion of any candidate, high entropy of 0.871 indicating precise classification, and a significant test favoring five classes over four. Yet the Bayesian information criterion actually preferred a two-class model, and the sample-size-adjusted variant preferred four. The six-class model failed outright, with a nonsignificant likelihood ratio test. The authors transparently report this disagreement, noting that the five-class solution should be considered an exploratory description requiring external replication rather than a definitive taxonomy.

The classes themselves are clinically evocative. Class 1, a small group of 26 patients, was defined by deep venous thrombosis together with cardiac disease and hypertension. Class 2, with 97 patients, was anchored by heart valve replacement and cardiac disease. Class 3, 55 patients, combined stroke, cardiac disease, and hypertension. Class 4, the largest at 151 patients, was characterized by hypertension and cardiac disease, while Class 5, 129 patients, displayed a broad multisystem cardiometabolic profile spanning cardiac disease, hypertension, diabetes, renal disease, and gastrointestinal disease. When the researchers cross-referenced medication use against these classes, a coherent pattern appeared: VKA use was concentrated in the thrombotic and valvular classes, where it reached 73.1 and 77.3 percent of patients respectively, while DOAC use peaked at 25.8 percent in the hypertensive-cardiac class. Antiplatelet monotherapy was more common in the cerebrovascular and multisystem classes.

That alignment between disease profile and drug choice is exactly what one would expect from rational prescribing, since the underlying indication drives anticoagulant selection, and it lends the latent classes face validity even though they were derived purely from diagnostic patterns. Stability testing through 100 bootstrap resamples yielded a mean adjusted Rand index of 0.607, a measure of classification agreement that the authors describe as moderate rather than high. Profile correlations were strong for the valvular and hypertensive-cardiac classes, above 0.96, but weaker for the small venous-thrombotic class and the cerebrovascular class, which is unsurprising given their smaller sizes and the inherent uncertainty of assigning patients probabilistically.

The clinical implications reach into one of the most persistent dilemmas in dental medicine: what to do with patients on blood thinners who need oral surgery. Stopping anticoagulation carries real dangers, including stroke and valve thrombosis, while continuing it raises the specter of uncontrolled bleeding. Existing evidence generally supports continuing antithrombotic therapy for many minor oral procedures with appropriate local hemostatic measures such as suturing, hemostatic dressings, and tranexamic acid mouthwashes, and the new study did not test any specific perioperative strategy. What it does provide is a richer description of the patients at the severe end of the bleeding spectrum, reinforcing that decisions must weigh the full comorbidity picture, the antithrombotic indication, the invasiveness of the procedure, and the available local measures together, ideally in interdisciplinary consultation.

The authors are careful about the limits of their design. Because the cohort contained only patients who had already bled badly enough to be hospitalized, with no comparator group of uneventful procedures, the study cannot identify risk factors, estimate medication-specific bleeding rates, or support causal claims. The admission INR of 1.28 reported for the cohort is a descriptive average across all patients, including those not on VKAs, and cannot be read as a measure of anticoagulation quality. Still, the scale of the dataset, its eleven-year span, the systematic dual-review chart abstraction, and the transparent handling of missing data give the findings real weight. Future work, the authors argue, should pair bleeding patients with matched controls, capture procedural denominators and medication exposure in the source population, and validate the multimorbidity classes in independent cohorts before they can be used to predict who is most likely to end up in a hospital bed after a trip to the dentist.

Subject of Research: Multimorbidity patterns and antithrombotic medication use in patients hospitalized for bleeding after dental surgery

Article Title: Latent multimorbidity classes and antithrombotic medication distributions among patients hospitalized for bleeding after dental surgery: a retrospective study

Article References: Wagner, J., Spille, J. H., Flor, J., Ouyang, Z., Conrad, J., Dörfer, C., Wiltfang, J., & Naujokat, H. (2026). Latent multimorbidity classes and antithrombotic medication distributions among patients hospitalized for bleeding after dental surgery: a retrospective study. Frontiers in Systems Biology, 6, Article 1917159. https://doi.org/10.3389/fsysb.2026.1917159

Image Credits: AI Generated

DOI: 10.3389/fsysb.2026.1917159

Keywords: latent class analysis, multimorbidity, dental surgery, postoperative bleeding, anticoagulants, DOACs, vitamin K antagonists, antiplatelet therapy, cardiovascular disease, retrospective study, oral surgery, hospitalization