The core insight behind CLATS is that alert optimization is not purely a statistical problem; it is an operational one. Most deployed deterioration-warning systems rely on static or ad hoc thresholds set when the software is installed. But hospitals are not static environments. Patient case-mix shifts between winter and summer, staffing levels rise and fall across shifts, and the workload a unit can absorb on a Tuesday afternoon differs from what it can absorb at three in the morning. A threshold that keeps missed events acceptably rare during calm periods may flood the same unit with alerts during a surge, or vice versa. The researchers argue that any sensible alert policy must therefore balance two competing quantities that most systems treat separately: the sensitivity of the detector, meaning its ability to catch true deterioration events, and the workload burden it imposes on the clinical staff who must respond.

CLATS addresses this trade-off with a deliberately transparent architecture that the authors describe as auditable and deterministic. The system pairs a calibrated predictive model with a two-threshold, three-tier triage policy. Patients whose deterioration risk score exceeds the upper threshold L2 land in the highest-alert tier and trigger immediate attention. Patients scoring between the lower threshold L1 and L2 fall into a middle tier, and those below L1 require no action. This tiered structure matters because it lets the optimizer treat false positives with nuance: an alert that merely asks a nurse to glance at a chart costs far less than one that demands an emergency response team. Rather than counting every false alarm equally, CLATS assigns tiered false-positive costs, alongside explicit penalties for missed events and for system overload situations in which the volume of alerts exceeds what the ward can realistically process.

The machinery that makes this work is a penalized, operations-aware cost function. In essence, the optimizer searches for the pair of thresholds that minimizes a weighted sum of harm: the penalty for each missed deterioration event, the accumulated costs of false alarms in each tier, and the burden of overload. Crucially, the search operates under a hard constraint, a fixed sensitivity target that the system must never fall below, no matter how attractive the cost savings. This design decision reflects a clinical reality that pure cost-benefit mathematics can obscure: in medicine, missing a true deterioration event is generally far more dangerous than responding to a few extra false ones, so the safety floor is enforced structurally rather than traded away when convenient.

Technically, the optimization proceeds in two phases. Initial thresholds are determined on validation data using isotonic calibration, a non-parametric technique that maps raw model scores into well-calibrated probabilities by fitting a monotonic, non-decreasing function. Isotonic calibration is prized in clinical prediction because it makes no assumptions about the shape of the relationship between scores and outcomes and produces probabilities that can be trusted as genuine risk estimates. On top of those calibrated scores, a discrete grid search enumerates candidate threshold pairs and identifies the combination that minimizes the operations-aware cost. Grid search is a blunt but dependable instrument: because the space of threshold pairs is finite and enumerable, every decision the system makes can be traced, reproduced, and explained to a hospital governance committee, a property many black-box alternatives lack.

What distinguishes CLATS from a one-time tuning exercise is its closed loop. Once deployed, the system does not let its thresholds fossilize. Instead, it operates in review cycles, re-estimating the optimal thresholds as new data accumulates and as staffing and patient-population conditions shift. The refinement process uses exponential smoothing, which blends the newly computed thresholds with the previous values while weighting recent observations more heavily, and guardrail constraints, which bound how far thresholds may move in a single cycle. The guardrails are a safeguard against instability: a sudden data anomaly or transient shift in the ward could otherwise cause a wild threshold swing that disrupts clinical workflows. With smoothing and bounded updates, the system adapts gradually and predictably, in the spirit of stochastic learning and adaptive control rather than abrupt reconfiguration.

The evaluation reported in the paper is synthetic, and the authors are unusually explicit about what that means for interpretation. Working with simulated data and an isotonic-calibrated CatBoost model, a gradient-boosted decision tree method known for strong performance on tabular clinical data, CLATS maintained a sensitivity of at least 0.989 across the tested conditions. Under the controlled data-generating assumptions of the simulation, the system produced no false positive alerts, a result that must be read carefully: it validates the optimization mechanism itself, not clinical performance, because real hospital data contain noise, label ambiguity, and distribution shift that simulations rarely capture. The authors state plainly that these findings support the feasibility of the CLATS mechanism rather than constituting clinical evidence, a caveat that responsible readers should carry forward.

The more operationally interesting result concerns the closed-loop dynamics. During simulated re-optimization cycles, the adaptive threshold updates reduced the operational cost objective by an average of 4.6 percent per review cycle, without reducing recall in this setting. Compounded across repeated cycles, savings of that magnitude suggest that static thresholds leave meaningful efficiency on the table, particularly in environments where patient acuity and staffing fluctuate. Because the cost function explicitly encodes workload and overload penalties, the improvements translate conceptually into fewer disruptive alarms and less wasted clinical attention, precisely the resources that alarm fatigue squanders. Prior literature has linked clinician workload to patient safety outcomes, and operations-management research has long documented how overburdened staff experience degraded performance, giving the workload-aware formulation an empirical footing beyond simple intuition.

Several design choices position CLATS for eventual real-world adoption, even though that step awaits validation on genuine longitudinal clinical data. The framework is model-agnostic: the CatBoost predictor in the paper could be swapped for any sufficiently accurate deterioration model, since the optimization layer interacts only with calibrated risk scores. The system is also governance-compatible, offering a transparent, deterministic basis for decisions that hospital safety committees and regulators can inspect line by line. That transparency addresses a persistent obstacle in clinical artificial intelligence deployment. Studies of real-world sepsis prediction systems have shown that technical accuracy alone does not guarantee clinical uptake; integration into routine workflows, trust, and auditability often decide whether a predictive tool helps or gathers dust. A framework whose every threshold decision can be explained and reproduced speaks directly to those implementation barriers.

The research lands at a moment when healthcare institutions are actively searching for ways to tame alert burden, from remote patient monitoring programs to decision-support systems in intensive care, and the bibliography of the paper reflects that urgency, spanning early warning score evaluations, alert reduction frameworks, and reinforcement learning approaches to healthcare operations. CLATS contributes a specific and disciplined piece to that puzzle: not a new predictor, but a principled mechanism for turning any calibrated predictor into an operationally sustainable alert policy. The caveats remain clear. The results are synthetic, the data-generating assumptions are idealized, and the authors themselves, along with their funders at the Gust Engineering and Applied Innovation Research Center, frame the work as a foundation for future validation. If that future validation on real hospital data holds, closed-loop, operations-aware threshold optimization could become a standard layer in the clinical monitoring stack, quietly retuning the alarm systems that clinicians have learned, at their peril, to tune out.

Subject of Research: A closed-loop adaptive optimization framework for clinical deterioration alert thresholds

Article Title: CLATS: a closed-loop adaptive triage system for operations-aware clinical alert optimization

Article References: Alzhouri, F., Daraghmeh, M., Agarwal, A., & Ebrahimi, D. (2026). CLATS: a closed-loop adaptive triage system for operations-aware clinical alert optimization. Cluster Computing, 29(14), Article 793. https://doi.org/10.1007/s10586-026-06611-x

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06611-x

Keywords: clinical alerts, alert fatigue, adaptive triage, machine learning, CatBoost, isotonic calibration, threshold optimization, hospital operations, sensitivity, closed-loop control, clinical decision support, predictive modeling

A new study published in Annals of Gastroenterological Surgery asks whether machine learning could tip that balance. Researchers from the University Hospital of Split in Croatia analyzed the records of 1,547 adults operated on for suspected appendicitis between January 2020 and June 2024, ultimately assembling a rigorously curated dataset of 623 patients. Of these, 66 had a histologically normal appendix, 248 had uncomplicated appendicitis, and 309 had complicated appendicitis confirmed by pathologists. Crucially, every case was verified by histopathology, eliminating the diagnostic guesswork that plagues many prior studies in which patients were classified by clinical or radiological impression alone.

The team built and compared four machine learning models: logistic regression, random forest, balanced random forest, and XGBoost, an extreme gradient boosting algorithm. All models relied exclusively on routinely available clinical and laboratory data—age, sex, symptom duration, fever, nausea, vomiting, rebound tenderness, pain migration, white blood cell count, neutrophil and lymphocyte percentages, platelet indices, C-reactive protein, and serum sodium. No imaging data were included. Missing values, which ranged from zero to 5.3 percent across variables, were imputed with a bagged trees algorithm carefully configured to prevent data leakage between training and test sets.

The methodological centerpiece of the study was a nested cross-validation design, with five inner and outer folds repeated ten times, yielding fifty independent test evaluations. Hyperparameters were tuned exclusively on inner validation data, and performance was then measured on untouched outer test folds, keeping performance estimates unbiased. On top of that, the researchers applied bootstrap resampling with 2,000 iterations to confirm that results were stable and not artifacts of a particular split of the data.

The most distinctive innovation, however, was threshold optimization. Rather than accepting a default probability cutoff of 0.5, the team deliberately shifted decision thresholds to meet six predefined minimum sensitivity targets, ranging from 0.95 to 0.995. Within each outer fold, thresholds were derived from pooled inner validation predictions, fixed, and then applied unchanged to the test set. The logic is clinical, not statistical: in appendicitis, missing a true case is far more dangerous than operating on a healthy appendix, so a decision support tool must be calibrated to catch virtually every case of appendicitis while identifying the small minority of patients who might safely avoid immediate surgery.

The results were revealing. For detecting acute appendicitis, logistic regression outperformed the more complex algorithms, achieving an area under the ROC curve of 0.765. At the strictest sensitivity target of 0.995, the model detected 99.8 percent of appendicitis cases, with specificity of just 0.038. Translated to a clinical scale, using the observed case distribution, this operating point would miss approximately two appendicitis cases per 1,000 patients while correctly flagging about four patients without appendicitis who could be spared surgery. At the more lenient sensitivity target of 0.95, roughly 41 appendicitis cases per 1,000 would be missed, but approximately 24 patients per 1,000 without appendicitis would avoid an unnecessary operation. The numbers lay bare the inherent trade-off between diagnostic safety and operative selectivity.

For distinguishing complicated from uncomplicated disease, the random forest model performed best, with an AUC of 0.785 and a maximum achievable sensitivity of 0.997 paired with specificity of 0.057. This capability carries real clinical weight: complicated appendicitis generally demands urgent surgery and carries higher morbidity and mortality, whereas some centers now attempt conservative, non-operative management of uncomplicated cases. A model that reliably identifies complicated disease could support more urgent operative decisions, temper enthusiasm for non-operative strategies, or prompt additional imaging and closer monitoring when the predicted risk is ambiguous.

To address the notorious black-box problem that undermines clinical trust in machine learning, the researchers applied SHAP analysis, a framework from explainable artificial intelligence that quantifies each feature’s contribution to individual predictions. For appendicitis detection, neutrophil percentage, the neutrophil-to-lymphocyte ratio, the platelet-to-lymphocyte ratio, sex, and lymphocyte percentage emerged as the most influential predictors. For complication prediction, C-reactive protein, age, symptom duration, and the neutrophil-to-lymphocyte ratio dominated. These rankings align closely with the pathophysiology of appendicitis, in which bacterial invasion of the appendiceal mucosa triggers a systemic inflammatory cascade, and they mirror the feature importances reported by other research groups using different algorithms and populations.

The authors are careful to frame what these models are and are not. The cohort consisted exclusively of adults already selected for surgery by a surgeon’s judgment, so the models should not be treated as stand-alone diagnostic tools for undifferentiated abdominal pain in the emergency department, nor as replacements for ultrasound or CT where those are routinely available. Instead, the researchers envision them as adjunctive decision-support and reassessment tools: a low-risk prediction should trigger repeat examination, laboratory reassessment, or additional imaging rather than discharge, particularly in settings with limited imaging access or after equivocal imaging findings. Notably, the superiority of the simple linear model over tree-based methods in this adult cohort contrasts with findings in pediatric populations, suggesting that inflammatory markers relate to appendicitis in a more linear fashion in adults and underscoring the need for population-specific models.

Limitations remain. The study was retrospective and single-center, and its performance may reflect local diagnostic pathways, surgeon thresholds, and imaging practices that do not transfer to other institutions. The dataset is also surgically enriched, meaning there are no true negative patients who avoided surgery, so specificity must be interpreted conditionally within the operated population. Still, the combination of histopathological verification, rigorous nested cross-validation, bootstrap stability testing, deliberate sensitivity-first threshold calibration, and transparent SHAP-based interpretability marks this work as a methodologically serious step toward machine learning that could meaningfully reduce negative appendectomies. The authors call for prospective, multicenter external validation before clinical deployment, along with future exploration of multimodal models integrating imaging and of tools to help select patients with uncomplicated appendicitis for non-operative management—a direction that could ultimately reshape how one of surgery’s most routine emergencies is decided.

Subject of Research: Machine learning with sensitivity-focused threshold optimization to reduce negative appendectomies in adults with suspected acute appendicitis

Article Title: Can Machine Learning Reduce Unnecessary Surgeries? A Retrospective Analysis Using Threshold Optimization to Prevent Negative Appendectomies in Adults

Article References: Males, I., Kumric, M., Boban, Z., Vrdoljak, J., Pecenkovic, D., Ivanda, M., Grahovac, M., Pogorelic, Z., & Bozic, J. (2026). Can Machine Learning Reduce Unnecessary Surgeries? A Retrospective Analysis Using Threshold Optimization to Prevent Negative Appendectomies in Adults. Annals of Gastroenterological Surgery, 10(5), 1631-1645. https://doi.org/10.1002/ags3.70225

Image Credits: AI Generated

DOI: 10.1002/ags3.70225

Keywords: machine learning, appendicitis, negative appendectomy, logistic regression, random forest, XGBoost, threshold optimization, SHAP explainability, clinical decision support, C-reactive protein, nested cross-validation, emergency surgery