Abstract Objective: Although guidelines have recommended standardized drug treatment for heart failure (HF), there are still many challenges in making the correct clinical decisions due to the complicated clinical situations of HF patients. Each patient would satisfy several recommendations, meaning the decision tree of HF treatment should be nonmutually exclusive, and the same patient would be allocated to several leaf nodes in the decision tree. In the current study, we aim to propose a way to ensemble a nonmutually exclusive decision tree for recommendation system for complicated diseases, such as HF. Methods: The nonmutually exclusive decision tree was constructed via knowledge rules summarized from the HF clinical guidelines. Then similar patients were defined as those who followed the same pattern of leaf node allocation according to the decision tree. The frequent medication patterns for each similar patient were mined using the Apriori algorithms, and we also carried out the outcome prognosis analyses to show the capability for the evidence-based medication recommendations of our nonmutually exclusive decision tree. Results: Based on a large database that included 29,689 patients with 84,705 admissions, we tested the framework for HF treatment recommendation. In the constructed decision tree, the HF treatment recommendations were grouped into two independent parts. The first part was recommendations for new cases, and the second part was recommendations when patients had different historical medication. There are 14 leaf nodes in our decision tree, and most of the leaf nodes had a guideline adherence of around 90%. We reported the top 10 popular similar patients, which accounted for 32.84% of the whole population. In addition, the multiple outcome prognosis analyses were carried out to assess the medications for one of the subgroups of similar patients. Our results showed even for the subgroup of the same similar patients that no one medication pattern would benefit all outcomes. Conclusion: In the present study, the methodology to construct a nonmutually exclusive decision tree for medication recommendations for HF and its application in CDSS was proposed. Our framework is universal for most diseases and could be generally applied in developing the CDSS for treatment.
Introduction: The Challenge of Clinical Complexity
Clinical guidelines for heart failure (HF) are comprehensive, yet applying them to individual patients remains a significant challenge. Traditional Clinical Decision Support Systems (CDSS) typically rely on "mutually exclusive" decision trees, where a patient follows a single path to one specific recommendation. However, real-world HF patients often present with overlapping conditions—such as congestion, renal impairment, and hypotension—simultaneously.
In clinical practice, a physician may need to follow multiple guideline recommendations at once. Standard models struggle with this "logical overlap," often forcing an artificial choice between paths. This study addresses the gap between idealized "if-this-then-that" logic and the multi-dimensional reality of treating complex chronic diseases.
Methods: A Three-Stage Framework
The researchers developed a novel framework to bridge the gap between guidelines and clinical practice, moving from rigid trees to a checklist-style logic.
- Knowledge Modeling (The NMEDT): Independent clinical rules were extracted from the 2016 ESC guidelines for HF. These were organized into a "Non-Mutually Exclusive Decision Tree" (NMEDT), allowing a single patient to trigger multiple branches and land in several leaf nodes simultaneously.
- Patient Similarity & Pattern Mining: "Similar patients" were defined as those who triggered the exact same combination of leaf nodes. The Apriori algorithm was then employed to identify the most frequent medication combinations prescribed for these specific subgroups within a real-world database.
- Prognostic Validation: To ensure recommendations were evidence-based, the team performed outcome prognosis analyses using propensity score adjustment to evaluate how different medication patterns affected clinical outcomes, including BNP levels and electrolyte balance.
Datasets and Processing
The study utilized a massive Electronic Health Record (EHR) database from the Chinese PLA General Hospital in Beijing, covering 2008 to 2018. The dataset included 29,689 unique patients and 84,705 admissions. To capture treatment changes over time, admissions were fragmented by day, resulting in 132,158 samples. Variables included demographics, physical examinations, laboratory results (e.g., creatinine, BNP), and detailed medication histories.
Key Insights: The NMEDT Innovation
The core innovation is the shift in logic: medical decision-making is often less like a branching path and more like a checklist. By allowing patients to be allocated to multiple leaf nodes (e.g., one for diuretics and another for ACEIs), the model captures the additive nature of heart failure therapy.
This approach maintains interpretability for clinicians while handling high complexity. Furthermore, the prognostic analysis revealed that no single drug pattern acts as a "silver bullet." A regimen might successfully lower BNP (improving heart function) while simultaneously increasing the risk of Acute Kidney Injury (AKI). This highlights the necessity for CDSS to provide a multidimensional view of treatment outcomes rather than a single "correct" answer.
Results and Clinical Adherence
The final NMEDT consisted of 14 leaf nodes. Analysis of real-world data showed that prescribing patterns had high adherence to guidelines—approximately 90% for most leaf nodes. The researchers identified the top 10 most common patient "patterns," which accounted for 32.84% of the total population. Detailed case studies of specific subgroups demonstrated how adding medications like Digoxin to standard regimens (ACEI/ARB+β+ARA) requires careful balancing of benefits against potential risks like hypernatremia.
Conclusion and Future Directions
This study proposes a universal methodology for constructing non-mutually exclusive decision trees for CDSS. While the framework proved effective for HF, limitations include single-center bias and the need to update the tree as guidelines evolve (e.g., incorporating SGLT2 inhibitors). Future work will focus on testing the framework on external datasets and adapting it for long-term outpatient management.
Summary updated