“Health” (AKIpress) – In nearly 9 out of 10 cases, doctors do not increase cholesterol-lowering therapy in patients with cardiovascular disease, even if cholesterol levels remain above target. Artificial intelligence helped to find out why this happens, analyzing more than 13 thousand medical records.
Professor, cardiologist and oncologist Igor Pershukov spoke about the results of the study:
Introduction and Problem Clinical guidelines clearly indicate the target level to which low-density lipoprotein cholesterol (LDL-C—the so-called “bad” cholesterol) should be reduced for secondary prevention of atherosclerotic cardiovascular disease (ASCVD). However, in practice, lipid-lowering therapy (LLT) is still not used actively enough. To bridge this gap between medical recommendations and actual clinical practice, It is important to understand two things: how often clinicians use the “treat-to-target strategy” and why they choose not to intensify therapy when needed.
Purpose of the study. To test whether large artificial intelligence language models (LLMs) can find data in medical records about the implementation of treatment strategies to achieve a target, and also identify the reasons why doctors do not increase LST in cases where a patient’s LDL-C level is higher than normal.
How was the study conducted? This was a retrospective cohort study covering patient data from January 1, 2019, to May 15, 2025, in a large integrated healthcare system in Northern California.
- Who participated: Adult patients aged 18 to 90 years who had at least two visits to their physician for ASCVD and whose LDL-C level remained elevated (≥70 mg/dL) at the last visit.
- How we worked with the data: The researchers downloaded records of recent visits to the doctor and divided them into three sets of data: for developing the model, for testing it, and for the final application – inference (automatic data analysis by a neural network).
What exactly was studied and how it was assessed The AI searched the records for documented target LDL-C levels and medical decisions on the further management of LST. The accuracy of the AI was tested in comparison with manual clinical marking (assessed by expert doctors) on four points:
1. Is the target LDL-C level recorded in the map?
2. What is its exact meaning.
3. Has the LST been adjusted?
4. Are the reasons for refusing to intensify (non-intensify) therapy indicated? The frequency of certain medical decisions and outcomes was assessed using the final data set (the inference set).
Results
- Total sample: A total of 28,113 unique patients referred for ASCVD were identified. Almost half of them – 13,810 people (49.1%) – had LDL-C levels above the target (mean age was 70.9 years (standard deviation – 11.4); among them, 7904 were men (57.2%)).
- AI accuracy: The assessments of the language model and the expert doctors coincided (no statistically significant differences were found between the AI and the experts, or between the experts themselves). Model accuracy scores (micro-weighted F1 measure) were found to be very high, ranging from 0.80 to 0.98.
- Application of treatment strategy: In the inference set (13,394 patients with elevated LDL-C and 13,354 corresponding clinical records), a treatment-to-target strategy was recorded in only 3756 records (28.1%). Physicians most often indicated an LDL-C level of <70 mg/dL as a goal (in 2280 records, or 17.1%).
- Failure to intensify therapy and its reasons: In 11,842 of all records analyzed (88.7%), lipid-lowering therapy was not intensified (i.e., treatment was not intensified). Moreover, at least one reason for such a decision was recorded by a doctor in 4788 cases (40.4%).
The most common reasons for refusing L strengtheningST:
1. Adverse events (side effects) reported by the patient – 1683 cases (12.6%).
2. The LDL-C level was already close to the target – 922 cases (6.9%).
3. The patient’s desire to avoid taking LST drugs – 914 cases (6.8%).
4. Preference to change lifestyle instead of intensifying drug therapy – 811 cases (6.1%).
5. Postponement of the decision until the results of repeated laboratory tests are received – 617 cases (4.6%).
Conclusions and practical significance. An analysis of medical records using modern large language models (LLM) showed that in almost 9 out of 10 cases when a patient with ASCVD and LDL-C levels above the target comes to an appointment, doctors do not intensify lipid-lowering therapy. The main obstacles are side effects, personal preferences of patients, attempts to adjust values due to lifestyle, cholesterol levels close to normal, as well as “therapeutic inertia” (for example, postponing a decision until retesting).
What does this give to medicine: Understanding these reasons allows us to fine-tune (personalize) digital health tools throughout the healthcare system. This will help to identify deviations from clinical recommendations in a timely manner and increase the effectiveness of secondary prevention of cardiovascular diseases.
Crime and incidents in the Telegram channel @svodka_akipress
