📄 medicine

A Predictive LASSO regression model to determine the potential predictors of medication adherence in myocardial infarction patients: A cross-sectional study

This cross-sectional study of 300 myocardial infarction patients utilized LASSO regression to identify ten significant predictors of poor medication adherence and developed a validated nomogram model to effectively predict adherence risks and guide clinical interventions.

Original authors: Mohammad Hossein Sharifi, Soorena Khorshidi, Somayeh Zarei, Mina Zarmehrparirouy, Reyhane Gashtasebi

Published 2026-06-30

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Mohammad Hossein Sharifi, Soorena Khorshidi, Somayeh Zarei, Mina Zarmehrparirouy, Reyhane Gashtasebi

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your heart is like a high-performance car engine. After a major breakdown (a heart attack, or Myocardial Infarction), the mechanic gives you a strict maintenance schedule: take these pills, eat this food, and exercise this much. If you follow the schedule perfectly, the engine runs smoothly. If you skip it, the engine might break down again, sometimes permanently.

This study is like a team of detectives trying to figure out why some people follow the maintenance schedule perfectly, while others forget or ignore it.

Here is the story of their investigation, broken down simply:

The Investigation Team and the Suspects

The researchers gathered a group of 300 drivers (patients) who had recently had their engines repaired. They wanted to find the "clues" that predicted who would stick to the maintenance plan and who wouldn't.

They didn't just guess; they used a very smart, computerized detective tool called LASSO. You can think of LASSO as a super-efficient filter. Imagine you have a giant bag of 39 different clues (like age, income, smoking habits, education, how much you trust yourself to follow rules, etc.). Most of these clues are just noise. LASSO's job is to sift through the bag, throw away the useless noise, and hand you only the 10 most important clues that actually matter.

The 10 Big Clues Found

After running the numbers, the study found that these 10 factors were the strongest predictors of whether a patient would be good or bad at taking their medicine:

Money: People who had enough money to cover their bills were more likely to follow the rules.
Where they lived: People living in the city (urban) did better than those in the countryside.
Smoking: Non-smokers were better at sticking to the plan than smokers.
Surgery History: People who had a specific type of heart surgery (CABG) tended to follow the rules better.
Self-Confidence: This is a big one. It's called "self-efficacy." If a patient believed they could handle their health, they did better. It's like believing you can actually change the oil in your car.
General Care: How well they took care of themselves in general.
Education: Higher levels of schooling helped.
Other Health Issues: Surprisingly, having other health problems (comorbidities) actually made people more likely to take their heart meds. (Maybe because they were already used to taking pills for other things!)
Hospital Visits: People who had been hospitalized more often were more likely to follow the rules.
Exercise: People who moved their bodies for 150–300 minutes a week did better.

The "Magic Map" (The Nomogram)

The researchers didn't just list these clues; they built a Magic Map (called a Nomogram).

Think of this map like a scorecard in a video game. You give the map the patient's details (e.g., "He smokes, lives in the city, and has low confidence"), and the map adds up points.

If the score is high, the map flashes a red warning light: "High Risk of forgetting meds!"
The study found that for the "worst-case scenario" combination of these factors, the map predicted an 81% chance that the patient would struggle to take their medicine.

How Good Was the Detective Work?

The researchers tested their Magic Map to see if it was accurate.

The Accuracy Test: They used a standard test called an ROC curve. The map scored 0.74, which is like getting a "B" or a "B+" in school. It's not perfect, but it's definitely good enough to be useful.
The Calibration: They checked if the map's predictions matched reality. It was very close, meaning the map wasn't just guessing; it was telling the truth.

The Bottom Line

The study concludes that if you want to know who is likely to forget their heart medicine, look at their confidence, their wallet, their smoking habits, and how much they exercise.

The researchers say this model is a powerful tool. It helps doctors look at a patient and say, "Hey, based on these 10 clues, this person is at high risk of forgetting their pills. Let's give them extra help before they have another heart attack."

Important Note: The study admits it has some limits. They only looked at patients for a short time (one year) and didn't track them for years afterward. Also, they couldn't get data from every single patient in the world, just a specific group in Shiraz, Iran. But for the group they studied, the "Magic Map" worked well.

Technical Summary: Predictive LASSO Regression for Medication Adherence in Myocardial Infarction Patients

Problem Statement
Myocardial infarction (MI) remains a leading cause of global morbidity and mortality, particularly in developing societies. While medical advances have improved post-MI survival, the effectiveness of long-term treatment is heavily contingent on medication adherence. Chronic low adherence to prescribed medications is a critical barrier to preventing recurrent cardiovascular events. Although factors such as self-efficacy, lifestyle changes, and demographic characteristics are known to influence adherence, there is a need for precise identification of the most relevant predictors among a wide array of potential variables to facilitate targeted clinical interventions. This study addresses the challenge of distinguishing significant predictors from a large set of candidate variables to improve the management of MI patients.

Methodology
This cross-sectional study, conducted between June and August 2024 in Shiraz, Iran, analyzed data from 300 adult patients (aged 30–65) diagnosed with MI within the preceding year. Participants were recruited via convenience sampling, excluding those with psychiatric disorders, dementia, or an inability to respond.

Data Collection: The study utilized a demographic and clinical questionnaire covering 39 variables, including socioeconomic status, clinical history (e.g., CABG, stents, comorbidities), and lifestyle factors.
Outcome Measurement: Medication adherence was assessed using the Morisky Medication Adherence Scale (MMAS-8). Scores were dichotomized into "good adherence" (score 8) and "poor adherence" (scores <8, combining low and moderate categories).
Self-Efficacy Measurement: The 16-item Sullivan self-efficacy questionnaire measured general care, disease symptom control, and diet management.
Statistical Analysis:
- Initial screening involved univariate analyses (Chi-square, Fisher's exact, Mann-Whitney U tests) to identify variables with $p < 0.05$ .
- LASSO Regression: To address multicollinearity and select the most parsimonious set of predictors, the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model was applied with 5-fold cross-validation. This machine learning technique minimized regression coefficients to exclude collinear variables.
- Model Validation: A predictive nomogram was constructed based on the LASSO-selected variables. Model performance was evaluated using Receiver Operating Characteristic (ROC) curves (AUC), calibration curves (Brier score), decision curve analysis, and clinical impact curves. All analyses were performed using R (version 4.5.1).

Key Results

Demographics: The cohort had a mean age of 54.20 ± 8.50 years, with 81% male participants. Notably, 73% of the total sample exhibited poor medication adherence.
Univariate Findings: Significant differences between good and poor adherence groups were observed in economics, physical activity, smoking, comorbidity, self-efficacy, general care, and diet management.
LASSO Variable Selection: Out of 39 initial variables, the LASSO model identified 10 significant predictors of poor medication adherence:
1. Economic status (income vs. expenses)
2. Urban/rural residence
3. Smoking status
4. History of Coronary Artery Bypass Graft (CABG)
5. Self-efficacy scores
6. General care scores
7. Education level
8. Comorbidity presence
9. Hospitalization frequency
10. Physical activity levels
Multivariable Insights: The multivariable logistic regression indicated that patients without comorbidities had 1.93 times higher odds of poor adherence, while non-smokers had 43% lower odds. Engaging in 150–300 minutes of weekly physical activity reduced the odds of poor adherence by 55%.
Model Performance:
- Nomogram: The model predicted a probability of 0.81 for high-risk poor adherence based on the selected predictors.
- Discrimination: The Area Under the Curve (AUC) was 0.742, indicating acceptable diagnostic accuracy.
- Calibration: The Brier score was 0.16, and the calibration curve showed good agreement between predicted and observed outcomes.
- Clinical Utility: Decision and clinical impact curves demonstrated that the model provided net clinical benefits and accurately identified high-risk individuals.

Significance and Claims
The authors claim that this study provides a robust, data-driven framework for identifying independent predictors of medication non-adherence in MI patients. By utilizing the LASSO regression model, the study successfully reduced a complex set of 39 variables to 10 key factors, offering a more precise tool for risk stratification than traditional methods.

The primary significance lies in the development of a validated nomogram that clinicians can use to identify patients at high risk for poor adherence (with a predicted probability of 0.81 for high-risk profiles). The authors assert that this model yields significant clinical benefits by enabling the timely implementation of targeted interventions. The study concludes that factors such as higher income, urban residence, non-smoking status, CABG history, higher education, frequent hospitalizations, greater self-efficacy, and specific levels of physical activity are remarkable predictors of good adherence. The authors position this work as a novel application of machine learning techniques in the context of MI management, offering profound insights for addressing adherence challenges.

Limitations
The authors acknowledge specific limitations, including the cross-sectional design which precludes causal inference, the lack of longitudinal data on adherence in years following the cardiac event due to clinic documentation constraints, and a sample size that may be insufficient for more granular sub-analyses.