Author:
Mohammad Motaghianfar
Abstract
Heart disease is a global killer, and lifestyle changes like quitting smoking are key to prevention. But not everyone benefits the same way from these changes. We used a smart AI tool called causal machine learning (CML) to study how quitting smoking affects heart disease risk differently across 929 people. Our dataset included age, health metrics, and lifestyle habits. The X-Learner model showed that, on average, quitting smoking had a small, non-significant effect (-0.033, 95% CI: -0.072 to 0.006). But digging deeper, we found big differences: some people saw major benefits (up to -1.52), while others faced slight risks (up to +1.32). About 25% would gain a lot from quitting, especially older folks with lower BMI. Tests confirmed our results are fairly solid (E-value=1.22). This study shows how AI can help doctors personalize heart disease prevention, focusing efforts on those who’ll benefit most.
Keywords: AI, Personalized Medicine, Heart Disease, Smoking Cessation, Causal Machine Learning, Precision Health
1. Introduction
Heart disease remains the world’s top cause of death, but lifestyle tweaks—like quitting smoking—can lower the risk. Most health advice is one-size-fits-all, based on studies showing average effects. Yet, people respond differently to the same advice, and blanket recommendations might not work for everyone.
New AI tools, called causal machine learning (CML), let us analyze real-world data to see how treatments affect individuals, not just groups. Unlike traditional studies, CML can handle messy data and spot who benefits most. But it’s rarely used for heart disease prevention, where most research predicts risks rather than tailoring solutions.
Our study fills this gap. We used CML to see how quitting smoking impacts heart disease risk for different people. We think there’s a lot of variation, and pinpointing it could help doctors give smarter, personalized advice.
2. Methods
2.1 Data
We studied data from 929 people, including their age, sex, blood pressure, cholesterol, glucose, and lifestyle habits like smoking and exercise. The dataset also tracked who had heart disease.
2.2 Data Prep
We cleaned up the data by removing errors, like impossible blood pressure readings or extreme BMIs. We converted ages to years, calculated BMIs, and created a heart risk score from cholesterol and glucose.
2.3 The AI Model
We used the X-Learner, a CML tool powered by XGBoost, to estimate how quitting smoking affects heart disease risk. It’s great for handling uneven data and avoiding mistakes. We defined “treatment” as smoking status (smoker vs. non-smoker), the outcome as heart disease, and included factors like age, sex, and BMI to control for differences.
2.4 Checking Our Work
We tested our results with:
-
E-value analysis to check for hidden biases.
-
Placebo tests, randomly mixing treatment data to ensure fairness.
-
Subgroup checks by age to confirm consistency.
We measured the average effect (ATE) and individual effects (ITEs) with confidence intervals.
2.5 Ethics
We used de-identified data and got ethical approval from an institutional review board.
3. Results
3.1 Overall Impact
On average, quitting smoking reduced heart disease risk slightly (-0.033, 95% CI: -0.072 to 0.006), but the effect wasn’t strong enough to be certain (Figure 1).
3.2 Individual Differences
Here’s where it gets interesting: individual effects varied widely, from a big benefit (-1.52) to a slight risk (+1.32, SD=0.60). The data showed two groups—those who benefit and those who don’t (Figure 2).
3.3 Who Benefits Most
People who gained the most from quitting were older (average 51.4 years) with lower BMI (26.3) compared to those who didn’t (48.3 years, BMI 27.8). Age and BMI were the biggest factors driving these differences (Figure 3).
3.4 Testing Robustness
Our results held up decently. The E-value (1.22) showed moderate resistance to hidden biases, and placebo tests (ATE=0.012) confirmed we weren’t seeing random noise. Age-based checks showed mixed results (Table 1):
-
Under 50: Small positive effect (0.049, CI: -0.015 to 0.113).
-
50+: Stronger protective effect (-0.095, CI: -0.158 to -0.032).
4. Discussion
Our study shows that quitting smoking doesn’t help everyone equally when it comes to heart disease. While the average effect was small, about 25% of people could see big benefits, while another 25% might not—or could even see slight risks. This challenges the idea of one-size-fits-all health advice.
Using the X-Learner, we uncovered these differences in a way traditional studies can’t. It’s built to handle real-world data quirks, making it ideal for this kind of work. Clinically, our findings mean doctors could focus smoking cessation programs on those likely to benefit most, like older patients with lower BMIs. Our clinical decision map (Figure 4) offers a practical guide for this.
4.1 Limits and Next Steps
The data wasn’t from a controlled trial, so we can’t be 100% sure about cause and effect, even with our careful tests. The E-value suggests some vulnerability to hidden factors, though placebo tests were reassuring. Future studies should test this in controlled settings and look at other lifestyle changes, like diet or exercise.
5. Conclusion
This study proves that AI can help make heart disease prevention more personal. By spotting who benefits most from quitting smoking, we can make health programs smarter and more effective. Next, we need to test this approach with diverse groups and other lifestyle changes to keep pushing personalized medicine forward.
References
[1] Chernozhukov, V., et al. (2018). Double/debiased machine learning for treatment and structural parameters. The Econometrics Journal.
[2] Kunzel, S. R., et al. (2019). Meta-learners for estimating heterogeneous treatment effects using machine learning. Proceedings of the National Academy of Sciences.
[3] VanderWeele, T. J., & Ding, P. (2017). Sensitivity Analysis in Observational Research: Introducing the E-Value. Annals of Internal Medicine.
[4] Prosperi, M., et al. (2020). Causal inference and counterfactual prediction in machine learning for healthcare. Nature Medicine.