Identifying Patients at Risk of Acute Kidney Injury Among Medicare Beneficiaries With Type 2 Diabetes Initiating SGLT2 Inhibitors: A Machine Learning Approach

Lanting Yang, Nico Gabriel, Inmaculada Hernandez, Scott Martin Vouri, Stephen E. Kimmel, Jiang Bian, Jingchuan GuoFrontiers in Pharmacology 2022 · DOI

TL;DR

This paper develops and validates a machine learning model to predict the risk of acute kidney injury (AKI) in patients with type 2 diabetes (T2D) who are prescribed SGLT2 inhibitors. The model, particularly a random forest approach, shows good predictive utility and identifies key risk factors, with the use of loop diuretics emerging as a significant predictor.

What Problem Does This Paper Address?

Type 2 diabetes (T2D) is a widespread condition associated with numerous complications, including kidney damage. Sodium-glucose cotransporter-2 inhibitors (SGLT2i) are a class of drugs that have shown benefits in managing T2D, particularly in preventing cardiovascular and renal decline. However, SGLT2i use has also been linked to an increased risk of acute kidney injury (AKI), potentially due to their diuretic effect leading to hypovolemia. While the association between SGLT2i and AKI is known, there's a gap in understanding which specific patient characteristics or comorbidities increase this risk. This knowledge is crucial for individualizing treatment, balancing the benefits of SGLT2i against their risks, and ultimately improving patient outcomes. This paper aims to address this by developing a predictive model to identify T2D patients on SGLT2i who are at higher risk of AKI.

Methods

The researchers employed a machine learning approach to build and validate their AKI prediction model. They utilized three common machine learning algorithms: Random Forests (RF), Elastic Net, and Least Absolute Shrinkage and Selection Operator (LASSO). The cohort was randomly split into training and testing sets. The models were trained on data from the year prior to SGLT2i initiation, using 65 potential predictor variables derived from Medicare claims data. Hyperparameter tuning was performed on the training set, and the models' performance was evaluated on the testing set. To identify important risk factors, a logistic regression model was constructed using features selected by LASSO.

Datasets

The study utilized a 5% random sample of Fee-for-Service Medicare claims data from 2012-2016. This dataset provided information on patient demographics, diagnoses, and medication use. The data was used to identify patients with T2D who initiated SGLT2i (canagliflozin, dapagliflozin, or empagliflozin) between April 2013 and December 2016. The dataset was split into training and testing sets, each containing 8,847 patients. No new benchmarks were introduced; the study leverages existing administrative claims data.

Key Insights and Ideas

  • Machine learning for AKI prediction in SGLT2i users is feasible and valuable: The study demonstrates that machine learning can effectively identify patients at increased risk of AKI when initiating SGLT2i, a critical step for personalized medicine.
  • Random Forest excels in risk stratification: The RF model proved to be the most effective in distinguishing between high and low-risk patients, showing a clear gradient in AKI incidence across different risk score percentiles.
  • Loop diuretics are a significant AKI risk factor in this context: The strong association between loop diuretic use and AKI incidence, even after accounting for other factors, highlights a potentially synergistic risk when combined with SGLT2i. This is a novel finding that warrants further investigation.
  • Early identification can guide clinical decisions: The development of a predictive tool could aid clinicians in balancing the benefits of SGLT2i (especially in heart failure) against the risk of AKI, potentially preventing adverse events.

Key Results

  • Overall AKI incidence: 1.10% of the 17,694 beneficiaries developed AKI over a median follow-up of 1.5 years.
  • Model performance: The Random Forest (RF) model achieved the highest predictive performance with a C-statistic of 0.72 (95% CI 0.68–0.76), outperforming LASSO and Elastic Net (both C-statistics = 0.69).
  • Risk stratification effectiveness: Patients in the top 10% of the RF risk score had an AKI incidence rate of 3.73%, compared to 0.47% in the bottom 50% risk group.
  • Key risk factors identified by LASSO: The most significant predictor of AKI was the use of loop diuretics (adjusted odds ratio [aOR]: 3.72, 95% CI 2.45-5.6). Other important factors included a history of AKI (aOR: 2.78), anemia (aOR: 2.56), and male sex (aOR: 1.61).

Limitations and Future Work

The authors acknowledge several limitations:

  • Limited clinical data: The use of claims data prevented the inclusion of crucial clinical information like estimated glomerular filtration rate (eGFR) and blood pressure.
  • Follow-up duration: The model predicts AKI risk within 1.5 years of SGLT2i initiation; longer-term prediction is needed.
  • External validation: The model requires validation in independent datasets and different populations.
  • Generalizability: The study focused on older adults with T2D using Fee-for-Service Medicare claims, which may limit generalizability to other demographics or insurance types.

Future work suggested includes:

  • Incorporating more robust data: Utilizing electronic health records (EHRs) with more detailed clinical data.
  • Updating the model: Regularly updating the model with more current data to reflect the evolving use of SGLT2i.
  • Developing clinical alert tools: Implementing machine learning-based alerts at the point of care to assist clinicians.
  • Investigating the combined use of SGLT2i and loop diuretics: Further research is needed to understand the safety profile and mechanisms behind this combination.

Summary updated

Abstract

<b>Introduction:</b> To predict acute kidney injury (AKI) risk in patients with type 2 diabetes (T2D) prescribed sodium-glucose cotransporter two inhibitors (SGLT2i). <b>Methods:</b> Using a 5% random sample of Medicare claims data, we identified 17,694 patients who filled ≥1 prescriptions for canagliflozin, dapagliflozin and empagliflozin in 2013-2016. The cohort was split randomly and equally into training and testing sets. We measured 65 predictor candidates using claims data from the year prior to SGLT2i initiation. We then applied three machine learning models, including random forests (RF), elastic net and least absolute shrinkage and selection operator (LASSO) for risk prediction. <b>Results:</b> The incidence rate of AKI was 1.1% over a median 1.5 year follow up. Among three machine learning methods, RF produced the best prediction (C-statistic = 0.72), followed by LASSO and elastic net (both C-statistics = 0.69). Among individuals classified in the top 10% of the RF risk score (i.e., high risk group), the actual incidence rate of AKI was as high as 3.7%. In the logistic regression model including 14 important risk factors selected by LASSO, use of loop diuretics [adjusted odds ratio (95% confidence interval): 3.72 (2.44-5.76)] had the strongest association with AKI incidence. <b>Disscusion:</b> Our machine learning model efficiently identified patients at risk of AKI among Medicare beneficiaries with T2D undergoing SGLT2i treatment.
BibTeX
@inproceedings{yang2022identifying,
  author = {Lanting Yang and Nico Gabriel and Inmaculada Hernandez and Scott Martin Vouri and Stephen E. Kimmel and Jiang Bian and Jingchuan Guo},
  title = {{Identifying Patients at Risk of Acute Kidney Injury Among Medicare Beneficiaries With Type 2 Diabetes Initiating SGLT2 Inhibitors: A Machine Learning Approach}},
  booktitle = {Frontiers in Pharmacology},
  year = {2022},
  doi = {10.3389/fphar.2022.834743},
}
Figure