Sequential Model for Predicting Patient Adherence in Subcutaneous Immunotherapy for Allergic Rhinitis

Yin Li, Yu Xiong, Wenxin Fan, Kai Wang, Qingqing Yu, Liping Si, Patrick van der Smagt, Jun Tang, Nutan Chen

Introduction

Subcutaneous Immunotherapy (SCIT) represents one of the few disease-modifying treatments for Allergic Rhinitis (AR), aimed at inducing long-term immune tolerance. Unlike symptomatic treatments, SCIT requires a significant commitment—typically three to five years of regular injections. This "marathon" nature of treatment often leads to poor adherence, with dropout rates reported between 25% and 90% in various clinical settings.

Predicting which patients are likely to discontinue treatment is crucial for clinical intervention. While traditional predictive models often treat adherence as a simple binary outcome (completion vs. dropout), they fail to capture the dynamic nature of a multi-year treatment plan. This study explores the application of sequential machine learning models to provide step-by-step predictions of both patient adherence and symptom progression, allowing for timely interventions.

Methodology

The researchers developed and compared two distinct sequential architectures to model the patient journey over a 36-month period:

  1. Long Short-Term Memory (LSTM): A recurrent neural network designed for time-series data, predicting next-step symptom scores and adherence status simultaneously.
  2. Sequential Latent-Variable Model (SLVM): Based on the Stochastic Latent Actor-Critic (SLAC) framework. This model treats adherence as an "action" within a "state-action" pair, allowing it to simulate how different adherence behaviors influence future health states.
Overview of the study workflow, including patient observation, follow-up categorization, and sequential model implementation.

The study utilized a retrospective dataset of 205 AR patients from the First People’s Hospital of Foshan, China. Data points included 14 baseline variables (e.g., age, gender, distance to clinic, IgE levels) and 11 longitudinal variables (e.g., nasal/ocular symptom scores and medication scores). Observations were recorded at specific intervals: 0, 4, 12, 18, 24, and 36 months.

Results and Discussion

The evaluation revealed a specialization trade-off between the two models. The LSTM proved more robust for binary adherence prediction, achieving an accuracy range of 66% to 84%. In contrast, the SLAC-based model excelled at predicting numerical symptom severity (VAS scores), yielding a lower Root Mean Square Error (RMSE) of 0.93 to 2.22, compared to the LSTM's 1.09 to 1.77.

A significant advantage of the SLAC model is its capability as a clinical simulator. By modeling adherence as an action, clinicians can input hypothetical scenarios—such as a patient missing several months of treatment—to visualize predicted symptom regressions. This "what-if" capability serves as a powerful tool for patient education and shared decision-making, potentially motivating patients to stay the course.

Interpretability analysis using the Captum library highlighted that "distance to clinic" was the most significant predictor of non-adherence. This suggests that logistical and geographical barriers are often more influential than biological factors in determining long-term treatment success. Furthermore, the data identified the one-year mark as a critical "danger zone" for dropouts, a trend exacerbated by the COVID-19 pandemic, which accounted for 25% of first-year dropouts in this cohort.

Conclusion

This research demonstrates the potential of sequential latent-variable models in managing chronic conditions like Allergic Rhinitis. While LSTMs are effective for identifying at-risk patients, SLVMs offer deeper insights into symptom progression and the consequences of non-adherence. Future work should focus on expanding the cohort size, standardizing observation intervals, and integrating these predictive tools into digital health platforms to provide real-time clinical support and improve long-term patient outcomes.

Summary updated

Abstract

Objective: Subcutaneous Immunotherapy (SCIT) is the long-lasting causal treatment of allergic rhinitis (AR). How to enhance the adherence of patients to maximize the benefit of allergen immunotherapy (AIT) plays a crucial role in the management of AIT. This study aims to leverage novel machine learning models to precisely predict the risk of non-adherence of AR patients and related local symptom scores in three years SCIT. Methods: The research develops and analyzes two models, sequential latent-variable model (SLVM) of Stochastic Latent Actor-Critic (SLAC) and Long Short-Term Memory (LSTM) evaluating them based on scoring and adherence prediction capabilities. Results: Excluding the biased samples at the first time step, the predictive adherence accuracy of the SLAC models is from 60\% to 72\%, and for LSTM models, it is 66\% to 84\%, varying according to the time steps. The range of Root Mean Square Error (RMSE) for SLAC models is between 0.93 and 2.22, while for LSTM models it is between 1.09 and 1.77. Notably, these RMSEs are significantly lower than the random prediction error of 4.55. Conclusion: We creatively apply sequential models in the long-term management of SCIT with promising accuracy in the prediction of SCIT nonadherence in AR patients. While LSTM outperforms SLAC in adherence prediction, SLAC excels in score prediction for patients undergoing SCIT for AR. The state-action-based SLAC adds flexibility, presenting a novel and effective approach for managing long-term AIT.
BibTeX
@inproceedings{li2024sequential,
  author = {Yin Li and Yu Xiong and Wenxin Fan and Kai Wang and Qingqing Yu and Liping Si and Patrick van der Smagt and Jun Tang and Nutan Chen},
  title = {{Sequential Model for Predicting Patient Adherence in Subcutaneous Immunotherapy for Allergic Rhinitis}},
  booktitle = {Frontiers in Pharmacology},
  year = {2024},
  doi = {10.3389/fphar.2024.1371504},
}
Figure