Sequential model for predicting patient adherence in subcutaneous immunotherapy for allergic rhinitis

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

Abstract

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.

Introduction

Subcutaneous Immunotherapy (SCIT) is a cornerstone treatment for allergic rhinitis (AR), aiming to induce long-term immune tolerance. However, its effectiveness is contingent on a multi-year commitment, typically lasting 3 to 5 years. Consequently, patient adherence is a significant challenge, with dropout rates historically ranging from 10% to 75%. Traditional predictive models are often "static," relying on baseline data to predict final outcomes. This study introduces sequential machine learning models that account for the temporal dynamics of treatment, providing real-time risk assessments and symptom forecasting.

Methodology

The researchers evaluated two primary sequential architectures using a retrospective dataset of 205 patients from the First People’s Hospital of Foshan, China. Data was collected at 0, 4, 12, 18, 24, and 36 months, including 14 baseline characteristics and longitudinal clinical data such as VAS symptom scores and medication scores.

Overview of the study workflow, illustrating the transition from clinical observation and follow-up to sequential latent variable modeling.

Two models were compared:

  1. Long Short-Term Memory (LSTM): A recurrent neural network designed to capture long-term dependencies in time-series data, predicting both adherence and symptom scores in parallel.
  2. Sequential Latent Variable Model (SLVM): Based on the Stochastic Latent Actor-Critic (SLAC) framework, this probabilistic model maps observations into a latent space. It treats the decision to continue treatment as an "action" that influences future symptom "states."

Key Insights and Results

The study revealed a distinct specialization between the models. The LSTM proved to be a superior "adherence alarm," achieving an accuracy range of 66%–84% in predicting dropouts. Conversely, the SLVM excelled as a "symptom forecaster," providing precise predictions for specific symptoms like nasal itching and congestion, with a Root Mean Square Error (RMSE) between 0.93 and 2.22.

Distribution of nasal itching scores across the patient cohort, representing one of the eleven clinical features modeled to track symptom progression.

The "Simulator" Advantage

A unique benefit of the SLVM is its capability to function as a clinical simulator. Because it is state-action-based, clinicians can input hypothetical scenarios—such as a patient missing several months of injections—to visualize the potential impact on their long-term prognosis. This allows for highly personalized treatment management.

Non-Clinical Drivers of Adherence

Interpretability analysis using Integrated Gradients highlighted that logistical factors, specifically the distance to the clinic, were more significant predictors of dropout than many biological markers. Additionally, the study noted that the COVID-19 pandemic created significant external barriers to treatment adherence that models must account for in real-world applications.

Conclusion

Sequential models offer a dynamic and effective approach to managing long-term immunotherapy. While LSTMs are highly effective at identifying patients at risk of dropping out, SLVMs provide the granular symptom forecasting and simulation capabilities necessary for personalized clinical intervention. Integrating these models into digital health platforms could significantly improve the success rates of SCIT for allergic rhinitis patients.

Summary updated

Abstract

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{yin2024sequential,
  author = {Li Yin 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