Research map / Time series / Foundation models for time series
Pharmacokinetics-Informed Neural Network for Predicting Opioid Administration Moments with Wearable Sensors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Detects oral opioid administrations and predicts their timing from wrist-sensor physiological signals by combining self-supervised pretraining with a pharmacokinetics-based auxiliary loss on relative plasma concentration.
Problem
Objectively monitoring oral opioid administration and predicting administration moments in inpatient and outpatient settings, where self-reports and toxicology are biased or impractical.
Approach
Pretrain a time-series model on a self-supervised task of detecting synthetic signal modifications, then fine-tune with a hybrid loss that adds an auxiliary RMSE term matching predicted relative plasma concentration to a single-compartment oxycodone pharmacokinetic curve, PINN-style.
Key results
- CTA-TCN with SSL and PK auxiliary task reached NMAE 0.09 in inpatient setting (SSL-only 0.13) and NMAE 0.11 with R2 0.77 in outpatient setting versus R2 0.29 without PK.
- Self-supervised pretraining improved F1 across all five models in both settings; CTA-TCN inpatient F1 rose from 0.73 to 0.79.
- Outpatient CTA-TCN F1 was 0.70 with SSL versus 0.65 without SSL, and NMAE 0.18 versus 0.27.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add self-supervised pre-training (ablation) | Time-series models · Inpatient | F1-score: 0.73 -> 0.79 | Helped |
| Add self-supervised pre-training (ablation) | Time-series models · Inpatient | NMAE: 0.21 -> 0.13 | Helped |
| Add self-supervised pre-training (ablation) | Time-series models · Outpatient | F1-score: 0.64 -> 0.70 | Helped |
| Add self-supervised pre-training (ablation) | Time-series models · Outpatient | NMAE: 0.27 -> 0.18 | Helped |
| Add pharmacokinetics-informed auxiliary loss (ablation) | CTA-TCN + SSL · Inpatient | NMAE: 0.13 -> 0.09 | Helped |
| Add pharmacokinetics-informed auxiliary loss (ablation) | CTA-TCN + SSL · Outpatient | NMAE: 0.18 -> 0.11 | Helped |
| Add pharmacokinetics-informed auxiliary loss (ablation) | Time-series models + SSL · Inpatient and outpatient | NMAE | Helped |
Surprising findings
- Pharmacokinetic auxiliary loss gave a large R2 gain in the outpatient setting but a non-uniform effect in inpatient results, where NMAE changed only slightly while R2 dropped from 0.79 to 0.53 for one configuration.
Limitations
- Data from a research-grade Empatica E4 device, impractical for general consumer use; outpatient labels rely on self-report only.
- Only oxycodone PK parameters were used; generalization to other opioids (fentanyl, morphine) is untested.
- Small sample size (42 participants, 227 administrations).
Experimental setup
Datasets: In-house inpatient E4 wearable data In-house outpatient E4 wearable data
Benchmarks: In-house inpatient opioid administration dataset In-house outpatient opioid administration dataset
Metrics: weighted F1-score specificity sensitivity normalized MAE R2
Baselines: BiLSTM LSTM-FCN ResNet InceptionTime CTA-TCN
Builds on: physics-informed neural networks CTA-TCN LSTM-FCN InceptionTime
Closest papers by meaning
- A Cautionary Tale on Integrating Studies with Disparate Outcome Measures for Causal Inference (NeurIPS 2025 · similarity 0.82)
- Data-Driven Discovery of Dynamical Systems in Pharmacology using Large Language Models (NeurIPS 2024 · similarity 0.81)
- Unraveling Pain Levels: A Data-Uncertainty Guided Approach for Effective Pain Assessment (AAAI 2024 · similarity 0.81)
- PEACE: A Dataset of Pharmaceutical Care for Cancer Pain Analgesia Evaluation and Medication Decision (NeurIPS 2024 · similarity 0.81)
- OpenMHC: Accelerating the Science of Wearable Foundation Models (NeurIPS 2026 · similarity 0.80)
- Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions (ICML 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.