PR #1028: TPC Model Implementation for ICU Length-of-Stay Prediction

docs/api/models.rst

@@ -196,6 +196,7 @@ API Reference
     models/pyhealth.models.GRASP
     models/pyhealth.models.MedLink
     models/pyhealth.models.TCN
+    models/pyhealth.models.TPC
     models/pyhealth.models.TFMTokenizer
     models/pyhealth.models.GAN
     models/pyhealth.models.VAE

docs/api/models/pyhealth.models.tpc.rst

@@ -0,0 +1,44 @@
+pyhealth.models.TPC
+===================
+
+Temporal Pointwise Convolution (TPC) model for ICU remaining length-of-stay prediction.
+
+Overview
+--------
+
+The TPC model combines grouped temporal convolutions with pointwise (1x1) convolutions to 
+capture both feature-specific temporal patterns and cross-feature interactions at each timestep. 
+The architecture is specifically designed for irregularly sampled multivariate time series in 
+intensive care settings.
+
+**Paper Reference:**
+Rocheteau, E., Liò, P., & Hyland, S. (2021). Temporal Pointwise Convolutional Networks for 
+Length of Stay Prediction in the Intensive Care Unit. In Proceedings of the Conference on 
+Health, Inference, and Learning (CHIL).
+
+**Key Features:**
+
+- Grouped temporal convolutions (one group per clinical feature)
+- Pointwise convolutions for cross-feature learning
+- Skip connections with hierarchical feature aggregation
+- Custom MSLE (Masked Mean Squared Logarithmic Error) loss
+- Monte Carlo Dropout for uncertainty estimation (extension)
+
+**Model Classes:**
+
+.. autoclass:: pyhealth.models.TPC
+    :members:
+    :undoc-members:
+    :show-inheritance:
+
+**Loss Functions:**
+
+.. autoclass:: pyhealth.models.MSLELoss
+    :members:
+    :undoc-members:
+    :show-inheritance:
+
+.. autoclass:: pyhealth.models.MaskedMSELoss
+    :members:
+    :undoc-members:
+    :show-inheritance:

docs/api/tasks.rst

@@ -213,6 +213,7 @@ Available Tasks
     DKA Prediction (MIMIC-IV) <tasks/pyhealth.tasks.dka>
     Drug Recommendation <tasks/pyhealth.tasks.drug_recommendation>
     Length of Stay Prediction <tasks/pyhealth.tasks.length_of_stay_prediction>
+    Remaining LoS (TPC MIMIC-IV) <tasks/pyhealth.tasks.length_of_stay_tpc_mimic4>
     Medical Transcriptions Classification <tasks/pyhealth.tasks.MedicalTranscriptionsClassification>
     Mortality Prediction (Next Visit) <tasks/pyhealth.tasks.mortality_prediction>
     Mortality Prediction (StageNet MIMIC-IV) <tasks/pyhealth.tasks.mortality_prediction_stagenet_mimic4>

docs/api/tasks/pyhealth.tasks.length_of_stay_tpc_mimic4.rst

@@ -0,0 +1,67 @@
+pyhealth.tasks.length_of_stay_tpc_mimic4
+===========================================
+
+Remaining ICU length-of-stay prediction task for MIMIC-IV with TPC-compatible preprocessing.
+
+Overview
+--------
+
+The RemainingLOSMIMIC4 task generates samples for predicting remaining ICU length of stay at 
+hourly intervals. Unlike traditional length-of-stay tasks that predict total stay duration at 
+admission, this task formulates the problem as a time-series regression where the model predicts 
+remaining hours at each timestep throughout the ICU stay.
+
+**Input Features:**
+
+- **Timeseries** ``(2F+2, T)``: Hourly clinical measurements with:
+
+  - Elapsed time channel (1)
+  - Feature values from chartevents and labevents (F channels)
+  - Decay indicators showing time since last measurement (F channels)
+  - Hour of day (1 channel)
+
+- **Static** ``(2,)``: Patient demographics (age, sex)
+
+- **Conditions**: ICD diagnosis codes from admission
+
+**Output:**
+
+- **Remaining LoS** ``(T,)``: Remaining hours in ICU at each timestep
+
+
+**Default Configuration:**
+
+- Prediction step size: 1 hour
+- Minimum history: 5 hours before predictions start
+- Minimum remaining stay: 1 hour
+- Maximum history window: 366 hours (15.25 days)
+- Clinical features: 17 chartevents + 17 labevents = 34 features
+
+**Clinical Features:**
+
+*Vital Signs (chartevents):*
+
+- Heart rate, blood pressure (systolic/diastolic/mean)
+- Respiratory rate, SpO2, temperature
+- Glasgow Coma Scale components
+- Urine output, weight
+
+*Laboratory Values (labevents):*
+
+- Hematology: WBC, platelets, hemoglobin, hematocrit
+- Chemistry: sodium, potassium, chloride, bicarbonate
+- Renal: BUN, creatinine
+- Metabolic: glucose, lactate
+- Liver: bilirubin, ALT
+API Reference
+-------------
+
+.. autoclass:: pyhealth.tasks.length_of_stay_tpc_mimic4.RemainingLOSMIMIC4
+    :members:
+    :undoc-members:
+    :show-inheritance:
+
+.. autoclass:: pyhealth.tasks.length_of_stay_tpc_mimic4.RemainingLOSConfig
+    :members:
+    :undoc-members:
+    :show-inheritance: