diff --git a/pyhealth/datasets/tuev.py b/pyhealth/datasets/tuev.py
index f7fdd667..7e8dacf9 100644
--- a/pyhealth/datasets/tuev.py
+++ b/pyhealth/datasets/tuev.py
@@ -81,6 +81,24 @@ def __init__(
             raise ValueError("subset must be one of 'train', 'eval', or 'both'")
         
         self.prepare_metadata()
+        
+        # Determine where the CSVs are located (shared directory or cache)
+        root_path = Path(root)
+        cache_dir = Path.home() / ".cache" / "pyhealth" / "tuev"
+        
+        # Check if CSVs exist in cache and not in shared location
+        use_cache = False
+        for table in tables:
+            shared_csv = root_path / f"tuev-{table}-pyhealth.csv"
+            cache_csv = cache_dir / f"tuev-{table}-pyhealth.csv"
+            if not shared_csv.exists() and cache_csv.exists():
+                use_cache = True
+                break
+        
+        # Use cache directory as root if CSVs are there
+        if use_cache:
+            logger.info(f"Using cached metadata from {cache_dir}")
+            root = str(cache_dir)
             
         super().__init__(
             root=root,
@@ -107,12 +125,16 @@ def prepare_metadata(self) -> None:
         - segment_id = a_ / a_1 / ...
         """
         root = Path(self.root)
+        cache_dir = Path.home() / ".cache" / "pyhealth" / "tuev"
 
         train_rows: list[dict] = []
         eval_rows: list[dict] = []
 
         for split in ("train", "eval"):
-            if os.path.exists(root / f"tuev-{split}-pyhealth.csv"):
+            # Check if metadata exists in either shared location or cache
+            shared_csv = root / f"tuev-{split}-pyhealth.csv"
+            cache_csv = cache_dir / f"tuev-{split}-pyhealth.csv"
+            if shared_csv.exists() or cache_csv.exists():
                 continue
             
             split_dir = root / split
@@ -153,7 +175,9 @@ def prepare_metadata(self) -> None:
                             }
                         )
 
-        root.mkdir(parents=True, exist_ok=True)
+        # Setup cache directory as fallback for metadata CSVs
+        cache_dir = Path.home() / ".cache" / "pyhealth" / "tuev"
+        cache_dir.mkdir(parents=True, exist_ok=True)
 
         # Write train metadata
         if train_rows:
@@ -162,9 +186,18 @@ def prepare_metadata(self) -> None:
                 ["patient_id", "record_id"], inplace=True, na_position="last"
             )
             train_df.reset_index(drop=True, inplace=True)
-            train_csv = root / "tuev-train-pyhealth.csv"
-            train_df.to_csv(train_csv, index=False)
-
+            
+            # Try shared location first, fall back to cache if no write permission
+            train_csv_shared = root / "tuev-train-pyhealth.csv"
+            train_csv_cache = cache_dir / "tuev-train-pyhealth.csv"
+            
+            try:
+                train_csv_shared.parent.mkdir(parents=True, exist_ok=True)
+                train_df.to_csv(train_csv_shared, index=False)
+                logger.info(f"Wrote train metadata to {train_csv_shared}")
+            except (PermissionError, OSError):
+                train_df.to_csv(train_csv_cache, index=False)
+                logger.info(f"Wrote train metadata to cache: {train_csv_cache}")
 
         # Write eval metadata
         if eval_rows:
@@ -175,8 +208,18 @@ def prepare_metadata(self) -> None:
                 na_position="last",
             )
             eval_df.reset_index(drop=True, inplace=True)
-            eval_csv = root / "tuev-eval-pyhealth.csv"
-            eval_df.to_csv(eval_csv, index=False)
+            
+            # Try shared location first, fall back to cache if no write permission
+            eval_csv_shared = root / "tuev-eval-pyhealth.csv"
+            eval_csv_cache = cache_dir / "tuev-eval-pyhealth.csv"
+            
+            try:
+                eval_csv_shared.parent.mkdir(parents=True, exist_ok=True)
+                eval_df.to_csv(eval_csv_shared, index=False)
+                logger.info(f"Wrote eval metadata to {eval_csv_shared}")
+            except (PermissionError, OSError):
+                eval_df.to_csv(eval_csv_cache, index=False)
+                logger.info(f"Wrote eval metadata to cache: {eval_csv_cache}")
     
     @property
     def default_task(self) -> EEGEventsTUEV: