diff --git a/src/transform/loop_unswitching.cc b/src/transform/loop_unswitching.cc
index aab8f7fdc..7767866bb 100644
--- a/src/transform/loop_unswitching.cc
+++ b/src/transform/loop_unswitching.cc
@@ -201,8 +201,8 @@ bool UsesLoopVarThroughLetBindings(
         auto it = let_bindings->find(var_node);
         if (it != let_bindings->end()) {
           // Check if the bound expression uses the loop variable
-          if (UsesVar(it->second,
-                      [&](const VarNode *v) { return v == loop_var.get(); })) {
+          if (UsesLoopVarThroughLetBindings(it->second, loop_var,
+                                            let_bindings)) {
             uses_loop_var = true;
           }
         }
diff --git a/testing/python/transform/test_tilelang_transform_loop_unswitching.py b/testing/python/transform/test_tilelang_transform_loop_unswitching.py
index 91ff355b9..212b57732 100644
--- a/testing/python/transform/test_tilelang_transform_loop_unswitching.py
+++ b/testing/python/transform/test_tilelang_transform_loop_unswitching.py
@@ -386,5 +386,19 @@ def expected(
     _check(before, expected)
 
 
+def test_no_hoist_multiple_let():
+    @tilelang.jit()
+    def get_fused_mapping_kernel(topk_idx: T.Tensor[(1,), T.int32]):
+        with T.Kernel():
+            _tmp1 = T.alloc_shared((1,), "int")
+            for i in T.serial(0, 4, 2):
+                _tmp2 = topk_idx[i]
+                T.assume(0 <= _tmp2 < 1)
+                if _tmp2 != -1:
+                    T.atomic_add(_tmp1[_tmp2], 1)
+
+    get_fused_mapping_kernel.compile()
+
+
 if __name__ == "__main__":
     tilelang.testing.main()