Create views for vasoactive infusion durations similar to MIMIC-III

concepts/durations/dopamine_durations.sql

@@ -0,0 +1,80 @@
+
+-- This query extracts durations of vasopressin administration. 
+-- It is heavily based on the MIMIC-III duration code
+-- Consecutive durations are numbered and administations are ended after 18 hours without 
+-- a charted value. 18 hours is a little arbitrary, but there are some vasopressin infusions
+-- that are long standing and non-titrated (and will have 14 hour gaps in charting)
+
+CREATE OR REPLACE FUNCTION SAFE_CAST_TO_NUMERIC(v_input text) RETURNS NUMERIC AS $$
+-- This casts a string to a numeric value without throwing an exception if the 
+-- sting is not 'numeric-like'. For BigQuery, use SAFE_CAST(x AS NUMERIC)
+DECLARE v_numeric_value NUMERIC DEFAULT NULL;
+BEGIN
+    BEGIN
+        v_numeric_value := v_input::NUMERIC;
+    EXCEPTION WHEN OTHERS THEN
+        RETURN NULL;
+    END;
+RETURN v_numeric_value;
+END;
+$$ LANGUAGE plpgsql;
+
+DROP MATERIALIZED VIEW IF EXISTS dopaminedurations;
+CREATE MATERIALIZED VIEW dopaminedurations AS 
+WITH drug_events AS (
+  SELECT
+      patientunitstayid
+    , infusionoffset
+    , drugrate
+    , CASE WHEN SAFE_CAST_TO_NUMERIC(drugrate) > 0 THEN 1 ELSE 0 END AS vasopressor_event
+    , CASE WHEN drugrate LIKE '%OFF%' THEN 1
+        WHEN SAFE_CAST_TO_NUMERIC(drugrate) = 0 THEN 1
+        ELSE 0 END AS stop_event
+  FROM
+    infusiondrug
+  WHERE
+    LOWER(drugname) LIKE 'dopamine%'
+), d0 AS (
+  SELECT
+    patientunitstayid 
+    , infusionoffset 
+    , drugrate 
+    , vasopressor_event
+    , stop_event
+    , LAG(InfusionOffset, 1) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS InfusionOffsetLag
+  FROM
+    drug_events 
+), d1 AS (
+  SELECT
+      patientunitstayid 
+      , infusionoffset 
+      , drugrate 
+      , vasopressor_event
+      , stop_event
+      , CASE 
+        -- Case 1: The row prior was a stop event. This is either part of a string of zeros
+        -- or the start of a new event
+          WHEN Lag(stop_event, 1, 1) OVER (PARTITION BY patientunitstayid  ORDER BY infusionoffset) = 1 THEN
+          	CASE WHEN stop_event = 1 THEN 0 ELSE 1 END
+        -- Case 2: If the row prior was more than 18 hours ago, end this is a new event
+          WHEN InfusionOffset - InfusionOffsetLag > (18*60) THEN 1
+        -- Otherwise, This is _not_ a new event.
+          ELSE 0
+       END AS new_event
+  FROM d0
+), d2 AS (
+  SELECT
+    d1.*
+    -- Generate a sequential integer for convience
+    , SUM(new_event) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS vasopressor_episode
+  FROM
+    d1
+)
+SELECT
+  patientunitstayid
+  , 'Dopamine' AS Drug
+  , vasopressor_episode
+  , MIN(infusionoffset) AS start_offset
+  , MAX(infusionoffset) AS end_offset
+FROM d2
+GROUP BY patientunitstayid, vasopressor_episode;

concepts/durations/epinepherine_durations.sql

@@ -0,0 +1,80 @@
+
+-- This query extracts durations of vasopressin administration. 
+-- It is heavily based on the MIMIC-III duration code
+-- Consecutive durations are numbered and administations are ended after 18 hours without 
+-- a charted value. 18 hours is a little arbitrary, but there are some vasopressin infusions
+-- that are long standing and non-titrated (and will have 14 hour gaps in charting)
+
+CREATE OR REPLACE FUNCTION SAFE_CAST_TO_NUMERIC(v_input text) RETURNS NUMERIC AS $$
+-- This casts a string to a numeric value without throwing an exception if the 
+-- sting is not 'numeric-like'. For BigQuery, use SAFE_CAST(x AS NUMERIC)
+DECLARE v_numeric_value NUMERIC DEFAULT NULL;
+BEGIN
+    BEGIN
+        v_numeric_value := v_input::NUMERIC;
+    EXCEPTION WHEN OTHERS THEN
+        RETURN NULL;
+    END;
+RETURN v_numeric_value;
+END;
+$$ LANGUAGE plpgsql;
+
+DROP MATERIALIZED VIEW IF EXISTS epinephrinedurations;
+CREATE MATERIALIZED VIEW epinephrinedurations AS 
+WITH drug_events AS (
+  SELECT
+      patientunitstayid
+    , infusionoffset
+    , drugrate
+    , CASE WHEN SAFE_CAST_TO_NUMERIC(drugrate) > 0 THEN 1 ELSE 0 END AS vasopressor_event
+    , CASE WHEN drugrate LIKE '%OFF%' THEN 1
+        WHEN SAFE_CAST_TO_NUMERIC(drugrate) = 0 THEN 1
+        ELSE 0 END AS stop_event
+  FROM
+    infusiondrug
+  WHERE
+    LOWER(drugname) LIKE 'epinephrine%'
+), d0 AS (
+  SELECT
+    patientunitstayid 
+    , infusionoffset 
+    , drugrate 
+    , vasopressor_event
+    , stop_event
+    , LAG(InfusionOffset, 1) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS InfusionOffsetLag
+  FROM
+    drug_events 
+), d1 AS (
+  SELECT
+      patientunitstayid 
+      , infusionoffset 
+      , drugrate 
+      , vasopressor_event
+      , stop_event
+      , CASE 
+        -- Case 1: The row prior was a stop event. This is either part of a string of zeros
+        -- or the start of a new event
+          WHEN Lag(stop_event, 1, 1) OVER (PARTITION BY patientunitstayid  ORDER BY infusionoffset) = 1 THEN
+          	CASE WHEN stop_event = 1 THEN 0 ELSE 1 END
+        -- Case 2: If the row prior was more than 18 hours ago, end this is a new event
+          WHEN InfusionOffset - InfusionOffsetLag > (18*60) THEN 1
+        -- Otherwise, This is _not_ a new event.
+          ELSE 0
+       END AS new_event
+  FROM d0
+), d2 AS (
+  SELECT
+    d1.*
+    -- Generate a sequential integer for convience
+    , SUM(new_event) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS vasopressor_episode
+  FROM
+    d1
+)
+SELECT
+  patientunitstayid
+  , 'Epinephrine' AS Drug
+  , vasopressor_episode
+  , MIN(infusionoffset) AS start_offset
+  , MAX(infusionoffset) AS end_offset
+FROM d2
+GROUP BY patientunitstayid, vasopressor_episode;

concepts/durations/norepinepherine_durations.sql

@@ -0,0 +1,80 @@
+
+-- This query extracts durations of vasopressin administration. 
+-- It is heavily based on the MIMIC-III duration code
+-- Consecutive durations are numbered and administations are ended after 18 hours without 
+-- a charted value. 18 hours is a little arbitrary, but there are some vasopressin infusions
+-- that are long standing and non-titrated (and will have 14 hour gaps in charting)
+
+CREATE OR REPLACE FUNCTION SAFE_CAST_TO_NUMERIC(v_input text) RETURNS NUMERIC AS $$
+-- This casts a string to a numeric value without throwing an exception if the 
+-- sting is not 'numeric-like'. For BigQuery, use SAFE_CAST(x AS NUMERIC)
+DECLARE v_numeric_value NUMERIC DEFAULT NULL;
+BEGIN
+    BEGIN
+        v_numeric_value := v_input::NUMERIC;
+    EXCEPTION WHEN OTHERS THEN
+        RETURN NULL;
+    END;
+RETURN v_numeric_value;
+END;
+$$ LANGUAGE plpgsql;
+
+DROP MATERIALIZED VIEW IF EXISTS norepinephrinedurations;
+CREATE MATERIALIZED VIEW norepinephrinedurations AS 
+WITH drug_events AS (
+  SELECT
+      patientunitstayid
+    , infusionoffset
+    , drugrate
+    , CASE WHEN SAFE_CAST_TO_NUMERIC(drugrate) > 0 THEN 1 ELSE 0 END AS vasopressor_event
+    , CASE WHEN drugrate LIKE '%OFF%' THEN 1
+        WHEN SAFE_CAST_TO_NUMERIC(drugrate) = 0 THEN 1
+        ELSE 0 END AS stop_event
+  FROM
+    infusiondrug
+  WHERE
+    LOWER(drugname) LIKE 'norepinephrine%'
+), d0 AS (
+  SELECT
+    patientunitstayid 
+    , infusionoffset 
+    , drugrate 
+    , vasopressor_event
+    , stop_event
+    , LAG(InfusionOffset, 1) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS InfusionOffsetLag
+  FROM
+    drug_events 
+), d1 AS (
+  SELECT
+      patientunitstayid 
+      , infusionoffset 
+      , drugrate 
+      , vasopressor_event
+      , stop_event
+      , CASE 
+        -- Case 1: The row prior was a stop event. This is either part of a string of zeros
+        -- or the start of a new event
+          WHEN Lag(stop_event, 1, 1) OVER (PARTITION BY patientunitstayid  ORDER BY infusionoffset) = 1 THEN
+          	CASE WHEN stop_event = 1 THEN 0 ELSE 1 END
+        -- Case 2: If the row prior was more than 18 hours ago, end this is a new event
+          WHEN InfusionOffset - InfusionOffsetLag > (18*60) THEN 1
+        -- Otherwise, This is _not_ a new event.
+          ELSE 0
+       END AS new_event
+  FROM d0
+), d2 AS (
+  SELECT
+    d1.*
+    -- Generate a sequential integer for convience
+    , SUM(new_event) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS vasopressor_episode
+  FROM
+    d1
+)
+SELECT
+  patientunitstayid
+  , 'Norepinephrine' AS Drug
+  , vasopressor_episode
+  , MIN(infusionoffset) AS start_offset
+  , MAX(infusionoffset) AS end_offset
+FROM d2
+GROUP BY patientunitstayid, vasopressor_episode;

concepts/durations/phenylepherine_durations.sql

@@ -0,0 +1,80 @@
+
+-- This query extracts durations of vasopressin administration. 
+-- It is heavily based on the MIMIC-III duration code
+-- Consecutive durations are numbered and administations are ended after 18 hours without 
+-- a charted value. 18 hours is a little arbitrary, but there are some vasopressin infusions
+-- that are long standing and non-titrated (and will have 14 hour gaps in charting)
+
+CREATE OR REPLACE FUNCTION SAFE_CAST_TO_NUMERIC(v_input text) RETURNS NUMERIC AS $$
+-- This casts a string to a numeric value without throwing an exception if the 
+-- sting is not 'numeric-like'. For BigQuery, use SAFE_CAST(x AS NUMERIC)
+DECLARE v_numeric_value NUMERIC DEFAULT NULL;
+BEGIN
+    BEGIN
+        v_numeric_value := v_input::NUMERIC;
+    EXCEPTION WHEN OTHERS THEN
+        RETURN NULL;
+    END;
+RETURN v_numeric_value;
+END;
+$$ LANGUAGE plpgsql;
+
+DROP MATERIALIZED VIEW IF EXISTS phenylephrinedurations;
+CREATE MATERIALIZED VIEW phenylephrinedurations AS 
+WITH drug_events AS (
+  SELECT
+      patientunitstayid
+    , infusionoffset
+    , drugrate
+    , CASE WHEN SAFE_CAST_TO_NUMERIC(drugrate) > 0 THEN 1 ELSE 0 END AS vasopressor_event
+    , CASE WHEN drugrate LIKE '%OFF%' THEN 1
+        WHEN SAFE_CAST_TO_NUMERIC(drugrate) = 0 THEN 1
+        ELSE 0 END AS stop_event
+  FROM
+    infusiondrug
+  WHERE
+    LOWER(drugname) LIKE 'phenylephrine%'
+), d0 AS (
+  SELECT
+    patientunitstayid 
+    , infusionoffset 
+    , drugrate 
+    , vasopressor_event
+    , stop_event
+    , LAG(InfusionOffset, 1) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS InfusionOffsetLag
+  FROM
+    drug_events 
+), d1 AS (
+  SELECT
+      patientunitstayid 
+      , infusionoffset 
+      , drugrate 
+      , vasopressor_event
+      , stop_event
+      , CASE 
+        -- Case 1: The row prior was a stop event. This is either part of a string of zeros
+        -- or the start of a new event
+          WHEN Lag(stop_event, 1, 1) OVER (PARTITION BY patientunitstayid  ORDER BY infusionoffset) = 1 THEN
+          	CASE WHEN stop_event = 1 THEN 0 ELSE 1 END
+        -- Case 2: If the row prior was more than 18 hours ago, end this is a new event
+          WHEN InfusionOffset - InfusionOffsetLag > (18*60) THEN 1
+        -- Otherwise, This is _not_ a new event.
+          ELSE 0
+       END AS new_event
+  FROM d0
+), d2 AS (
+  SELECT
+    d1.*
+    -- Generate a sequential integer for convience
+    , SUM(new_event) OVER (PARTITION BY patientunitstayid ORDER BY infusionoffset) AS vasopressor_episode
+  FROM
+    d1
+)
+SELECT
+  patientunitstayid
+  , 'Phenylephrine' AS Drug
+  , vasopressor_episode
+  , MIN(infusionoffset) AS start_offset
+  , MAX(infusionoffset) AS end_offset
+FROM d2
+GROUP BY patientunitstayid, vasopressor_episode;