CNAM-154 Added filter to remove patients without outcomes

CNAM-154 Added comments to config file

Author: danielpes

src/main/resources/config/filtering-default.conf

@@ -41,22 +41,23 @@ default = {
 
   }
   mlpp_parameters = {
-    bucket_size = [30]  # in days
-    lag_count = [10]
-    min_timestamp = ${default.dates.study_start}
-    max_timestamp = ${default.dates.study_end}
-    include_death_bucket = false
+    bucket_size = [30]  # Number of days of each bucket of time
+    lag_count = [10]  # Number of lags to be created
+    min_timestamp = ${default.dates.study_start}  # Min timestamp to be considered by the time buckets
+    max_timestamp = ${default.dates.study_end}  # Max timestamp to be considered by the time buckets
+    include_death_bucket = false  # If false, the row corresponding to the death bucket is filled with 0s
 
     exposures = {
-      min_purchases = 1
-      start_delay = 0
-      purchases_window = 0
-      only_first = false
-      filter_lost_patients = false
-      filter_diagnosed_patients = true
-      diagnosed_patients_threshold = 0
-      filter_delayed_entries = true
-      delayed_entry_threshold = 12
+      min_purchases = 1  # Number of drug purchases within <purchases_window> to form an exposure
+      start_delay = 0  # Months after the drug purchases for the exposure to start
+      purchases_window = 0  # Period in months to look for multiple drug purchases
+      only_first = false  # If true, only the first exposure is kept for each <patient, molecule> pair
+      filter_never_sick_patients = false  # if true, patients who never got a target disease are removed
+      filter_lost_patients = false  # if true, patients with a trackloss are removed
+      filter_diagnosed_patients = true  # if true, patients with an early diagnostic are removed
+      diagnosed_patients_threshold = 0  # number of months after study start for a diagnostic to be considered "early"
+      filter_delayed_entries = true  # if true, patients who entered the study (i.e. first drug purchase) after a delay are removed
+      delayed_entry_threshold = 12  # delay in months for the delayed entries filter
     }
   }
 }

src/main/scala/fr/polytechnique/cmap/cnam/filtering/mlpp/MLPPConfig.scala

@@ -13,8 +13,9 @@ object MLPPConfig {
     startDelay: Int,
     purchasesWindow: Int,
     onlyFirst: Boolean,
+    filterNeverSickPatients: Boolean,
     filterLostPatients: Boolean,
-    filterDiagnosedPatients: Boolean,
+    filterEarlyDiagnosedPatients: Boolean,
     diagnosedPatientsThreshold: Int,
     filterDelayedEntries: Boolean,
     delayedEntryThreshold: Int
@@ -33,8 +34,9 @@ object MLPPConfig {
     startDelay = conf.getInt("exposures.start_delay"),
     purchasesWindow = conf.getInt("exposures.purchases_window"),
     onlyFirst = conf.getBoolean("exposures.only_first"),
+    filterNeverSickPatients = conf.getBoolean("exposures.filter_never_sick_patients"),
     filterLostPatients = conf.getBoolean("exposures.filter_lost_patients"),
-    filterDiagnosedPatients = conf.getBoolean("exposures.filter_diagnosed_patients"),
+    filterEarlyDiagnosedPatients = conf.getBoolean("exposures.filter_diagnosed_patients"),
     diagnosedPatientsThreshold = conf.getInt("exposures.diagnosed_patients_threshold"),
     filterDelayedEntries = conf.getBoolean("exposures.filter_delayed_entries"),
     delayedEntryThreshold = conf.getInt("exposures.delayed_entry_threshold")

src/main/scala/fr/polytechnique/cmap/cnam/filtering/mlpp/MLPPExposuresTransformer.scala

@@ -15,8 +15,9 @@ object MLPPExposuresTransformer extends ExposuresTransformer {
   private def filterLostPatients = MLPPConfig.exposureDefinition.filterLostPatients
   private def filterDelayedEntries  = MLPPConfig.exposureDefinition.filterDelayedEntries
   private def delayedEntryThreshold = MLPPConfig.exposureDefinition.delayedEntryThreshold
-  private def filterDiagnosedPatients = MLPPConfig.exposureDefinition.filterDiagnosedPatients
+  private def filterEarlyDiagnosedPatients = MLPPConfig.exposureDefinition.filterEarlyDiagnosedPatients
   private def diagnosedPatientsThreshold = MLPPConfig.exposureDefinition.diagnosedPatientsThreshold
+  private def filterNeverSickPatients = MLPPConfig.exposureDefinition.filterNeverSickPatients
 
   val outputColumns = List(
     col("patientID"),
@@ -33,9 +34,9 @@ object MLPPExposuresTransformer extends ExposuresTransformer {
   implicit class ExposuresDataFrame(data: DataFrame) {
 
     /**
-      * Drops patients whose got a target disease before periodStart + delay (default = 0)
+      * Drops patients who got a target disease before periodStart + delay (default = 0)
       */
-    def filterDiagnosedPatients(doFilter: Boolean): DataFrame = {
+    def filterEarlyDiagnosedPatients(doFilter: Boolean): DataFrame = {
 
       if (doFilter) {
         val window = Window.partitionBy("patientID")
@@ -104,6 +105,29 @@ object MLPPExposuresTransformer extends ExposuresTransformer {
       }
     }
 
+    /**
+      * Drops patients who never had a target disease event
+      */
+    def filterNeverSickPatients(doFilter: Boolean): DataFrame = {
+
+      if (doFilter) {
+        val window = Window.partitionBy("patientID")
+
+        val filterColumn: Column = max(
+          when(
+            (col("category") === "disease") &&
+            (col("eventId") === "targetDisease") &&
+            (col("start") <= MLPPConfig.maxTimestamp), lit(1)
+          ).otherwise(lit(0))
+        ).over(window).cast(BooleanType)
+
+        data.withColumn("filter", filterColumn).where(col("filter")).drop("filter")
+      }
+      else {
+        data
+      }
+    }
+
     def withExposureStart(minPurchases: Int = 1, intervalSize: Int = 6,
         startDelay: Int = 0, firstOnly: Boolean = false): DataFrame = {
 
@@ -142,8 +166,9 @@ object MLPPExposuresTransformer extends ExposuresTransformer {
 
     input.toDF
       .filterDelayedEntries(filterDelayedEntries)
-      .filterDiagnosedPatients(filterDiagnosedPatients)
+      .filterEarlyDiagnosedPatients(filterEarlyDiagnosedPatients)
       .filterLostPatients(filterLostPatients)
+      .filterNeverSickPatients(filterNeverSickPatients)
       .where(col("category") === "molecule")
       .withExposureStart(
         minPurchases = minPurchases,

src/test/scala/fr/polytechnique/cmap/cnam/filtering/mlpp/MLPPExposuresTransformerSuite.scala

@@ -77,7 +77,7 @@ class MLPPExposuresTransformerSuite extends SharedContext {
 
     // When
     import MLPPExposuresTransformer.ExposuresDataFrame
-    val result = input.filterDiagnosedPatients(true).select("patientID", "category")
+    val result = input.filterEarlyDiagnosedPatients(true).select("patientID", "category")
 
     // Then
     import RichDataFrames._
@@ -100,7 +100,7 @@ class MLPPExposuresTransformerSuite extends SharedContext {
 
     // When
     import MLPPExposuresTransformer.ExposuresDataFrame
-    val result = input.filterDiagnosedPatients(false)
+    val result = input.filterEarlyDiagnosedPatients(false)
 
     // Then
     import RichDataFrames._
@@ -135,6 +135,37 @@ class MLPPExposuresTransformerSuite extends SharedContext {
     assert(result === expected)
   }
 
+  "filterNeverSickPatients" should "remove patients who never have a target disease" in {
+    val sqlCtx = sqlContext
+    import sqlCtx.implicits._
+
+    // Given
+    val input = Seq(
+      ("Patient_A", "molecule", "", makeTS(2006, 1, 1)),
+      ("Patient_A", "molecule", "", makeTS(2006, 3, 1)),
+      ("Patient_A", "disease", "targetDisease", makeTS(2006, 6, 1)),
+      ("Patient_B", "molecule", "", makeTS(2006, 5, 1)),
+      ("Patient_B", "molecule", "", makeTS(2007, 1, 1)),
+      ("Patient_C", "molecule", "", makeTS(2006, 11, 1))
+    ).toDF("patientID", "category", "eventId", "start")
+
+    val expected = Seq(
+      ("Patient_A", "molecule", "", makeTS(2006, 1, 1)),
+      ("Patient_A", "molecule", "", makeTS(2006, 3, 1)),
+      ("Patient_A", "disease", "targetDisease", makeTS(2006, 6, 1))
+    ).toDF("patientID", "category", "eventId", "start")
+
+    // When
+    import MLPPExposuresTransformer.ExposuresDataFrame
+    val result = input.filterNeverSickPatients(true)
+
+    // Then
+    import RichDataFrames._
+    result.show
+    expected.show
+    assert(result === expected)
+  }
+
   "withExposureStart" should "add a column with the start of the default MLPP exposure definition" in {
     val sqlCtx = sqlContext
     import sqlCtx.implicits._