fix: 🐛 Ensure mortality for cancer is larger than the general population

Previously, the mortality for people with and without cancer were assessed separately, and mortality for people with cancer was independent of age. This lead to some weird results, where mortality for some older ages could be greater if a person did not have cancer, than if it had it. This commit fixes this bug by considering that cancer mortality is always additional to a baseline (non-cancer related) age-adjusted mortality.

seed/natural_history_params.yaml

@@ -477,21 +477,6 @@ ALL_TYPES:
   # NOTE: The following transitions do not come from Kim et al. (2017)        #
   # ######################################################################### #
 
-  # Undetected cancer mortalities
-  # We take the steepest year of mortality from Carmo & Luiz (2011) and assume
-  # this is the constant mortality for undetected cancers. In fact, we do not
-  # know the true mortality rates for undetected cancers, but this approach at
-  # least ensures that detecting a cancer will always be as good as or better
-  # than not detecting it.
-  "LOCAL_UNDETECTED,DECEASED":
-    0: 0.0042
-
-  "REGIONAL_UNDETECTED,DECEASED":
-    0: 0.0287
-
-  "DISTANT_UNDETECTED,DECEASED":
-    0: 0.0821
-
   # Detected cancer cure
   # We assume all cancers get cured exactly at the 60th month after detection.
   # This a very rough approximation.

seed/survival_curves.yaml

@@ -7,34 +7,52 @@
 # Kaplan-Meier curves were extracted from Carlo & Luiz (2011) Figure 1. We used the WebPlotDigitizer tool (available from: https://web.eecs.utk.edu/~dcostine/personal/PowerDeviceLib/DigiTest/index.html) to extract the approximate data points.
 
 
-I:
-  0: 1.000
-  12: 0.980
-  24: 0.953
-  36: 0.937
-  48: 0.901
-  60: 0.856
-
-II:
-  0: 1.000
-  12: 0.909
-  24: 0.744
-  36: 0.634
-  48: 0.554
-  60: 0.512
-
-III:
-  0: 1.000
-  12: 0.752
-  24: 0.456
-  36: 0.345
-  48: 0.271
-  60: 0.244
-
-IV:
-  0: 1.000
-  12: 0.426
-  24: 0.152
-  36: 0.083
-  48: 0.060
-  60: 0.059
+curves_by_stage_at_detection:
+  I:
+    0: 1.000
+    12: 0.980
+    24: 0.953
+    36: 0.937
+    48: 0.901
+    60: 0.856
+
+  II:
+    0: 1.000
+    12: 0.909
+    24: 0.744
+    36: 0.634
+    48: 0.554
+    60: 0.512
+
+  III:
+    0: 1.000
+    12: 0.752
+    24: 0.456
+    36: 0.345
+    48: 0.271
+    60: 0.244
+
+  IV:
+    0: 1.000
+    12: 0.426
+    24: 0.152
+    36: 0.083
+    48: 0.060
+    60: 0.059
+
+# Age composition, in years, of the sample for each survival curve. The 0 key
+# indicates participants below the age of 65 years, and the 65 key indicates
+# participants aged 65 years and above. See Carmo & Luiz (2011), Table 1.
+sample_by_age:
+  I:
+    0: 842
+    65: 100
+  II:
+    0: 905
+    65: 228
+  III:
+    0: 950
+    65: 266
+  IV:
+    0: 120
+    65: 29

src/hpv_progression_model/model.py

@@ -236,12 +236,9 @@ def next(self):
                     infections_to_keep.add(infection)
             self.infections = infections_to_keep
 
-            # Add non-cancer death probability to the transition probabilities
-            # NOTE: Cancer states have their own mortality curves that also
-            # include mortality from other causes
-            if self.state not in CANCER_STATES:
-                if self.non_cancer_death_probability >= RNG.random():
-                    self.state = HPVInfectionState.DECEASED
+            # Apply non-cancer related death probability
+            if self.non_cancer_death_probability >= RNG.random():
+                self.state = HPVInfectionState.DECEASED
 
             # Do not consider infections for deceased individuals
             if self.state == HPVInfectionState.DECEASED: