Closed bschilder closed 2 months ago
res_coded <- HPOExplorer::gpt_annot_codify(annot = gpt_annot)
# always=3, death=5: 3x5=15
res_coded$annot_weighted[death==15]
hpo_name severity_score_gpt hpo_id
<ord> <num> <char>
1: Atrophy/Degeneration affecting the central nervous system 54.00000 HP:0007367
2: Anencephaly 52.00000 HP:0002323
3: Hydranencephaly 52.00000 HP:0002324
4: Fetal akinesia sequence 50.66667 HP:0001989
5: Arrhinencephaly 47.33333 HP:0002139
6: Iniencephaly 47.33333 HP:0034205
7: Congenital neuroblastoma 46.00000 HP:0006742
8: obsolete Glioblastoma 44.66667 HP:0100843
9: Lethal short-limbed short stature 42.66667 HP:0008909
10: Grade IV preterm intraventricular hemorrhage 41.33333 HP:0030751
11: Acute necrotizing encephalopathy 39.33333 HP:0006965
12: Exencephaly 37.33333 HP:0030769
13: Lethal infantile mitochondrial myopathy 37.33333 HP:0009069
14: Glioblastoma multiforme 34.66667 HP:0012174
15: Lethal skeletal dysplasia 34.66667 HP:0005716
16: 1-minute APGAR score of 0 33.33333 HP:0030927
17: Anaplastic thyroid carcinoma 32.66667 HP:0011779
18: Lethal short-trunk short stature 32.66667 HP:0011404
19: Amyotrophic lateral sclerosis 32.00000 HP:0007354
20: Fatal liver failure in infancy 32.00000 HP:0006583
21: Acrania 29.33333 HP:0030716
22: Ventilator dependence with inability to wean 29.33333 HP:0005946
23: Esophageal neoplasm 28.00000 HP:0100751
24: Esophageal carcinoma 26.66667 HP:0011459
25: Mortality/Aging 26.66667 <NA>
26: T-cell lymphoma/leukemia 26.66667 HP:0005517
27: Absent aortic valve cusps 26.00000 HP:0031569
28: Pulmonary necrosis 26.00000 HP:0033243
29: Cardiac arrest 25.33333 HP:0001695
30: Absent left sided atrioventricular connection 24.00000 HP:0011547
31: Congenital malformation of the right heart 24.00000 HP:0011723
32: Cardiac rhabdoid tumor 23.33333 HP:0034559
33: Respiratory failure 21.33333 HP:0002878
34: Unguarded tricuspid valve 21.33333 HP:0030719
35: Univentricular heart with absent left sided atrioventricular connection 21.33333 HP:0011549
36: obsolete Severe T lymphocytopenia 21.33333 HP:0005379
37: Aortic valve atresia 20.00000 HP:0010883
38: Pulmonary valve atresia 20.00000 HP:0010882
39: Aborted sudden cardiac death 18.00000 HP:0031628
40: Bilateral lung agenesis 18.00000 HP:0005944
41: Cardiorespiratory arrest 18.00000 HP:0006543
42: Intrauterine fetal demise of one twin after midgestation 18.00000 HP:0030753
43: Suicide behaviors 18.00000 HP:5200330
44: Hypovolemic shock 17.33333 HP:0031274
45: Shock 17.33333 HP:0031273
46: Ascending aortic dissection 16.66667 HP:0004933
47: Bilateral renal agenesis 16.00000 HP:0010958
48: Zygotic cleavage failure 14.00000 HP:0033336
49: Abnormal preimplantation embryonic development 12.00000 HP:0033335
50: Death in infancy 12.00000 <NA>
51: Neonatal death 12.00000 <NA>
52: Prenatal death 12.00000 <NA>
53: Stillbirth 12.00000 <NA>
54: Respiratory arrest 10.66667 HP:0005943
55: Therapeutic abortion 10.66667 HP:0030449
56: Ventricular fibrillation 10.66667 HP:0001663
57: Age of death 10.00000 <NA>
58: Death in adolescence 10.00000 <NA>
59: Death in adulthood 10.00000 <NA>
60: Death in childhood 10.00000 <NA>
61: Death in early adulthood 10.00000 <NA>
62: Death in late adulthood 10.00000 <NA>
63: Death in middle age 10.00000 <NA>
64: Severity 10.00000 <NA>
65: Sudden cardiac death 10.00000 HP:0001645
66: Sudden death 10.00000 HP:0001699
hpo_name severity_score_gpt hpo_id
Many of the death phenotypes (eg 'Death in middle age') only have a composite severity score of 10 bc they're necessarily associated with any of the other severity annotations, partly bc they're so vague.
Interestingly, it rates 'Suicide behaviours' as higher on severity score because it says that this can sometimes (rarely) cause impaired mobility. I suppose that's true in cases where a suicide attempt was unsuccessful. though this highlights the trickiness of assigning "direct causality" to behavioural traits.
For comparison, I selected from other traits with a severity score of 10. This includes some mild physical malformations like syndactyly, which conceptually is definitely not as severe as death.
Looks like Lazarin 2014 suffered from some of these same issues: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0114391
We made two noteworthy observations during this exercise. Alpha-thalassemia, specifically Hb Barts disease, invariably causes fetal or neonatal death. While this would be presumed to be a Profound condition, it does not meet the criteria above due to lack of observable intellectual disability. This may suggest the Profound category criteria to be too strict. Likewise, familial dysautonomia carries a non-negligible risk of intellectual disability. But because a minimal 25% risk is proposed for any characteristic, the disease is also categorized as Severe, rather than Profound.
Options
- Add another tier of weights to which only 'death' metric belongs. Currently, our highest weight is 5, so death would be bumped to a weight of 6.
- Keep as-is and note the limitation in the paper. This avoids us fiddling with the severity score to account for edge scenarios like this. We could note that phenotypes that directly describe death are obvious enough that people using our ranking system would understand these are severe (despite their low rank).
Decided to go with option 1 (bumping death up to 6). This, in combination with using Profound/Severe/Moderate/Mild categories, should allow us to capture the importance of death in the rankings and categories.
@NathanSkene mentioned that @KittyMurphy showed examples where some phenotypes were lowly ranked in terms of composite severity scores but nevertheless always caused death. Could you share these examples @KittyMurphy and let us know how many of them there are ?