Table 1 Characteristics of the Danish and US-VA datasets
From: A deep learning algorithm to predict risk of pancreatic cancer from disease trajectories
General cohort information | Danish dataset | US-VA dataset |
|---|---|---|
Dataset timeline | 1977–2018 | 1999–2020 |
Total n patients | 8,110,706 | 2,962,383 |
Male (%) | 4,030,504 (49.7%) | 2,538,762 (85.7%) |
Female (%) | 4,080,202 (50.3%) | 423,621 (14.3%) |
Median n disease codes per patient | 22 | 188 |
Median length of trajectory in years | 23.0 | 12.0 |
PC cohort information | ||
Total n patients | 23,985 | 3,869 |
Male (%) | 11,880 (49.5%) | 3,741 (96.7%) |
Female (%) | 12,105 (50.5%) | 128 (3.3%) |
Median n disease codes per patient | 18 | 121 |
Median length of trajectory in years | 17.0 | 8.0 |
Median age at PC diagnosis | 70.0 | 68.0 |
n disease codes 0–3 months pre-PC | 95,358 | 125,305 |
n disease codes 3–6 months pre-PC | 27,131 | 56,198 |
n disease codes 6–12 months pre-PC | 38,109 | 97,911 |
n disease codes >12 months pre-PC | 480,830 | 1,188,199 |
