Skip to main content
. Author manuscript; available in PMC: 2022 Jan 1.
Published in final edited form as: J Biomed Inform. 2020 Dec 3;113:103652. doi: 10.1016/j.jbi.2020.103652

TABLE 3a.

UKB PheRS estimate and diagnostics using beta estimates from MGI

Time threshold (in years prior to diagnosis)
0 years 1 year 2 years 5 years
Odds Ratio 1.66 1.51 1.57 1.56
Odds Ratio 95% CI (1.62, 1.70) (1.48, 1.55) (1.51, 1.63) (1.50, 1.62)
P-value 0.00E+00 1.61E-245 1.48E-111 1.31E-108
AUC 0.699 0.657 0.606 0.603
AUC 95% CI 0.679, 0.719 0.635, 0.679 0.583, 0.629 0.583, 0.623
HL Stat, P-value 45.44, 3.04e-07 89.75, 5.55e-16 100.72, 0.00e+00 38.71, 5.56e-06
Brier score 0.00168 0.00168 0.00167 0.00167
Top %-ile OR (95% CI)
 1st percentile 15.77 (14.51, 20.05) 21.24 (19.47, 27.44) 14.07 (12.70, 17.93) 1.75 (1.19, 2.58)
 2nd percentile 20.14 (18.84, 25.67) 12.20 (11.40, 15.93) 9.60 (8.78, 12.23) 10.20 (9.34, 12.99)
 5th percentile 1.28 (0.90, 1.88) 0.71 (0.55, 0.89) 0.77 (0.63, 0.90) 0.62 (0.47, 0.77)
 10th percentile 1.28 (0.90, 1.88) 0.71 (0.55, 0.89) 0.77 (0.63, 0.90) 0.62 (0.47, 0.77)
 25th percentile 1.28 (0.90, 1.88) 0.71 (0.55, 0.89) 0.77 (0.63, 0.90) 0.62 (0.47, 0.77)

Abbreviations: CI, confidence interval; OR, odds ratio; HL, Hosmer-Lemeshow Goodness of Fit test; %-ile, percentile; PheRS, phenotype risk score

Notes:

- The estimates and diagnostic values in this table correspond to pancreatic cancer PheRS constructed in UKB using association estimates obtained from time-restricted co-occurrence analysis in the matched MGI discovery cohort.

- The odds ratio, corresponding 95% confidence interval estimate, and p-value come from a logistic GLM model for the PheRS on pancreatic cancer, adjusted for, sex, birthyear, genotyping array, and the first four principal components of the genotype data.

- The AUC and corresponding 95% confidence interval are from an unadjusted ROC model for pancreatic cancer case/control status.

- The Hosmer-Lemeshow test statistic and p-value and the Brier score are from a matched, unadjusted logistic GLM model for pancreatic cancer case/control status.

- The percentile-based odds ratios and confidence intervals come from a Firth-corrected logistic regression model on pancreatic cancer case/control status adjusted for sex, birthyear, genotyping array, and the first four principal components of the genotype data.