Skip to main content

Table 2 Cox regression multivariable prediction model for participants with a FOBT result (either positive or negative) N = 191,081, 1676 events

From: The use of electronic healthcare records for colorectal cancer screening referral decisions and risk prediction model development

VariableHazard RatioObserved CoefficientBootstrapped Standard ErrorzP > z[95% Confidence Intervals]
FOBT Result Positive (reference category negative FOBT result)70.1734.2510.05774.19<0.0014.1394.363
Smoking Status
Ex-smoker (reference category non-smoker)1.1410.1320.0502.610.0090.0330.230
Current smoker (reference category non-smoker)1.2650.2350.0902.610.0090.0580.411
((Alcohol + 1)/100)2*3.1471.1802.670.0080.8355.460
((Alcohol + 1)/100)3*−4.1771.557−2.680.007−7.229−1.125
Sex Female (reference category male)0.850−0.1620.054−2.990.003−0.269−0.056
Age/10 *5.8592.0642.840.0051.8149.904
(Age/10)2*−0.4190.154−2.710.007−0.722−0.116
Previous Negative BCSP FOBTs*0.862−0.1490.049−3.050.002−0.245−0.053
Family History of Gastrointestinal Cancer1.5600.4440.1682.640.0080.1150.774
  1. Abbreviations: CI confidence intervals, FOBT faecal occult blood test (specifically guaiac). The continuous variables (Age/10) has been centred at 6.639, (Age/10)2at 44.077, ((Alcohol + 1)/100)2at 0.011, ((Alcohol + 1)/100)3at 0.001, Previous negative BCSP FOBTs at 0.507. A ‘*’ indicates that the variable is treated as continuous.
  2. Survival Probability
  3. \( S(2)={0.9932}^{\exp \left(4.25{x}_1+0.13{x}_2+0.23{x}_3+3.15\left(\ {\left(\frac{x_4+1}{100}\right)}^2-0.011\right)-4.18\left(\ {\left(\frac{x_4+1}{100}\right)}^3-0.001\right)-0.16{x}_5+5.86\left(\frac{x_6}{10}-6.639\right)-0.42\left(\ {\left(\frac{x_6}{10}\right)}^2-44.077\right)-0.15\left({x}_7-0.507\right)+0.44{x}_8\right)} \)
  4. 0.9932  baseline CRC free survival at 2 years S0(2) (the re-estimated shrunken baseline CRC free survival at 2 years was also 0.9932 when rounded) the heuristic shrinkage factor was 0.998.
  5. Where S(2) is the survival probability at 2 years (probability of not being diagnosed with colorectal cancer/polyps)
  6. Event Probability
  7. P = 1 – S(2).
  8. Where P is the probability of colorectal cancer/polyp being diagnosed within 2 years of the latest FOBT date; x1Latest FOBT result; x2ex-smoker; x3current smoker; x4alcohol consumption; x5sex; x6age at FOBT; x7Number of previous negative BCSP FOBTs < 80 fL; x8Family History of GI Cancer.
  9. The dataset derived for the multivariable modelling analysis had 1676 colorectal cancers and polyp diagnoses (sample population = 191,081) and considered 17 degrees of freedom in the model building process giving 98.59 events per variable. The final model had 10 degrees of freedom with an AIC of 34,050.33 and BIC 34,104.77 (N = 1676 when calculating BIC). Overall model fit was assessed using adjusted R2which was 0.600 (bootstrapped CI 100 reps: 0.580, 0.622) and adjusted D was 2.509. Regular R2was 0.602 with a D statistic of 2.519. The linear predictor from the final model had a mean of − 0.021 and a standard deviation of 1.630 (range: -446.458 to 5.048, IQR: -0.235 to 0.781).