Skip to main content

Table 4 Results of the full and parsimonious multivariable logistic regression models, with prediction formulae

From: The Value of Preseason Screening for Injury Prediction: The Development and Internal Validation of a Multivariable Prognostic Model to Predict Indirect Muscle Injury Risk in Elite Football (Soccer) Players

  Full model Parsimonious model (after variable selection) Final model after adjustment (shrinkage) for overfitting
Candidate prognostic factors β 95% CI OR 95% CI p value β 95% CI OR 95% CI p value Adjusted β Adjusted OR
Anthropometrics             
Age at PHE (years) 0.095 0.032 to 0.159 1.100 1.032 to 1.172 0.003 0.091 0.034 to 0.148 1.095 1.035 to 1.159 0.002 0.065* 1.068
BMI (kg/m2) − 0.078 − 0.249 to 0.093 0.925 0.780 to 1.098 0.372        
Past medical history             
Freq. of previous IMIs in 3 years prior to PHE 0.235 0.037 to 0.507 1.265 0.964 to 1.661 0.090 0.168 0.015 to 0.350 1.182 0.986 to 1.419 0.0 0.120* 1.128
Most recent previous IMI in 3 years prior to PHE             
 Never Ref Ref Ref Ref Ref - - - - - - -
 < 6 months 0.043 − 0.892 to 0.978 1.044 0.410 to 2.660 0.928 - - - - - - -
 6–12 months − 0.463 − 1.317 to 0.392 0.630 0.268 to 1.480 0.289 - - - - - - -
 > 12 months − 0.308 − 1.056 to 0.440 0.735 0.348 to 1.553 0.420 - - - - - - -
Musculoskeletal examination            - -
 PROM hip internal rotation difference (deg.) 0.008 − 0.029 to 0.044 1.008 0.971 to 1.045 0.682 - - - - - - -
 PROM hip external rotation difference (deg.) 0.024 − 0.011 to 0.059 1.024 0.989 to 1.061 0.180 - - - - - - -
 Hip flexor length difference (deg.) 0.026 − 0.032 to 0.083 1.026 0.969 to 1.087 0.382 - - - - - - -
 Hamstring length/neural mobility difference (deg.) − 0.007 − 0.083 to 0.068 0.993 0.920 to 1.070 0.846 - - - - - - -
 Calf muscle length difference (deg.) 0.018 − 0.033 to 0.069 1.018 0.967 to 1.072 0.493 - - - - - - -
Lower extremity power             
 CMJ power (watts) 0.000 0.000 to 0.001 1.000 1.000 to 1.001 0.394 - - - - - - -
Model Intercept  − 1.448 − 4.564 to 1.668 - - - − 2.384 − 3.558 to − 1.211 - - - − 1.786** -
Model performance statistics Apparent performance (95% CI) Apparent performance (95% CI)—before validation Optimism-adjusted performance with 95% CI—after validation   
NagelkerkeR2 0.120 0.089 0.064   
Calibration slope 1.000 (0.608 to 1.392) 1.000 (0.557 to 1.443) 0.718 (0.275 to 1.161) - -
CITL 0.000 (− 0.233 to 0.233) 0.000 (− 0.230 to 0.230) − 0.009 (− 0.239 to 0.239) - -
C-index 0.670 (0.609 to 0.731) 0.641 (0.580 to 0.703) 0.589 (0.528 to 0.651) - -
  1. Factors in bold indicate significance at the 0.157 level (equivalent to Akaike’s Information Criterion)
  2. β Beta (regression) coefficient, SE standard error, CI confidence interval, OR odds ratio, PHE periodic health examination, Freq. frequency, IMI indirect muscle injury, deg. degrees, BMI body mass index, kg/m2 kilograms/body height squared, ref reference category
  3. β values are expressed per one-unit increase for all continuous variables, and according to category for the most recent IMI within 3 years prior to PHE
  4. *Adjusted regression value after multiplication with uniform shrinkage factor of 0.718
  5. **Re-estimated model intercept after internal validation
  6. Prediction formula of parsimonious model (used during internal validation procedure). The predicted probability of a player sustaining an I-IMI during a season can be calculated using the following: Probability = 1/(1 + exp(2.384 − 0.091 × age − 0.168 × freq. of previous IMIs within 3 years prior to PHE)). If desired, a percentage risk score can be obtained by multiplying the probability × 100.
  7. Prediction formula for final model (for use on new datasets). The predicted probability of a player sustaining an I-IMI during a season can be calculated using the following: Probability = 1/(1 + exp(1.786 − 0.065 × age − 0.120 × freq. of previous IMIs within 3 years prior to PHE)). Note: exp = exponentiate. If desired, a percentage risk score can be obtained by multiplying the probability × 100