Končni model logistične regresije in nomogram

1 Priprava podatkov

2 Multipla logistična regresija

Izvedemo multiplo logistično regresijo samo na predktorjih s p < 0.05 iz LASSO modela.

Predictor	Level	OR	CI95	p
AcuteKidneyFailure	AcuteKidneyFailure=1.0	2.133	1.492–3.048	< 0.001
BIMA	BIMA=1.0	4.756	2.820–8.021	< 0.001
DiabetesOnInsuline	DiabetesOnInsuline=1.0	1.891	1.305–2.740	< 0.001
HospitalizationBeforeSurgeryMo	HospitalizationBeforeSurgeryMo=1.0	1.633	1.165–2.289	0.004
IABPPreoperatively	IABPPreoperatively=1.0	0.108	0.015–0.789	0.028
PerifernoArterijskoObolenje	PerifernoArterijskoObolenje=1.0	1.848	1.279–2.670	0.001
PleuralEffusion	PleuralEffusion=1.0	2.477	1.751–3.504	< 0.001
RethoracotomyYN	RethoracotomyYN=1.0	2.617	1.598–4.287	< 0.001
Transfusion	Transfusion=1.0	2.150	1.521–3.038	< 0.001
BMI	BMI	1.147	1.109–1.187	< 0.001
CockcraftGaultIndexPreop	CockcraftGaultIndexPreop	0.993	0.987–0.998	0.007

2.1 Forest plot

2.2 Vrednotenje modela

2.2.1 ROC analiza

2.2.2 Vrednotenje napovednih vrednosti

Youden index: 0.458 (threshold 0.037)

	Pred 0	Pred 1
True 0	3897	1236
True 1	56	130

Medical-oriented classification measures:

Sensitivity    : 0.6989
Specificity    : 0.7592
Pos Pred Value : 0.0952
Neg Pred Value : 0.9858
Accuracy       : 0.7571

Classification_report (sklearn):
              precision    recall  f1-score   support

     class 0     0.9858    0.7592    0.8578      5133
     class 1     0.0952    0.6989    0.1675       186

    accuracy                         0.7571      5319
   macro avg     0.5405    0.7291    0.5127      5319
weighted avg     0.9547    0.7571    0.8337      5319

3 Nomogram

Nomogram multiple logistične regresije.