Evaluating the impact of prediction models: lessons learned, challenges, and recommendations

Table 1 Predictors and regression coefficients of the original and updated models

Original model		Updated model
Predictor	Regression coefficients		Predictor
Age (years)	− 0.022	− 0.017	Age (years)
Female gender	0.46	0.36	Female gender
Current smoking	− 0.63	− 0.50	Current smoking
History of PONV or motion sickness	0.76	0.60	History of PONV or motion sickness
Lower abdominal or middle ear surgery	0.61	0.48	Abdominal or middle ear surgery^a
Isoflurane and/or nitrous oxide anesthesia^b	0.72	0.35	Inhalational anesthesia^b
	–	− 1.16	Outpatient surgery^c
Intercept	0.15	0.12	Intercept
	Model performance characteristics^d
Model discrimination as C-statistic (95% CI)	0.62 (0.60–0.64)	0.68 (0.66–0.70)	Model discrimination as C-statistic (95% CI)
Calibration slope (95% CI)	0.57 (0.48–0.66)	1.00 (0.89–1.10)	Calibration slope (95% CI)

PONV postoperative nausea and vomiting, CI confidence interval
^aIn the updated model, the predictor included lower abdominal, upper abdominal, and laparoscopic surgery in addition to middle ear surgery
^bAs compared to intravenous anesthesia using propofol
^cPredictor not included in the original model, but added in the update of the model
^dModel performance was validated in a subset of patients (between March 2006 and February 2007) treated by anesthesiologists of the care-as-usual group of the cluster-randomized trial [19]

ISSN: 2397-7523