All other predictors were highly statistically significant. After testing the prespecified interactions, a significant

interaction (P < 0.001) was found between location of arrest and presenting rhythm in both models and so alternative categorisations for interactions between location click here of arrest and presenting rhythm were considered. All other interaction terms were non-significant. The non-linear relationships between age and outcome are illustrated in Fig. 1. For ROSC greater than 20 min, the relationship with age was flat up to around age 60, with a rapid decrease in the odds of ROSC greater than 20 min at older ages. Hospital survival decreased across the full age range, although this relationship was steeper at older ages. Results of the model validation, based on models fitted in the development dataset, are shown in Table 3. Discrimination and accuracy were better for hospital survival

(c index ∼0.81, R-squared 0.21–0.24) than for ROSC greater than 20 min (c index ∼0.73, R-squared 0.11–0.17). Calibration was generally good, supported visually by calibration plots (Fig. 2), although there was some evidence of worse calibration for ROSC greater than 20 min in the validation dataset. Model performance was generally SP600125 cost well preserved in the validation datasets compared with the development dataset, particularly for hospital survival. Model accuracy was also compared across age groups (Supplemental Fig. 2). Although there was some variation in outcomes (consistent with chance) in the age groups with smaller sample sizes, overall the Ketotifen model fit was good across all age groups. Interactions between age and other predictors were considered but were found to be unnecessary. The final models for ROSC greater than 20 min and hospital survival, refitted to the full dataset, are shown in Supplemental Tables 3 and 4, respectively. The shrinkage factors were 0.964 and 0.970, respectively,

indicating very little overfitting. A spreadsheet for automatic calculation of the predicted probability of ROSC greater than 20 min and hospital survival is provided as online supplemental material. Based on a relatively simple dataset, we have developed a risk model with good discrimination (c index > 0.8) for predicting survival to hospital discharge following an in-hospital cardiac arrest attended by a hospital-based resuscitation team. This model validated well in subsequent data, including external validation in data from 21 hospitals not included in the development dataset. A risk model for ROSC greater than 20 min performed less well, being potentially more sensitive to inter-hospital variation in the organisation and delivery of resuscitation practice, but still demonstrated acceptable discrimination (c index > 0.7).