Scholarly article on topic 'A prediction tool for initial out-of-hospital cardiac arrest survivors'

A prediction tool for initial out-of-hospital cardiac arrest survivors Academic research paper on "Clinical medicine"

CC BY-NC-SA
0
0
Share paper
Academic journal
Resuscitation
OECD Field of science
Keywords
{Resuscitation / Prediction-tool / "Out-of-hospital cardiac arrest"}

Abstract of research paper on Clinical medicine, author of scientific article — S. Aschauer, G. Dorffner, F. Sterz, A. Erdogmus, A. Laggner

Abstract Aim Improvement in predicting survival after out-of-hospital cardiac arrest is of major medical, scientific and socioeconomic interest. The current study aimed at developing an accurate outcome-prediction tool for patients following out-of-hospital cardiac arrests. Methods This retrospective cohort study was based on a cardiac arrest registry. From out-of-hospital cardiac arrest patients (n =1932), a set of variables established before restoration of spontaneous circulation was explored using multivariable logistic regression. To obtain reliable estimates of the classification performance the patients were allocated to training (oldest 80%) and validation (most recent 20%) sets. The main performance parameter was the area under the ROC curve (AUC), classifying patients into survivors/non-survivors after 30 days. Based on rankings of importance, a subset of variables was selected that would have the same predictive power as the entire set. This reduced-variable set was used to derive a comprehensive score to predict mortality. Results The average AUC was 0.827 (CI 0.793–0.861) for a logistic regression model using all 21 variables. This was significantly better than the AUC for any single considered variable. The total amount of adrenaline, number of minutes to sustained restoration of spontaneous circulation, patient age and first rhythm had the same predictive power as all 21 variables. Based on this finding, our score was built and had excellent predictive accuracy (the AUC was 0.810), discriminating patients into 10%, 30%, 50%, 70%, and 90% survival probabilities. Conclusion The current results are promising to increase prognostication accuracy, and we are confident that our score will be helpful in the daily clinical routine.

Academic research paper on topic "A prediction tool for initial out-of-hospital cardiac arrest survivors"

Contents lists available at ScienceDirect

Resuscitation

journal homepage: www.elsevier.com/locate/resuscitation

Clinical paper

A prediction tool for initial out-of-hospital cardiac arrest survivors*

S. Aschauerc, G. Dorffnera, F. Sterzb *, A. Erdogmusa, A. Laggnerb

CrossMark

a Centre for Medical Statistics, Informatics and Intelligent Systems, Medical University of Vienna, Spitalgasse 23, Vienna 1090, Austria b Department of Emergency Medicine, Medical University of Vienna, WahringerGilrtel 18-20, Vienna 1090, Austria

c Division of Cardiology andAngiology, Department of Internal Medicine II, Medical University of Vienna, WahringerGilrtel 18-20, Vienna 1090, Austria

ARTICLE INFO

Article history:

Received 8 April 2014

Received in revised form 4June 2014

Accepted 4 June 2014

Keywords: Resuscitation Prediction-tool Out-of-hospital cardiac arrest

ABSTRACT

Aim: Improvement in predicting survival after out-of-hospital cardiac arrest is of major medical, scientific and socioeconomic interest. The current study aimed at developing an accurate outcome-prediction tool for patients following out-of-hospital cardiac arrests.

Methods: This retrospective cohort study was based on a cardiac arrest registry. From out-of-hospital cardiac arrest patients (n = 1932), a set of variables established before restoration of spontaneous circulation was explored using multivariable logistic regression. To obtain reliable estimates of the classification performance the patients were allocated to training (oldest 80%) and validation (most recent 20%) sets. The main performance parameter was the area under the ROC curve (AUC), classifying patients into survivors/non-survivors after 30 days. Based on rankings of importance, a subset of variables was selected that would have the same predictive power as the entire set. This reduced-variable set was used to derive a comprehensive score to predict mortality.

Results: The average AUC was 0.827 (CI 0.793-0.861) for a logistic regression model using all 21 variables. This was significantly better than the AUC for any single considered variable. The total amount of adrenaline, number of minutes to sustained restoration of spontaneous circulation, patient age and first rhythm had the same predictive power as all 21 variables. Based on this finding, our score was built and had excellent predictive accuracy (the AUC was 0.810), discriminating patients into 10%, 30%, 50%, 70%, and 90% survival probabilities.

Conclusion: The current results are promising to increase prognostication accuracy, and we are confident that our score will be helpful in the daily clinical routine.

© 2014 The Authors. Published by Elsevier Ireland Ltd. This is an open access article under the CC BY-NC-SA license (http://creativecommons.org/licenses/by-nc-sa/3.0/).

1. Introduction

Out-of-hospital cardiac arrest is one of the major health problems in the world with a global incidence of 55 adult out-of-hospital cardiac arrests per 100.000 person-years and a poor survival rate of between 2% and11%.1 Despite considerable effort over the last decades,2-4 a valid and applicable scoring system to assess patient survival after out-of-hospital cardiac arrest is not available. Hence, healthcare professionals are required to base crucial decisions on their own experience and impressions, which have been shown to have limited accuracy.5 Accurate risk prediction in the out-of-hospital cardiac arrest population is of great value. It can facilitate

* A Spanish translated version of the summary of this article appears as Appendix in the final online version at http://dx.doi.org/10.1016/j.resuscitation.2014.06.007.

* Corresponding author.

E-mail address: fritz.sterz@meduniwien.ac.at (F. Sterz).

conversations with families, enable quality-of-care assessments and improve research due to precise patient stratification.

The objective of the current study was to improve outcome prediction after cardiac arrest, to compare a multivariable approach with a univariable approach and to assess possible nonlinear dependencies between variables and outcomes. The variables analysed in the current study encompassed patient characteristics as well as resuscitation characteristics. In the end, we wanted to identify the variables with the highest predictive power to derive an out-of-hospital cardiac arrest prediction score.

2. Methods

The current study is based on a prospectively designed and conducted cardiac arrest registry. Since 1991 the registry, approved by the institutional ethical review board, included more than 4200 patients who were resuscitated following cardiac arrest and who were admitted to the department of emergency medicine at a large university hospital.

http://dx.doi.org/10.1016/j.resuscitation.2014.06.007

0300-9572/© 2014 The Authors. Published by Elsevier Ireland Ltd. This is an open access article under the CC BY-NC-SA license (http://creativecommons.org/licenses/ by-nc-sa/3.0/).

Table 1

List of all variables analysed. Variables marked with an asterisk (*

were only available for witnessed out-of-hospital cardiac arrest patients.

Variable name

Description

sex age bmi

diabetes smoker myocinfarct

hypertension

heartfail

opcpre

nyhapre

noflow*

min2srosc*

cause firstaid*

nodefi

adrenaline

shockable

defireaction

cpc30d mortality

Sex of the patient

Age of the patient

Body mass index

Previous diagnosis of diabetes

Patient is an actual smoker

Patient previously had a myocardial

infarction

Previous diagnosis of coronary artery disease

Previous diagnosis of hypertension Previous diagnosis of heart failure Previous diagnosis of cerebral vascular insufficiency

Previous diagnosis of chronic obstructive pulmonary disease OPC score prior to cardiac arrest

NYHA score prior to cardiac arrest

Minutes between cardiac arrest and first aid (length of "no flow" time) Minutes between cardiac arrest and return of spontaneous circulation Main cause of cardiac arrest First aid performed by physician, family member, paramedic or layman Number of defibrillation shocks Amount of adrenaline administered Shockability of rhythm at first defibrillation

Reaction to the first defibrillation

CPC score 30 days after cardiac arrest Mortality 30 days following cardiac arrest

Male = 0, female = 1 In years, at the time of cardiac arrest Weight (kg)/size (m2) Diabetes = 1, no diabetes = 0 Smoker=1, non-smoker=0 Infarction= 1, no infarction = 0

CAD= 1, no CAD = 0

Hypertension = 1, no hypertension = 0 Heart failure = 1, no heart failure = 0 CV1 = 1, no CV1 = 0

COPD = 1, no COPD = 0

Score 1-5

Score 1-5

1n minutes

1n minutes

Cardiac reason = 1, non-cardiac = 0 Physician, family member, paramedic or layman = 1, other or none = 0 Total count of shocks applied Total amount (inmg) Shockable = 1, non-shockable = 0

Not shockable = 0, shockable and VT/VF (as reaction to first defi) = 1, shockable and PEA = 2, shockable + asystole = 3, shock-

able + SR/RHY/SVES/VES/AVES + no pulse = 4, shockable + pulse = 5 Score 1-5

1fcpc30d = 5: 1, otherwise: 0

Binary Metric Metric Binary Binary Binary

Binary

Binary Binary Binary

Binary

Ordinal, treated as metric

Ordinal, treated as

metric

Metric

Metric

Binary Binary

Metric Metric Binary

Ordinal, treated as metric

Ordinal

Binary, target variable

Cardiac arrest was defined as the absence of signs of circulation. Treatment of patients until restoration of spontaneous circulation and post-resuscitation care were given according to international guidelines and recommendations.6

2.1. Data source

The data for all of the patients were prospectively documented according to the "Utstein Style Criteria", which are the recommended guidelines for cardiac arrest and cardiopulmonary resuscitation outcome reports.7 These guidelines include structured protocols on demographic factors, chronic pre-arrest health conditions, and resuscitation. The data on the out-of-hospital cardiac arrests were documented on case report forms and entered into the cardiac arrest registry by specially trained study nurses or physicians. Range and consistency checks were performed. The data were obtained through communication with the dispatch centre, the emergency physicians and paramedics on the scene, bystanders, relatives, and if possible, the patients themselves.

2.2. Selection of participants

The study population included all patients who were successfully resuscitated following out-of-hospital cardiac arrest and who were admitted to our department between 2000 and 2012.This comprises a total of 1932 cases. Older data was not included in order to avoid a bias due to major changes in practice (e.g. standard inclusion of therapeutic hypothermia) around the turn of the century.

2.3. Outcome and predictors

Twenty-one variables determined prior to the return of spontaneous circulation were selected and deemed to have high statistical power to predict survival after 30 days (Table 1).

2.4. Statistical method

We chose the framework of machine learning as the most appropriate to address all of the questions aimed at by this study. This approach, in particular, entailed the following (see also Fig. 1). We first split the available data set of 1932 out-of-hospital cardiac arrest cases into a training set—to be used for optimising the prediction and selecting the most predictive variables—and a validation set—to be used for validating the final prediction tool. To simulate a prospective validation we performed this split along the time axis. For the training set, we selected the oldest roughly 80% (1533) cases while holding out the remaining, most recent, 399 cases for the validation set. The ratio of cases in the training set vs. cases in the validation set was about 4:1, as is common in machine learning.

Subsequently, the training and validation sets were divided into witnessed out-of-hospital cardiac arrests, and non-witnessed out-of-hospital cardiac arrests. Only those out-of-hospital cardiac arrest patients were selected for whom all 21 (or 18, for the non-witnessed out-of-hospital cardiac arrests) variable values were available. 1n other words, missing data were treated by omitting the entire case affected. All of the subsequent steps of analysis were performed independently on the two separate training and

Total (from 1/1/2000) 2783

Inhospital: Out of hospital

851 1932

Training: Validation

1533 399

Witnessed: Non witnessed: Witnessed: Non witnessed 1315 218 336 63

All 21 variables: All 18 variables: All 21 variables: All 18 variables: 1068 174 291 25

Fig. 1. Flow diagram of the number of cases at the different analysis phases.

validation sets of witnessed and non-witnessed out-of-hospital cardiac arrest patients, respectively. At first, we normalised all of the variables in the training set (including the binary ones) to zero means and unit standard deviations, mainly to make the coefficients in the linear logistic regression comparable to each other. Afterwards, we subjected all of the prediction methods to a 10-fold cross-validation using the training set. This means that we divided the training data set into 10 partitions, applied each classification method 10 times to the data from 9 partitions, and used the respective 10th partition to test the performance. From this series of 10 classification tasks, we derived confidence interval figures for all of the performance parameters in a straightforward manner using the mean of each parameter and its respective standard error of the mean. Then, we followed standard practice semi-heuristic methods for feature selection based on a ranking of the variables according to single prediction performance and the absolute value of the coefficient in the linear logistic regression. The main performance parameter was the area under the curve (AUC) of the ROC curve when viewing the problem as a two-class classification problem. Note that this approach is identical to standard C statistics in a dichotomous classification. For comparing nonlinear and linear methods, we used a standard linear logistic regression in terms of a perceptron with sigmoid outputs for the former and multilayer perceptrons (neural networks) with one hidden layer for the latter. Using the criteria mentioned above, the minimum set of variables was selected that showed the same performance on the training set as the entire set. To avoid selection bias in the estimation using the reduced set of variables, the cross-validation of the regression models was then repeated using all of the cases for which the values of the reduced set of variables were available. The final logistic regression formula for this variable subset (an average over all 10 models from the cross-validation) was then used to derive a simplified score by assigning points to value ranges of each variable such that those points can easily be added without a calculator and compared to a table of score ranges to yield one of five possible probabilities for mortality: 0.1, 0.3, 0.5, 0.7, or 0.9. The final classifiers, both with the full and reduced sets of variables as well as the derived score, were then validated on the validation sets.

3. Results

3.1. Study population

After selecting the data based on the chosen variables, the final training set included 1068 patients with witnessed

out-of-hospital cardiac arrests and 174 patients with non-witnessed out-of-hospital cardiac arrests. The corresponding validation sets contained 291 witnessed out-of-hospital cardiac arrest patients and 25 non-witnessed out-of-hospital cardiac arrest patients, respectively. The training sets and the validation sets were comparable regarding patients' characteristics in terms of the chosen variables, as displayed in Table 2 for the witnessed cases and (supplementary data) for the non-witnessed ones.

3.2. Prediction of mortality

The best prediction results could be achieved with simple linear logistic regression. The average AUC (the mean over 10 cross-validation runs, as measured for each of the 10 hold-out sets) was 0.827 (CI 0.793-0.861) for witnessed out-of-hospital cardiac arrest and 0.713 (CI 0.587-0.838) for non-witnessed out-of-hospital cardiac arrest. These results were consistently better (or at least were not significantly worse) than any neural network (multilayer perceptions with 2-10 hidden units) that was tested (results not shown). Therefore, all of the remaining results reported concern the linear predictor version.

For witnessed out-of-hospital cardiac arrests, the total AUC using all of the variables was significantly better than the AUC for any single considered variable (see Fig. 2). This was not the case for non-witnessed out-of-hospital cardiac arrests, for which prediction on the single variable adrenaline achieved a mean AUC of 0.780 (CI 0.650-0.802) (see supplementary files). Although this appears to be better than prediction using all of the variables (see Fig. 2B), it was not significantly different based on a two-tailed paired t-test with a = 0.05. For witnessed out-of-hospital cardiac arrests, the single most predictive variable was also adrenaline (AUC: 0.730 [CI 0.689-0.772]).

3.3. Ranking and selection of variables

The right side of Figs. 2 depicts a ranking of the variables by absolute value of regression coefficient for witnessed and non-witnessed out-of-hospital cardiac arrest patients, respectively. For witnessed cases, the ranking of the variables in Fig. 2A was generated by selectively including variables one by one, beginning with the variable with the highest regression coefficient, minutes to SROSC, and comparing the performances of the limited number of variable predictors with the prediction based on all of the variables. The result is shown in Fig. 2B. For non-witnessed cases, the same type of analysis was not possible, given that the single most predictive variable, adrenaline, was not statistically inferior to using all of the variables.

3.4. The score

From Fig. 2B, it can be concluded that using the four variables min2srosc, age, shockable, and adrenaline, one achieves practically the same prediction performance as using all 21 variables. Therefore, these four variables were used to devise the main prediction score. There were two ways of carrying out this step. The first was to use the final logistic equation—after repeating model estimation with an extended training set of n = 1095 and the reversal of the original normalisation of each variable—to yield the following values of the regression

Y = 0.0284 x min 2srosc + 0.0355 x age-1.4608 x shockable

+ 0.1528 x adrenaline (1)

where adrenaline, min2srosc shockable and age are numerical values. Via the logistic function, probabilities for mortality can be assigned to different ranges of Y as follows:

Table 2

Descriptive statistics of all variables for witnessed out-of-hospital cardiac arrest patients in the training (n = 1068) and validation (n = 291) sets.

Training set Median 25% percentile 75% percentile Percent 1 Percent 0

sex 27.53% 72.47%

age 59 49 69 0 0

bmi 26.12 23.88 29.22 0 0

diabetes 16.20% 83.80%

smoker 30.90% 69.10%

myocinfarct 12.92% 87.08%

cad 21.91% 78.09%

hypertension 32.21% 67.79%

heartfail 11.05% 88.95%

cvi 5.99% 94.01%

copd 9.74% 90.26%

opcpre 1 1 1

nyh5pre 1 1 2

noflow 1 0 6.5

min2srosc 20 10 30

cause 69.76% 30.24%

firstaid 34.18% 65.82%

nodefi 2 0 4

adrenaline 2 0 4

defireaction 1 0 2

shockable 59.83% 40.17%

cpc30d 3 1 5

mortality 39.89% 60.11%

Test set Median 25% percentile 75% percentile Percent 1 Percent 0

sex 27.84% 72.17%

age 61 50 71

bmi 26.23 24.11 29.41

diabetes 20.62% 79.38%

smoker 31.62% 68.39%

myocinfarct 14.09% 85.91%

cad 24.74% 75.26%

hypertension 41.92% 58.08%

heartfail 14.78% 85.22%

cvi 4.81% 95.19%

copd 6.53% 93.47%

opcpre 1 1 2

nyh5pre 1 1 2

noflow 1 0 5

min2srosc 19 12 32

cause 62.54% 37.46%

firstaid 49.49% 50.52%

nodefi 1 0 3

adrenaline 1 0 3

defireaction 1 0 2

shockable 54.30% 45.70%

cpc30d 3 1 5

mortality 42.27% 57.73%

Table.if Y <1.3320thenp(mortality) = 0.1else if

Y< 2.3129thenp(mortality) = 0.3else if Y< 3.1238thenp(mortality) = 0.5else if Y< 4.1046thenp(mortality) = 0.7elsep(mortality) = 0.9

The second, simplified version of the score was derived by assigning simple point values to several subranges of each variable. The best subranges and associated points were found heuristically by roughly dividing the entire range of each variable into subranges of approximately equal numbers of cases. Points were chosen to be round numbers after multiplying the original score by 10. Following an optimisation of the thresholds and points assigned using the training set, the final scoring system was as shown in Table 3.

3.5. Validation

The area under the ROC curves for prediction using the validation set (n = 297 after including cases that had missing values on some of the other, no longer needed, variables) was 0.827 for all of the variables and 0.810 for the best four variables, which were both within the confidence interval estimated based on the training set. Fig. 3 depicts the ROC curve of the regression formula (1) on the validation set (left panel), as well as a comparison of predicted

probabilities and true frequencies of mortality on the validation set when using the simplified score (right panel).

4. Discussion

In the current study, we have created a simple prediction tool for initial survivors of out-of-hospital cardiac arrest. The number of minutes to sustained return of spontaneous circulation, the age of the patient, the first rhythm, and the amount of adrenaline administered were shown to have high statistical power to accurately predict survival after 30 days in out-of-hospital cardiac arrest patients. Note that despite the fact that amount of adrenaline and minutes to sustained return of spontaneous circulation are correlated (r = 0.4, p < 0.001), both variables appear to provide independent information regarding mortality. To enable quick and simple prediction immediately after sustained spontaneous circulation, we converted these results into an applicable bedside tool that allows discrimination between 10%, 30%, 50%, 70% and 90% survival probabilities for out-of-hospital cardiac arrest patients.

However, for our prediction tool, we considered suggested methodological standards for the development and evaluation

Fig. 2. (A) Results for witnessed out-of-hospital cardiac arrest patients. Top panel: mean areas under the ROC curve (AUC, blue squares) together with 95% confidence intervals (horizontal lines) for a multivariable prediction using all variables ("all") compared with univariate predictions using each individual variable, ranked by mean AUC. Bottom panel: mean coefficients (blue squares) for each variable in a multivariable logistic regression using all variables, ranked by absolute value of the coefficient. Positive coefficients correspond to a positive contribution towards the probability of death; negative coefficients correspond to a negative contribution. (B) Results for witnessed out-of-hospital cardiac arrest patients. Mean AUC (blue quadrate) and corresponding confidence intervals (horizontal lines) forthe prediction based on the combination of the k best variables are given. Each square corresponds to the prediction performance when using all of the variables from the left up to the one listed below. The last value is the performance when all of the variables are used. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

of prediction scores.8 To our knowledge, this is the first out-of-hospital cardiac arrest prediction score that has been developed from such a large cohort and that can be calculated immediately after the restoration of sustained spontaneous circulation without the need for laboratory markers. Patients included in our study had to survive the initial resuscitation because we wanted to focus on the post-spontaneous circulation period and not on the resuscitation time-span like other published scores.9,10 The fact that only four variables are needed to assess the out-of-hospital cardiac arrest survival probability makes the tool easily applicable and clinically useful.

Previous risk assessments predominantly focused on single factors to predict the survival of out-of-hospital cardiac arrest patients, e.g., witnessed or non-witnessed cardiac arrest, bystander cardio-pulmonary resuscitation, age or primary rhythm,11-14 but had little impact on individual survival prediction.

The results of the current study demonstrate that a multivariable approach assessing factors for out-of-hospital cardiac arrest survival prediction is superior to a univariate approach (AUC: 0.82 vs. AUC: 0.704 under the ROC curve). There are only a few published papers that have performed multivariable analyses assessing

variables for out-of-hospital cardiac arrest survival prediction. Consistent with our study, age and the time to sustained spontaneous circulation were strong survival predictors.15,16 For several reasons previous studies had limited clinical application. The OHACA score, for example, requires laboratory markers and variables, including the no-flow and low-flow times, which often are not easy to distinguish and which have been shown to be less accurate.17 Other scores fail to accurately predict out-of-hospital cardiac arrest survival18 or are based on small sample sizes and require elaborate calculations.15

Clinical prediction rules have become popular in modern medicine and have been recognised as powerful tools to improve clinical decision making.19,20 For out-of-hospital cardiac arrest patients, our outcome prediction tool can give support in conversations with relatives and can help physicians choose among the increasing number of post-cardiac arrest treatment options such as emergency extracorporeal life support or haemodialysis. In addition to clinical benefit, our tool can help to enhance the precision of clinical research. Accurate survival estimation enables better stratification of patients. Selecting out-of-hospital cardiac arrest patients according to their survival probability can therefore yield

ROC curve on validation set

Predicted precentage of death

Fig. 3. The ROC curve for the predictor based on the best four variables on the test set (n = 291, AUC = 0.810) and a scatter plot depicting the validation of the simplified score on the extended test set. The predicted probability for death along the x-axis is compared to the empirical relative frequency of mortality predicted in the test set along the y-axis.

new research knowledge and can identify new patient subgroups that might profit from a specific treatment. However, we want to emphasise that we believe that it should not be used to make end-of-life or termination-of-resuscitation decisions.

This new score is predominately suitable for witnessed out-of-hospital cardiac arrests. We also looked for predictive variables in the non-witnessed population, and we found that the single variable adrenaline is as good as consideration of all 18 variables regarding outcome prediction. Unfortunately, it was not possible to derive a robust survival prediction score for non-witnessed out-of-hospital cardiac arrests, which can be explained by the ambiguous onset time of the out-of-hospital cardiac arrest.

In future we want to focus on improvement of the tool's prediction accuracy. Beside external validation, we want to incorporate early available laboratory markers to increase prediction accuracy. Hence, we hope that in the future we will be able to give reliable outcome estimations for non-witnessed out-of-hospital cardiac arrests as well.

Our study has several limitations. First, our score was validated internally by a holdout strategy and not externally. Although from the literature it is known that an n-fold cross-validation leads to

Table 3

The simplified score based on assigning points to subranges of each variable. Comparing the score with the thresholds below yields the probability for mortality.

Predictor Points Predictor Points

1. Age group 3. Minutes until SROSC

>80 32 >100min 35

>70 27 >50min 21

>50 23 >40min 13

>60 20 >30min 10

>40 16 >20min 7

<40 11 >10min 4

>0 min 1

2. Adrenalin administered 0min 0

>10mg 24

>5 mg 12 4. Shockable rhythm?

>4mg 7 Yes -15

>3 mg 5 No 0

>2mg 4

>1 mg 2

>0mg 1

0 mg 0 Total score

Total score Probability for mortality

<13 10%

13-22 30%

23-30 50%

31-40 70%

>40 90%

unbiased estimates of model performance, it is still important that our prediction score will be validated by other institutions before usage in routine clinical practice.21 Second, the data were collected over a period of time in the context of evolving guidelines. Thus, medical treatment could have affected the outcomes and consequently might affect the accuracy of our score. Third, since only cases with all available variables were considered, a resulting selection bias might have slightly skewed the results. On the other hand, however, our score is only applicable for cases with available variable values and thus such a bias can be considered minimal. Additionally, we performed a pre-selection of variables presumed to have predictive power regarding outcome.

5. Conclusion

A valid and robust survival prediction score for out-of-hospital cardiac arrest patients has been developed. Due to its accuracy and applicability, our prediction tool can supply physicians with critical information at a very early stage, and we hope that it will find its way into clinical practice.

Funding

Conflict of interest statement

There was no support from any organisation for the submitted work; Georg Dorffner is CEO and shareholder of the clinical trial service provider The Siesta Group and a part-time employee of Philips-Respironics. The other authors have no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; there were no other relationships or activities that could appear to have influenced the submitted work.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.resuscitation. 2014.06.007.

References

1. Berdowski J, Berg RA, Tijssen JGP, Koster RW. Global incidences of out-of-hospital cardiac arrest and survival rates: systematic review of 67 prospective studies. Resuscitation 2010;81:1479-87.

2. Eisenberg M, Hallstrom A, Bergner L. The ACLS score predicting survival from out-of-hospital cardiac arrest. JAMA 1981;246:50-2.

3. Wilcox-Gök VL. Survival from out-of-hospital cardiac arrest. A multivariate analysis. Med Care 1991;29:104-14.

4. Waalewijn RA, de Vos R, Tijssen JG, Koster RW. Survival models for out-of-hospital cardiopulmonary resuscitation from the perspectives of the bystander, the first responder, and the paramedic. Resuscitation 2001;51:113-22.

5. Madl C, Kramer L, Domanovits H, et al. Improved outcome prediction in unconscious cardiac arrest survivors with sensory evoked potentials compared with clinical assessment. Crit Care Med 2000;28:721-6.

6. Nolan JP, Soar J, Zideman DA, et al. European Resuscitation Council Guidelines for Resuscitation 2010 Section 1. Executive summary. Resuscitation 2010;81:1219-76.

7. Jacobs I, Nadkarni V, Bahr J, et al. Cardiac arrest and cardiopulmonary resuscitation outcome reports: update and simplification of the Utstein templates for resuscitation registries: a statement for healthcare professionals from a task force ofthe International Liaison Committee on Resuscitation (American Heart Association, European Resuscitation Council, Australian Resuscitation Council, New Zealand Resuscitation Council, Heart and Stroke Foundation of Canada, InterAmerican Heart Foundation, Resuscitation Councils of Southern Africa). Circulation 2004;110:3385-97.

8. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review and suggested modifications of methodological standards. JAMA 1997;277:488-94.

9. Morrison LJ, Visentin LM, Kiss A, et al. Validation of a rule for termination of resuscitation in out-of-hospital cardiac arrest. N Engl J Med 2006;355: 478-87.

10. Verbeek PR, Vermeulen MJ, Ali FH, Messenger DW, Summers J, Morrison LJ. Derivation of a termination-of-resuscitation guideline for emergency medical technicians using automated external defibrillators. Acad Emerg Med 2002;9:671-8.

11. Sasson C, Rogers MAM, Dahl J, Kellermann AL. Predictor ofsurvival from out-of-hospital cardiac arrest: a systemati review and meta-analysis. Circ Cardiovasc Qual Outcomes 2010;3:63-81.

12. Eisenburger P, Sterz F, Haugk M, et al. Cardiac arrest in public locations—an independent predictor for better outcome. Resuscitation 2006;70:395-403.

13. Herlitz J, Svensson L, Engdahl J, et al. Characteristics of cardiac arrest and resuscitation by age group: an analysis from the Swedish Cardiac Arrest Registry. Am J Emerg Med 2007;25:1025-31.

14. Dumas F, Rea TD. Long-term prognosis following resuscitation from out-of-hospital cardiac arrest: role of aetiology and presenting arrest rhythm. Resuscitation 2012;83:1001-5.

15. Hayakawa K, Tasaki O, Hamasaki T, et al. Prognostic indicators and outcome prediction model for patients with return of spontaneous circulation from cardiopulmonary arrest: the Utstein Osaka Project. Resuscitation 2011;82: 874-80.

16. Adrie C, Cariou A, Mourvillier B, et al. Predicting survival with good neurological recovery at hospital admission after successful resuscitation of out-of-hospital cardiac arrest: the OHCA score. Eur Heart J 2006;27:2840-5.

17. Harvey AL, Gerard WC, Rice GF, Finch H. Actual vs perceived EMS response time. Prehosp Emerg Care 1999;3:11-4.

18. Pircher IR, Stadlbauer K-H, Severing AC, et al. A prediction model for out-of-hospital cardiopulmonary resuscitation. Anesth Analg 2009;109:1196-201.

19. Reilly BM, Evans AT. Translating clinical research into clinical practice: impact of using prediction rules to make decisions. Ann Intern Med 2006;144: 201-9.

20. Adams ST, Leveson SH. Clinical prediction rules. BMJ 2012;344:d8312-22.

21. Steyerberg EW, Harrell FE, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001;54:774-81.