Clinical background
Deciding whether cardiopulmonary resuscitation should be attempted for an in-hospital cardiac arrest is one of the most difficult conversations in hospital care. Patients and relatives often overestimate the probability of a good outcome after cardiac arrest, and clinicians' informal predictions have proved barely better than chance [4]. The GO-FAR score was developed precisely to provide an objective, pre-arrest basis for these conversations rather than relying on intuitive judgement. The instrument addresses one specific question: what is the probability that the patient survives to discharge with intact neurological function (Cerebral Performance Category, CPC, grade 1) if an in-hospital cardiac arrest occurs and CPR is started?
The score is not a tool for deciding who should receive CPR. It is a basis for the conversation, particularly in patients with multiple comorbidities in whom the expected benefit of CPR may be very low.
Calculating the GO-FAR score
The GO-FAR score is the sum of 13 weighted variables assessed before any cardiac arrest:
where is the weight for variable and indicates whether the variable is present, with the exception of age (several levels) and neurologically intact at baseline (a negative weight). The variables and their points are:
| Variable | Points |
|---|---|
| Age <70 years | 0 |
| Age 70–74 years | 2 |
| Age 75–79 years | 5 |
| Age 80–84 years | 6 |
| Age ≥85 years | 11 |
| Neurologically intact at baseline: yes | −15 |
| Admission diagnosis: major trauma | 10 |
| Admission diagnosis: acute stroke | 8 |
| Metastatic or haematological cancer | 7 |
| Septicaemia present before the arrest | 7 |
| Admitted with a non-cardiac medical diagnosis | 7 |
| Hepatic failure | 6 |
| Admitted from a skilled nursing facility | 6 |
| Hypotension or haemodynamic instability before the arrest | 5 |
| Renal failure or dialysis dependence | 4 |
| Respiratory failure before the arrest | 4 |
| Pneumonia | 1 |
The only variable that reduces the score is being neurologically intact at baseline, which subtracts 15 points. This reflects the fact that intact neurological status on admission is the single strongest predictor of a good outcome.
The derivation cohort consisted of 51,240 adult inpatients with an index cardiac arrest between January 2007 and December 2009 at 366 hospitals in the US Get With The Guidelines-Resuscitation registry [1]. The data were split into training (44.4 per cent), test (22.2 per cent) and validation (33.4 per cent) sets. The outcome measure was survival to discharge with CPC 1 after CPR for in-hospital cardiac arrest. Multivariable methods were used to select the best independent predictors, and a series of candidate models was tested against the test data to find the one that best classified patients into four probability bands.
Interpretation in practice
The total score translates into four probability bands for neurologically intact survival to discharge:
| Total score | Band | Observed rate of good outcome in the derivation | Clinical management |
|---|---|---|---|
| ≥24 | Very low (<1 per cent) | 0.9 per cent | The expected benefit of CPR is minimal. This is a strong argument for discussing withholding CPR with the patient and relatives. |
| 14–23 | Low (1–3 per cent) | 1.7 per cent | Low probability of a good outcome. The conversation should include the fact that CPR is unlikely to lead to neurologically intact survival, but the decision must be weighed against the patient's values. |
| −5 to 13 | Average (>3–15 per cent) | 9.4 per cent | No strong signal in either direction. The decision should be guided primarily by the patient's preferences and the overall clinical picture. |
| ≤−6 | Above average (>15 per cent) | 27.5 per cent | A relatively favourable prognosis in the event of cardiac arrest. CPR is generally justified unless other reasons argue against it. |
In the derivation cohort, 9.4 per cent of patients were classified as very low risk and 18.9 per cent as low risk, so more than a quarter of patients fell into the two lowest bands [1]. It is these patients that the instrument is primarily intended to identify.
Validation and performance
The derivation study achieved a c-statistic of 0.78 in the validation data [1]. A systematic review and meta-analysis from 2023 that included five external validation studies of GO-FAR reported a pooled AUROC of 0.78 (95 per cent CI 0.69–0.85), the best performance among all pre-arrest prognostic models for in-hospital cardiac arrest [3]. Comparable models such as PAM and PAR had pooled AUROCs of 0.59 and 0.62 respectively.
External performance varies considerably between populations, however. In a US validation at an academic trauma centre in Minneapolis (403 adults with in-hospital cardiac arrest, 2009–2018), the AUC fell to 0.68 [2]. The rate of good outcome in this cohort was higher than in the derivation: 5.3 per cent in the below-average probability group, 22.5 per cent in the average group and 34.1 per cent in the above-average group. The higher survival may partly be explained by the cohort being younger (mean age 60.3 years) and containing a high proportion of trauma patients.
In an external validation at a community hospital in Michigan (397 cases, 2016–2019), a c-statistic of 0.726 was measured for GO-FAR with good calibration, that is, the observed outcomes fell within the predicted ranges in all four risk bands [4]. Predictive ability fell markedly, however, when the outcome was broadened to include CPC 2 or survival to discharge alone without regard to neurological function. This indicates that GO-FAR is specifically aimed at predicting the best neurological outcome and is not suited to predicting survival in general.
A Swedish external validation at Karolinska University Hospital (717 patients with in-hospital cardiac arrest in Stockholm County, 2013–2014) found an AUROC of 0.82 (95 per cent CI 0.78–0.86) in a complete-case analysis, indicating good discrimination [5]. Calibration analysis showed, however, that GO-FAR systematically underestimated the probability of neurologically intact survival in this population, in which the observed rate of good outcome was 22 per cent compared with about 10 per cent in the derivation cohort. The authors therefore recommended caution in clinical use without first updating the model for the local population.
An ILCOR systematic review from 2022 evaluated 13 different pre-arrest scoring tools and found that GO-FAR was examined in seven studies with a negative predictive value for CPC 1 of between 95.0 and 99.2 per cent [6]. The certainty of the evidence was judged to be very low overall, however, and no tool was considered sufficiently reliable to support clinical decision-making on its own.
Limitations
GO-FAR was derived and validated exclusively for in-hospital cardiac arrest and must not be applied outside hospital. The instrument applies to adults and has not been validated in children.
The score is designed to predict survival with CPC 1, that is, neurologically intact or with only minimal deficits. It is not intended to predict survival with moderate neurological impairment (CPC 2–3), an outcome that some patients and relatives may regard as acceptable. In the US validation in Michigan, the c-statistic fell to 0.56 when the outcome was broadened to survival to discharge regardless of neurological status [4].
Several variables are assessed at admission, but some (hypotension or haemodynamic instability before the arrest, septicaemia, respiratory failure) may develop during the hospital stay. In the original study these variables were assessed within two hours before the arrest, while external validations have often used admission data, which may explain some of the variation in performance [4].
The Swedish validation showed that GO-FAR systematically underestimates the probability of a good outcome in a population with higher baseline survival than the US derivation cohort [5]. This is a calibration problem that may lead to patients being judged to have a worse prognosis than they actually do, which in a clinical context may mean that CPR is advised against unnecessarily.
The instrument is not a decision rule. The ILCOR review concluded that the certainty of evidence for all pre-arrest scoring tools is very low and that no tool is sufficiently reliable to govern clinical practice on its own [6]. GO-FAR should be used as an input to shared decision-making, not as an automatic decision to write a DNACPR order.
External validation in a Nordic cohort
The Swedish validation at Karolinska University Hospital is the only external validation in a Nordic population and the one most directly relevant to Swedish health care [5]. It showed good discrimination but systematic underestimation of the probability of a good outcome, meaning that the absolute probability bands in the calculator are probably too low for Swedish patients. The observed rate of neurologically intact survival in the Stockholm cohort was 22 per cent, compared with about 10 per cent in the derivation cohort. A clinician using GO-FAR in Swedish clinical practice should be aware that the score indicates a lower bound for the probability rather than an exact estimate, and that a patient who falls into the average band in a Swedish setting probably has a somewhat higher probability of a good outcome than the band indicates.
References
- Ebell MH, Jang W, Shen Y, Geocadin RG et al. Development and validation of the Good Outcome Following Attempted Resuscitation (GO-FAR) score to predict neurologically intact survival after in-hospital cardiopulmonary resuscitation. JAMA Intern Med 2013. PMID: 24018585
- Rubins JB, Kinzie SD, Rubins DM. Predicting outcomes of in-hospital cardiac arrest: retrospective US validation of the Good Outcome Following Attempted Resuscitation score. J Gen Intern Med 2019. PMID: 31512185
- Grandbois van Ravenhorst C, Schluep M, Endeman H et al. Prognostic models for outcome prediction following in-hospital cardiac arrest using pre-arrest factors: a systematic review, meta-analysis and critical appraisal. Crit Care 2023. PMID: 36670450
- Maravelas R, Aydemir B, Vos D et al. External validation of GO-FAR 2 calculator for outcomes after in-hospital cardiac arrest with comparison to GO-FAR and trial of expanded applications. Resusc Plus 2023. PMID: 37711682
- Piscator E, Göransson K, Bruchfeld S et al. Predicting neurologically intact survival after in-hospital cardiac arrest: external validation of the Good Outcome Following Attempted Resuscitation score. Resuscitation 2018. PMID: 29723607
- Lauridsen KG, Djärv T, Breckwoldt J et al. Pre-arrest prediction of survival following in-hospital cardiac arrest: a systematic review of diagnostic test accuracy studies. Resuscitation 2022. PMID: 35933060
