CART Score (Cardiac Arrest Risk Triage)

Clinical background

Cardiac arrest on the ward is often preceded by hours of abnormal vital signs, but these signs are missed or interpreted too late in clinical practice. The purpose of the CART score is to systematise the assessment of deterioration on the ward and thereby provide a data-driven basis for when prompt action, primarily activation of the rapid response team or transfer to intensive care, is justified. Unlike older early warning scores such as MEWS, which rest on expert opinion, CART is derived from statistical modelling of actual vital signs in inpatients.

Calculating the CART score

The CART score is the sum of points for four variables: respiratory rate, heart rate, diastolic blood pressure and age. The scoring rests on regression coefficients from a multivariable logistic model, multiplied by a factor of 9 to give manageable whole numbers:

$$\text{CART} = f_{\text{RR}}(\text{Respiratory rate}) + f_{\text{HR}}(\text{Heart rate}) + f_{\text{DBP}}(\text{Diastolic blood pressure}) + f_{\text{age}}(\text{Age})$$

where the point levels are:

The score ranges from 0 to 57. In clinical use, the highest score during the current ward stay is taken, not the most recent single value, since patients may have intermittent abnormalities.

The derivation cohort consisted of 47,427 patients on hospital wards (including telemetry units, excluding the ICU) at the University of Chicago Medical Center between November 2008 and January 2011 [1]. Of these, 88 patients had a cardiac arrest on the ward, 2,820 were transferred to intensive care and 44,519 served as controls. Vital signs were collected from electronic health records and the model was derived by stepwise logistic regression with backward elimination. The primary outcome was cardiac arrest on the ward, defined as loss of a palpable pulse with resuscitation started. Vital signs within 30 minutes before the arrest were excluded to ensure a time margin in which intervention could occur.

Interpretation in practice

The CART score is continuous and has no single defined action threshold in the original study. The authors instead present performance at various cut-offs and compare it with the MEWS threshold of >4, which was established practice. At a specificity of 89.9% (corresponding to MEWS >4), a CART score >17 had a sensitivity of 53.4% for cardiac arrest, compared with 47.7% for MEWS [1]. At a CART score >20, a sensitivity of 47.7% was maintained with a specificity of 91.9%, which would have meant 890 fewer alerts during the study period with unchanged detection.

In the Philippine validation study, a threshold of ≥12 was recommended for identifying patients at risk of cardiac arrest or ICU transfer, based on the greatest accuracy 8 hours before the event [2]. At this threshold, specificity was 80.4% and sensitivity 66.7%.

Clinically, low scores (below about 12) mean that the risk of imminent cardiac arrest is low and that routine monitoring can continue. Intermediate scores (about 12–17) should prompt increased frequency of observations and clinical review of the patient, with particular attention to the respiratory rate and diastolic blood pressure, the variables that contribute the highest point levels. High scores (>17) justify immediate assessment with a view to activating the rapid response team or transfer to intensive care, especially if the score is rising over time.

Validation and performance

In the derivation cohort, the AUC for cardiac arrest was 0.84 for CART compared with 0.76 for MEWS (p = 0.001) [1]. For ICU transfer the AUC was 0.71 for CART compared with 0.67 for MEWS (p < 0.001). CART identified patients who arrested a median of 48 hours before the event at the threshold giving 90% specificity, compared with 42 hours for MEWS, but the difference was not statistically significant.

In a review from the same research group, based on over 59,000 ward patients, CART was the best of all the scores tested for predicting cardiac arrest (AUC 0.83), ICU transfer (AUC 0.77) and a composite outcome (AUC 0.78) [3]. Single-parameter systems, such as the MERIT criteria, had the lowest predictive accuracy.

The only published external validation of the original CART score (with the four variables) was conducted as a case-control study at the Philippine Heart Center with 82 patients [2]. Here the AUC for CART was 0.74 for the composite outcome of cardiac arrest or ICU transfer 8 hours before the event, compared with 0.71 for MEWS. The difference was not statistically significant. For ICU transfer alone, CART was numerically superior (AUC 0.82 vs 0.71), while MEWS was numerically superior for cardiac arrest alone (AUC 0.71 vs 0.67). The study was small and retrospective, which limits generalisability.

A further developed electronic version, eCART, which uses 33 time-dependent variables including laboratory values, has been validated in larger cohorts. In a retrospective study of 32,537 postoperative patients, eCART was significantly more accurate than both NEWS and MEWS for predicting a composite outcome of cardiac arrest, ICU transfer or death on the ward (AUC 0.79 vs 0.76 vs 0.75) [4]. This version is not identical to the CART score that this calculator implements, however, and requires electronic calculation with laboratory data.

Limitations

The CART score was derived at a single academic hospital in the USA and has so far been externally validated only in one small Philippine study [2]. Its generalisability to other health systems, other patient populations and other ward structures is therefore uncertain.

The score rests solely on vital signs and age. It takes no account of overall clinical assessment, the underlying diagnosis, laboratory values or trends in vital signs over time. A patient with sepsis and a rising lactate but normal vital signs may have a low CART score despite a high risk. Conversely, postoperative patients may have physiological abnormalities such as pain-related tachycardia or residual effects of anaesthesia that do not reflect deterioration, which can lead to falsely raised scores [4].

The respiratory rate, the variable contributing the highest point levels, is also the vital sign most often measured least reliably on hospital wards. An incorrectly estimated respiratory rate can strongly affect the score. In the original study, missing values were imputed with the last known value, which reflects clinical practice but may conceal deterioration if the parameter is not measured at all.

The score applies to adult inpatients on hospital wards. It is not validated for patients in intensive care, on postoperative recovery units or in the emergency department. Patients on a palliative pathway or with a DNR decision should not be assessed with CART, since the score aims to trigger life-saving interventions.

References

1. Churpek MM, Yuen TC, Park SY, et al. Derivation of a cardiac arrest prediction model using ward vital signs. Crit Care Med 2012;40(7):2102–8. PMID: 22584764
2. Tan ADA, Permejo CC, Torres MCD. Modified Early Warning Score vs Cardiac Arrest Risk Triage Score for Prediction of Cardiopulmonary Arrest: A Case–Control Study. Indian J Crit Care Med 2022;26(7):780–5. PMID: 36864863
3. Churpek MM, Yuen TC, Edelson DP. Risk stratification of hospitalized patients on the wards. Chest 2013;143(6):1758–65. PMID: 23732586
4. Bartkowiak B, Snyder AM, Benjamin A, et al. Validating the Electronic Cardiac Arrest Risk Triage (eCART) Score for Risk Stratification of Surgical Inpatients in the Postoperative Setting: Retrospective Cohort Study. Ann Surg 2019;269(6):1059–63. PMID: 31082902