ACTION ICU Score for intensive care in NSTEMI

Clinical background

In NSTEMI, the proportion of patients admitted to an intensive care unit is considerably higher than the proportion who actually develop complications requiring intensive care. In the derivation cohort, 43% of patients were cared for in the ICU, but only 14% developed a complication warranting intensive care [1]. The decision between ICU and telemetry is in practice often governed by local routines and individual physician preference rather than by the patient's risk profile. The ACTION ICU score for intensive care in NSTEMI was developed to provide a structured basis for precisely this triage decision, and fills a niche that earlier risk scores do not cover: GRACE and TIMI are designed to predict mortality or ischaemic events, not the need for intensive care [1].

Calculating the ACTION ICU score

The score rests on nine variables, all known on arrival at hospital. The variables and their point values are:

$$\text{ACTION ICU} = \text{age} + \text{creatinine} + \text{HR} + \text{SBP} + \text{troponin} + \text{heart failure} + \text{ST depression} + \text{revasc.} + \text{lung disease}$$

The score ranges from 0 to 19. The weighting reflects the strength of each variable's association with the outcome: heart failure is assigned 5 points because it is the single strongest predictor, while previous revascularisation is protective and scores 1 point when absent [1].

The derivation cohort consisted of 29,973 patients $\geq 65$ years with NSTEMI without cardiogenic shock or cardiac arrest on arrival, enrolled at 499 hospitals in the USA between April 2011 and December 2012. The data came from the ACTION Registry linked to Medicare. The outcome was defined as the development of cardiac arrest, shock, high-degree AV block requiring pacing, respiratory failure, stroke or death during the index admission. Of the cohort, 14.3% developed a complication requiring intensive care [1].

Interpretation in practice

The score translates into an estimated risk of complications requiring intensive care during the current admission. In the derivation cohort, a score $\leq 2$ corresponded to a risk below 5%, a score $\leq 5$ to a risk below 10%, and a score $\geq 12$ to a risk above 30% [1]. Of the patients, 14.6% had a score $\leq 2$, 48.5% a score $\leq 5$, and 11.3% a score $\geq 12$ [1].

In the derivation cohort, 478 of 4,282 patients with complications died without ever having been cared for in the ICU. Among patients with a score $\geq 12$, 4.4% died without ICU care, compared with 0.4% among those with a score $\leq 2$ [1]. This indicates that a low score identifies a group in whom forgoing the ICU is safe, while a high score marks patients in whom the absence of intensive care may have consequences.

Validation and performance

In the derivation cohort, the continuous model achieved a c-statistic of 0.73 (95% CI 0.72–0.74) and the integer-based score 0.72 (95% CI 0.71–0.73) [1]. Calibration was good, with small differences between observed and expected probability across all deciles, including in subgroups by sex and age. Internal validation with bootstrapping showed negligible optimism: the optimism-corrected c-statistic was essentially unchanged [1]. The score performed better than a recalibrated GRACE score for predicting the need for intensive care (c-statistic 0.69) and also outperformed the TIMI mortality score (0.65) and the CRUSADE bleeding score (0.71) [1].

External validation has been carried out in three Latin American cohorts with varying results. In a Mexican single-centre study of 345 NSTEMI patients, an AUC of 0.77 (95% CI 0.64–0.97) was achieved with a complication rate of 14.2%, comparable to the derivation cohort [2]. At a cut-off of 5 points, sensitivity was 87.7% and the negative predictive value 95.4%, confirming the ability of the score to identify low-risk patients [2]. In a Brazilian cohort of 1,263 patients the complication rate was considerably lower (4.9%) and the c-statistic only 0.55 (95% CI 0.47–0.63), suggesting that the score performs less well in populations with a low event rate and different demographics [3]. A Colombian study of 1,062 patients found a complication rate of 7.1% and proposed a lower cut-off (3 points) for that population [4].

In summary, external validations suggest that the score retains its ability to identify low-risk patients, but that discrimination varies with the composition of the population and the event rate. The original cohort consisted solely of patients $\geq 65$ years, and the score has not been validated in younger populations in larger studies.

Limitations

The score applies to initially haemodynamically stable NSTEMI patients. Patients presenting with cardiogenic shock, cardiac arrest or respiratory failure on arrival already have an indication for intensive care and should not be scored with the tool [1, 2].

The derivation cohort was limited to patients $\geq 65$ years, and the score has not been systematically validated in younger patients. The age and serum creatinine variables have fixed thresholds that do not allow for continuous relationships, and the creatinine threshold of 1.1 mg/dL is the same for men and women despite known sex differences in reference ranges. The troponin threshold of 12 times the upper limit of normal presupposes that the laboratory has a defined upper limit of normal, which can vary between hospitals.

The Brazilian validation showed that the score performed close to chance in a population with a low complication rate and younger patients (mean age 62.3 years) [3]. This underlines that the score cannot be transferred uncritically to populations that differ markedly from the derivation cohort.

A common error is to use the score as an absolute decision aid without weighing in clinical judgement. The score quantifies risk but does not replace clinical intuition in borderline cases, particularly when the patient has factors not covered by the model, such as acute kidney injury, sepsis or another concurrent acute illness.

References

1. Fanaroff AC et al. Risk Score to Predict Need for Intensive Care in Initially Hemodynamically Stable Adults With Non–ST-Segment-Elevation Myocardial Infarction. J Am Heart Assoc. 2018. PMID: 29802146
2. Alanís-Naranjo JM et al. Usefulness of the ACTION ICU score to predict complications requiring critical care in Mexican patients with non-ST-segment elevation myocardial infarction. Arch Cardiol Mex. 2025. PMID: 40763819
3. Guimarães PO et al. Clinical outcomes and need for intensive care after non-ST-segment-elevation myocardial infarction. Eur J Intern Med. 2020. PMID: 32089424
4. Vasquez-Rodriguez JF et al. Risk of complications after a non-ST segment elevation acute myocardial infarction in a Latin-American cohort: An application of the ACTION ICU score. Heart Lung. 2023. PMID: 36183629