ABC | Volume 112, Nº6, June 2019

Original Article Carvalhal et al. Does GRACE Score modulate invasiveness? Arq Bras Cardiol. 2019; 112(6):721-726 Table 2 – Logistic regression univariate and multivariate associations between the candidate’s predictive variables and invasive strategy Univariate Analysis Multivariate Analysis Model 1 Model 2 OR (95% CI) p Value OR (95% CI) p Value OR (95% CI) p Value Positive Tn 2.7 (1.8 - 3.8) < 0.001 2.5 (1.7 - 3.7) < 0.001 2.6 (1.8 - 3.8) < 0.001 ST-deviation 2.0 (1.2 - 3.2) 0.006 1.8 (1.1 - 3.1) 0.026 1.8 (1.1 - 2.9) 0.026 Hemoglobin 1.2 (1.1 - 1.4) 0.001 1.2 (1.1 - 1.4) < 0.001 -- Age 0.98 (0.97-0.99) 0.013 -- 0.09 0.98 (0.96 - 0.99) 0.002 CRUSADE 0.98 (0.97-0.99) 0.018 -- 0.29 -- The 5 variables on this table are the ones that reached statistical significance in univariate analysis. Model was derived by the initial inclusion of all 5 variables (full model) and Model 2 only included typical risk prediction variables (did not include hemoglobin and CRUSADE Score). Positive Tn = Troponin change to a level beyond the 99th percentile. The incidence of death during hospitalization was 5.1% (29 individuals). GRACE Score accurately predicted mortality, with an AUC of 0.87 (95% CI = 0.80 - 0.94; p < 0.001). The propensity score for invasive strategy also predicted mortality (AUC = 0.64; 95% CI = 0.56 - 0.72), but had a lower accuracy in comparison with GRACE Score (p < 0.001) - Figure 1B. Discussion The present study found a dissociation between the risk predicted by a probabilistic model and the physician's choice towards invasive strategy in patients with non-ST-elevation acute coronary syndromes. GRACE Score was the probabilistic model utilized in this analysis, a well-validated and accurate tool for prediction of death in ACS. 9,10 The study took place in an environment whose team of physicians has the duty to calculate GRACE Score for risk stratification and decision making. In spite of that, GRACE Score was not higher in individuals who underwent an invasive strategy, in comparison with patients of a selective strategy. Our findings reproduce behavioral science experiments where decisions are not well driven by knowledge . 5 Contrary to GRACE Score, some patients’ characteristics were independently associated with decision and were utilized to build a propensity score for invasive strategy. This score had a prognostic value lower than GRACE Score. Therefore, we found a paradox in which the variables that determined an invasive approach had a weaker association with prognosis in comparison with a true prognostic model that was not related to this decision. Our findings are in line with previous evidences of dissociation between risk and intensity of treatment, the so‑called risk-treatment paradox. 12-14 This phenomenon takes place when management has a risk/benefit trade-off, and the size of beneficial effect correlates with risk of unintended consequences. In this case, individuals who mostly need the treatment are the ones who most discourage the physician’s decision . 15 For example, older ages were associated with a more conservative strategy, despite being the most important risk predictors in GRACE Score. 16,17 Traditionally, medical judgment is based on intuition and experience, the so-called gestalt. This non-structured method of decision is vulnerable to cognitive bias. 18,19 Possibly, in elderly patients, a kind of nihilistic view makes the sense of risk surpass the sense of beneficial effect, while there is more enthusiasm towards young individuals, making the sense of benefit surpass the sense of risk. The utilization of a probabilistic model tends to avoid under- or overestimation of probabilities due to cognitive bias. Instead, it allows the quantification and balance of the risk/benefit ratio. Secondly, it is proved in different scenarios that the estimation of probabilities under uncertainty is more accurate when a probabilistic model is utilized instead of gestalt. 19 Indeed, in acute coronary syndromes, GRACE Score has shown to have better accuracy than the physician’s opinion. 20,21 Our data validates this concept, since GRACE Score was more accurate in relation to the propensity score for invasiveness. However, a mental reluctance of specialists to utilize a mathematical model, at the expense of unstructured judgment, has been reported. 22 Our observation is peculiar because it arises from an environment in which GRACE Score is systematically calculated and registered in the chart. In spite of that, physicians did not seem to be influenced by the predictive model, a phenomenon illustrated by GRACE Score being virtually identical in invasive and non‑invasive groups. One could find only natural that physicians sometimes overrule GRACE Score based on patients' individualities and preferences. However, this should not be frequent enough to totally blunt the contrast of risk between the selective and invasive groups. In our observations, positive troponin and ST-deviation were independent predictors of invasive strategy. They are both part of the 8 variables in GRACE Score, which were not associated with decision. This may be an indication that medical decision tends to be more univariate than multivariate, more deterministic than probabilistic. 14 Probably, either a positive troponin or an ST-deviation would lead them to opt for the invasive strategy, as opposed to a multivariable probabilistic approach. Also, in our first model, low hemoglobin was independently associated with a more conservative strategy. Considering that hemoglobin is not a 724