Gupta perioperative risk for myocardial infarction or cardiac arrest (MICA)

Clinical background

Perioperative myocardial infarction and cardiac arrest are uncommon but serious complications of non-cardiac surgery, with a high mortality once they occur. The decision on how far preoperative cardiological investigation should go, and how the risk should be communicated to the patient, requires a quantified risk estimate. The dominant instrument before Gupta's model was the Revised Cardiac Risk Index (RCRI), which rests on six clinical variables and gives a broad risk class. The RCRI has, however, been criticised for relatively low discrimination and for not taking detailed account of the risk profile of the procedure. Gupta's perioperative risk of myocardial infarction or cardiac arrest (MICA) was developed specifically to address these weaknesses, in particular to incorporate the surgical subcategory as a strong driver of risk and thereby to give a more differentiated, individualised risk percentage [1].

Calculating Gupta's perioperative risk

The model is a logistic regression with five predictors: age, functional status, ASA class, preoperative creatinine and type of procedure. It is calculated as:

$$x = 0{.}02 \times \text{age} + \beta_{\text{functional status}} + \beta_{\text{ASA class}} + \beta_{\text{creatinine}} + \beta_{\text{type of procedure}} - 5{.}25$$

$$\text{risk ($$

where each $\beta$ is the coefficient for the chosen category within each variable. Age contributes linearly at 0.02 per year of life. Functional status scores 0 for independent, 0.65 for partially dependent and 1.03 for totally dependent. The ASA class contributes negative coefficients for the lower classes (I: −5.17; II: −3.29; III: −1.92; IV: −0.95; V: 0), which means that healthier patients receive a lower risk. A preoperative creatinine >1.5 mg/dL adds 0.61. The coefficients for the type of procedure range from −1.61 (breast surgery) to +1.60 (aortic surgery) and constitute the single strongest driver of risk in the model.

The derivation cohort consisted of 211,410 patients in the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) from 2007, a prospective multicentre database covering more than 250 hospitals. The outcome was perioperative myocardial infarction or cardiac arrest, which occurred in 1,371 patients (0.65%). The model was then validated on NSQIP 2008 with 257,385 patients [1].

Interpretation in practice

Unlike indices that give risk classes, Gupta's model delivers a continuous probability as a percentage. There are no established, generally accepted thresholds within the calculator itself for when risk becomes "high" or "low", and the 2022 ESC guidelines likewise define no fixed cut-offs at which the model should trigger specific measures [3]. The risk figure must therefore be interpreted in its clinical context:

It is important that the risk figure be used to support shared decision-making, not as an autonomous decision tool. A patient with a low risk figure but a recent myocardial infarction or unstable angina should be managed according to the clinical picture, not according to the calculator.

Validation and performance

In the derivation study the model showed a c-statistic of 0.884 in the development cohort and 0.874 in the temporal validation cohort (NSQIP 2008). For comparison, the RCRI applied to the same 2008 data gave a c-statistic of 0.747 [1].

External validation has shown considerably poorer performance. Fronczek et al. validated NSQIP MICA and the RCRI in 870 patients aged ≥45 years undergoing non-cardiac vascular surgery at a tertiary centre, with systematic postoperative measurement of high-sensitivity troponin T and diagnosis according to the Third Universal Definition of Myocardial Infarction. The composite outcome occurred in 8.7%. NSQIP MICA had a c-statistic of 0.64 (95% CI 0.57–0.70) compared with 0.60 for the RCRI. Calibration was poor: the model systematically underestimated risk (P<0.001) [2].

Alrezk et al. tested Gupta MICA on NSQIP 2012 data restricted to geriatric patients (≥65 years, n = 172,905). The AUC fell to 0.70, a deterioration of approximately 17% compared with its original performance in NSQIP 2007. Calibration was inadequate: the model underestimated risk in the low-risk categories and overestimated it in the high-risk categories, with a median deviation of −0.73 percentage points in geriatric patients [4].

In summary, the model performs well in the population from which it was derived, but discrimination falls markedly on external validation, particularly in geriatric cohorts and in vascular surgery. Calibration is a recurring problem, with systematic underestimation of the observed risk.

Limitations

The model was derived from NSQIP data from 2007 and reflects the surgical technique and patient population of that time. Surgical outcomes have improved since then, which means that the coefficients may be outdated. Alrezk et al. point out that models should be updated at regular intervals to reflect changing practice [4].

The geriatric population is a particular problem. The assumption of linearity for age means that the model does not capture the non-linear increase in risk seen in older patients, which contributes to the poorer performance and the systematic underestimation of risk in this group [4].

Fronczek et al. showed that when perioperative myocardial infarctions are diagnosed with systematic troponin measurement under the current universal definition, the c-statistic falls sharply compared with the derivation study, which relied on clinically detected infarctions [2]. This is a fundamental limitation: the model was trained on an outcome definition that misses asymptomatic troponin rises, which account for a substantial proportion of perioperative myocardial injury in modern practice.

The model lacks several clinically relevant predictors: ischaemic heart disease, heart failure, diabetes, stroke and anaemia are not included. The RCRI includes several of these, which partly explains why the RCRI is still used alongside it. On the other hand, Gupta's model includes the type of procedure in detail, which the RCRI does not.

The creatinine threshold of 1.5 mg/dL (corresponding to approximately 133 µmol/L) is the same as in the RCRI and is a crude binary classification that takes no account of the eGFR or of sex differences in creatinine.

References

1. Gupta PK, Gupta H, Sundaram A, et al. Development and validation of a risk calculator for prediction of cardiac risk after surgery. Circulation. 2011;124(4):381–7. PMID: 21730309
2. Fronczek J, Polok K, Devereaux PJ, et al. External validation of the Revised Cardiac Risk Index and National Surgical Quality Improvement Program Myocardial Infarction and Cardiac Arrest calculator in noncardiac vascular surgery. Br J Anaesth. 2019;123(4):421–429. PMID: 31256916
3. Halvorsen S, Mehilli J, Cassese S, et al. 2022 ESC Guidelines on cardiovascular assessment and management of patients undergoing non-cardiac surgery. Eur Heart J. 2022;43(39):3826–3924. PMID: 36017553
4. Alrezk R, Jackson N, Al Rezk M, et al. Derivation and validation of a geriatric-sensitive perioperative cardiac risk index. J Am Heart Assoc. 2017;6(11):e006648. PMID: 29146612