Predicting the outcome of autologous cell therapy on ischemic heart disease; a potential role of risk factors on cell therapy efficacy?

Abstract Background Cell therapy has emerged as a promising additional therapy after myocardial infarction (MI). We have previously shown that interactions between risk factors and cell therapy influence efficacy, suggesting that only certain patient populations benefit from cell therapy after MI. Purpose We aim to generate and validate a prediction model for cell therapy after MI for both functional outcome and clinical endpoints. Methods Data from the REPAIR-AMI and REGENERATE-AMI trial were used to model the effect of autologous bone marrow-derived mononuclear cell transplantation after acute MI. First, a prediction model was generated based on the REPAIR-AMI trial. The outcome measure for the model was difference in ejection fraction (EF) over time. Prespecified variables of interest were age, sex, weight, hypertension, hyperlipidaemia, diabetes, history of smoking and baseline EF, taking into account their interaction per therapy (placebo vs cells). This model resembles the gain in function over time after cell therapy, compared to the placebo group. Validation of this model was performed in the REGENERATE-AMI trial dataset, to test whether interactions that influence the therapy are similar across trials. Finally, the model was tested for its ability to predict clinical outcome (i.e. MACE) in REPAIR-AMI. Results The generated prediction model on REPAIR-AMI data included multiple interactions with therapy. Internally, the model functioned appropriately to predict the difference in EF (n=186, R2=0.17, p=9x10–9). In the validation step, our model showed similar predictive capacity in REGENERATE-AMI (n=94, R2=0.19, p=1.4x10–5) specifically in the cell therapy group (R2=0.31, p=1.9x10–5) (see Figure). A multivariable analysis on the combined datasets revealed that patients with a smoking history or low baseline EF were more likely to have an increased functional response after cell therapy. Applying this prediction model to clinical endpoint data showed that a generated benefit score (based on cardiac function) translates to differences in event rate. The high benefit group yielded a significantly different event rate between groups (45% in placebo vs 15% in cell therapy group, p=0.001). Conclusion Based on real life clinical data of two autologous cell therapy trials we have generated a clinically useful prediction model for the response to cell therapy after MI. Particular factors (smoking, baseline EF) showed relevant interactions. Not only functional