Ann: CHF P3 Trial Accrues Sufficient Primary Endpoint Events, page-26

I had another read, you are correct. The enrichment was with both treated and untreated patients but only with baseline NYHA class III patients. That was because;

....in early 2017, the DREAM-HF trial data were reviewed by an independent statistical consultant who was blinded to the treatment groups to assess the rate of TCEs for the entire study population and as stratified by baseline NYHA class (separately for class II and III patients). The mean follow-up time was similar for baseline class III and II patients (≈15.4 months) as were the number of patients in each subgroup (class III, 147 patients; class II, 123 patients; class III/II ratio, 1.19). The sponsor was provided blinded-to-treatment summaries of the results. Strikingly, the number of TCEs was 3.5-fold higher for baseline NYHA class III patients than for class II patients (P=0.0014 by Kaplan-Meier log-rank analysis).

It is the NYHA class III patients who contributed the most MACE but they also had 3.5 times more terminal cardiac events than class II patients, that is they died and then their accumulated MACE events were censored, or removed, from the trial.

This significant excess of TCEs in the more vulnerable class III HF patients and their subsequent censoring of recurrent nonfatal HF-MACE from the trial are critical factors because the TCE-censored class III patients had a significantly higher rate of primary end point events before the time of censoring than did the TCE-censored class II patients. Over time, the expected result of disproportionately losing class III patients would be an important reduction in the number, as well as prolongation of the timing of future recurrent nonfatal HF-MACE.

From https://www.ahajournals.org/doi/10.1161/CIRCRESAHA.119.314951