By considering the resource-constrained affordable wearable or portable health monitoring devices, in this paper, we present a lightweight digital compressed ECG sensing with recognition of fast/slow and regular/irregular heartbeat patterns by using beat-to-beat intervals (BBIs) directly computed from compressed sensing (CS) measurements without ECG reconstruction process. For extracting BBIs, we presented a fast straightforward R-peak detection method in the CS domain without using sets of search-back detection rules unlike other methods. On the standard MIT-BIH arrhythmia database, the CS-ECG based R-peak detection method had an average accuracy of 99.47% with false positives of 381 beats and false negatives of 195 beats for a total of 109021 beats. The CS-ECG based HR classification method, with three classes of bradycardia, tachycardia and normal, had an accuracy of 96.75%, 91.18% and 99.02% based on the feature of number of BBIs and an accuracy of 95.64%, 82.35% and 98.81% based on the feature of average BBIs. The CS-ECG based regular/irregular rhythm (RIR) recognition method had a sensitivity (SE) of 93.17%, specificity (SP) of 96.87% and accuracy (ACC) of 94.83% based on the variation between the BBIs that are determined by comparing the difference between the successive BBIs with a predefined BBI threshold. The proposed method can reduce computational resources and energy reduction of 75% with the CS-based data reduction factor of 4 that has a great potential in energy-efficiency of battery operated wearable devices in long-term continuous health monitoring environments. © 2023 IEEE.