Background: Radium-223 (Ra-223), an α–particle-emitting isotope, inhibits bony metastases and prevents patients from skeletal-related events in metastatic castration-resistant prostate cancer (mCRPC). We retrospectively reviewed the treatment response, predictive factors, and adverse events (AEs) of Ra-223 before the National Health Insurance reimbursement in a Taiwanese tertiary institute. Methods: Patients treated with Ra-223 before January 2019 were enrolled and categorized into progressive disease (PD) and clinical benefits (CB) groups. Laboratory data before and after the treatment were collected, and spider plots concerning percentage changes of alkaline phosphatase (ALP), lactate dehydrogenase (LDH), and prostate-specific antigen (PSA) were prepared and calculated statistically. CB/PD, baseline ALP, LDH, and PSA levels were also adopted as stratification factors for overall survival (OS). Results: Among 19 patients included, 5 (26.3%) and 14 (73.4%) belonged to the PD and CB groups, respectively, with no significant difference observed in the baseline laboratory data. The percentage changes in ALP, LDH, and PSA levels after Ra-223 treatment were statistically significant among the two groups (ALP: CB 54.3 ± 21.4% vs PD 77.6 ± 11.8%, p = 0.044; LDH: CB 88.2 ± 22.8% vs PD 138.3 ± 49.0%, p = 0.046; PSA: CB 97.8 ± 61.7% vs PD 277.0 ± 101.1%, p = 0.002). The trends of LDH between the two groups in spider plot were separated significantly. There were no differences in the AEs between the two groups. CB had a longer median OS than the PD group (20.50 months vs 9.43 months, p = 0.009). Patients with LDH <250 U/L at baseline tended to have longer OS but without significance. Conclusion: The CB rate of Ra-223 was 73.7%. No predictive factor for treatment response was obtained from pretreatment data. The mean percentage changes in ALP, LDH, and PSA levels compared with baseline significantly differed between the CB and PD groups, especially the LDH levels. The CB and PD groups showed different OS, with LDH levels exhibiting the potential to predict OS.