To observe the effect of high-flow nasal cannula oxygen therapy (HFNC) in patients with chronic obstructive pulmonary disease (COPD) and mild hypercapnia, and to evaluate the early predictive ability of physiological parameters in these patients. A retrospective cohort study was conducted based on Medical Information Mart for Intensive Care-IV (MIMIC-IV) updated in September 2020 and the data of adult patients with COPD and mild hypercapnia [45 mmHg (1 mmHg = 0.133 kPa) < arterial partial pressure of carbon dioxide (PaCO2) ≤ 60 mmHg] from 2008 to 2019 were collected. These patients were assigned to the HFNC group or non-invasive ventilation (NIV) group according to whether they received HFNC or NIV. Baseline data such as gender, age, body mass index (BMI), simplified acute physiology score II (SAPS II), Charlson comorbidity index (CCI) and physiological parameters were collected. A propensity score matching was conducted according to the baseline data of the HFNC group patients. The 48-hour and 28-day intubation rates, 28-day mortality, length of intensive care unit (ICU) stay, the length of hospital stay, and the changes in physiological parameters within 48 hours after treatment were compared between the two groups. The receiver operating characteristic curve (ROC curve) was drawn and the ratio of heart rate over pulse oxygen saturation (HR/SpO2) and ROX index [SpO2/(inhaled oxygen concentration, FiO2×respiratory rate, RR)] were analyzed to predict the 24-hour and 48-hour intubation rates. A total of 524 520 inpatient records were screened and 153 patients were included, while 37 patients in the HFNC group and 116 patients in NIV group. There were 31 patients in the HFNC group and 84 patients in the NIV group remained after propensity score matching according to the baseline data. There were no significant differences in the baseline data of gender, age, BMI, SAPS II, CCI score, physiological parameters and prognosis data except the length of ICU stay. The length of ICU stay in HFNC group was significant longer than that of the NIV group [days: 4.6 (3.1, 10.0) vs. 3.1 (1.6, 5.8), P < 0.05]. HR and RR at 40-48 hours were significantly lower than those at 0-8 hours after treatment only in the HFNC group [HR (bpm): 84.1±12.2 vs. 91.1±16.4, RR (times/min): 19.8±4.9 vs. 21.6±4.1, both P < 0.05]. Both in the HFNC group and NIV group the pH increased (7.42±0.08 vs. 7.36±0.05 and 7.41±0.06 vs. 7.36±0.05, both P < 0.05) and PaCO2 decreased significantly [mmHg: 46.3 (39.5, 51.0) vs. 49.8 (45.5, 54.0) and 46.0 (40.5, 51.5) vs. 49.5 (45.5, 55.3), both P < 0.05]. The HR, PaO2 were higher in the HFNC group than those in the HFNC group at 40-48 hours after treatment [HR (bpm): 91.1±15.4 vs. 84.1±12.2, PaO2 (mmHg): 99.5 (86.0, 132.3) vs. 85.8 (76.5, 118.0), both P < 0.05], PaO2/FiO2 were lower in the HFNC group than that in the HFNC group at 40-48 hours after treatment [mmHg: 223.8 (216.5, 285.0) vs. 278.0 (212.3, 306.0), P < 0.05]. Both HR/SpO2 and ROX index at 4 hours after treatment had predictive value for 24-hour and 48-hour intubation in the HFNC group. The areas under ROC curve (AUC) of HR/SpO2 at 4 hours after treatment in the HFNC group were larger than those of ROX index for predicting 24-hour and 48-hour intubation (24-hour: 0.649 vs. 0.574, 48-hour: 0.692 vs. 0.581, both P < 0.01); the 95% confidence interval (95%CI) of 4 hours HR/SpO2 and for ROX index predicting 24 hours and 48 hours intubation were 0.497-0.780, 0.567-0.799, 0.450-0.694 and 0.454-0.716, respectively. The high sensitivity of HR/SpO2 and ROX index in predicting 24-hour and 48-hour intubation were 84.6%, 92.9%, 88.2% and 94.4%, respectively, and the low specificity were 52.3%, 23.7%, 54.7% and 29.6%, respectively. HFNC can be used in COPD patients with mild hypercapnia, but it cannot replace NIV. The accuracy of ROX index at 4 hours after HFNC treatment in predicting intubation in COPD patients with mild hypercapnia is poor.