Objectives: A survival analysis was conducted on patients with COPD receiving long-term oxygen therapy (LTOT) to compare two different statistical methods.
Methods: We used a multivariate crude (observed) survival model (Cox) and a multivariate relative survival model (Hakulinen). Only the latter is able to correct the survival by adjusting it to the normal life expectancy of the studied patients.
Patients: Two hundred fifty-two hypoxemic COPD patients (207 male) requiring LTOT were included. Mean PaO2 was <50 mm Hg before oxygen therapy. Mean age was >69 years (SE: 9.9). They had severe bronchial obstruction: mean FEV1 was <33% (10.6) of predicted values, with some CO2 retention: mean PaCO2 was 45.6 (7.1) mm Hg. By December 31, 1995, 189 patients had died (75%) and 13 (5%) were unavailable for follow-up.
Results: The overall crude survival was poor: 80.9% after 1 year, 67.1% after 2 years, 34.7% after 5 years, and 7.1% after 10 years. In the crude multivariate analysis (Cox), the negative prognostic factors were age and hypercapnia. The overall relative survival (Hakulinen), corrected for life expectancy, was 82.8% after 1 year, 70.8% after 2 years, 41.5% after 5 years, and 10.25% after 10 years. In the final multivariate relative model, age was no longer significant and the only bad prognostic factor was hypercapnia with a relative risk of 1.97 (1.16 to 3.34).
Conclusion: This work shows the inadequacy of the Cox observed survival model when it comes to appreciating the real prognostic impact of age, because of the confusing factor associated with a normal life expectancy.