A multivariate model for predicting hospital admissions for patients with decompensated chronic obstructive pulmonary disease.

PURPOSE

To develop a method for predicting hospital admissions for patients with decompensated chronic obstructive pulmonary disease treated in an emergency department.

METHODS

A 4-year survey including training and validation periods was conducted. Stepwise logistic regression was used to develop a multivariate model using information from the patient's previous visits and results of baseline pulmonary function tests.

MEASUREMENTS AND MAIN RESULTS

During the first 2 years, there were 693 visits to the emergency department for decompensated chronic obstructive pulmonary disease. The patient was admitted to the hospital on 210 occasions (30.3%). Logistic regression showed that the probability of admission was related to the following: the admission and relapse rates for previous visits, the proportion of previous discharges from the emergency department in which "conservative therapy" was given, the highest baseline post-bronchodilator forced expiratory volume in 1 second within 3 years of entry, and the highest baseline pre-bronchodilator forced expiratory volume in 1 second-vital capacity ratio. A relapse was defined as an unscheduled return to the emergency department within 48 hours. "Conservative therapy" was any treatment regimen that did not include parenteral medications. During the next 2 years, the model was validated with patients not previously treated at this medical center. Seventy-six (28.3%) of 269 episodes resulted in hospital admission. The logistic model was used to categorize each visit during the validation phase. "High-risk" visits had calculated probabilities of admission greater than .208, while "low-risk" visits had values that were less. The admission rate for 98 low-risk visits (8.2%) was much lower than the rate for 171 high-risk visits (39.8%).

CONCLUSIONS

A multivariate model can be used to identify patients with decompensated chronic obstructive pulmonary disease who are unlikely to need hospitalization. This model could be used to select episodes of decompensated chronic obstructive pulmonary disease for treatment at home.