PURPOSE: COPD is a common cause of hospitalization in the US. Many COPD patients are treated in community hospitals that are less represented in the current literature and their severity on admission is not well characterized. We sought to develop and validate a mortality predictive model using a large US clinical database which consists of clinical data from both teaching and non-teaching hospitals.
METHODS: We analyzed 57, 791 COPD admissions during 2004-2005 across 79 teaching and 112 non-teaching hospitals. We randomly split the study population into a derivation cohort (n=28,895; deaths=711) and a validation cohort (n=28,896; deaths=658). We developed a logistic regression model using age, gender, laboratory findings, vital signs, and comorbidities as covariates. We validated the model internally with 200 boot strap reiterations and externally with validation cohort. We used c-statistic to assess model fit. The results are presented as odds ratios and 95% confidence intervals.
RESULTS: Median age was 72 years old (IQR: 63-79). There were 55% women. The overall crude mortality was 2.4%. Predictors (p <.0001) with odds ratios greater than 2 included severe altered mental status (5.92, 4.54-7.71), moderate altered mental status (2.09, 1.79-2.43), pulse <50 or >129 per minute (2.08, 1.74-2.50), albumin < 2.5 g/dl (2.44, 1.83-3.27), pH arterial <7.21 (2.14, 1.56-2.93), BUN >40 mg/dl (2.42,2.03-2.90), AST >100 U/L (2.01, 1.38-2.93), and respiratory cancer or metastatic cancer (2.23, 1.77-2.82). The model c-statistic was .82. It calibrated well for both derivation and validation cohorts.
CONCLUSION: Admission physiological assessments indicating acidosis, hypoalbumenia, cardiac, neurological, and renal dysfunctions predict COPD mortality. Pathophysiological variables commonly measured on admission can generate a parsimonious and clinically plausible model.
CLINICAL IMPLICATIONS: This model is developed and validated with a large clinical database using information routinely available on admission. It can be used for risk stratification and outcome studies among COPD patients.
DISCLOSURE: Ying Tabak, No Financial Disclosure Information; No Product/Research Disclosure Information