Friday, August 1, 2008

Predicting in-hospital mortality from CHF

Journal of the American College of Cardiology has a look at predictors of in-hospital mortality for patients hospitalized with heart failure. The study uses data from the OPTIMIZE-HF registry which was a large, US-based quality improvement study/program (i.e. getting patients hospitalized with CHF on more evidence-based therapies, discharged on beta-blockers, etc. - see here) involving 259 hospitals (both academic and community) and 48,000 patients (mean age 73 years, both patients with systolic dysfunction and preserved systolic function were included). The database included ~50 variables: demographics, comorbidities, laboratory (hemoglobin, serum Na, etc.), drug categories (on diuretics, digoxin, etc.), and things like weight, vital signs, etc. In-hospital mortality was 3.8% (about 1800 patients) for the entire cohort.

Using the database, they derived a multivariate prediction model of in-hospital mortality which contained 18 variables. The strongest univarite predictors were serum creatinine (in-hospital mortality increased by 18% for every 0.3mg/dl increase in creatinine), age, and blood pressure (higher being more protective).

They then derived a relatively simple point-system based on the factors which most powerfully predicted mortality (the above 3 plus heart rate, serum sodium, presence or absence of systolic dysfunction, and whether or not CHF was the primary reason for hospitalization) and created a mortality risk nomogram based on that point system (available here - click on the prediction nomogram pdf). (Of note, the model only included patients with complete data so this was based on ~40,000 patients/~1300 deaths.) The model was validated with a within-cohort sample, as well as with data from other large CHF registries, with pretty good results (C-statistics greater than 0.7). As an example, an 85 year old with a pulse of 110, systolic BP of 90, serum Na of 120, serum creatinine of 2.5, and systolic dysfunction would have a ~40% chance of in-hospital mortality based on this model.

To rephrase that, of 100 patients presenting with those characteristics, about 40% would die each hospitalization. I rephrased that to highlight the obvious limits of such models - they can tell us really accurately what will happen to a population of patients but are really limited in telling us what will happen to the patient in front of us. One further caveat about these models is that since it comes from a large QI study there is reason to think that this may overestimate prognosis - patients are likely to do worse outside of such an environment (this is one of the reasons it is helpful to have it validated in outside cohorts, which was done, all of which however were large study registries....). Despite that, they can be used as clinical 'data points' (one of many) in helping us to counsel patients/families as to what to expect. More than this though these are best used as screening tools to identify patients/families with 'acute' palliative care needs (psychosocial/family assessment, prognostic and goals of care conversations, advance care planning, symptom assessment, etc.).

HT to Bob Arnold.

ResearchBlogging.orgABRAHAM, W., FONAROW, G., ALBERT, N., STOUGH, W., GHEORGHIADE, M., GREENBERG, B., OCONNOR, C., SUN, J., YANCY, C., YOUNG, J. (2008). Predictors of In-Hospital Mortality in Patients Hospitalized for Heart FailureInsights From the Organized Program to Initiate Lifesaving Treatment in Hospitalized Patients With Heart Failure (OPTIMIZE-HF). Journal of the American College of Cardiology, 52(5), 347-356. DOI: 10.1016/j.jacc.2008.04.028

Pallimed | Blogger Template adapted from Mash2 by Bloggermint