Wednesday, March 22, 2006
Circulation has published an article about the derivation and validation of the Seattle Heart Failure Model (an online version of the model is available here--it's kind of fun to play around with). It is a survival model based on clinical and laboratory characteristics. It can uses relatively easily identifiable clinical (age, weight, NYHA class, etc) and measured data (hemoglobin, uric acid, ejection fraction, cholesterol, etc). Interestingly it also includes interventions (both drug and device) and you can see how the addition of these affects prognosis if you add them. It gives you mean survival in years, as well as expected 1, 2, and 3 year survival in percentages.
The Circulation article describes its development. It used data from 6 cohorts of patients (outpatients with mostly left-sided systolic heart failure), which was gathered in other controlled trials; they used multivariate analyses to derive the prediction model using one trial (n=1100), then validated it using the patient data from the 5 other trials (n=10,000). Of note they mention that the biggest univariate predictor of death was diuretic use (in milligrams of furosemide equivalents per kilogram of body mass). These are graphs from both the derivation and validation cohorts (predicted vs. actual mortality), showing generally very tight fits.
I'm not really qualified to judge the validity of this, but I'm curious to see how it is going to be used. Its feature letting you "see" the mortality benefit from adding in a device is curious, and open to misinterpretation it seems. Its use in the palliative world, obviously, would be to identify those patients at particularly high risk of death and thus "in need" of palliative-oriented services--but what this model adds over usual clinical assessment is unclear as of now. Also, this used data from controlled trials which generally have highly select patient groups in them--not necessarily healthy patients (many involved NYHA Class 3 & 4 patients)--but perhaps ones with fewer comorbidities etc (I have not looked at the in-/exclusion criteria for these trials of course). A prospective multicenter validation would be nice. Also, when looking at the graphs, one gets the impression that most of the subjects in these trials had estimated and actual survivals well over 1 year, which makes me wonder how accurate this will be for the sickest patients with the potentially shortest estimated survivals...