Clinical Laboratory Strategies: June 25, 2009

Strategies logo

Predicting Risk of Mortality Using BMP
Algorithm using common lab test found to be highly predictive
By Genna Rollins

Risk prediction is important in determining appropriate treatments and in providing patients with information to make decisions about their care. Yet with the exception of certain robust and well-accepted models, such as the Framingham Risk Score, risk prediction algorithms generally have had limited adoption and application in clinical practice. However, new research indicates that a model incorporating the basic metabolic panel along with patient age and sex has strong predictive ability for death at 30-days, 1-year, and 5-years. This issue of Strategies explores those findings.

The basic metabolic panel (BMP) is a common blood test in both the inpatient and ambulatory care settings, and elements of it, such as glucose and creatinine, have been shown individually to be predictive of poor prognosis. However, new research by University of Utah and Intermountain Healthcare researchers is the first to explore the predictive utility of the entire BMP measurement (Am Heart J 2009; 157:946-54).

The findings could lead to a new way for physicians and patients to understand and manage their risks, according to senior author Jeffrey Anderson, MD, professor of internal medicine at the University of Utah School of Medicine and associate chief of cardiology and vice chair for research at Intermountain Medical Center in Salt Lake City. "We measure BMP for virtually everybody, but as clinicians we’re not able to deal simultaneously with every component of BMP at once. However, the lab can do that, so our thought was why not combine what’s already in the hands of the lab and make a risk profile for mortality," he said. "The result could then be used to refine clinicians’ general impression of the risk profiles for their patients."

The study involved data from >279,000 adult patients treated at any Intermountain Healthcare facility between 1999 and 2005 who had at least one BMP result and subsequent follow-up. Follow-up data were available at 30-days for 275, 982 patients, 1-year for 229,113, and 5-years for 41,282. In addition to Intermountain Healthcare’s electronic data warehouse, the researchers accessed State of Utah death certificates and the Social Security death master file to develop a virtually complete mortality follow-up record. The researchers then randomly divided the study cohort into two independent populations. The training group used to develop the risk model included data from 167,635 patients, while the group used to test the model included data from 111,702 patients. The model was validated further internally using data from the Intermountain Heart Collaborative Study and Intermountain Healthcare patients with a diagnosis of primary heart failure, and externally using NHANES III data.

The various BMP components were categorized as low, normal, or high based on the standard range of normal, with creatinine ≥2 mg/dL further characterized as very high. The researchers used logistic regression to develop risk scores, which were subsequently adjusted for age and sex. Receiver operator characteristic curves were used to determine areas under the curve (AUC) for the models. A scoring algorithm with an integer risk score for each BMP metric used, along with sex and age, was determined to be the best at distinguishing clinically relevant and equally distributed divisions in risk. Thirty-days, 1-year and 5-year mortality risk AUCs for the training group were 0.887, 0.850, and 0.858, respectively; AUCs for the testing group at 30-days, 1-year and 5-years was 0.882, 0.848, and 0.847, respectively. In external validation using NHANES III data, BMP risk scores were significantly predictive of death at 30-days, 1-year and 5-years (all p <0.0001). The researchers also studied the algorithm in various sub-populations, including inpatient, ambulatory care, emergency, cardiovascular and heart failure patients. Thirty-day AUCs ranged from 0.894 in emergency patients to 0.732 in heart failure patients; 1-year AUCs ranged from 0.85 in emergency patients to 0.719 in heart failure patients; and 5-year AUCs ranged from 0.854 in emergency patients to 0.733 in cardiovascular patients. Based upon these findings, the authors conclude that "the BMP risk score is precise, easily computed, and provides valuable information regarding the composite meaning of the BMP, a routinely measured blood test, for short-term (30-days), intermediate-term (1-year) and long-term (5-years) mortality."

Detecting Critical Physiological Parameters

The robust performance of the BMP score, while notable, is not totally unforeseen, according to Larry Allen, MD, MHS, assistant professor of medicine at the University of Colorado Denver. "Although the formal use of BMP in risk prediction is novel, it should not be a surprise that something like BMP ends up having a lot of predictive value. Patients who are not doing well, and who are at high risk for death, particularly in the short run, will have derangements in the fundamental biologic processes BMP measures, like kidney function and acidosis," he observed. In an editorial accompanying the study, Allen called the findings "remarkable—a simple laboratory panel routinely performed on millions of patients each year provides highly significant, quantifiable information about risk of death.... The authors should be commended for this ironic finding, that although enormous resources are being poured into identification of novel genomic risk markers, we have not adequately assessed what we already have at our fingertips every day."

The authors have expanded on this study by exploring the addition of CBC to the BMP risk score. Further analysis also is aimed at better understanding the significance of the algorithm. "We want to discern what diseases it is marking, or whether it’s just a general indicator of health status. We think it’s a broad indicator, but we want to get a little more information, so we’re looking at common diseases to see which ones it picks up the most," said Anderson.

The authors envision a time when the BMP risk score could be added to routine lab reports to aide physicians and families in deciding upon treatment pathways; Intermountain Healthcare is piloting such an approach. "The most important validation would be to see if this made any difference in outcomes. Right now we don’t know what it can do in terms of changing a patient’s risk, but if the physician is alerted to the risk it might make him more or less aggressive in terms of modifying or treating modifiable risks," said Anderson. "Short of that, we feel it can be helpful in educating patients. The more prognostic information that is available the more it can help patients and families plan."

Rate this page:       
Page Access:

Clinical Laboratory Strategies Podcast
Predicting Risk of Mortality Using BMP©