Coming Soon to a Lab Near You: The Holy Grail
Coming Soon to a Lab Near You: The Holy Grail
Abstract & Commentary
By Allan J. Wilke, MD, Associate Professor of Family Medicine, University of Alabama at Birmingham School of Medicine Huntsville Regional Medical Campus, Huntsville. Dr. Wilke reports no financial relationship to this field of study.
Synopsis: The combination of a CBC and a BMP can predict death.
Source: Horne BD, et al. Exceptional mortality prediction by risk scores from common laboratory tests. Am J Med 2009;122:550-558.
These investigators from the University of Utah and the Intermountain Medical Center set an ambitious agenda: to develop a mortality risk scoring schema, which could be applied to the general population, using easily available and commonly obtained laboratory tests. They chose the complete blood count (CBC) and the basic metabolic profile (BMP), which measures concentrations of sodium, potassium, chloride, bicarbonate, blood urea nitrogen, creatinine, glucose, and calcium. They had experience doing this, having previously developed the Complete Blood Count Risk Score1 for use in patients with heart disease and the Basic Metabolic Profile Risk Score2 for use in the general population. They had a powerful ally: the Intermountain Healthcare electronic records. They linked the laboratory results in that database to the state of Utah's death certificates database and the Social Security death master file.
Data from four populations were used to develop and validate the risk score: two from the Intermountain records, another from the Third National Health and Nutrition Examination Survey (NHANES III), and the fourth, a group undergoing coronary angiography. The first Intermountain group was used to develop the risk models and the other three to validate them. The third group was chosen to provide a population that was more demographically similar to the rest of the U.S. population, and the fourth was chosen to provide a "high risk for death" population. Age in the two Intermountain groups averaged 55 years (range, 18-103 years) and 42% were male. For the 30-day mortality model, there were 71,921 subjects in the development group and 47,458 in the validation group. The numbers in the 1-year and 5-year models were smaller, but substantial. The blood samples were obtained from inpatient, outpatient, and emergency department settings in similar proportions in the two Intermountain groups.
Six separate risk models were evaluated, differentiating by gender and by 30-day, 1-year, and 5-year mortality. The results of each component of the CBC and the BMP were divided into quintiles. Patient ages were divided into 10-year intervals, except for the age 18-29 group and the age ≥ 80 group. In each of the six risk models, after appropriate statistical analysis, each component of the CBC and the BMP, each age interval, and both genders were assigned an integer point value. (Note: The actual table is 75 x 6 and not published here, but is in the article and is "freely available for academic use." More on that below.) The models, adjusted for age and gender, were compared using receiver operating characteristic curves.
The investigators discovered that the risk scores performed better in women and outpatients. They could not explain the gender difference, but attributed the difference in patient location to the "additional risk determinants" existing with hospitalized patients, i.e., they were sick with something. The risk models predicted death well in both the NHANES III and angiography groups. Sensitivity and specificity varied appropriately as different cutoff points were chosen: the higher the cutoff, the lower the sensitivity and the higher the specificity. Surprisingly, in the angiography group, the risk scores were independent of the patients' comorbidities and whatever treatment they received. Not surprisingly, age was a powerful predictor; however, age and gender alone were not as powerful as CBC and BMP without age and gender. Other powerful predictive components were: hematocrit (Hct) ≤ 34.6, white blood cell (WBC) count ≥ 11.3, red blood cell (RBC) distribution width (RDW) ≥ 14.4, bicarbonate ≤ 23, calcium ≤ 8.5, glucose ≥ 126, and creatinine ≥ 1.3. RDW ≥ 14.4 was the strongest single risk factor, except for being 80 years or older.
Commentary
This study provides more support for the value of electronic medical records and connected databases. It simply could not have been done without them (or could not have been done simply). There are a few questions and concerns to keep in mind. Are the people in the Intermountain Healthcare system sufficiently similar to the rest of Americans to apply the results generally? The results with the NHANES III group support this. This was an observational study, not a randomized controlled trial, but the very large number of subjects probably compensates for this. We don't know if correction of abnormal lab values improves mortality or what happens when there is repeat testing. This would make an interesting follow-up study.
Some of the predictors make sense. A low Hct suggests anemia. A high WBC suggests infection or inflammation. An elevated glucose points to diabetes and an elevated creatinine to kidney disease. I don't know what to make of the RDW finding. When I've noticed it, it's usually been in the situation of acute blood loss, and I've chalked it up to the patient's bone marrow pumping out immature RBCs, whose size is greater than mature RBCs. RDW is not something that I've routinely paid a great deal of attention to in the past, but I will from now on.
The authors argue for inclusion of one of their risk scores, based on age, gender, CBC, and BMP, in the patient reports you receive from your lab, similar to the estimated glomerular filtration rate calculations that we've grown used to seeing over the last decade. They are, after all, simple formulas, and it would not cost much to program a lab computer to print an additional line on the report. Four of the authors of this paper are named as inventors on a patent protecting these risk scores. The conflict of interest is obvious, but that doesn't mean that the risk scores aren't useful. It does mean that they need to be independently validated on other populations prior to widespread adoption. If and when they are adopted, how will we use them? Currently, we use the Framingham Risk score to predict the development of cardiovascular disease over the next 10 years and use 20% as the cutoff for intervention. If I received notification that based on a person's CBC and BMP that person had a 20% risk of death in the next 5 years, I would examine the patient and review the medical record, looking for modifiable risk factors.
In 1994, the Journal of the American Medical Association published one of a series of articles by the McMaster group on evidence-based medicine.3 This one looked at articles about prognosis, and posed several questions that require an affirmative answer to accept the findings of the study as valid. They are:
1. Was there a representative and well-defined sample of patients at a similar point in the course of the disease?
2. Was follow-up sufficiently long and complete?
3. Were objective and unbiased outcome criteria used?
4. Was there adjustment for important prognostic factors?
In this study, the population was a very large collection of patients in the Intermountain Healthcare system. There was no single disease; instead, there was a single outcome, death, which is objective and not subject to bias. The follow-up was up to 5 years for some patients and as complete as possible based on state and federal death databases. Important prognostic factors (gender and age) were used to adjust the results. On this basis, I believe the results are valid.
P.S.: Yes, I did check my risk score. Inquiring minds want to know.
References
1. Anderson JL, et al. Usefulness of a complete blood count-derived risk score to predict incident mortality in patients with suspected cardiovascular disease. Am J Cardiol 2007;99:169-174.
2. May HT, et al. Superior predictive ability for death of a basic metabolic profile risk score. Am Heart J 2009;157:946-954.
3. Laupacis A, et al. Users' guides to the medical literature. V. How to use an article about prognosis. Evidence-Based Medicine Working Group. JAMA 1994;272:234-237.
The combination of a CBC and a BMP can predict death.Subscribe Now for Access
You have reached your article limit for the month. We hope you found our articles both enjoyable and insightful. For information on new subscriptions, product trials, alternative billing arrangements or group and site discounts please call 800-688-2421. We look forward to having you as a long-term member of the Relias Media community.