Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Can risk scores based on common laboratory tests predict short-term mortality?
prospective cohort study with ⩽5 years of follow-up.
Intermountain Healthcare in Utah, USA.
derivation (n = 71 921) and internal validation (n = 47 458) populations were patients ⩾18 years of age (mean age 55 y, 58% women) who had had a complete blood count (CBC) and basic metabolic profile (BMP) done. External validation cohorts were patients from the Third National Health and Nutrition Examination Survey (NHANES III, n = 16 372, mean age 48 y, 54% women) and patients having coronary angiography (mean age 61 y, 54% men).
Description of prediction guide:
the Intermountain Risk Score included, from the CBC, haematocrit, red cell distribution width, corpuscular volume, platelet count, platelet volume, corpuscular haemoglobin concentration, and total white blood cell count; from the BMP, serum concentrations of sodium, potassium, bicarbonate, creatinine, glucose, and calcium; and age. Each component was divided into quintiles, and a value ranging from 0 to 5 (which varied by sex and follow-up period) was assigned to each level based on results of logistic regression analysis. Age was divided into 7 categories (values −5 to 8). The individual values were summed to calculate the risk score (maximum score 27–33). Thresholds for dividing the population into low, moderate, and high risk were those that provided >90% sensitivity or specificity. For example, for 30-day mortality, a score <15 for both sexes indicated low mortality risk, whereas a score ⩾20 for women or ⩾19 for men indicated high risk.
all-cause mortality at 30 days, 1 year, and 5 years.
The area under the receiver operating characteristic curve (c-statistic) for the Intermountain Risk Score in predicting mortality was 0.83–0.90 in the derivation population, 0.82−0.89 in the internal validation population, and 0.78–0.87 in the angiography population. In most cases, the risk score was more accurate in women than in men. In the NHANES III population, risk scores were strongly associated with death at 1 and 5 years. The table shows relative risks for death by category of risk.
The Intermountain Risk Score, based on common laboratory tests, was highly predictive of short-term mortality.
Horne BD, May HT, Muhlestein JB, et al. Exceptional mortality prediction by risk scores from common laboratory tests. Am J Med 2009;122:550–8.
▸ Clinical impact ratings: GP/FP/Primary care 7/7; IM Ambulatory care 5/7; Cardiology 5/7; Endocrine 5/7; Haematology/Thrombosis 5/7; Public health 5/7
The study by Horne et al is an example of data in search of clinical usefulness. The authors are probably correct that much information in the CBC and BMP is not used by clinicians.1 It is laudable to attempt to use all data; however, this study does not seem to return information in a clinically useful form.
A major drawback is linking the data in multitest panels to all-cause mortality. Although other established risk measures (eg, Framingham for heart disease and APACHE for critical care outcomes) have limitations, they assess risk of conditions that are specific and amenable to intervention. All-cause mortality is too broad an outcome to apply indiscriminately. The authors suggest that their risk score be used to broadly encourage “surveillance and detection” but could have better informed clinicians about what to survey or detect.
Part of the limitation in using the Intermountain Risk Score may be because the CBC and BMP results used to develop the score were obtained from inpatients or emergency patients—not an ideal population for comparison with healthy ambulatory patients for whom the score might have the most utility.
The authors attempted to validate the score by testing it in 2 other populations. However, these validations seem to highlight the difficulties in using the risk score. The angiographic population had risk scores “associated with death at 30 days and 1 year,” which is not surprising because such patients are typically suspected of having coronary disease. In contrast, the predictive power for the generally healthy NHANES III population was weaker than for the Intermountain validation population, which is also not surprising.
To summarise, calculation of the Intermountain Risk Score should be used with caution because it has the potential for more adverse than beneficial consequences. Presenting such a score runs the risk of heightened anxiety for the patient, urgency for the doctor to find some elusive diagnosis, and ultimately, frustration and expense.
Source of funding: no external funding.