Clinical research study
Exceptional Mortality Prediction by Risk Scores from Common Laboratory Tests

https://doi.org/10.1016/j.amjmed.2008.10.043Get rights and content

Abstract

Background

Some components of the complete blood count and basic metabolic profile are commonly used risk predictors. Many of their components are not commonly used, but they might contain independent risk information. This study tested the ability of a risk score combining all components to predict all-cause mortality.

Methods

Patients with baseline complete blood count and basic metabolic profile measurements were randomly assigned (60%/40%) to independent training (N = 71,921) and test (N = 47,458) populations. A third population (N = 16,372) from the Third National Health and Nutrition Examination Survey and a fourth population of patients who underwent coronary angiography (N = 2558) were used as additional validation groups. Risk scores were computed in the training population for 30-day, 1-year, and 5-year mortality using age- and sex-adjusted weights from multivariable modeling of all complete blood count and basic metabolic profile components.

Results

Area under the curve c-statistics were exceptional in the training population for death at 30 days (c = 0.90 for women, 0.87 for men), 1 year (c = 0.87, 0.83), and 5-years (c = 0.90, 0.85) and in the test population for death at 30 days (c = 0.88 for women, 0.85 for men), 1 year (c = 0.86, 0.82), and 5 years (c = 0.89, 0.83). In the test, the Third National Health and Nutrition Examination Survey, and the angiography populations, risk scores were highly associated with death (P <.001), and thresholds of risk significantly stratified all 3 populations.

Conclusion

In large patient and general populations, risk scores combining complete blood count and basic metabolic profile components were highly predictive of death. Easily computed in a clinical laboratory at negligible incremental cost, these risk scores aggregate baseline risk information from both the popular and the underused components of ubiquitous laboratory tests.

Section snippets

Study Populations and End Points

This study's primary aim was to develop and validate risk scores for mortality that aggregate all of the independent risk information contained within the complete blood count and the basic metabolic profile into a useful and intuitive metric for clinicians.

The primary outcome of this study was incident all-cause mortality. Death outcomes were determined from Intermountain Healthcare electronic records (covering an integrated delivery system of 22 hospitals and many clinics and employed

Training and Test Populations

Demographics of the training population were as follows: age, 55.0 ± 20.0 years (range: 18-103 years); female, 58.4%; inpatient, 48.5%; outpatient, 40.3%; and emergency, 11.2%. Results were similar for the test population for age (55.2 ± 19.9 years), sex (58.0% were female), patient care setting, and laboratory test component values (all P >.12 compared with the training population).

Sex-specific risk score values are provided in Table 2 (overall values are in Supplemental Table 1, available

Discussion

Optimally, a general medical risk score would be intuitive to clinicians, would apply to any patient, group, or individual, and would not complicate (by time or expense) the care-delivery process—being computed outside of the clinical setting and provided on standard clinical reports. The components of this ideal risk score would not have qualitative or subjective measures, would be simple to obtain, would use commonly obtained well-characterized tests, and would not use specialty-specific risk

Study Strengths and Limitations

Because the study was observational, it might be limited by confounders and unmeasured variables may have influenced study findings. The study used only a baseline measurement of complete blood count and basic metabolic profile, so repeated measurement over time may provide additional research ability to stratify risk and clinical opportunities to ameliorate risk. Use of quintiles of component measures instead of other statistical methods of categorization may have made the risk scores more

Conclusions

In large, independent patient populations across a large integrated health care system and in a nationally representative general US population, the Intermountain Risk Score provided exceptional stratification of mortality by simultaneously modeling the components of the complete blood count and basic metabolic profile together with age and sex. Although only a few of the components of these panels are routinely used clinically, each component provided meaningful contribution to risk, including

Cited by (89)

  • Intermittent fasting and changes in clinical risk scores: Secondary analysis of a randomized controlled trial

    2023, International Journal of Cardiology: Cardiovascular Risk and Prevention
  • Artificial intelligence in heart failure and transplant

    2023, Artificial Intelligence in Clinical Practice: How AI Technologies Impact Medical Research and Clinics
View all citing articles on Scopus

Funding: This study was funded by internal institutional funds.

Conflict of Interest: BDH, HTM, BSR, and JLA are named as inventors on a patent protecting the risk scores; the authors have no other potential conflicts of interest to report.

Authorship: All authors had access to the data and played a role in writing this manuscript.

Trial Registration: Database registry of the Intermountain Heart Collaborative Study: NCT00406185 (ClinicalTrials.gov).

Current address: Clinical Development and Medical Affairs, GlaxoSmithKline, 5 Moore Drive, Research Triangle Park, NC 27709.

View full text