EHR-based tool may help predict life expectancy in older patients with diabetes
The tool uses sex, body mass index, serum creatinine level, dementia, metastatic cancer, peripheral vascular disease, albuminuria, home oxygen use, wheelchair use, current smoking, and the interaction between age and heart failure to generate a risk score.
A tool based on data from the electronic health record (EHR) may help clinicians estimate life expectancy in older adults with diabetes, a recent study found.
To develop and validate the tool, researchers used predictive modeling based on survival analysis in a retrospective cohort of adults in the Kaiser Permanente Northern California health system who had diabetes and were at least 65 years of age. The cohort was followed from 2015 through 2019 and was randomly split into a training set for model development (n=87,085) and a test set for validation (n=24,311). The tool was also validated in a 2010 cohort with 10-year follow-up through 2019 (n=89,563) and a 2019 cohort with two-year follow-up through 2020 (n=152,357). The main measures were demographics, diagnoses, utilization and procedures, medications, behaviors and vital signs, and mortality. The results were published May 30 by the Journal of General Internal Medicine.
The mean age of patients in the training set was 75 years, 49% were women, and 48% were racial and ethnic minorities; these demographics were similar in all three test sets. Overall, 23% of those in the 2015 cohort died during five years of follow-up, 47% of the 2010 cohort died during 10 years of follow-up, and 9% of the 2019 cohort died during two years of follow-up. The researchers used 94 candidate variables to develop a mortality prediction model, which was distilled into a life expectancy model with 11 input variables that was used as a risk-scoring tool.
The variables included in the tool were sex, dementia, metastatic cancer, peripheral vascular disease, body mass index, serum creatinine level, albuminuria, home oxygen use, wheelchair use, current smoking, and interaction between age (five-year categories) and heart failure. The tool generated a score that ranged from 0 to 22 and had good discrimination in the test set (C-statistic, 0.78), the 2010 cohort (C-statistic, 0.74), and the 2019 cohort (C-statistic, 0.81). Good calibration was seen when both observed and predicted survival curves were compared.
The authors noted that their tool, which they named the Life Expectancy Estimator for Older Adults with Diabetes (LEAD), was not validated in patients younger than age 65 years and was less accurate over longer periods, among other limitations. They also cautioned that the tool's results should be considered in a broader context along with other factors that influence individualized care, such as patient preferences.
“This simple tool is trained in a diverse cohort of men and women with diabetes, demonstrated good discrimination and calibration in several validation and temporal cohorts, and had flexible (vs. fixed) outcome periods,” the authors concluded, adding that it was designed for ease of use on paper or via an EHR or app and does not depend on values from a special clinical examination. “In addition to guiding diabetes treatment goals or identifying patients that could be screened for overtreatment, LEAD also has utility to compare predicted life expectancy across populations, inform the need for cancer screening and long-term and advanced care, resource allocation, and determine eligibility for clinical trials,” the authors wrote.