Article Text
Abstract
Background Although we embarked on the era of the global increase in the ageing population, the recent declines in life expectancy simultaneously experienced by many high-income countries highlight a need for an accurate prediction model for estimating individual, rather than average, risk for mortality in older adults, based on readily accessible information about individuals’ lives, health and environment. Using advanced computer intensive statistical learning methods, we derived, evaluated and validated a prediction model of the 10-year risk for all-cause mortality in older adults from the general population.
Methods The model was developed using a prospective population-based cohort of English adults aged ≥50 years old from English Longitudinal Study of Ageing study. Having included a large pool of predictors, we employed cox proportional hazards model with regularisation by the least absolute shrinkage and selection operator (Cox-Lasso) to identity the most robust predictors of mortality and quantify their relative contribution to all-cause mortality in the next 10 years. The model was internally validated using Harrell’s optimism-correction procedure followed by external validation in the Health and Retirement Study, which is a nationally representative, longitudinal survey of adults aged ≥50 years old in the United States. The model’s prediction accuracy was evaluated with calibration, discrimination, sensitivity and specificity.
Results For model development, the sample comprised 9154 individuals; of these, 1240 (13.4%) died during the 10-year follow-up with an average length of survival of 70.2 months (SD=35.4). For external validation, the sample included 2575 individuals; of these, 491 (19.1%) died during the 10-year follow-up with an average length of survival of 77.7 months (SD=36.5). The prediction model selected 13 (15.5% of n=84) prognostic factors, which included increasing age, male gender, low accumulated wealth, comorbid health conditions (i.e., previous diagnoses of cancer, chronic lung disease or stroke), functional difficulties (i.e., difficulty walking 100 yards, or doing work around house and garden) and worsening memory. External validation demonstrated good discrimination (c-index=0.69), calibration (calibration slope β=0.80), specificity (73.2%) and sensitivity (72.4%).
Discussion Our model is likely to provide accurate estimates of individual 10-year risk of mortality using information that is often available in patients’ reports. It is calibrated for individuals aged 50–75 years living in the UK but generalises reasonably well to other populations with similar underlying characteristics. The developed prediction model could be used to communicate risk to individuals and their families (if appropriate), guide strategies for risk reduction and design future studies targeting high risk subpopulations.