You are here: Home / Publications / Publications Database / Can Screening for Self-Rated Health Improve Prediction of Whether Adolescent Patients Will be Diagnosed With a Chronic Disease in Young Adulthood? A Longitudinal Cohort Study

Can Screening for Self-Rated Health Improve Prediction of Whether Adolescent Patients Will be Diagnosed With a Chronic Disease in Young Adulthood? A Longitudinal Cohort Study

Allen, Chenoa D.; McNeely, Clea A.; & Ehrenthal, Deborah. (2017). Can Screening for Self-Rated Health Improve Prediction of Whether Adolescent Patients Will be Diagnosed With a Chronic Disease in Young Adulthood? A Longitudinal Cohort Study. Presented at the Journal of Adolescent Health, New Orleans, LA.

Allen, Chenoa D.; McNeely, Clea A.; & Ehrenthal, Deborah. (2017). Can Screening for Self-Rated Health Improve Prediction of Whether Adolescent Patients Will be Diagnosed With a Chronic Disease in Young Adulthood? A Longitudinal Cohort Study. Presented at the Journal of Adolescent Health, New Orleans, LA.

Octet Stream icon 6822.ris — Octet Stream, 3 kB (4072 bytes)

Purpose
Self-rated health (SRH) is measured with a single question asking respondents to rate their general health. SRH is easy to measure, has measurement equivalence across racial/ethnic groups among adolescents, and in population surveys consistently predicts physical and mental health, health care utilization, and mortality. Not surprisingly, there is growing interest in the clinical utility of SRH to identify individual adolescents’ risk of chronic illness and health care utilization. Although SRH consistently predicts health status in epidemiologic studies of populations, it may not reliably and validly predict risk for individuals in clinical settings. Studies of the clinical utility of SRH suffer from major methodological limitations. First, existing studies control for too few predictors. The appropriate question is whether measuring SRH provides additional information over and above clinical variables already collected by providers. Second, few studies use appropriate statistical methods. Models to longitudinally predict individual outcomes should be evaluated on their ability to 1) correctly predict the proportion that will experience the outcome (calibration), 2) correctly identify which individuals in the dataset will experience the outcomes (discrimination), and 3) assess improvements in discrimination with the addition of a single indicator (reclassification). In this paper we apply long-standing and newly developed methods to determine if clinical assessment of SRH can increase healthcare providers’ ability to identify which adolescents will be heavy health care utilizers or have a chronic disease in five years.

Methods
Data come from Waves 1 (ages 11-21), 3 (ages 18-28), and 4 (ages 24-34) of the National Longitudinal Study of Adolescent to Adult Health. Outcome variables are 1) hospitalized in the past five years (W3; except for childbirth); 2) prescription medication for more than one condition in the past year (W4); 3) diagnosis of cancer, diabetes, or heart disease; 4) diagnosis of ashma; and 5) diagnosis of migraine. Predictor variables (W1) included demographic characteristics, health status, and health behaviors recommended for preventive clinical screening by the GAPS protocol. We estimated logistic regression models (with and without SRH) and conducted post-hoc analyses to assess the internal calibration (calibration graphs), internal discrimination (Somers D, ROC curves, c-index), and improvements in discrimination for SRH (net reclassification index, integrated discrimination improvement, mean risk difference) using bootstrap sampling in Stata 14.

Results
After controlling for health status and health behaviors typically measured in clinical encounters, SRH did not improve identification of which adolescents would experience chronic illness or utilize health care in young adulthood, even though the association between SRH and chronic disease (but not health care utilization) was statistically significant.

Conclusions
Measuring adolescents’ SRH in the clinical encounter does not increase our ability to predict chronic disease or health care utilization for individual patients in young adulthood, over and above measures already recommended by GAPS. More broadly, estimation of a new variable’s capacity to predict long-term health outcomes should be based on measures of calibration, discrimination, and reclassification, not simply on effect sizes and p-values.




CONF

Socirty for Adolescent Health and Medicine


Allen, Chenoa D.
McNeely, Clea A.
Ehrenthal, Deborah



2017



60

2, Supplement 1

S60




Journal of Adolescent Health

New Orleans, LA

1054-139X

10.1016/j.jadohealth.2016.10.302



6822