Return to search

The use of clinical, behavioral, and social determinants of health to improve identification of patients in need of advanced care for depression

Indiana University-Purdue University Indianapolis (IUPUI) / Depression is the most commonly occurring mental illness the world over. It poses
a significant health and economic burden across the individual and community. Not all
occurrences of depression require the same level of treatment. However, identifying
patients in need of advanced care has been challenging and presents a significant bottleneck
in providing care. We developed a knowledge-driven depression taxonomy comprised of
features representing clinical, behavioral, and social determinants of health (SDH) that
inform the onset, progression, and outcome of depression. We leveraged the depression
taxonomy to build decision models that predicted need for referrals across: (a) the overall
patient population and (b) various high-risk populations. Decision models were built using
longitudinal, clinical, and behavioral data extracted from a population of 84,317 patients
seeking care at Eskenazi Health of Indianapolis, Indiana. Each decision model yielded
significantly high predictive performance. However, models predicting need of treatment
across high-risk populations (ROC’s of 86.31% to 94.42%) outperformed models
representing the overall patient population (ROC of 78.87%). Next, we assessed the value
of adding SDH into each model. For each patient population under study, we built
additional decision models that incorporated a wide range of patient and aggregate-level
SDH and compared their performance against the original models. Models that
incorporated SDH yielded high predictive performance. However, use of SDH did not yield
statistically significant performance improvements. Our efforts present significant
potential to identify patients in need of advanced care using a limited number of clinical
and behavioral features. However, we found no benefit to incorporating additional SDH
into these models. Our methods can also be applied across other datasets in response to a
wide variety of healthcare challenges.

Identiferoai:union.ndltd.org:IUPUI/oai:scholarworks.iupui.edu:1805/17765
Date30 May 2018
CreatorsKasthurirathne, Suranga N.
ContributorsJones, Josette, Grannis, Shaun, Biondich, Paul, Purkayastha, Saptarshi, Vest, Joshua
Source SetsIndiana University-Purdue University Indianapolis
Languageen_US
Detected LanguageEnglish
TypeDissertation

Page generated in 0.0016 seconds