APPLYING MULTILEVEL MODELS TO HEALTH SERVICES RESEARCH FOR CHILDREN WITH MENTAL HEALTH PROBLEMS

Open Access
Author:
Gifford, Elizabeth Joanne
Graduate Program:
Health Policy and Administration
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
October 11, 2004
Committee Members:
  • E Michael Foster, Committee Chair/Co-Chair
  • Marianne Messersmith Hillemeier, Committee Member
  • D Wayne Osgood, Committee Member
  • Rebecca Wells, Committee Member
  • Linda Marie Collins, Committee Member
Keywords:
  • inpatient
  • mental health
  • substance abuse
  • multilevel modeling
  • children
  • services
Abstract:
This dissertation is composed of three studies that use multilevel modeling to examine variation in health services for youth with mental health and substance abuse problems. The first study examines whether the racial/ethnic variation in use of medication among youth with attention problems can be explained by factors related to where youth live. Data for this analysis come from the evaluation of the Comprehensive Community Mental Health Services for Children and Their Families program (CCMHSCF) and the Area Resource File (ARF). Our results suggest that among boys with attention problems, blacks and Hispanics were roughly 15 percent less likely to receive medication than whites (p<.001). Among girls with attention problems, blacks were 12 percent less likely than whites to receive medication (p<.05). The second and third studies examine sources of variation in inpatient care for youth with mental health and substance abuse problems using the Medicaid claims data from the Tennessee Impact Study. The second study uses a Bayesian cross-classified multilevel model to examine determinants of inpatient length of stay. Our results suggest that about 4 percent of the variation in LOS is explained at the patient-level while 42 percent is explained at the facility-level. These analyses also demonstrate that having a cross-classified data structure rather than a completely nested data structure improve the precision of our patient-level variance estimate by 84 percent. The third study uses an multilevel event history model to examine determinants of receipt of follow-up services following discharge from an inpatient stay. Our results suggest that relative to youth with mental health problems, the hazard of receiving aftercare services was 26 percent lower for youth with substance abuse problems. Relatively little (9%) of the variation in aftercare services was determined at the facility level, and 16 percent was explained by patient and family characteristics. Together, these studies demonstrate several features of multilevel modeling that have potential wide scale use in the area of health services research.