Results from the maximum likelihood estimation of the logistic TPFs defined in expression (1) are given in Table 2. Coefficients measure the change in log-odds of transition per unit change in the independent variable. While both functions are significant below p=.0001 on a chi-square test of fit relative to a null model, our results are similar to those of other investigators in showing limited success in predicting nursing home use. With cutoffs set at the observed transition prevalence rate, sensitivity and specificity for the CN TPF are 57% and 76% respectively, while the same quantities for NC TPF are 63% and 62%. Areas under the ROC curves are .67 and .68 respectively.
In considering the control variables (the interaction terms involving the service levels), one observes first that the coefficients for the service levels themselves are of the expected sign (negative for the CN function, positive for NC), indicating that community services indeed reduce risk of admission to nursing homes and facilitate exit once admitted, thus reducing use overall. Observe that in both T`PFs, the coefficients for nursing care are considerably larger than that for the other services, indicating a much greater direct effect on nursing home use per hour of service than is the case for the other services.
Considering first the CN function, the impact of nursing hours on risk for those requiring wheelchairs (a proxy for major health impairment) is strong and statistically significant, but weaker and insignificant for others. Home health aide services are seen to significantly enhance chances of remaining in the community for those cognitively impaired, while personal care aide services significantly enhance prospects for those more severely impaired in ADL functioning. Housekeeping services were particularly effective for those impaired in IADL, as would be expected.
Turning now to the NC function, available nursing services are a strong and significant predictor of return to the community for those not in wheelchairs (bearing in mind that the nursing home sample was selected to be good prospects for discharge within 90 days), but less so for others. Personal care services significantly increase discharge prospects for those more impaired in ADL, home health aide services for those more impaired in cognition and housekeeping for those severely impaired in IADL.
As we noted previously, the effects of the exogenous covariates on nursing home use have been considered extensively elsewhere (Garber and MaCurdy, 1989; Greene and Ondrich, 1990). For sake of space, they will not be considered at length here. Suffice it to say that for those in the community, African-Americans, Hispanic-Americans, homeowners, those with greater income and with more children are at significantly lower risk of nursing home admission. Conversely, being older, living alone, being more impaired in ADL, IADL or cognition, or being in an area with a larger nursing home bed supply, all increase risk of admission.
For those in a nursing home, prospects of return to the community are significantly enhanced by homeownership, being in the Channeling demonstration treatment group at a financial control site, and reporting better health. Prospects are reduced by having lived alone at time of admission and -being more impaired in IADL functioning or cognition.