Abstract
Rationale, aims and objectives. Some objective indicators like symptoms, toxicity, performance status, rate of hospitalization or re-employment have been already employed in scientific literature as proxies of Quality of Life assessment, and, in spite of the intrinsic limitations of such a measurement, they represent a valuable source of information in all the situations where a formal assessment is impossible, due to budget, time or human resources constrains. We concentrate here on some models for the analysis of frequency of hospitalization data and we discuss an application to the Hearth Muscle Disease Study Group data.
Methods. A sample of 235 patients with dilated cardiomyopathy (DCM) prospectively treated at the Department of Cardiology (Trieste, Italy) have been observed during a period of 18 years,
from 1978 to 1992 and data regarding hospitalization history were collected. The hospitalization process depends on the time since the last event, and usually is a function of a set of explanatory
variables, such as the current state of the patient, treatments he has been receiving and the severity of disease. We propose here a semi-Markov representation of the hospitalization process, and we analyze data regarding DCM, implementing Exponential, Weibull, and Cox models; in Cox models we take care also of the stratification according to the duration or to the levels of the state factor.
Results. The probability of experiencing a second hospitalization within one year after the first one is estimated about 0.50, and within two years about 0.30. After this point the probability
remains constant at a 0.10 level. The same pattern is observed for the second hospitalization, while things are getting worse after the third hospitalization, when the probability of not having a subsequent hospitalization is about 0.10 within one year. Betablockers have a strong influence in enlarging the time interval spent between an hospitalization and the other.
Conclusions. The hospitalization process can be viewed only as a rough approximation of the good standing of the patient. However, for diseases like DCM can be reasonable, because of
the relatively fast increment in the worsening conditions of the patients and the consequently high chances of observing new hospitalizations up to the absorbing state (the death). Moreover a
very detailed modeling of the process leads to extract as much information as possible from the data.