Joint Modeling of Longitudinal and Time-to-Event Data

Book Examples

  • Scleroderma Lung Study (SLS)
    Download Data for the Scleroderma Lung Study
    The Scleroderma Lung Study, a 13-center double-blind, randomized, placebocontrolled trial sponsored by the National Institutes of Health, was designed to evaluate the effectiveness and safety of oral cyclophosphamide for one year in patients with active, symptomatic scleroderma-related interstitial lung disease (Tashkin et al., 2006). The study was initiated with 158 patients equally distributed into the two treatment groups. Patients were followed for a total of two years.The data set is used in Chapter 2 to illustrate the analysis of longitudinal measurements assuming ignorable missing data mechanisms, in Example 4.11 to illustrate a joint model that handles both intermittent and monotone non-ignorable missing data, and in Example 5.1 for a joint analysis of longitudinal data and competing risks. It is also used to show the application of robust joint models to reduce the impact of outlying %FVC measurements in Chapter 5.

  • Stroke Study: the NINDS recombinant t-PA trial
    Download Data for the Stroke Study
    The National Institute of Neurological Disorders and Stroke (NINDS) rt-PA stroke study is a randomized, double-blind trial of intravenous recombinant tissue plasminogen activator (t-PA) in patients with acute ischemic stroke (the NINDS rt-PA stroke study group, 1995). The NIH stroke data are used in Chapter 2 to illustrate standard approaches to longitudinal analysis assuming ignorable missingness. The modi_ed Rankin scale data are used in Example 5.3 to illustrate joint analysis for ordinal longitudinal data with missing values caused by competing risks. The Barthel index data are used in Example 7.5 for variable selection in joint models.

  • Enable II Study
    Download Data for the Enable II Study
    The ENABLE II study is a randomized clinical trial comparing a nurse-led, phone-based palliative care to the usual care in 322 advanced cancer patients (N = 161 in each group). Example 4.7, has been developed to analyze the data.

  • Milk Protein Trial
    Download Data for the Milk Protein Trial
    In this study, a total of 79 cows were randomized into three diet groups, barley (N = 25), mixed barley-lupins (N = 27), and lupins (N = 27). This data set is analyzed in Example 4.8 to examine the impact of missing data on the statistical inference of diet e_ect, and in Example 7.2 to illustrate the use of an index to measure the overall sensitivity of milk protein estimation in the neighborhood of a model that assumes ignorable missingness.

  • ACTG Study
    Download Data for the ACTG Study
    AIDS Clinical Trials Group (ACTG) Protocol 175 is a randomized trial conducted on 2467 HIV-infected patients to compare four therapies: zidovudine alone, zidovudine + didanosine, zidovudine + zalcitabine, and didanosine alone. This study is used in Example 4.12 to illustrate a joint model in which the association between CD4 counts and time to progression or death is more appropriately characterized.

  • Medfly Fecundity Data
    Download Data for the Medfly Fecundity Study
    The joint model is applied to a dataset of female Mediterranean fruit flies (medflies) on whom daily egg production was recorded until death. In the illustration given in Example 4.14, a joint analysis is used to characterize the number of eggs laid each day for a subset of most fertile medes. Additionally, Example 4.14 uses an arti_cially created incomplete dataset of the same medies population to illustrate the performance of the joint model when there are irregularly spaced, incomplete longitudinal measurements and censored survival times.

  • Bladder Cancer Study
    Download Data for the Bladder Cancer Study
    The bladder cancer study included 85 patients with superficial bladder tumors. At the study entry, the tumors were removed transurethrally. This data set is used in Example 4.16 to illustrate a joint model for longitudinal outcomes with informative observation times.

  • Renal Graft Failure Study
    Download Data for the Renal Graft Failure Study
    In this study, 407 patients with chronic kidney disease received renal transplantations in the hospital of the Catholic University of Leuven (Belgium) between 1983 and 2000, and were followed until graft failure or being censored. This data set is used in Example 6.1 to illustrate a joint model for multivariate longitudinal outcomes and survival data.

  • PAQUID Study
    Download Data for the PAQUID Study
    This is a prospective cohort study initiated in 1988 to evaluate normal versus pathological aging among subjects 65 and older in France. This study is used in Example 6.2 to illustrate a latent-class joint model that predicts dementia using the three cognitive measures.

  • Rat Data
    Download Data for the Rat Study
    Verdonck et al. (1998) conducted a randomized study on male rats to evaluate the craniofacial growth after inhibiting endogenous testosterone production by cecapeptyl. This data set is used in Example 7.1 to illustrate a local inuence approach that examines the sensitivity of parameter estimation in a neighborhood of the ignorable missing data assumption.

  • AIDS Clinical Trial
    Download Data for the AIDS Clinical Trial
    This is a multicenter, randomized, open-label, community-based clinical trial comparing the ecacy and safety of two antiretroviral drugs, didanosine (ddI) and zalcitabine (ddC). As we have discussed in other examples, missing data due to death could lead to biased estimates of CD4 trajectory. Furthermore, when the interest is to select important predictors for CD4 counts, the estimation bias due to missing data would compromise the validity of commonly used variable selection procedures. To address this issue, a variable selection method has been developed within the framework of joint models and the approach is illustrated in Example 7.4.