Explore chapters and articles related to this topic
Neural Network Survival Analysis
Published in Prabhanjan Narayanachar Tattar, H. J. Vaman, Survival Analysis, 2022
Prabhanjan Narayanachar Tattar, H. J. Vaman
We have seen the working of a neural network in previous sections. The technique has been developed with the intent of determining weights such that predicted values are as close as possible to the actuals. Note however that the purpose of determining the weights is not to identify the significant/causal covariates and this gives an impression that a neural network is a black box. This is even more so because the multiple layers mix up the impact of all the variables and once the mixing happens, the original impact of the variables looses traceability. In brief, NN as a methodology is concerned about prediction problems and not impact of specific covariates. Most of the applications in clinical trials studies relate to survival analysis. Ripley and Ripley (2007)[95] note the following attempts to apply the (classification) NN for survival data: Consider survival time up to a fixed time t as an indicator variable. Ignore all observations which were censored before time t and setup the classification NN.Bin the survival times into one of the k-time intervals and use the NN with multiple outputs.Create k separate NNs for the binning done in the previous step.
Structural Equation Modeling with Longitudinal Data
Published in Douglas D. Gunzler, Adam T. Perzynski, Adam C. Carle, Structural Equation Modeling for Health and Medicine, 2021
Douglas D. Gunzler, Adam T. Perzynski, Adam C. Carle
Censoring is a characteristic that distinguishes survival analysis from other types of statistical analysis. Censoring occurs when incomplete information is available about the survival outcome for some of the individuals in a study. For example, one is examining a survival outcome over a pre-specified time interval using a large electronic health registry from a single institution. Typical reasons for censoring may include an individual may (1) not have experienced the event within the time interval, (2) have moved and be lost to follow-up, or (3) have received care at a different institution for the event. Some individuals may be observed over a shorter time interval than other individuals (depending on when data for an individual is first recorded in the registry), giving less of a time frame for the event to occur.
Prognosis: Studies of disease course and outcomes
Published in Milos Jenicek, Foundations of Evidence-Based Medicine, 2019
As already mentioned in Chapter 6, one of the most important aspects of a good diagnosis is its prognostic value. If a diagnosis does not indicate whether a patient will do better or worse than other individuals, appropriate therapeutic decisions cannot be made. Prognostic studies and their results give meaning to diagnostic methods. Matthew et al.55 summarize their critical review of biostatistical methodology in the domain of prognosis as follows: Survival analysis should be carried-out by the Kaplan– Meier method;Median survival time should be reported;Confidence intervals should be provided as a measure of variability;A log-rank test should be used to compare two or more survival curves;Stratified analysis and the Cox model should be used to define the impact of multiple prognostic factors on survival.
Budget impact analysis of a bovine pericardial aortic bioprosthesis versus mechanical aortic valve replacement in adult patients with aortic stenosis in Romania
Published in Journal of Medical Economics, 2023
João L. Carapinha, Vlad A. Iliescu, Lucian Florin Dorobantu, Adina Turcu-Stiolica, Jens Deckert, Andrea White, Adham Salem, Catalina Parasca
Partitioned survival analysis is associated with several limitations that can potentially impact its accuracy. Primarily, partitioned survival analysis operates under the assumption that the survival functions it models are independent, an assumption that may not hold in real-world scenarios. For instance, in situations where mortality rates differ between BPAB and MV patients, this model’s inability to accommodate such variations could lead to inaccuracies. Further, these models struggle to handle complex dependencies, particularly when projecting beyond the known data period the varying risks of thromboembolic events and disabling strokes among different patients. The model relies on published research data and observational studies, which when extrapolated, might not accurately reflect long-term trends and could introduce biases. Notwithstanding these limitations, we decided to model the budget impact of BPAB and MV over a five-year period to alleviate the long-term limitations associated with partitioned survival analysis, especially in cost-effectiveness analyses.
Bacterial and fungal infections: a frequent and deadly complication among critically ill acute liver failure patients
Published in Infectious Diseases, 2023
Félicie Belicard, Kieran Pinceaux, Estelle Le Pabic, Valentin Coirier, Flora Delamaire, Benoît Painvin, Mathieu Lesouhaitier, Adel Maamar, Pauline Guillot, Quentin Quelven, Pauline Houssel, Karim Boudjema, Florian Reizine, Christophe Camus
Logistic regression was used to identify risk factors for infections. Log linearity of continuous variables was tested and, if necessary, variables were categorised according to clinically relevant thresholds or at median values. Non-collinear variables, that reached p values smaller than 0.2 by univariate analysis, were considered for the multivariate model. Then, variables were selected using a descending step-by-step procedure to keep in the final model only variables associated with a p value < .05. To handle missing values as potential confounders, multiple imputation with chained equation was used. To assess the effect of infection on 28-day survival, as previously suggested [28,29], age and paracetamol aetiology were included in a multivariable Cox model. Survival rates were reported using the Kaplan–Meier method and compared by the log-rank test. Results of logistic regression analysis and survival analysis were expressed as odds ratio (OR) and hazard ratio (HR), respectively, with their 95% confidence intervals (95% CI).
On maximum likelihood estimation of the semi-parametric Cox model with time-varying covariates
Published in Journal of Applied Statistics, 2020
Survival analysis involves following subjects for an observation period in anticipation an event of interest will occur and modeling the time to this event of interest by a statistical model. If the event does not occur within this observation period, the time to event for this subject is right censored. The Cox model [9] is the corner stone of modern survival analysis allowing the natural logarithm of the hazard ratio to be a linear function of covariates. It has been applied in areas as diverse as biomedical science, industrial life testing [21] and finance [24]. Crowley and Hu [12] extend the model to include time-varying covariates whose values change for a subject while they are in the study. These time-varying covariates can also be used to relax the inherited proportional hazards assumption of the Cox model with time-fixed covariates [11].