Project Details
Projekt Print View

Non-parametric and semi-parametric inference for a longitudinal multi-state model with an application to migraine research

Subject Area Epidemiology and Medical Biometry/Statistics
Term from 2016 to 2020
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 316904537
 
Final Report Year 2020

Final Report Abstract

Migraine and tension-type headache pathologies provide recurrent data examples of great interest in medical research studies. The challenging issue in studying these diseases is to recognize the main trigger factors and the premonitory symptoms that increases or decreases the risk to carry the pathology, or to identify which is the most effective combination of acute medications. In particular, the latter consists of analyzing whether a specific treatment may reduce the probability of disease or may increase the probability of being in disease-free state. Aim of this proposal is to address the above mentioned issues by means innovative statistical approaches based on causal analysis and recurrent multi-state patterns analysis. Our first study was to assess the prevalence of neck discomfort and sensitivity to sensory stimuli as well as the use of acute medication on migraine-free days, in presence of timedependent confounding and to analyze the impact of these symptoms on the risk of a forthcoming migraine attack and to investigate whether neck discomfort and sensitivity to lights, noises, or odors could be indicators for taking acute medication early during the premonitory phase of a migraine attack. Time-varying confounding affected by past treatment is a common issue in longitudinal migraine data, but is often ignored or dealt with conventional advanced statistical methods such as Cox regression, random effects model, generalized estimation equation which provide biased estimator in this specific scenario, as they are inadequate in controlling time varying confounding affected by past exposure. The marginal structural model represents an innovative and appealing approach which controls for time-dependent confounding and well mimics the ideal situation given by randomized controlled trials. Indeed this approach corrects for the non-random assignment of acute medication by up-weighting patients whose treatment and covariates histories are underrepresented compared to what would have been observed if treatment had been randomized. Our second study focuses on the impact of triptans (with respect to other compounds) on migraine relief, accounting for potential time-varying confounders, i.e., time-dependent covariates that are risk factors for migraine and also predicts subsequent use of acute medication. We use the inverse-probability-of-treatment weighting, building a pseudo-population where treatment is un-confounded with the covariates. Two model specifications based on two different time interpretations for the marginal structural models and the Cox regression models are considered and compared. Our last study based on the recurrent multi-state patterns analysis relies on estimating non-Markov transition probabilities, non-Markov state occupation probabilities and transition hazards within a multi-state model. This is a strategic approach which properly takes into account for the recurrent events and the correlation of each event within the individuals' patterns history. The estimation of the related measures of interest and corresponding estimated variability of the same measures through confidence bands are investigated. Observed covariates are also included through the use of an “ad-hoc” procedure, allowing for dependent observations and based on pseudo-observation value approach with an extension of generalized linear regression methods. These, although being key-strategies, are absent in migraine studies and still rarely reported in the literature of medical studies, since most of them assume the independence of the individuals' events within the history. Note that our present research work is based on population-based analysis; however it is important to underline that among subjects there is a high level of variability, that means that each individual has his or her own trigger factors and premonitory symptoms phase that act and impact of the onset and on the duration of headache. This motivates further investigations on individual-based analysis.

Publications

  • “Indicators for acute medication intake-results from a diary-based cohort study with migraineuers”, Cephalalgia Reports 3, 2020
    Di Termini S., Wöber C., Brannath W.
    (See online at https://doi.org/10.1177%2F2515816320944928)
 
 

Additional Information

Textvergrößerung und Kontrastanpassung