Evaluation of Degree of Dependency After Stroke.

Study Design It is a prospective, longitudinal, multicenter and community study, with a 2-year follow-up period (from 01.01.2017 to 31.12.2018) of patients who suffered stroke in the Community of Catalonia, Terres De l'Ebre County from the population-based register through specific ICD-9 diagnostic and procedure codes.

Data collection methods Probabilistic sample: all consecutive stroke cases up to reaching the previously calculated sample size. Study will be carried out according the common clinical practice.

Primary outcome: The primary outcome was dependence occurring within the 2-year follow-up after the stroke episode. Assessment of the patients' degree of dependency is essential in determining nursing care needs, planning nursing intervention, helping increase patients' abilities, and creating proper discharge plans. The European Council [12,13] defines dependence as the state in which people, due to causes linked to the lack or loss of physical, psychological, or intellectual autonomy, are in need of assistance and/or significant help to carry out common activities of daily life. In primary care, the nurses in charge are trained of data collection. This situation needs of formal care provided by health or/and social workers, private or public.

Secondary outcomes:

1. Propose a suitable instrument with predictive power propose for identifying patients in higher risk for needing formal care from health and/or social care providers.
2. Measure the time elapsed from the hospital discharge to first contact with health primary care services, with social services, application for recognition of dependence degree, and get effective certification.
3. Know the newly diagnosed cases of dependence after stroke.

Statistical analysis All statistical tests will be performed as intention-to-treat. Prognostic factors' estimates will be adjusted by mixed-effects regression models. Possible confounding or effect-modifying factors will be taken into account. Predictions of dependence risk were based on Cox proportional-hazard regression models. Data analysis information extracted included the adjusted risk estimates and 95% confidence intervals (CI) and all statistical tests were two sided at the 5% significance level.

All potential predictors were considered in a multivariate logistic regression, and a backward step selection procedure was carried out to pick the variables that composed the best model. Subsequently, design of a predictive model of multivariate Cox regression analysis was utilized to define the weight of each of the pathologies in the dependence. To assign the weight according to the hazard ratio (HR) value, we took into account only those with a HR ≥1.2 in the multivariate model approximating the value of HR to the nearest whole number:

• HR between 1.20 and 1.49 scored a 1.
• HR between 1.50 and 2.49 was a 2.
• HR between 2.50 and 3.49 received 3, and so on. The final score for each patient will be made up of the sum of their scores. We will use ROC curves and the AUC to assess the ability of this tool to stratify patients and predict dependence. To ensure internal validity, we will perform a ten-fold cross-validated multivariate regularized logistic regression to predict dependence status based on all other variables. We will plot the receiver operating characteristic (ROC) curves and compute the area under curve (AUC) to assess the prediction power of the models. In a next phase, there will be a prospective study of validation in the cohort of patients with an episode of stroke along 2018 year.