Assessing hidden bias in the estimation of causal effect in longitudinal data by using a matching estimator with Rosenbaum bounds