Quantile difference estimation with censoring indicators missing at random