Bias correction for nonignorable missing counts of areal HIV new diagnosis

Public health data, such as HIV new diagnoses, are often left-censored due to confidentiality issues. Standard analysis approaches that assume censored values missing at random lead biased estimates and inferior predictions. Motivated by the Philadelphia areal counts of diagnosis for which all less than or equal 5 suppressed, we propose two methods reduce adverse influence missingness on predictions imputation diagnoses. One is likelihood-based method integrates mechanism into likelihood function, other a nonparametric algorithm matrix factorization imputation. Numerical studies data demonstrate proposed can significantly improve prediction based data. We also compare their robustness model misspecification find both appear be robust prediction, while performance depends specification.