Multilevel Poisson sample selection models and alternative methods for estimating hospital effects on long-term outcomes
Hospital care focuses on improving patients’ long-term quality of life, yet hospital quality metrics typically focus on short-term processes. Attempting to understand a patient’s long-term process introduces sample selection bias since patients must survive the hospitalization in order to observe post-hospitalization outcomes. As a result, proper analysis of long-term outcomes should account for clustering, due to the hierarchical structure of hospital data, as well as sample selection bias. The objective of this paper was to evaluate random effect parameter estimation and higher-level ranking of long-term count outcomes and short-term selection processes in the presence of cluster and selection bias by comparing multilevel Poisson models, multilevel zero-inflated Poisson models, and multilevel Poisson sample selection models (MPSSMs) in a series of simulations. We simulated an outcome resembling a post-discharge Poisson count with a pre-specified selection process determining a patient’s hospitalization survival with each hospital having a unique effect on both processes. In order to clarify the methodology, we also analyzed a real-world hospital dataset involving a count outcome conditioned on the selection process of hospital survival. Across all simulations, the random effect parameter estimates were directly compared and the empirical Bayes estimates were extracted, ranked, and compared using the Spearman rank correlation. Results show that the MPSSM produces more accurate random effect parameter estimates and higher-level empirical Bayes ranks. When modeling multilevel effects on long-term count outcomes observed after a short-term selection process, higher-level effects are more reliably measured using MPSSMs.
KeywordsSelection bias Multilevel Poisson models Multilevel zero-inflated Poisson models Multilevel Poisson sample selection models Empirical Bayes ranking
This work was supported by VA HSR&D IIR 11-109. The authors’ opinions do not necessarily represent the views of the US Department of Veterans Affairs or the US Government.
- Cooper, D.J., Rosenfeld, J.V., Murray, L., Arabi, Y.M., Davies, A.R., D’Urso, P., Kossmann, T., Ponsford, J., Seppelt, I., Reilly, P., Wolfe, R., DECRA Trial Investigators, Australian and New Zealand Intensive Care Society Clinical Trials Group: Decompressive craniectomy in diffuse traumatic brain injury. N. Engl. J. Med. 364(16), 1493–1502 (2011). doi: 10.1056/NEJMoa1102077 CrossRefPubMedGoogle Scholar
- Hastie, T., Tibshirani, R., Friedman, J.: The Elements of Statistical Learning—Data Mining, Inference, and Prediction, 2nd edn. Springer, New York (2009)Google Scholar
- Hollander, M., Wolfe, D.A.: Nonparametric Statistical Methods, 2nd edn. Wiley, New York (1999)Google Scholar
- Iwashyna, T.J., Odden, A., Rohde, J., Bonham, C., Kuhn, L., Malani, P., Chen, L., Flanders, S.: Identifying patients with severe sepsis using administrative claims: patient-level validation of the angus implementation of the international consensus conference definition of severe sepsis. Med. Care 52(6), e39–e43 (2014). doi: 10.1097/MLR.0b013e318268ac86 CrossRefPubMedCentralPubMedGoogle Scholar
- Mikkelsen, M.E., Christie, J.D., Lanken, P.N., Biester, R.C., Thompson, B.T., Bellamy, S.L., Localio, A.R., Demissie, E., Hopkins, R.O., Angus, D.C.: The adult respiratory distress syndrome cognitive outcomes study: long-term neuropsychological function is survivors of acute lung injury. Am. J. Respir. Crit. Care Med. 185(12), 1307–1315 (2012). doi: 10.1164/rccm.201111-2025OC CrossRefPubMedCentralPubMedGoogle Scholar
- Miranda, A., Rabe-Hesketh, S.: Maximum likelihood estimation of endogenous switching and sample selection models for binary, ordinal, and count variables. Stata J. 6(3), 285–308 (2006)Google Scholar
- Rabe-Hesketh, S., Skrondal, A., Pickles, A.: GLLAMM Manual. University of California-Berkeley Division of Biostatistics Working Paper Series. Working Paper 160. http://www.bepress.com/ucbbiostat/paper160. (2004). Accessed 2 July 2014