Health Outcome Algorithm Inventory

This file contains a summary of literature reviews conducted in 2012-2013 by the Mini-Sentinel Applied Surveillance Core on International Classification of Diseases, Ninth Revision (ICD-9) algorithms to identify health outcomes of interest in administrative data, focusing on algorithms that have undergone validation through medical records review. The Mini-Sentinel Applied Surveillance Core examined the available literature and consulted with experts in an attempt to locate algorithms to identify specific HOIs. This inventory was produced to assist investigators in selecting algorithms to identify health outcomes of interest in administrative data. 

It should be noted that not all algorithms identified are of sufficient quality for all purposes. Not every algorithm studied by every source was documented. Search strategies to identify algorithms were developed to maximize efficiency. Some relevant studies may have been missed.

DISCLAIMER: The FDA does not specifically endorse use of any of these algorithms, and cannot guarantee the accuracy of the information in this library. The inclusion of an algorithm in this inventory does not imply that its performance is considered adequate for use. Not all data have been cross-validated by multiple investigators, interpretations of investigators may differ, and this was not peer-reviewed in the traditional sense. Algorithms may need to be customized for some purposes, or updated since codes and definitions can change over time. The use of an algorithm in a prior protocol should not be taken to infer that the algorithm has sufficient performance to be used without validation. Expanding knowledge of Sentinel data sources may also lead to changes in which algorithms might be considered optimal from those used in prior protocols. Some protocol algorithms were also used to maximize sensitivity with anticipated validation through medical record reviews, so the algorithms may not be optimal for investigations with no anticipated outcome validation. In addition, good performance of an algorithm in a prior study does not necessarily mean that the algorithm will perform as well in a new population. The potential need for validation of outcomes should always be considered. Investigators must take responsibility for choosing and justifying algorithms used in their work, and determining whether outcome validation is necessary. Further guidance for conducting and reporting on pharmacoepidemiologic safety studies that use electronic healthcare data, including validating health outcomes, may be found within the Best Practices for Conducting and Reporting Pharmacoepidemiologic Safety Studies Using Electronic Healthcare Data guidance document provided by the FDA.