of the reporting poison control center, inadequate data are provided to demonstrate cause and effect. This is because the case information is secondary from hospital or other sources and the center rarely has access to complete case data. TESS data are therefore most useful as a signal along with other data sources, but cannot be utilized for policy making. This problem is exacerbated under HIPAA regulations, which decrease information flow to the poison control center once the patient has been admitted to a health care facility. In addition, the characteristics of the fatal cases differ proportionally for variables such as intent, type of poisoning, and demographics. Although one source of underreporting by TESS is attributable to out-of-hospital deaths, there is also substantial underreporting for fatal hospitalized cases. Despite the smaller numbers, there are also fatal cases detected through TESS surveillance that are not detected by death records. Comparison between death certificate data and medical examiner data has shown that these sources do not wholly overlap (Landen et al., 2003; Linakis and Frederick, 1993; Soslow and Wolf, 1992). A single study of fatal cases in the MedWatch system also found poor overlap with death certificate data (Chyka, 2000).
Direct review of hospital charts has demonstrated that only about 20 to 30 percent of poisoning cases managed in the emergency department are reported to poison control centers (Blanc et al., 1993b; Harchelroad et al., 1990; Hoyt et al., 1999). Successful case detection of medically treated cases by the DAWN system appears to be in a similar range (Roberts, 1996). In contrast with these patterns, surveillance based on the National Health Interview Survey yielded lower population estimates for poisoning incidence than those derived from TESS data (Polivka et al., 2002). This analysis was limited to pediatric cases ages 5 and younger.
Linkages among the various datasets are limited. The United States does not have a universal identification number that is used in medical records and surveys allowing for interlinking of disparate datasets. Although such linkages are desirable, no such identification system is likely to be developed and applied in the foreseeable future.
The available data suggest that no single surveillance source can provide a universal data source from which to draw a complete picture of all aspects of poisoning and drug overdose morbidity and mortality. The strengths and limitations of each source should be taken into account in interpreting surveillance data.