1989). Certain datasets are not coded by either ICD-9 or ICD-10 criteria, such as the TESS poison control center data system (see Chapter 7). Within the ICD coding scheme, provision has been made to differentiate among different types of poisoning and drug overdose events and among different categories of intent. There is no defined grouping of ICD codes that establishes a single category subsuming all poisoning events.
Some of the problems described may be magnified in the ICD-10, in which injury coding (including for poisoning) has changed substantially. Specifically, in ICD-10 it may not be possible for “intent” codes to separate out ethanol-related toxicity from toxic syndromes caused by alcohol substitutes such as methanol. Added instructions for coding deaths further impact definitions by requiring the principal cause of death to be categorized as due to a selected group of mental and behavioral disorders if such a disorder appears among contributing causes in a poisoning death. Thus an acute acetaminophen fatality in a chronic ethanol abuser (if this was listed as a contributing cause) would be coded in ICD-10 as a primary alcohol-related death (World Health Organization, 1992–1994). CDC recently added a series of special “U” codes (allowed for in the ICD-10 scheme) to capture terrorism-related fatalities. Some of these new codes also could be relevant to poisoning, such as U01.7 for terrorism involving chemical weapons (http://www.cdc.gov/nchs).
As a precursor to pooling data for multiple years for each source, the proportion of poisoning observations for each year was examined to ensure that poisoning estimates were relatively stable during the period. Each source was analyzed separately, but demographic variable recoding was done on each to create consistent categorical variables for gender, age, race, and region. Estimates associated with relative standard errors (ratio of the standard error to its estimate) >.3 or based on a small sample size were retained, but are noted. Sampling weights included in the data files were rescaled to an annual timeframe and used to create population estimates. SUDAAN, the standard computerized statistical package that calculates estimated rates taking into account the sampling weights built into the design of each survey, was used whenever possible to adjust for the multistage sampling design of the surveys.