California Department of Public Health logo: three likenesses of people colored blue, green, and orange  
carbon monoxide detector

Carbon Monoxide Poisoning

Join our list

Get updates on our project activities and new features of our website. Sign up for our newsletter here.

Contact Us

California Environmental Health Tracking Program

850 Marina Bay Pkwy, P-3
Richmond, CA 94804

(510) 620-3038
E-Mail Us
Last Edited: 6/24/2014

Carbon Monoxide Poisoning: Measurement and Limitations

To understand who is at risk for carbon monoxide (CO) poisoning in California, it is important to look at who has visited the emergency department or been hospitalized for CO poisoning (morbidity), who has died from CO poisoning (mortality), and who has certain risk factors that increase their chances of experiencing CO poisoning.

In our data query, we provide data and information specific to hospitalizations due to CO poisonings. 

Learn more about CO poisoning data sources and limitations

This section contains the following topics:

CO Poisoning Classification

Cases of CO poisoning hospitalizations are identified by using the following 9th revision of the International Classification of Disease (ICD-9) codes: 

ICD-9CM Code
Code description
Toxic effect of carbon monoxide
Carbon monoxide from incomplete combustion of other domestic fuels
Carbon monoxide from other sources
Unspecified carbon monoxide
Codes used to classify records as to intent and fire-relatedness
E890.0- E899.9
Accidental injury caused by fire and flames
Accidental injury – transportation related
Accidental injury – solids, liquids gases and vapors,
Accidental injury -- Falls
Accidental injury – includes natural and environmental factors; water-related and other accidents


Cases with intentional causes (e.g. suicide, homicide) are excluded from this case definition.  Cases with unintentional causes or unknown intent are further identified as having one of three types of CO poisoning:

  1. Unintentional, fire-related: cases when the CO poisoning was due to accidental injury caused by fire and flames

  2. Unintentional,  non-fire related: cases when the CO poisoning was due to accidental injuries that were transportation related, from solids, liquids, gas, and vapors, from falls, or from natural and environmental factors

  3. Unknown intent: cases when the CO poisoning was due to an unknown cause


Back to top 


Data Sources

Hospitalization and Emergency Department Data

Since 1986, the Office of Statewide Health Planning and Development (OSHPD) has been responsible for routinely collecting data on hospital discharges from every licensed acute care hospital in California, excluding federal hospitals.  Each year, OSHPD compiles the data from all hospitals to create the Patient Discharge Database (PDD).  Since 2005, OSHPD has been routinely collecting data on emergency department (ED) visits from California hospitals.  These datasets include information like age, gender, race/ethnicity, and diagnosis.

PDD and ED data are available for hospitals to assess their level of performance.  The data are also available to some California State and Federal government programs and some academic institutions to be used for research and other public health purposes.

To find out more about what data is available from OSHPD, go to their website:

To request public use data from OSHPD, go here:


Back to top


Multiple Causes of Death Data

For California, the Center for Health Statistics (CHS) is responsible for coordinating and overseeing the collection, management, and dissemination of public health and vital statistics data in conjunction with other State agencies, local government agencies, and other customers.

The Multiple Causes of Death Files (MCOD) were created by the National Center for Health Statistics  and include underlying, immediate, intermediate, and contributing causes of death and demographic data. Each record may include up to 20 causes of death derived from California death certificates. All causes of death are coded according to the International Classification of Diseases.

To find out more about what data are available from CHS, go to their website:  

Back to top

Population Data

To calculate rates within a certain population and timeframe, one uses population data (i.e. denominator data) that represent that population during the specified period of time.  The U.S. Census is the only comprehensive source of population data and is collected every ten years.  The last census was collected in 2010 and the next collection will be in 2020.  To estimate the population during the interim (2011-2020), demographers use various algorithms to project how the population has changed.


For the CO poisoning rates presented on our CO poisoning data query system, we use population data from the California Department of Finance (DOF) For years 2000-2010: DOF uses the 2000 and 2010 population data collected by the U.S. Census Bureau and takes into account births, deaths, and migration for California to estimate the intercensal California population by county, race/ethnicity, age, and gender for 2000-2010 (State of California, Department of Finance, Race/Hispanics Population with Age and Gender Detail, 2000–2010. Sacramento, California, September 2012.).Years 2011-2012: DOF used the 2010 population data collected by the U.S. Census Bureau and takes into account survival rates, migration patterns, and fertility rates for California to project the California population by county, race/ethnicity, age, and gender for 2010-2060 (State of California, Department of Finance, Report P-3: State and County Population Projections by Race/Ethnicity, Detailed Age, and Gender, 2010-2060. Sacramento, California, January 2013.)

Rates presented by other data providers (e.g. CDC) may differ slightly because they may use population projections calculated by other methods.


Back to top


Statistical Concepts


The number of hospitalizations, ED visits, or deaths among California residents is calculated by summing the total number of events (i.e., hospital discharges for CO poisoning) for a given time period (e.g. year) and geography (e.g. county).  The hospitalization event is based on the date of discharge.

The events are per discharge or visit, and not per person, since some people can be admitted to the hospital or visit the ED more than once in a given year.

Hospitalizations due to CO poisoning are a rare event.  Thus, we aggregate the data in four-year increments and provide a four-year annual average. Deaths due to CO poisoning are also a rare event.  Thus, we aggregate all the data, 2000-2010.


Back to top



For ED visits due to CO poisoning, we calculated annual rates.  For hospitalizations due to CO poisoning, we aggregate the data into 4-year increments and calculate a rate over that four year period. For deaths due to CO poisoning, we aggregate all the data (2000-2010) and calculate a rate over the entire eleven year period.

  • Crude rates

Crude rates (i.e. unadjusted) are calculated by taking the total number of events for a given time period (e.g. 2005-2008) and geography (e.g. county) and dividing by the total underlying population for the same time period and geography.  The rates are then multiplied by 1,000,000 and expressed as X hospitalizations per 1,000,000 California residents.


  • Age-adjusted rates

Age-adjusted rates take into account the age-distribution of a population and are calculated to allow for direct comparisons between two or more populations at one point in time or between a single population at two or more points in time.  Crude rates measure the true risk for a population, while age-adjusted rates are useful as a relative index of risk.

Using the direct method of age-adjustment, crude rates are weighted to be comparable to a standard population.  For the age-adjusted hospitalization rates presented here, we use the U.S. Census 2000 population as the standard population (

Below are the steps taken to calculate the age-adjusted rates:

    1. Group the numerator (i.e. hospitalizations, a) and denominator (i.e. population, b) data into 19 age group strata (0-4 years old, 5-9 years old, …, 75-79 years old, 85 years old and over)
    2. For each stratum, divide the numerator by the denominator (a/b)
    3. For each stratum, multiply a/b by the stratum-specific weight of the standard population (a/b*w). The weight is calculated by taking the total number of people in each stratum of the standard population and dividing by the total number of people in the entire standard population
    4. Sum a/b*w across all strata and multiply by 1,000,000.The total is the age-adjusted rate per 1,000,000 California residents


Back to top


Confidence Intervals

Given the data at hand, to understand the range of possible values for the true rate, we calculate the 95% confidence interval for each rate.  Statisticians have developed a large number of methods for calculating these confidence intervals.  Usually the result is the same no matter which method is used, although when numbers of events are small, the results may differ.

The method developed by Tiwari, Clegg, and Zou (Methods in Medical Research, 2006; 15:547-569), applies specifically to the situation of age-adjusted health outcomes such as hospitalizations due to CO poisoning.  We have chosen to use this method for calculating confidence intervals for the CEHTP Portal, but it is important that users understand that others in the field may not be using the same method.


Back to top


Common Limitations with the Data

Purpose of data collection differs

Per legislation, hospitals are required to report patient data to the California State Office of Statewide Health Planning and Development (OSHPD).  The data are collected for purposes of assessing healthcare quality, services, and insurance coverage and not reported specifically for public health surveillance.  Although the data are being used for public health purposes, the limitations associated with how the data is collected must be kept in mind and when possible, accounted for.

An example of such a limitation has to do with how a patient’s principal diagnosis gets coded.  Since the diagnosis codes are recorded by hospitals for reimbursement purposes, the code might have been different if the purpose for recording was for primarily public health surveillance.  Since we identify events based solely on the principal diagnosis code, we must keep this limitation in mind.


Back to top


Low-level of resolution

At this time, hospitals are not mandated to report patient addresses.  The level of geographic resolution of the data is limited to state, county, and zip code.  Currently, the CEHTP web portal only displays county-level counts and rates, and more local-level or neighborhood-level patterns of disease currently cannot be examined.  Thus, the data are not as useful to users who might like to examine the rates of hospitalization in their immediate neighborhoods.

Zip code-level data have a higher geographic resolution than county-level data and are of more value to some users.  However, at this time, there is not a reliable zip code-level population data source available, so rates are difficult to calculate.


Back to top