SAMMEC - Methodology

The Methodology page describes the methods and data sources used to estimate the health and economic consequences of smoking. For more detailed information, review the original source material in References.

Estimating the Health Consequences of Smoking

The Adult and MCH SAMMEC programs derive smoking-attributable mortality (SAM) using an attributable-fraction formula (Lilienfeld and Lilienfeld, 1980) originally described by Levin (1953). The smoking-attributable fractions (SAFs) of deaths for 19 diseases where cigarette smoking is a cause are calculated using sex-specific smoking prevalence and relative risk (RR) of death data for current and former smokers aged 35 and older. Infant mortality SAFs are calculated using estimates of maternal smoking prevalence and RR of death for four perinatal conditions caused by smoking. Refer to the International Classification of Disease (ICD) Table for a complete list of diseases included in the Adult and MCH SAMMEC programs. SAFs for each disease and sex are derived from the following formula:

Measure	Adult SAMMEC	MCH SAMMEC
p0	Percentage of adult never smokers in study group	Percentage of maternal nonsmokers in study group
p1	Percentage of adult current smokers in study group	Percentage of maternal smokers in study group
p2	Percentage of adult former smokers in study group	Not applicable
RR1	Relative risk of death for adult current smokers relative to adult never smokers	Relative risk of death for infants of maternal smokers relative to infants of maternal nonsmokers
RR2	Relative risk of death for adult former smokers relative to adult never smokers	Not applicable
Adult: 35-64 years of age and 65+ years of age

Note: Because MCH SAMMEC calculates the impact of maternal smoking, data for former smokers are not included in the SAF.

Refer to the Glossary of Terms for current smoker, former smoker, never smoker, and other SAMMEC-related definitions.

The prevalence data used by Adult SAMMEC for smokers aged 35�64 years and 65 years and older comes from two different sources. National smoking prevalence estimates were obtained from the National Health Interview Survey (NHIS), and state-specific prevalence estimates were obtained from the Behavioral Risk Factor Surveillance System (BRFSS). Users may also employ custom-level smoking prevalence data from other sources, such as the Tobacco Use Supplement (TUS) of the Current Population Survey (CPS).

National maternal smoking estimates used by MCH SAMMEC for 1999 to 2004 were obtained from Vital Statistics from the National Center for Health Statistics (NCHS), with some exceptions. California does not report smoking during pregnancy on the birth certificate, and hence, prevalence estimates were derived from the California Maternal and Infant Health Assessment (MIHA) for the entire time frame. For 1999, Indiana, New York State, and South Dakota maternal smoking estimates were obtained from the Behavioral Risk Factor Surveillance System (BRFSS).

In 2003, the birth certificate was revised, and new maternal smoking questions were asked. The states� adoption of the new birth certificate has not yet been universal. In 2003, Pennsylvania and Washington implemented the revised birth certificate; and in 2004, Florida, Idaho, Kentucky, New York State (excluding New York City), South Carolina, and Tennessee implemented the revised birth certificate. These states� maternal smoking measures are not comparable to data from the previous birth certificate and are excluded from the national estimate of the prevalence of maternal smoking in SAMMEC. However, they are used in deriving state SAMs.

Adult SAMMEC uses unpublished age-adjusted RR estimates for persons aged 35 years and older from the second wave of the American Cancer Society's Cancer Prevention Study (CPS-II), 6-year follow-up (Thun et al., 1997). Separate RR data are used for smokers aged 35�64 years and those aged 65 years and older for ischemic (coronary) heart disease and cerebrovascular disease. The RR of death from smoking drops dramatically after age 65 for these two conditions.

MCH SAMMEC uses unpublished RR estimates for short gestation/low birth weight, sudden infant death syndrome (SIDS), respiratory distress (syndrome) - newborn (RDS), and other respiratory conditions among newborns obtained from a meta analysis of epidemiological studies ( Gavin et al. 2001).

To estimate SAM, Adult SAMMEC multiplies the age- and sex-specific SAFs by the number of deaths for each smoking-related disease. Specifically, the numbers of deaths by sex and 5-year age categories are multiplied by the SAF.

Summing across age categories provides the sex-specific estimate of SAM for each disease. Total SAM is the sum of the sex-specific SAM estimates. National and state-level mortality data were obtained from the National Center for Health Statistics (NCHS).

The national smoking-attributable mortality (SAM) estimates may differ from the previously published estimates in two ways. First, SAMMEC uses updated data and second, fire deaths caused by smoking and second-hand smoke deaths are not reflected in the SAMMEC smoking-attributable mortality estimates.

MCH SAMMEC applies the SAF to infant mortality data derived from CDC's WONDER-linked infant birth and death records maintained on the Web at http://wonder.cdc.gov. Both Adult and MCH SAMMEC use the formula shown above.

The SAM estimates for each age category, stratified by sex and grouped by underlying disease category, are multiplied by the remaining life expectancy (RLE) (http://www.cdc.gov/nchs/fastats/lifeexpec.htm) of people at the midpoint of each age range, and the resulting numbers for all age categories are summed to obtain YPLL by sex (CDC, 1986). The total YPLL is the sum of the male and female YPLL within each disease category.

Estimating the Health-Related Economic Consequences of Smoking

The estimated economic impact of smoking is based on smoking-attributable health care expenditures and human capital (productivity losses). Adult SAMMEC allows users to estimate adult smoking-attributable expenditures (SAEs) and productivity losses associated with smoking-related premature deaths for each state and the nation. Adult SAMMEC does not include morbidity-related productivity losses.

MCH SAMMEC allows users to estimate smoking-attributable infant health care costs at delivery for a range of population characteristics. Morbidity-related productivity losses are also not included in MCH SAMMEC.

The SAMMEC application can be used to estimate the health care costs of smoking. Smoking-attributable health care expenditures (SAEs) are the excess personal health care costs of smokers and former smokers compared with those of never smokers. The health care expenditure components of the Adult and MCH SAMMEC programs are based on similar econometric methods. However, the programs have different estimation capabilities. Adult SAMMEC allows users to estimate state-level expenditures, while MCH SAMMEC allows users to calculate neonatal expenditures by state and by several demographic characteristics of mothers within each state.

The Adult SAMMEC application includes pre-calculated national and state-level expenditure data for adults aged 18 years and older. The pre-calculated estimates include:

Total personal health care expenditures were obtained from the state health care expenditure files provided by the Centers for Medicare and Medicaid Services (CMS), and are available at http://www.cms.hhs.gov/NationalHealthExpendData/. Adult SAMMEC provides estimates for each of five expenditure categories: ambulatory care, hospital care, prescription drugs, nursing home care, and other care (including home health, nonprescription drugs, and nondurable medical products). Expenditures for health care, dental care, and vision care products are excluded from the totals.

The expenditure smoking-attributable fractions (SAFs) denote the proportion of annual personal health care expenditures that could be avoided if smoking were eliminated from the population. Adult SAMMEC uses expenditure SAFs from Miller et al. (1999) for each of the five expenditure categories.

Miller et al. calculated expenditure SAFs of expenditures for ambulatory care, hospital care, prescription drug, and other care (including home health care, vision care, and durable and nondurable medical equipment) by using models that alternatively included and excluded the influence of smoking history on expenditures. Expenditures were estimated by using a two-step econometric model specified by Duan et al. (1983) to account for the large proportion of individuals who have no medical expenditures in any given year. First, the probability of a person having positive expenditures for each category was estimated on the basis of that person's smoking history, demographic characteristics, other risk behaviors, and other variables. Second, given that expenditures were positive, the levels of expenditures for each category were estimated.

Two sets of estimates were used to calculate the SAF for each expenditure category. The first set of estimates were for all individuals, including smokers. The second set of estimates were calculated after setting the smoking history variables to zero and holding all other factors constant. This generated expenditure estimates as if smoking were eliminated from the study population. SAFs were derived by dividing the difference in the expenditure estimates by the estimates that included smoking history.

Models for each expenditure category were applied to data from the 1987 National Medical Expenditure Survey (NMES). The national expenditure estimates were translated to state-specific estimates using 1992�1993 survey data from the TUS of the Current Population Survey (CPS), 1993 income and insurance data from the March CPS, and 1993 BRFSS data (1994 for Wyoming).

Miller et al. (1999) estimated nursing home expenditure SAFs using the nursing home component of the NMES. Smoking-attributable expenditures were derived from a preliminary model that estimated the probability of people being admitted to a nursing home, given their smoking history. Because of data limitations, multiple admissions and length of stay were not considered.

Smoking-attributable fractions (SAFs) of expenditures for 2004 were calculated using the Miller et. al (1999) methodology. National Health Interview Survey (NHIS) data (2000-2004) were used to estimate the SAFs for ambulatory care, prescription drugs, hospitalization, other, and total expenditures. NHIS (2000-2004) does not have data on nursing home expenditures, therefore; we used the total SAF of expenditures to represent nursing home portion of SAF.

For the MCH SAMMEC application, statistical analysis was used to estimate the relationship between a woman smoking during pregnancy and the probability that her infant will be admitted to a neonatal intensive care unit (NICU) and the infant's length of stay (Adams et al., 2002). Each equation controls for factors that could influence resource utilization, including mother�s age, race, education, region, marital status, insurance coverage, number of children, receipt of prenatal care, and maternal drinking during pregnancy. The analysis was based on pooled 2001/2002 data from the Pregnancy Risk Assessment Monitoring System (PRAMS), an ongoing surveillance system of the CDC and participating states (Colley-Gilbert et al., 1999). The analysis indicated that while maternal smoking was not significantly associated with the probability of infant admission to an NICU it was associated with a longer length of stay among those infants admitted to an NICU. Additional analyses indicated positive associations of smoking with NICU admission for some subgroups--those very preterm (greater than 20 but less than 32 weeks gestation) and born to moderate (20 to 29 cigarettes a day) or heavy (30 to 40 cigarettes a day) smokers.

Although PRAMS contains data on maternal demographics, risk factors, and health care service use, it does not contain data on health care expenditures. To convert the smoking-attributable use of hospital resources by infants into dollar expenditures, MCH SAMMEC uses cost estimates for all inpatient services (accommodations, physician, ancillary and pharmaceutical services) provided to infants during their initial hospital stay. The source for these estimates was private sector claims data provided by the Medstat MarketScan� database for 2004. The estimated nightly costs for infants not admitted to an NICU was $1,222 and for infants who were admitted to an NICU was $3,223.

The MarketScan� data showed that infants admitted to an NICU only spent 61% of the total stay in the NICU but their nightly costs were higher than infants never admitted to an NICU even when in a regular nursery bed. The $3,223 noted above reflects a weighted average of the NICU and non-NICU expenses per night for infants with an NICU admission (0.39 X $2,482 + 0.61 X $3,696 = $3,223).

Once these dollar values were assigned to each infant in the PRAMS sample, statistical models were used to predict total expenditures for infants at delivery for each of the population subgroups identified in MCH SAMMEC, including both mothers who smoke and mothers who do not smoke. The models were then used to predict neonatal expenditures for infants of mothers who smoke in two different ways. In the first estimate, the smoking variable was set to "1" if the mother was a smoker, and in the second estimate, the smoking variable was set to "0" as if the mother was a nonsmoker. The difference between these two estimates reflect the expected expenditures of infants exposed to maternal smoking versus those expected for these infants if they were not exposed. This difference constitutes estimated "smoking-attributable neonatal expenditures".

The difference between these two estimates reflect the expected neonatal expenses of infants exposed in utero versus those expected for infants if they were not exposed. This difference constitutes estimated "smoking-attributable neonatal expenditures".

To derive the SAF for neonatal expenditures, MCH SAMMEC used 2003 birth certificate data from NCHS. The birth certificate data included demographic information such as, age, education, insurance, marital status, prenatal care, and race, as well as maternal smoking prevalence for each population subgroup within 49 states and the District of Columbia. The maternal smoking measure for two of these states, Pennsylvania and Washington, were based on the 2003 revised birth certificate question on smoking. The maternal smoking measure for California was obtained from that state�s weighted 2003 Maternal and Infant Health Assessment (MIHA) data.

The SAE per maternal smoker was calculated for each population subgroup by dividing their smoking attributable expenditures by the number of maternal smokers in the group. This measure is an estimate of the infant healthcare costs at delivery averted if the mother quit smoking during pregnancy. Users can compare this estimate to the cost of a smoking cessation intervention per patient.

Smoking-attributable productivity losses are defined as the present value of foregone future earnings (PVFE) from paid labor and of foregone future imputed earnings from unpaid household work.

Adult SAMMEC uses age-specific PVFE estimates for 2000 from Haddix et al. (2003), which they estimated assuming a 1% productivity growth rate and a 3% discount rate. Sex-weighted estimates were used to eliminate the impact of gender bias in compensation and occupational attainment. The 2000 data were updated to 2004 data using employment cost index estimates for total compensation from the Bureau of Labor Statistics. For each smoking-related disease, SAM by sex and 5-year age category is multiplied by the PVFE.

The sex-specific estimate of smoking-attributable productivity losses for each disease category is then determined by adding the costs for all age categories and the total smoking-attributable productivity losses is determined by adding the sex-specific productivity loss estimates.

Rate Calculations

SAMMEC calculates age-adjusted SAM rates and YPLL rates for persons aged 35 years and older. These rates were standardized to the age distribution of the United States population in 2000, typically referred to as the Year 2000 Standard Population (see Klein and Schoenborn, 2001).

Crude rates defined as the total number (e.g., of deaths) divided by the population are often used to express risk. Since risk varies by demographic characteristics, such as age, populations with older age distributions will tend to have higher crude death rates. As a result, when one compares crude rates across groups (e.g., states) one cannot be sure whether apparent differences are due to differences in the age distribution of the groups or to actual differences in mortality risk. Age-adjustment is a statistical technique designed to reduce differences in crude rates that result from differences in populations' age distribution.

We estimated age-adjusted SAM and YPLL rates using a method suggested through correspondence with RN Anderson (see Anderson and Rosenberg, 1998). First, the crude SAM and YPLL rates by sex and 5-year age category were calculated. Second, the standard weights were calculated by dividing the standard population estimates by sex and 5-year age category by the total standard population aged 35 and older. Third, the age-specific and sex-specific crude rates were multiplied by the standard weights and summed over age for males and then for females to get age-adjusted SAM and YPLL rates by sex. Both the total SAM and YPLL rates (for both sexes) were obtained by multiplying the age-specific crude death rates by the standard weights and summing over age. The age-adjusted rates are expressed per 100,000 population.

Five-Year Reports

Previous CDC publications including the most recent Surgeon General�s Report (The Health Consequences of Smoking: A Report of the Surgeon General, USDHHS, 2004) have presented SAMMEC estimates for the period 1995 through 1999. In addition to updating estimates of smoking-attributable mortality (SAM), years of potential life lost (YPLL), and lifetime future productivity losses for the period 1997 through 2004, we have added SAM and YPLL age-adjusted rate reports for this period. To discourage the use of SAMMEC as a surveillance system, we recommend that users employ the five-year average estimates in their work.

Adult and MCH SAMMEC calculate the number of deaths for 23 adult and infant disease categories that contain diagnoses that are causally related to smoking (USDHHS, 2004). These diseases are categorized according to diagnosis codes found in the 10th revision of the International Classification of Disease (ICD 9-10).

The mortality data were first categorized according to ICD-10 codes. Prior to 1999, SAMMEC estimates were derived from mortality data categorized according to ICD-9 codes. Because of changes in the classification of deaths from ICD-9 to ICD-10, death counts for 1999 and later are not directly comparable with death counts in previous years. Users planning to compare 1999 or later estimates with those generated in previous years must adjust the number of deaths data using comparability ratios published by the National Center for Health Statistics. If estimating results for 1998 or earlier, users should use the comparability ratios to adjust death counts before entering the number of deaths data into SAMMEC.

Other Smoking-Attributable Health Consequences

Published national SAMMEC estimates have typically included smoking-attributable deaths from residential fires and lung cancer deaths attributable to exposure to secondhand smoke (CDC, 2005). These estimates were not calculated within the SAMMEC application but were obtained directly from other sources.

Smoking-attributable burn deaths were obtained from the National Fire Protection Association (Hall, 2007). The National Fire Protection Association (NFPA) publishes estimates of the average annual number of civilian deaths attributed to smoking-material fires in the United States. These estimates are based on information reported to U.S. municipal fire departments and on information obtained from NFPA and National Fire Incidence Reporting System (NFIRS) surveys. In 2005, estimated 800 civilian deaths were attributed to smoking-material fires. The long-term trend on civilian deaths attributed to smoking material has been down by 75% from 1980 to 2005 (Hall, 2007). The estimates exclude non-residential and vehicle fire deaths attributable to smoking, approximately 3 to 4% of all smoking-attributable fire deaths. Exposure to secondhand smoke has been estimated to cause about 3,400 lung cancer deaths and between 22,700 to 69,600 heart disease deaths in the United States each year (CDC, 2006).

Deaths attributed to smoking-related residential fires and exposure to secondhand smoke may be added to state SAM estimates, in accordance with the proportion of the U.S. population that resides in the state.

Limitations

The methodology used to calculate smoking-attributable deaths has some limitations as discussed below.

The attributable-fraction methodology calculates smoking-attributable deaths using smoking prevalence and number of deaths for the current year. However, most smoking-attributable deaths are the result of smoking in previous decades, during which smoking rates were higher. During periods where smoking prevalence is declining, the attributable-fraction (AF) methodology will tend to understate the number of deaths caused by smoking. Conversely, when smoking prevalence is increasing, the AF formula may overstate the number of deaths caused by smoking.

Adult SAMMEC uses smoking prevalence estimates and relative risk estimates to calculate SAFs for each cause of death for adults aged 35�64 years and for those aged 65 years and older. The mortality and smoking prevalence data are available by age and may be readily summarized by 5-year age categories. However, for most diseases-specific deaths, age-specific RRs for smoking are unavailable. Therefore, summary estimates are calculated for all ages combined or for two age categories (35-64 years, and = 65 years) and not 5-year categories of age.

For most diseases, the mortality RRs associated with smoking remain stable across age categories. For such diseases, applying a single SAF estimate to all age categories is not problematic. In contrast, the RR for death from ischemic heart disease (IHD) � also referred to as coronary heart disease (CHD) � and cerebrovascular disease (CVD) decline substantially after the age of 65 (USDHHS, 1989; Thun et al., 1997; Thun et al., 2000).

Current smoking prevalence rates decline with age and the excess risks for IHD and CVD decline rapidly after smoking cessation. These findings of decreased RR and prevalence estimates indicate that the SAF also declines with age. In contrast, the rate of death from IHD and CVD increases with age, and about three-fourths of deaths from these conditions occur among people older than 65. So, while the number of deaths from CHD and CVD increases rapidly after the age of 65, the proportion of these deaths attributable to smoking drops markedly within this age bracket. The calculations performed within Adult SAMMEC account for differences in age category-specific risks for IHD and CVD. For IHD and CVD, Adult SAMMEC calculates separate SAFs for smokers aged 35�64 years and 65 years and older, by using different prevalence rates and RRs for the two age groups.

The RR estimates in Adult SAMMEC have been adjusted to account for the influence of age, but have not been adjusted for other risk factors, such as alcohol consumption. Two studies have been conducted using nationally representative survey data in combination with the CPS-II data to determine the impact of other risk and demographic factors on estimates of the RRs for death from smoking and on SAFs. Thun et al. (2000) and Malarcher et al. (2000) both found that accounting for other risks and demographic characteristics had little impact on these estimates.

Although the CPS-II cohort includes more than 1.2 million men and women, it is not representative of the U.S. population. The CPS-II population contains somewhat more whites and persons in the middle class than the U.S. population. In an examination of the external validity of RR estimates obtained from a non-representative cohort, Szklo (1998) suggested that RR estimates obtained from one population could be applied to other populations if the smoking histories, product usage, underlying susceptibility, and distributions of confounding variables of the populations are similar. In a review of several U.S. and international cohort studies of coronary heart disease, Chambless et al. (1990) found the RR for death from smoking within the various populations were within a narrow range if common definitions of smoking were used.

The external validity of the smoking-attributable death estimates calculated from SAMMEC has been tested twice with data from Oregon. McAnulty et al. (1994) and Thomas et al. (2001) conducted separate analyses of Oregon death certificate data and compared the number of deaths for which smoking was cited as a contributing factor with state estimates of the number of smoking-attributable deaths generated by SAMMEC. McAnulty et al. examined 1989�1990 physician reported deaths and found 10,072 were attributable to smoking according to the death certificates, compared with 10,351 smoking-attributable deaths calculated by using SAMMEC. The difference of 279 deaths was less than 3% of the total. Using 1989-1996 death certificate data, Thomas et al. compared smoking-attributable death estimates from physician reports with estimates generated by SAMMEC and found a difference of only 61 deaths between the 42,839 smoking-attributable deaths reported by physicians and 42,778 deaths estimated using SAMMEC.

The estimates in Adult SAMMEC do not account for deaths from cigar smoking, pipe smoking, and smokeless tobacco use.

The health-related economic costs of smoking estimates in Adult SAMMEC also have some limitations as described below.

The productivity loss estimates were based on lifetime future earnings data that were weighted by sex to remove the effects of gender discrimination. As a result, productivity losses are likely to be understated because men have higher average earnings than do women and are more likely to die from a smoking-attributable disease.

The productivity loss estimates are also understated because they do not include the value of work missed because of smoking-related illness, other smoking-related absenteeism, excess work breaks, or the effects of secondhand smoke.

Additional limitations relate to the calculation of 1998 and 2004 smoking-attributable expenditures. The 1998 and 2004 SAEs were derived by applying 1993 smoking-attributable fractions of expenditures to 1998 and 2004 personal health care expenditure data. Changes in the health care system, economic and demographic characteristics, and risk behaviors between 1993 and 2004 may have influenced the SAFs.

More importantly, the expenditure SAFs were derived from survey data that included limited information on other risk behaviors that could influence the study results, particularly alcohol consumption. To test the potential confounding effects of other risk behaviors, Miller et al. (1999) used the NHIS to estimate the impact of smoking on health care utilization rates while controlling for alcohol consumption and other risks. They found that the inclusion of alcohol did not substantially change the smoking-attributable proportions of health care utilization, but that the SAF for hospital care utilization was twice as high as the hospital expenditure SAF. This relationship was also present in the 1987 NMES results. These results suggested smokers had less intensive hospital stays than did nonsmokers. Warner et al. (1999) noted this anomaly, re-examined the NMES data, and found smokers did not have lower average expenditures than did nonsmokers. Thus the hospital expenditures SAFs in Miller et al. (1999) are likely to be substantially understated.

Finally, the nursing home SAFs were derived from estimates of the probability of admission given a person's smoking history. However, these SAFs do not reflect the effect of smoking on multiple admissions and length-of-stay because of data limitations. The inclusion of these factors may affect the estimates in unknown ways. For example, Nusselder et al. (2000) found smoking to be positively associated with disability duration.

Overall, the proportion of medical expenditures attributable to smoking by Miller et al. is within the 6�14% range of published estimates (Warner et al. 1999; Max, 2001).

In a CDC study comparing the completeness of data collected from six PRAMS states with that of data from birth certificates, the completeness of ascertainment for maternal smoking rates among white women ranged from 71�82% using birth certificate data and 86�90% using data from PRAMS questionnaires (Dietz et al. 1998). When applied to all women, the results from both sources potentially underestimate maternal smoking prevalence. However, a strength of the birth certificate files is that they are the only source of comparable, population-based measures available for all states over multiple time periods.

Smoking-attributable infant deaths are estimated by using the attributable-fraction methodology also employed in the Adult SAMMEC application. Maternal smoking status is provided on birth certificate data for 49 states and the District of Columbia (omits California). However, smoking status is obtained through maternal self reports, and the prevalence of maternal smoking may be substantially understated (USDHHS, 2001). The estimate of smoking for California, derived from the state's Maternal and Infant Health Assessment (MIHA) data, is similar to the estimate derived from PRAMS as it is based on a sample of women giving birth in the state.

Another limitation is that very few smoking-attributable infant deaths occur in small states and the SAM report rounds the number of these deaths to the nearest whole number, whereas the YPLL estimates are based on the SAM calculation, not the rounded estimate. Therefore the health outcome report may indicate zero deaths for a particular infant condition yet show a positive value for YPLL because of that condition.

For the expenditure component of MCH SAMMEC, the empirical models for the relationship of maternal smoking to infant outcomes were extrapolated to the full national birth certificate files based on analysis of 2001/2002 data from the 27 states participating in PRAMS during these years. While the PRAMS sample was chosen to be representative of live births in those states, they are not representative of births in the nation as a whole but do account for approximately 47 percent of all U.S. births in 2002. Moreover, since most of the data in the models are available in each state's birth certificate records, population data are used in the extrapolation process. Since the extrapolation for California uses only a sample of mothers delivering live births in the state, extensive demographic sub-grouping results in sample sizes (i.e., number of births) too small to provide stable expenditure SAF estimates. Users should not analyze subgroups in which fewer than 200 births occurred in any state, and the software warns the end-user when this occurs. The use of weighted survey data in California also results in a slightly lower number of total births than reported in birth certificate files.

The need to use private sector cost estimates to assign dollar values to resource use by infants is another limitation. Given that the underlying estimates of smoking-attributable resource use from the PRAMS data were based on NICU admission and infant length of stay measures, estimates of the costs of these resources were needed. The MarketScan� data is one of the few databases that can be used to estimate costs separately for infants who were admitted to a NICU and for those who were not. These 2004 data are from a large, convenience sample of privately insured women, which does not represent all pregnant women. Further, dollar values are based on private sector payment rates and do not represent the rates paid by public programs, such as Medicaid. However, women covered by Medicaid are likely to have more complicated deliveries than private insured, states significantly increased their Medicaid fees for obstetrical services in the 1990s (Norton and Zuckerman, 2000) and Medicaid pregnancies were found to cost more when �priced� at private rates in three study states (Adams et al, 2001).

Users entering their own expenditure data should use the most current data available, use actual payments rather than charges, include payments for all inpatient services, and use data specific to the payer of interest. (For example, if you are interested in Medicaid costs, obtain data on Medicaid amounts paid for all inpatient services for infants at delivery.)

The estimates of smoking-attributable neonatal expenditures calculated in MCH SAMMEC do not include estimates of costs related to spontaneous abortion, ectopic pregnancy, or other maternal conditions shown to be related to smoking during pregnancy; infant health care costs related to smoke exposure in utero that occur after the delivery hospitalization, such as additional readmissions in the first year of life; or costs related to environmental tobacco smoke exposure of infants and children.

The method used to estimate the SAF when overwriting smoking prevalence as based on the birth certificate data has limits. Currently, the application simply increases the SAF in proportion to the increase in prevalence as entered by the user. Preliminary analyses conducted with more complex methods resulted in SAFs similar to those obtained through proportional adjustment. Be aware that the larger the change in prevalence entered into MCH SAMMEC, the greater its imprecision in measuring the SAF.

Methodology

Estimating the Health Consequences of Smoking

Estimating the Health-Related Economic Consequences of Smoking

Rate Calculations

Five-Year Reports

Smoking-Related Causes of Death in SAMMEC

Other Smoking-Attributable Health Consequences

Limitations