State Trends in Health Risk Factors and Receipt of Clinical Preventive Services Among US Adults During the 1990s FREE

It is well established that premature mortality can be reduced through changes in health risk factors and timely receipt of clinical preventive services. For instance, there is a strong scientific consensus that tobacco use, excessive alcohol use, physical inactivity, obesity, and failure to use safety belts increase mortality risk^1- 6 and that timely receipt of adult immunizations, screening for breast and cervical cancer, and screening for high blood cholesterol levels can reduce the risk of premature death.^7- 11

Population-based monitoring of progress toward health goals is a critical part of the assessment function of public health and a key role for health agencies.¹² At the national level, monitoring of health trends is well established through the Healthy People initiative that began in 1980.^13- 15 Such information is essential for guiding current and future efforts to improve health. State improvements in preventive health depend on many factors, including socioeconomic status and other variables (eg, employment status, income level, and insurance status), state and local programs and policies, clinical standards of care, and the presence of consumers who are informed about prevention and health care services.

Although tracking progress toward national health goals is helpful, the use of national estimates is limited because most preventive health care activities, policies, and other efforts occur at the state or community level. Important state trends may exist that are obscured by national data.^16- 17 State mortality rates vary widely,¹⁸ as do the prevalence of health risk factors and receipt of clinical preventive services¹⁹; state mortality rates are known to correlate strongly with state risk factor estimates for certain measures, such as coronary heart disease²⁰ and lung cancer (Suzanne Proctor, MSPH, written communication, National Center for Health Statistics, June 25, 2001).

A few studies have examined state-specific trends for health risk factors and receipt of clinical preventive services during part of the 1990s,^17,21- 26 but they focused on trends for specific measures (eg, obesity and mammography screening) rather than on a more comprehensive examination of multiple measures. These studies used different methods and years and did not account for the role of changes in demographics, thus limiting their usefulness. The purpose of this study was to provide a comprehensive overview of state trends during the 1990s, using a standard statistical approach, for 5 adult health risk factors and 6 clinical preventive services that are known to have a substantial impact on preventable morbidity and mortality in the United States. The trends we examined include cigarette smoking, binge alcohol use, physical inactivity, obesity, and safety belt use, as well as receipt of mammography, cervical cancer screening, colorectal cancer screening, cholesterol screening, and vaccination for influenza and for pneumococcal disease.

Data used in this study came from the Behavioral Risk Factor Surveillance System (BRFSS) from 1991 through 2000. Details of the BRFSS have been published elsewhere.^27- 29 The BRFSS is a state-based system of health surveys coordinated by the Centers for Disease Control and Prevention. Begun in 15 states in 1984, by 1994 all states were participating in the BRFSS. Data are obtained monthly by state health departments through telephone surveys of randomly selected persons aged 18 years or older.

The system obtains self-reported information primarily on health risk factors related to chronic disease and injury, including health risk behaviors, receipt of clinical preventive services, and health care access. In most states, the BRFSS is the sole source for these data. Results from 30 methodological studies suggest that most measures included in the BRFSS are both reliable and valid.³⁰

For the study period from 1991 through 2000, the total sample size increased from 87 846 to 182 444 and the median state sample size increased from 1790 to 3338. Annual state sample sizes ranged from 1188 to 7543. Median annual response rates, based on persons actually reached by telephone, ranged from 84.1% in 1991 to 59.6% in 2000.

A change in the survey design in 1993 resulted in data collection at different intervals for certain topics. In all years, questions were asked on cigarette smoking, obesity, mammography, and cervical cancer screening (Papanicolaou test); in odd-numbered years for alcohol use, safety belt use, colorectal cancer screening, cholesterol screening, pneumococcal vaccination, and influenza vaccination; and in even-numbered years for physical inactivity (Table 1). Because of variation in the year in which states began to participate in the BRFSS and because of missing data for certain years, trend data are only available for selected measures for Arizona, Arkansas, Kansas, and Nevada and are not available for the District of Columbia and Wyoming.

To provide stable estimates of state trends, analyses were restricted to topics that were included on the survey by 1993 and for which data were collected for 4 or more years. Data on safety belt use were collected only through 1997. The questions used to define receipt of a Papanicolaou test changed in 1992; thus, analyses for these questions were restricted to 1992 through 2000.

Definitions for the 5 health risk factors and the 6 clinical preventive services are listed in Table 1. We used definitions that are consistent with those used to measure national Healthy People 2000 objectives^{14- 15,31- 32} or that are commonly used within the respective subject areas.

Except for cigarette smoking, mammography, and colorectal cancer screening, the wording of survey questions was highly consistent across all years for all measures. In 1996, a question that was used to define current cigarette smoking was changed to make BRFSS questions comparable with other major national surveys (Table 1). Based on data from the National Health Interview Survey, this change increased the estimate of smoking prevalence by about 1 percentage point.³³ Because of this question's change, we adjusted model-based prevalence estimates downward by 1 percentage point for the year 2000.

The introductory sentence to the mammography screening question was altered slightly in 1992 to describe a mammogram as a radiograph that involves pressing the breast between 2 plastic plates. This change resulted in a slight decrease in the prevalence of mammography use,³⁴ but we did not adjust our data because the wording change occurred only in 1 year and the overall effects were minor. Because of changes in screening recommendations, the colorectal cancer screening question referred to proctoscopy in 1993 and 1995, proctoscopy or sigmoidoscopy in 1997, and sigmoidoscopy or colonoscopy in 1999. We were unable to examine trends for fecal occult blood testing, another method used for colorectal cancer screening, because these questions were not included until 1997.

To estimate alcohol use, we examined state trends in binge drinking, which was defined as consumption of 5 or more alcoholic beverages on 1 or more occasions in the past 30 days. Binge drinking is strongly associated with many health risk behaviors and adverse outcomes, including alcohol-impaired driving, unprotected sexual activity, and acute alcohol poisoning.^2,35

There is no consensus on the most appropriate survey questions for assessing moderate or vigorous physical activity.³⁶ Because of this lack of consensus, we chose to examine trends in physical inactivity, which we defined as participating in no leisure-time physical activity in the past 30 days. Using guidelines from the World Health Organization,³⁷ we defined obesity as having a body mass index of at least 30 kg/m². Safety belt use was defined as always using a safety belt because this definition produces estimates similar to those obtained from observational surveys.^38- 39

Recommendations differ on the optimal age for beginning routine breast cancer screening.⁴⁰ In this study, we examined trends in receipt of mammography in the past 2 years for women aged 40 years or older. For receipt of cervical cancer screening within the past 3 years, analyses were restricted to women aged 18 years or older with an intact uterus. Cholesterol screening was defined as receipt of a blood cholesterol test within the past 5 years among persons aged 18 years or older.³² For receipt of influenza vaccination in the past year and ever receiving a pneumococcal vaccination, analyses were restricted to persons aged 65 years or older.^31- 32

Because all 11 of the outcomes we modeled were prevalences, we used logit models to evaluate state trends. Using logit models allowed us to evaluate trends while controlling for the effects age, sex, race/ethnicity, and education level have on the various outcomes. Thus, the unit of analysis in all models was the individual survey respondent. Respondents' answers were used to define dichotomous outcome variables. The independent variables in the models were year, age, sex, race/ethnicity (white non-Hispanic, black non-Hispanic, Hispanic, or other), and education level (less than high school graduate, high school graduate, and some college/college graduate). SUDAAN was used to take into account the complex survey design in the modeling.⁴¹

To determine the appropriateness of a linear trend, we added year² (a quadratic term) to the models and examined its P value, setting the significance level at P<.01 because of the large sample sizes and the number of states. If the P value for the quadratic term was not less than .01, we assumed the trend was linear. Of the 525 total state trends examined, a total of 458 (87%) met our assumption of linearity. For these models, we removed the quadratic term, refit the models, and used the estimated β coefficients for year to calculate odds ratios (ORs). These ORs reflect the average annual change per year (or per 2 years) in overall state odds estimates for each outcome, controlling for the effects of demographic changes on each outcome. Odds ratios with 99% confidence intervals (CIs) that excluded the null value of 1 were considered statistically significant. β Coefficients with P<.01 were considered statistically significant.

For the 67 trends (13%) that violated the linearity assumption, we treated year as a categorical variable and used the contrast between the estimate available from the last year of the decade with the estimate from the first year of the decade. The β value for the contrast was used to calculate an OR that compared the overall odds estimate from the last year with the overall odds estimate from the first year of the decade. These ORs represent the average change across the entire decade using the first and last data points available, controlling for the effects of demographic changes on each outcome. β Coefficients with P<.01 were considered statistically significant.

Finally, we used our models to determine the predicted prevalence at the start and end of the decade for each measure in each state. Estimates were indirectly standardized to the age, sex, race/ethnicity, and education level of a typical person based on the pooled annual BRFSS sample population from all states.^41- 42 Although these model-based values are not the actual prevalence estimates, they allow for appropriate comparisons across states over time and are similar to published state estimates.³⁰

State trends during the 1990s for the 5 health risk factor measures are summarized in Table 2. Median state estimates showed a substantial increase in both obesity and safety belt use and a slight increase in binge alcohol use, and were essentially unchanged for physical inactivity and cigarette smoking.

Smoking prevalence decreased significantly only in Minnesota; in contrast, it was unchanged in 32 states and increased in 14 states primarily located in the Midwest (7 states) and South (4 states). Binge alcohol use declined in Arizona, Minnesota, and Pennsylvania, remained unchanged in 25 states, and increased in 19 states. Increases were concentrated primarily in the South (8 states) and Midwest (6 states), although baseline prevalence estimates were generally low in southern states. The prevalence of physical inactivity decreased in 11 states, increased in 3 states (Arizona, Minnesota, and Montana), and was unchanged in 34 states; about half the states registering declines were in the South. The prevalence of obesity increased in all states, and safety belt use increased in 39 of 47 states.

The pattern for receipt of clinical preventive services was different from that for health risk factors (Table 3). Median state estimates increased substantially for mammography, influenza immunization, and pneumococcal immunization, as most states experienced increases for these measures. In contrast, median estimates increased only slightly for colorectal cancer, cervical cancer, and cholesterol screening. Receipt of mammography for women aged 40 years or older in the prior 2 years increased significantly in 43 of 47 states (all except Alaska, Colorado, Minnesota, and Washington). Receipt of cervical cancer screening increased in 8 of 48 states, all of which were in the Northeast or South (Connecticut, Delaware, Kentucky, Maryland, Massachusetts, New York, Pennsylvania, and Rhode Island).

The use of colorectal cancer screening increased in 13 of 49 states, with 11 of these states in the Northeast or South. Trends in the receipt of cholesterol screening demonstrated a mixed picture, as increases occurred in 13 of 47 states (11 of which were in the South), but declines in screening occurred in Iowa, Minnesota, North Dakota, South Dakota, and Washington state. The trend in the increased use of vaccinations among persons aged 65 years or older was almost nationwide. Increases in the receipt of influenza vaccination in the past year occurred in 44 of 49 states, and the percentage of the older population who had ever received pneumococcal vaccination increased in 48 of 49 states.

With the exception of Minnesota, state trends in cigarette smoking during the 1990s were discouraging. After adjusting for changes in demographics, nearly 30% of states experienced an increase in smoking during this period, with half of these states located in the Midwest. Only Utah achieved the Healthy People 2000 goal for adult smoking prevalence of 15%,¹⁴ but this state had already attained this goal before the start of the decade. Reasons for the decline in smoking in Minnesota are unclear from these data, and there were few state or local tobacco control policies implemented in this state during the 1990s.^43- 44 The lack of a reduction in smoking in nearly all states indicates the continued need for major tobacco prevention and control efforts, such as increasing excise taxes and smoking cessation activities,⁴⁵ because cigarette smoking remains the leading cause of preventable death in the United States.⁶

State trends in binge alcohol use during the 1990s were disturbing. Although per capita consumption of alcohol-containing beverages decreased from 1991 through 1998 in most states,⁴⁶ national alcohol survey data suggest that binge alcohol use leveled off between 1990 and 1995 after decreasing in earlier years.⁴⁷ Reasons for the increase in binge drinking in 19 states, and for the decline of such drinking in Arizona and Pennsylvania,⁴⁸ are unknown. Minnesota's decrease in binge drinking during the 1990s may have resulted from that state's extensive educational and policy efforts directed toward reducing drinking and driving, fetal alcohol exposure, and adolescent drinking.^49- 51

Physical inactivity decreased in 11 states. The reasons for this improvement are unclear but may have resulted from state and local health promotion programs designed to improve cardiovascular health. In contrast, physical inactivity prevalence increased or was unchanged in 37 states. For obesity, all states experienced a significant increase from 1991 through 2000, confirming findings from previous studies on the growing nationwide epidemic of obesity.^24,52- 55 Multiple approaches for increasing physical activity and reducing obesity are needed, including educational programs, policies, and environmental interventions to increase physical activity and improve dietary patterns.^3,24

Although data were available only through 1997, adult safety belt use increased in most states. Even among some states that showed no improvement (ie, Hawaii, Maryland, New Jersey, and Virginia), safety belt use was already high in 1991 and remained so during the study period. By the end of 1997, California, Hawaii, North Carolina, New Mexico, and Oregon had achieved the Healthy People 2000 goal of 85% for safety belt use.¹⁴ The extensive enactment of safety belt laws during the 1980s and 1990s and changing societal norms about safety belt use have probably been the major contributors to the increase in safety belt use.⁵⁶ Despite the increases, enhanced efforts to improve safety belt use are still warranted in most states.

State trends of increasing mammography screening among women aged 40 years or older are very encouraging and confirm findings from earlier studies.^22,57- 59 Educational campaigns directed toward health care practitioners and the general public, state mandates for insurance coverage of mammograms, and programs for providing mammography services to low-income women have all played a role in increasing breast cancer screening in nearly all states.²² Efforts to increase mammography are needed in Alaska, Colorado, Minnesota, and Washington.

State trends in cervical cancer screening are difficult to interpret. Median estimates were high in 1991 and increased only slightly by 2000; however, most states reached the Healthy People 2000 goal of 85% for receipt of a Papanicolaou test in the prior 3 years.¹⁴ The situation is further complicated because during the 1990s, various organizations issued differing recommendations on frequency and age groups for cervical cancer screening.^22,60- 61 More research is needed to delineate state-specific trends among subpopulations of women to ascertain whether changes in screening are occurring among women who are at different risks for developing cervical cancer.

Colorectal cancer screening, defined as receipt of proctoscopy, sigmoidoscopy, or colonoscopy among persons aged 50 years or older, increased in 13 states during the 1990s. Despite these improvements, fewer than half of persons in this age group in all states received this type of screening in 1999. Unfortunately, trend data on receipt of fecal occult blood testing, another means of screening for colorectal cancer, are not available, as these questions were not included on the BRFSS until 1997. Data from 1997 and 1999 indicate that about 20% of adults aged 50 years or older reported undergoing fecal occult blood testing in the past year.^26,62 Because colorectal cancer is the second leading cause of cancer-related deaths in the United States,⁶³ increased efforts directed toward the public and health care practitioners are needed to increase the use of colorectal cancer screening.

Cholesterol screening trends among states were variable. Although receipt of screening increased in 13 states, the nationwide median for all states increased only modestly. Decreases in the receipt of screening in Iowa, Minnesota, North Dakota, South Dakota, and Washington are of special concern. Because heart disease is the leading cause of death in the United States and because high blood cholesterol levels increase the risk for this disease, major efforts are needed to improve cholesterol screening and lower cholesterol levels.²³

State trends in adult vaccination were highly encouraging. Forty-eight of 49 states met the Healthy People 2000 goal of 60% for influenza vaccination, although only 8 of 49 states reached the goal of 60% for pneumococcal vaccination.¹⁴ (Data are not available from the BRFSS to ascertain influenza or pneumococcal disease vaccination among persons with selected chronic conditions that warrant receipt of these vaccines.) Reasons for these increases probably include a greater acceptance of using preventive medical services by practitioners and consumers, increased delivery and administration of vaccines from multiple sources, and Medicare reimbursement for vaccination.²⁵

While examining state trends for specific health risk factors and clinical preventive service measures is important, it is also useful to consider state trends across the entire spectrum of measures. New York showed the greatest positive change during the 1990s, with improvements in 8 measures. Delaware, Kentucky, and Maryland improved in 7 measures, but these increases were tempered somewhat by increases in binge alcohol use in Delaware and Maryland and in smoking in Kentucky. In contrast, Alaska had statistically significant improvements for no measures, North and South Dakota had 4 measures that improved and 4 that worsened, and Iowa had 5 measures that improved and 4 that worsened.

Reasons for the state-to-state differences in trends across all 11 measures cannot be ascertained from these data. Contributing factors may include changes and differences in states' economies, health or other policies, migration patterns, socioeconomic status, educational or mass media efforts, social norms, and health-related activities by voluntary, professional, and private organizations.

This study has several limitations. First, we relied on self-reports. Depending on the measure, self-reports can result in overestimates or underestimates compared with other data sources, such as health care records, physiological measures, or biochemical verification. Social desirability⁶⁴ can lead to reporting of behavior in a more favorable light, probably resulting in underestimates of obesity, binge drinking, and smoking. Social desirability,⁶⁴ unfamiliarity with medical terms, and telescoping⁶⁵ (recalling that events occurred more recently than they actually occurred) can affect the validity of self-reported data on clinical preventive services and probably result in overestimates.^66- 67 For example, self-reports overestimate prevalence of receipt of mammography⁶⁷ and underestimate prevalence of obesity.⁶⁸ The effects of social desirability, unfamiliarity with medical terms, and telescoping on state trends during the 1990s is unknown.

As mentioned previously, there were wording changes to questions on smoking, mammography, and colorectal cancer screening; adjustments were made only for smoking. The impact of these wording changes on trends is unclear, although any effect would be consistent across states. The effect of the 1992 wording change for mammography occurred only for 1 year and probably had little effect.

The most substantial changes in question wording occurred for colorectal cancer screening, and this question applied to different examinations across the study period. How well the average adult can distinguish between terms used to describe different medical instruments for examining the colon is unknown. Thus, trends in receipt of colorectal cancer screening should be interpreted cautiously, although as with other measures, the effects of wording changes on trends should be consistent across states.

Households without telephones were excluded, which probably resulted in slight underestimates of risk factors and overestimates of receipt of clinical preventive services for most measures.⁶⁹ Typical of other telephone surveys conducted during the 1990s,⁷⁰ response rates declined for nearly all states. Weighting procedures used in the BRFSS partially adjust for nonresponse; however, the effect of declines in response rates on trends is not known.

Because of different baseline values, states with high estimates early in the decade were unlikely to experience increases to the same extent as states with lower estimates (similarly, states with low estimates were unlikely to experience declines to the same extent as those with higher initial estimates). This finding was evident in states with high initial values for safety belt use and receipt of cervical cancer screening. States had different sample sizes, and some measures were applicable only to subpopulations with smaller numbers of respondents (eg, vaccination among persons aged ≥65 years and mammography among women aged ≥40 years). States with consistently larger sample sizes in general were probably more likely to have statistically significant differences compared with states with smaller sample sizes. This may, for example, partially explain the lack of significant changes found for most measures in Alaska, which was a state with smaller sample sizes in many years.

The numbers of years with data and, consequently, the number of data points per measure differed (eg, 10 data points for mammography and 4 for colorectal cancer screening). The effect of differing numbers of data points on our results is not clear; however, for some measures, the assumption of linearity was based on a limited number of data points. Finally, because of the lack of linearity of data for 67 models, we used a different approach for our regression analyses that relied only on data from the first and last years. Thus, in these analyses, we were only able to compare estimates from the beginning and end of the decade but could not examine trends over the entire decade.

State monitoring of preventive health measures is an invaluable role for public health,¹² especially given the heterogeneity of state trends. One of the most important purposes of the Healthy People 2000 and 2010 initiatives was to create health goals to be used throughout the nation^14- 15 and not merely as part of an exercise in federal data collection and reporting. The BRFSS is, in turn, an invaluable data source for monitoring states' progress towards reaching national objectives related to health risk factors and the receipt of clinical preventive services.

This study demonstrates a mixed picture of state progress in adult preventive health during the 1990s. With the exception of safety belt use, there was limited progress in improving health risk factors. The trends in smoking, obesity, and binge alcohol use are especially worrisome and will require strong efforts to reduce the extent of these problems. Progress in increasing the use of clinical preventive services has been substantial, especially for the receipt of mammography and for influenza and pneumococcal vaccination. Our findings suggest that increased efforts are especially needed to increase the use of colorectal cancer and cholesterol screening.